Re: [libvirt] [Qemu-ppc] [RFC PATCH qemu] spapr_pci: Create PCI-express root bus by default
On Tue, 2016-11-01 at 13:46 +1100, David Gibson wrote:
On Mon, Oct 31, 2016 at 03:10:23PM +1100, Alexey Kardashevskiy wrote:
On Fri, 28 Oct 2016 18:56:40 +1100 Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
At the moment sPAPR PHB creates a root buf of TYPE_PCI_BUS type. This means that vfio-pci devices attached to it (and this is a default behaviour) hide PCIe extended capabilities as the bus does not pass a pci_bus_is_express(pdev->bus) check. This changes adds a default PCI bus type property to sPAPR PHB and uses TYPE_PCIE_BUS if none passed; older machines get TYPE_PCI_BUS for backward compatibility as a bus type is used in the bus name so the root bus name becomes "pcie.0" instead of "pci.0". Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> --- What can possibly go wrong with such change of a name? From devices prospective, I cannot see any. libvirt might get upset as "pci.0" will not be available, will it make sense to create pcie.0 as a root bus and always add a PCIe->PCI bridge and name its bus "pci.0"? Or create root bus from TYPE_PCIE_BUS and force name to "pci.0"? pci_register_bus() can do this. --- hw/ppc/spapr.c | 5 +++++ hw/ppc/spapr_pci.c | 5 ++++- include/hw/pci-host/spapr.h | 1 + 3 files changed, 10 insertions(+), 1 deletion(-) diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c index 0b3820b..a268511 100644 --- a/hw/ppc/spapr.c +++ b/hw/ppc/spapr.c @@ -2541,6 +2541,11 @@ DEFINE_SPAPR_MACHINE(2_8, "2.8", true); .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ .property = "mem64_win_size", \ .value = "0", \ + }, \ + { \ + .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ + .property = "root_bus_type", \ + .value = TYPE_PCI_BUS, \ }, static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index, diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c index 7cde30e..2fa1f22 100644 --- a/hw/ppc/spapr_pci.c +++ b/hw/ppc/spapr_pci.c @@ -1434,7 +1434,9 @@ static void spapr_phb_realize(DeviceState *dev, Error **errp) bus = pci_register_bus(dev, NULL, pci_spapr_set_irq, pci_spapr_map_irq, sphb, &sphb->memspace, &sphb->iospace, - PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS); + PCI_DEVFN(0, 0), PCI_NUM_PINS, + sphb->root_bus_type ? sphb->root_bus_type : + TYPE_PCIE_BUS); Shouldn't we ensure that sphb->root_bus_type is either TYPE_PCIE_BUS or TYPE_PCI_BUS ? Yes, I think so. In fact, I think it would be better to make the
On Fri, Oct 28, 2016 at 12:07:12PM +0200, Greg Kurz wrote: property a boolean that just selects PCI-E, rather than this which exposes qemu (semi-)internal type names on the comamnd line. Sure, a "pcie-root" boolean property should do. However this is not my main concern, I rather wonder if we have to have
On 31/10/16 13:53, David Gibson wrote: pci.0 when we pick PCIe for the root. Right. I've added Andrea Bologna to the CC list to get a libvirt perspective.
Thanks for doing so: changes such as this one can have quite an impact on the upper layers of the stack, so the earliest libvirt is involved in the discussion the better. I'm going to go a step further and cross-post to libvir-list in order to give other libvirt contributors a chance to chime in too.
Andrea, To summarise the issue here: * As I've said before the PAPR spec kinda-sorta abstracts the difference between vanilla PCI and PCI-E * However, because within qemu we're declaring the bus as PCI that means some PCI-E devices aren't working right * In particular it means that PCI-E extended config space isn't available The proposal is to change (on newer machine types) the spapr PHB code to declare a PCI-E bus instead. AIUI this still won't make the root complex guest visible (which it's not supposed to be under PAPR), and the guest shouldn't see a difference in most cases - it will still see the PAPR abstracted PCIish bus, but will now be able to get extended config space. The possible problem from a libvirt perspective is that doing this in the simplest way in qemu would change the name of the default bus from pci.0 to pcie.0. We have two suggested ways to mitigate this: 1) Automatically create a PCI-E to PCI bridge, so that new machine types will have both a pcie.0 and pci.0 bus 2) Force the name of the bus to be pci.0, even though it's treated as PCI-E in other ways. We're trying to work out exactly what will and won't cause trouble for libvirt.
Option 2) is definitely a no-no, as we don't want to be piling up even more hacks and architecture-specific code: the PCI Express Root Bus should be called pcie.0, just as it is on q35 and mach-virt machine types. Option 1) doesn't look too bad, but devices that are added automatically by QEMU are an issue since we need to hardcode knowledge of them into libvirt if we want the rest of the PCI address allocation logic to handle them correctly. Moreover libvirt now has the ability of building a legacy PCI topology without user intervention, if needed to plug in legacy devices, on machines that have a PCI Express Root Bus, which makes the additional bridge fully redundant... ... or at least it would, if we actually had a proper PCIe-to-PCI bridge; AFAIK, though, the closest we have is the i82801b11-bridge that is Intel-specific despite having so far been abused as a generic PCIe-to-PCI bridge. I'm not even sure whether it would work at all on ppc64. Moving from legacy PCI to PCI Express would definitely be an improvement, in my opinion. As mentioned, that's already the case for at least two other architectures, so the more we can standardize on that, the better. That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which. Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation: 1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch 2) switch between legacy PCI and PCI Express based on a machine type option. libvirt would be able to find out whether the option is available or not, and default to either <controller type='pci' model='pci-root'/> or <controller type='pci' model='pcie-root'/> based on that. In order to support multiple PHBs properly, those would have to be switchable with an option as well 3) create an entirely new machine type, eg. pseries-pcie or whatever someone with the ability to come up with decent names can suggest :) That would make ppc64 similar to x86, where i440fx and q35 have different root buses. libvirt would learn about the new machine type, know that it has a PCI Express Root Bus, and behave accordingly Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using PCI Express. Option 2) with default to legacy PCI and option 3) would work just fine with existing libvirt versions AFAICT, but wouldn't of course expose the new capabilities. Option 3) is probably the one that will be less confusing to users; we might even decide to take the chance and fix other small annoyances with the current pseries machine type, if there's any. On the other hand, it might very well be considered to be too big a hammer for such a small nail. -- Andrea Bolognani / Red Hat / Virtualization
On 16/11/16 01:02, Andrea Bolognani wrote:
On Tue, 2016-11-01 at 13:46 +1100, David Gibson wrote:
On Mon, Oct 31, 2016 at 03:10:23PM +1100, Alexey Kardashevskiy wrote:
On 31/10/16 13:53, David Gibson wrote:
On Fri, Oct 28, 2016 at 12:07:12PM +0200, Greg Kurz wrote:
On Fri, 28 Oct 2016 18:56:40 +1100 Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
At the moment sPAPR PHB creates a root buf of TYPE_PCI_BUS type. This means that vfio-pci devices attached to it (and this is a default behaviour) hide PCIe extended capabilities as the bus does not pass a pci_bus_is_express(pdev->bus) check.
This changes adds a default PCI bus type property to sPAPR PHB and uses TYPE_PCIE_BUS if none passed; older machines get TYPE_PCI_BUS for backward compatibility as a bus type is used in the bus name so the root bus name becomes "pcie.0" instead of "pci.0".
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> ---
What can possibly go wrong with such change of a name? From devices prospective, I cannot see any.
libvirt might get upset as "pci.0" will not be available, will it make sense to create pcie.0 as a root bus and always add a PCIe->PCI bridge and name its bus "pci.0"?
Or create root bus from TYPE_PCIE_BUS and force name to "pci.0"? pci_register_bus() can do this.
--- hw/ppc/spapr.c | 5 +++++ hw/ppc/spapr_pci.c | 5 ++++- include/hw/pci-host/spapr.h | 1 + 3 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c index 0b3820b..a268511 100644 --- a/hw/ppc/spapr.c +++ b/hw/ppc/spapr.c @@ -2541,6 +2541,11 @@ DEFINE_SPAPR_MACHINE(2_8, "2.8", true); .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ .property = "mem64_win_size", \ .value = "0", \ + }, \ + { \ + .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ + .property = "root_bus_type", \ + .value = TYPE_PCI_BUS, \ },
static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index, diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c index 7cde30e..2fa1f22 100644 --- a/hw/ppc/spapr_pci.c +++ b/hw/ppc/spapr_pci.c @@ -1434,7 +1434,9 @@ static void spapr_phb_realize(DeviceState *dev, Error **errp) bus = pci_register_bus(dev, NULL, pci_spapr_set_irq, pci_spapr_map_irq, sphb, &sphb->memspace, &sphb->iospace, - PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS); + PCI_DEVFN(0, 0), PCI_NUM_PINS, + sphb->root_bus_type ? sphb->root_bus_type : + TYPE_PCIE_BUS);
Shouldn't we ensure that sphb->root_bus_type is either TYPE_PCIE_BUS or TYPE_PCI_BUS ?
Yes, I think so. In fact, I think it would be better to make the property a boolean that just selects PCI-E, rather than this which exposes qemu (semi-)internal type names on the comamnd line.
Sure, a "pcie-root" boolean property should do.
However this is not my main concern, I rather wonder if we have to have pci.0 when we pick PCIe for the root.
Right.
I've added Andrea Bologna to the CC list to get a libvirt perspective.
Thanks for doing so: changes such as this one can have quite an impact on the upper layers of the stack, so the earliest libvirt is involved in the discussion the better.
I'm going to go a step further and cross-post to libvir-list in order to give other libvirt contributors a chance to chime in too.
Andrea,
To summarise the issue here: * As I've said before the PAPR spec kinda-sorta abstracts the difference between vanilla PCI and PCI-E * However, because within qemu we're declaring the bus as PCI that means some PCI-E devices aren't working right * In particular it means that PCI-E extended config space isn't available
The proposal is to change (on newer machine types) the spapr PHB code to declare a PCI-E bus instead. AIUI this still won't make the root complex guest visible (which it's not supposed to be under PAPR), and the guest shouldn't see a difference in most cases - it will still see the PAPR abstracted PCIish bus, but will now be able to get extended config space.
The possible problem from a libvirt perspective is that doing this in the simplest way in qemu would change the name of the default bus from pci.0 to pcie.0. We have two suggested ways to mitigate this: 1) Automatically create a PCI-E to PCI bridge, so that new machine types will have both a pcie.0 and pci.0 bus 2) Force the name of the bus to be pci.0, even though it's treated as PCI-E in other ways.
We're trying to work out exactly what will and won't cause trouble for libvirt.
Option 2) is definitely a no-no, as we don't want to be piling up even more hacks and architecture-specific code: the PCI Express Root Bus should be called pcie.0, just as it is on q35 and mach-virt machine types.
Option 1) doesn't look too bad, but devices that are added automatically by QEMU are an issue since we need to hardcode knowledge of them into libvirt if we want the rest of the PCI address allocation logic to handle them correctly.
Moreover libvirt now has the ability of building a legacy PCI topology without user intervention, if needed to plug in legacy devices, on machines that have a PCI Express Root Bus, which makes the additional bridge fully redundant...
... or at least it would, if we actually had a proper PCIe-to-PCI bridge; AFAIK, though, the closest we have is the i82801b11-bridge that is Intel-specific despite having so far been abused as a generic PCIe-to-PCI bridge. I'm not even sure whether it would work at all on ppc64.
Moving from legacy PCI to PCI Express would definitely be an improvement, in my opinion. As mentioned, that's already the case for at least two other architectures, so the more we can standardize on that, the better.
That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which.
Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
2) switch between legacy PCI and PCI Express based on a machine type option. libvirt would be able to find out whether the option is available or not, and default to either
<controller type='pci' model='pci-root'/>
or
<controller type='pci' model='pcie-root'/>
based on that. In order to support multiple PHBs properly, those would have to be switchable with an option as well
3) create an entirely new machine type, eg. pseries-pcie or whatever someone with the ability to come up with decent names can suggest :) That would make ppc64 similar to x86, where i440fx and q35 have different root buses. libvirt would learn about the new machine type, know that it has a PCI Express Root Bus, and behave accordingly
Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using
How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make? I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
PCI Express. Option 2) with default to legacy PCI and option 3) would work just fine with existing libvirt versions AFAICT, but wouldn't of course expose the new capabilities.
Option 3) is probably the one that will be less confusing to users; we might even decide to take the chance and fix other small annoyances with the current pseries machine type, if there's any. On the other hand, it might very well be considered to be too big a hammer for such a small nail.
-- Alexey
On Thu, Nov 17, 2016 at 01:02:57PM +1100, Alexey Kardashevskiy wrote:
On 16/11/16 01:02, Andrea Bolognani wrote:
On Tue, 2016-11-01 at 13:46 +1100, David Gibson wrote:
On Mon, Oct 31, 2016 at 03:10:23PM +1100, Alexey Kardashevskiy wrote:
On 31/10/16 13:53, David Gibson wrote:
On Fri, Oct 28, 2016 at 12:07:12PM +0200, Greg Kurz wrote:
On Fri, 28 Oct 2016 18:56:40 +1100 Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> > At the moment sPAPR PHB creates a root buf of TYPE_PCI_BUS type. > This means that vfio-pci devices attached to it (and this is > a default behaviour) hide PCIe extended capabilities as > the bus does not pass a pci_bus_is_express(pdev->bus) check. > > This changes adds a default PCI bus type property to sPAPR PHB > and uses TYPE_PCIE_BUS if none passed; older machines get TYPE_PCI_BUS > for backward compatibility as a bus type is used in the bus name > so the root bus name becomes "pcie.0" instead of "pci.0". > > Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> > --- > > What can possibly go wrong with such change of a name? > From devices prospective, I cannot see any. > > libvirt might get upset as "pci.0" will not be available, > will it make sense to create pcie.0 as a root bus and always > add a PCIe->PCI bridge and name its bus "pci.0"? > > Or create root bus from TYPE_PCIE_BUS and force name to "pci.0"? > pci_register_bus() can do this. > > > --- > hw/ppc/spapr.c | 5 +++++ > hw/ppc/spapr_pci.c | 5 ++++- > include/hw/pci-host/spapr.h | 1 + > 3 files changed, 10 insertions(+), 1 deletion(-) > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c > index 0b3820b..a268511 100644 > --- a/hw/ppc/spapr.c > +++ b/hw/ppc/spapr.c > @@ -2541,6 +2541,11 @@ DEFINE_SPAPR_MACHINE(2_8, "2.8", true); > .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ > .property = "mem64_win_size", \ > .value = "0", \ > + }, \ > + { \ > + .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \ > + .property = "root_bus_type", \ > + .value = TYPE_PCI_BUS, \ > }, > > static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index, > diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c > index 7cde30e..2fa1f22 100644 > --- a/hw/ppc/spapr_pci.c > +++ b/hw/ppc/spapr_pci.c > @@ -1434,7 +1434,9 @@ static void spapr_phb_realize(DeviceState *dev, Error **errp) > bus = pci_register_bus(dev, NULL, > pci_spapr_set_irq, pci_spapr_map_irq, sphb, > &sphb->memspace, &sphb->iospace, > - PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS); > + PCI_DEVFN(0, 0), PCI_NUM_PINS, > + sphb->root_bus_type ? sphb->root_bus_type : > + TYPE_PCIE_BUS);
Shouldn't we ensure that sphb->root_bus_type is either TYPE_PCIE_BUS or TYPE_PCI_BUS ?
Yes, I think so. In fact, I think it would be better to make the property a boolean that just selects PCI-E, rather than this which exposes qemu (semi-)internal type names on the comamnd line.
Sure, a "pcie-root" boolean property should do.
However this is not my main concern, I rather wonder if we have to have pci.0 when we pick PCIe for the root.
Right.
I've added Andrea Bologna to the CC list to get a libvirt perspective.
Thanks for doing so: changes such as this one can have quite an impact on the upper layers of the stack, so the earliest libvirt is involved in the discussion the better.
I'm going to go a step further and cross-post to libvir-list in order to give other libvirt contributors a chance to chime in too.
Andrea,
To summarise the issue here: * As I've said before the PAPR spec kinda-sorta abstracts the difference between vanilla PCI and PCI-E * However, because within qemu we're declaring the bus as PCI that means some PCI-E devices aren't working right * In particular it means that PCI-E extended config space isn't available
The proposal is to change (on newer machine types) the spapr PHB code to declare a PCI-E bus instead. AIUI this still won't make the root complex guest visible (which it's not supposed to be under PAPR), and the guest shouldn't see a difference in most cases - it will still see the PAPR abstracted PCIish bus, but will now be able to get extended config space.
The possible problem from a libvirt perspective is that doing this in the simplest way in qemu would change the name of the default bus from pci.0 to pcie.0. We have two suggested ways to mitigate this: 1) Automatically create a PCI-E to PCI bridge, so that new machine types will have both a pcie.0 and pci.0 bus 2) Force the name of the bus to be pci.0, even though it's treated as PCI-E in other ways.
We're trying to work out exactly what will and won't cause trouble for libvirt.
Option 2) is definitely a no-no, as we don't want to be piling up even more hacks and architecture-specific code: the PCI Express Root Bus should be called pcie.0, just as it is on q35 and mach-virt machine types.
Option 1) doesn't look too bad, but devices that are added automatically by QEMU are an issue since we need to hardcode knowledge of them into libvirt if we want the rest of the PCI address allocation logic to handle them correctly.
Moreover libvirt now has the ability of building a legacy PCI topology without user intervention, if needed to plug in legacy devices, on machines that have a PCI Express Root Bus, which makes the additional bridge fully redundant...
... or at least it would, if we actually had a proper PCIe-to-PCI bridge; AFAIK, though, the closest we have is the i82801b11-bridge that is Intel-specific despite having so far been abused as a generic PCIe-to-PCI bridge. I'm not even sure whether it would work at all on ppc64.
Moving from legacy PCI to PCI Express would definitely be an improvement, in my opinion. As mentioned, that's already the case for at least two other architectures, so the more we can standardize on that, the better.
That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which.
Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
2) switch between legacy PCI and PCI Express based on a machine type option. libvirt would be able to find out whether the option is available or not, and default to either
<controller type='pci' model='pci-root'/>
or
<controller type='pci' model='pcie-root'/>
based on that. In order to support multiple PHBs properly, those would have to be switchable with an option as well
3) create an entirely new machine type, eg. pseries-pcie or whatever someone with the ability to come up with decent names can suggest :) That would make ppc64 similar to x86, where i440fx and q35 have different root buses. libvirt would learn about the new machine type, know that it has a PCI Express Root Bus, and behave accordingly
Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using
How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make? I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
IIUC, it's because when libvirt wants to attach a PCI device, it will just try to attach it to bus pci.0, which will no longer exist.
PCI Express. Option 2) with default to legacy PCI and option 3) would work just fine with existing libvirt versions AFAICT, but wouldn't of course expose the new capabilities.
Option 3) is probably the one that will be less confusing to users; we might even decide to take the chance and fix other small annoyances with the current pseries machine type, if there's any. On the other hand, it might very well be considered to be too big a hammer for such a small nail.
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Thu, 2016-11-17 at 13:02 +1100, Alexey Kardashevskiy wrote:
That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which. Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation: 1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
[...]
Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make?
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type. Even if no newer pseries-x.y were to be added after introducing pseries-pcie, you could still easily create guests that use either root bus, so no loss in functionality. -- Andrea Bolognani / Red Hat / Virtualization
On 18/11/16 19:17, Andrea Bolognani wrote:
On Thu, 2016-11-17 at 13:02 +1100, Alexey Kardashevskiy wrote:
That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which.
Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
[...]
Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using
How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make?
Existing libvirt versions assume that pseries guests have
If libvirt is using just "pseries" (without a version), then having a virtual PCIe-PCI bridge (and "pci.0" always available by default) will do it. If libvirt is using a specific version of pseries, then it already knows that <=2.7 has pci.0 as a root, pcie.0 otherwise. libvirt has a knowledge what QEMU version has what, right? In what scenario will an additional machine type help?
a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges.
So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type.
I do not know about i440<->q35 but in pseries the difference is going to be very simple. For example, we did not change the machine type when we switched from default OHCI to XHCI, switching from PCI to PCIe does not sound like we need a whole new machine type for this either.
Even if no newer pseries-x.y were to be added after introducing pseries-pcie, you could still easily create guests that use either root bus, so no loss in functionality.
I could do this with the existing pseries if the machine had a "root but type" property. -- Alexey
On Mon, 2016-11-21 at 13:12 +1100, Alexey Kardashevskiy wrote:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch [...] Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make? Existing libvirt versions assume that pseries guests have If libvirt is using just "pseries" (without a version), then having a virtual PCIe-PCI bridge (and "pci.0" always available by default) will do it.
Please don't. Any device that is included in the guest by default and can't be turned off makes libvirt's life significantly more difficult (see below).
If libvirt is using a specific version of pseries, then it already knows that <=2.7 has pci.0 as a root, pcie.0 otherwise. libvirt has a knowledge what QEMU version has what, right?
It doesn't yet, that's the point :) We *could* add such knowledge to libvirt[1], but *existing* libvirt versions would still not know about it, which means that upgrading QEMU withough upgrading libvirt will result in failure to create new guests.
In what scenario will an additional machine type help?
Because then libvirt could learn that pseries-x.y <-> pci.0 pseries-pcie-x.y <-> pcie.0 the same way it already knows that pc-i440fx-x.y <-> pci.0 pc-q35-x.y <-> pcie.0 and choosing between one or the other would be, I think, much easier for the user as well.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y. Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between
a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course. those two machine types than just the root bus type. I do not know about i440<->q35 but in pseries the difference is going to be very simple. For example, we did not change the machine type when we switched from default OHCI to XHCI, switching from PCI to PCIe does not sound like we need a whole new machine type for this either.
The change from OHCI to XHCI only affected the *default* USB controller, which libvirt tries its best not to use anyway: instead, it will prefer to use '-M ...,usb=off' along with '-device ...' and set both the controller model and its PCI address explicitly, partially to shield its users from such changes in QEMU. Let's not forget that libvirt is a management layer, and as such it needs to have as much control as possible over the virtual hardware presented to the guest, mostly for guest ABI compatibility purposes. Defaulting to XHCI instead of OHCI, same as pretty much all other defaults, is good for people who run QEMU directly, but libvirt needs to be able to control such knobs itself in order to effectively perform the task it was designed for. Moreover, we're talking about a more fundamental change here: the PCI Root Bus is not just any random device, it's the one fundation upon which the entire PCI hierarchy is built. Based on whether the machine exposes a PCI Express Root Bus or a legacy PCI Express Root Bus, libvirt will create very different PCI hierarchies, eg. using several ioh3420 devices instead of a single pci-bridge device. I'm still not sure if that enough to warrant an entirely new machine type, but it definitely has a more far-reaching impact than simply flipping the default USB controller from OHCI to XHCI.
Even if no newer pseries-x.y were to be added after introducing pseries-pcie, you could still easily create guests that use either root bus, so no loss in functionality. I could do this with the existing pseries if the machine had a "root but type" property.
That was indeed one of my original proposals ;) Just note, once again, that the default for this property would have to be "off" or we would run into the same issues described above. [1] Even though, as I mentioned earlier in the thread, performing version checks on machine types is frowned upon as it turns into a minefield as soon as backports and downstream machine types enter the picture... It's much better to expose some flag through QMP for libvirt to introspect -- Andrea Bolognani / Red Hat / Virtualization
On 22/11/16 00:08, Andrea Bolognani wrote:
On Mon, 2016-11-21 at 13:12 +1100, Alexey Kardashevskiy wrote:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
[...] Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using
How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make?
Existing libvirt versions assume that pseries guests have
If libvirt is using just "pseries" (without a version), then having a virtual PCIe-PCI bridge (and "pci.0" always available by default) will do it.
Please don't. Any device that is included in the guest by default and can't be turned off makes libvirt's life significantly more difficult (see below).
If libvirt is using a specific version of pseries, then it already knows that <=2.7 has pci.0 as a root, pcie.0 otherwise. libvirt has a knowledge what QEMU version has what, right?
It doesn't yet, that's the point :)
We *could* add such knowledge to libvirt[1], but *existing* libvirt versions would still not know about it, which means that upgrading QEMU withough upgrading libvirt will result in failure to create new guests.
In what scenario will an additional machine type help?
Because then libvirt could learn that
pseries-x.y <-> pci.0 pseries-pcie-x.y <-> pcie.0
the same way it already knows that
pc-i440fx-x.y <-> pci.0 pc-q35-x.y <-> pcie.0
and choosing between one or the other would be, I think, much easier for the user as well.
a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges.
So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type.
I do not know about i440<->q35 but in pseries the difference is going to be very simple.
For example, we did not change the machine type when we switched from default OHCI to XHCI, switching from PCI to PCIe does not sound like we need a whole new machine type for this either.
The change from OHCI to XHCI only affected the *default* USB controller, which libvirt tries its best not to use anyway: instead, it will prefer to use '-M ...,usb=off' along with '-device ...' and set both the controller model and its PCI address explicitly, partially to shield its users from such changes in QEMU.
Ok. Always likes this approach really. We should start moving to this direction with PHB - stop adding the default PHB at all when -nodefaults is passed (or -machine pseries,pci=off ?) and let libvirt manage PHBs itself (and provide another spapr-phb type like spapr-pcie-host-bridge or add a "pcie_root_bus_type" property to the existing PHB type). What will be wrong with this approach?
Let's not forget that libvirt is a management layer, and as such it needs to have as much control as possible over the virtual hardware presented to the guest, mostly for guest ABI compatibility purposes. Defaulting to XHCI instead of OHCI, same as pretty much all other defaults, is good for people who run QEMU directly, but libvirt needs to be able to control such knobs itself in order to effectively perform the task it was designed for.
Moreover, we're talking about a more fundamental change here: the PCI Root Bus is not just any random device, it's the one fundation upon which the entire PCI hierarchy is built. Based on whether the machine exposes a PCI Express Root Bus or a legacy PCI Express Root Bus, libvirt will create very different PCI hierarchies, eg. using several ioh3420 devices instead of a single pci-bridge device.
I'm still not sure if that enough to warrant an entirely new machine type, but it definitely has a more far-reaching impact than simply flipping the default USB controller from OHCI to XHCI.
Even if no newer pseries-x.y were to be added after introducing pseries-pcie, you could still easily create guests that use either root bus, so no loss in functionality.
I could do this with the existing pseries if the machine had a "root but type" property.
That was indeed one of my original proposals ;)
Just note, once again, that the default for this property would have to be "off" or we would run into the same issues described above.
[1] Even though, as I mentioned earlier in the thread, performing version checks on machine types is frowned upon as it turns into a minefield as soon as backports and downstream machine types enter the picture... It's much better to expose some flag through QMP for libvirt to introspect -- Andrea Bolognani / Red Hat / Virtualization
-- Alexey
On Tue, Nov 22, 2016 at 01:26:49PM +1100, Alexey Kardashevskiy wrote:
On 22/11/16 00:08, Andrea Bolognani wrote:
On Mon, 2016-11-21 at 13:12 +1100, Alexey Kardashevskiy wrote:
1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
[...] Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using
How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make?
Existing libvirt versions assume that pseries guests have
If libvirt is using just "pseries" (without a version), then having a virtual PCIe-PCI bridge (and "pci.0" always available by default) will do it.
Please don't. Any device that is included in the guest by default and can't be turned off makes libvirt's life significantly more difficult (see below).
If libvirt is using a specific version of pseries, then it already knows that <=2.7 has pci.0 as a root, pcie.0 otherwise. libvirt has a knowledge what QEMU version has what, right?
It doesn't yet, that's the point :)
We *could* add such knowledge to libvirt[1], but *existing* libvirt versions would still not know about it, which means that upgrading QEMU withough upgrading libvirt will result in failure to create new guests.
In what scenario will an additional machine type help?
Because then libvirt could learn that
pseries-x.y <-> pci.0 pseries-pcie-x.y <-> pcie.0
the same way it already knows that
pc-i440fx-x.y <-> pci.0 pc-q35-x.y <-> pcie.0
and choosing between one or the other would be, I think, much easier for the user as well.
a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges.
So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type.
I do not know about i440<->q35 but in pseries the difference is going to be very simple.
For example, we did not change the machine type when we switched from default OHCI to XHCI, switching from PCI to PCIe does not sound like we need a whole new machine type for this either.
The change from OHCI to XHCI only affected the *default* USB controller, which libvirt tries its best not to use anyway: instead, it will prefer to use '-M ...,usb=off' along with '-device ...' and set both the controller model and its PCI address explicitly, partially to shield its users from such changes in QEMU.
Ok. Always likes this approach really. We should start moving to this direction with PHB - stop adding the default PHB at all when -nodefaults is passed (or -machine pseries,pci=off ?) and let libvirt manage PHBs itself (and provide another spapr-phb type like spapr-pcie-host-bridge or add a "pcie_root_bus_type" property to the existing PHB type).
What will be wrong with this approach?
Hm, that's a good point. If were removing the default PHB entirely, that I would consider a possible case for a new machine type. Putting construction of the PHBs into libvirt's hand could make life simpler there. Although it would make it a bit of a pain for people starting qemu by hand. I think this option is worth some thought. -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Wed, 2016-11-23 at 16:02 +1100, David Gibson wrote:
The change from OHCI to XHCI only affected the *default* USB controller, which libvirt tries its best not to use anyway: instead, it will prefer to use '-M ...,usb=off' along with '-device ...' and set both the controller model and its PCI address explicitly, partially to shield its users from such changes in QEMU. Ok. Always likes this approach really. We should start moving to this direction with PHB - stop adding the default PHB at all when -nodefaults is passed (or -machine pseries,pci=off ?) and let libvirt manage PHBs itself (and provide another spapr-phb type like spapr-pcie-host-bridge or add a "pcie_root_bus_type" property to the existing PHB type). What will be wrong with this approach? Hm, that's a good point. If were removing the default PHB entirely, that I would consider a possible case for a new machine type. Putting construction of the PHBs into libvirt's hand could make life simpler there. Although it would make it a bit of a pain for people starting qemu by hand. I think this option is worth some thought.
Note that libvirt always runs QEMU with -nodefaults, so we could just remove the default PHB if that flag is present, and leave it in if that's not the case. The idea itself sounds good, but of course it will require more work from the libvirt side than simply making the PCIe machine type behave like q35 and mach-virt. Moreover, we already have an RFE for supporting additional PHBs, we could solve both issues in one fell swoop and have the libvirt guest XML be a more faithful representation of the actual virtual hardware, which is always a win in my book. That will be even more important if it turns out that the rules for PCIe device assignment (eg. what device/controller can be plugged into which slot) are different for PCIe PHBs than they are for q35/mach-virt PCIe Root Bus. I've asked for clarifications about this elsewhere in the thread. -- Andrea Bolognani / Red Hat / Virtualization
On Fri, Nov 25, 2016 at 03:36:25PM +0100, Andrea Bolognani wrote:
On Wed, 2016-11-23 at 16:02 +1100, David Gibson wrote:
The change from OHCI to XHCI only affected the *default* USB controller, which libvirt tries its best not to use anyway: instead, it will prefer to use '-M ...,usb=off' along with '-device ...' and set both the controller model and its PCI address explicitly, partially to shield its users from such changes in QEMU. Ok. Always likes this approach really. We should start moving to this direction with PHB - stop adding the default PHB at all when -nodefaults is passed (or -machine pseries,pci=off ?) and let libvirt manage PHBs itself (and provide another spapr-phb type like spapr-pcie-host-bridge or add a "pcie_root_bus_type" property to the existing PHB type). What will be wrong with this approach? Hm, that's a good point. If were removing the default PHB entirely, that I would consider a possible case for a new machine type. Putting construction of the PHBs into libvirt's hand could make life simpler there. Although it would make it a bit of a pain for people starting qemu by hand. I think this option is worth some thought.
Note that libvirt always runs QEMU with -nodefaults, so we could just remove the default PHB if that flag is present, and leave it in if that's not the case.
The idea itself sounds good, but of course it will require more work from the libvirt side than simply making the PCIe machine type behave like q35 and mach-virt.
Yeah, but the latter basically just won't work.
Moreover, we already have an RFE for supporting additional PHBs, we could solve both issues in one fell swoop and have the libvirt guest XML be a more faithful representation of the actual virtual hardware, which is always a win in my book.
Right. And the general recomendation for PAPR guests is that each passed through device (or rather, each passed through iommu group) have a separate virtual PHB on the guest. With this rework libvirt could switch over to implementing that recommendation.
That will be even more important if it turns out that the rules for PCIe device assignment (eg. what device/controller can be plugged into which slot) are different for PCIe PHBs than they are for q35/mach-virt PCIe Root Bus. I've asked for clarifications about this elsewhere in the thread.
-- Andrea Bolognani / Red Hat / Virtualization
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On 11/21/2016 07:08 AM, Andrea Bolognani wrote:
If libvirt is using a specific version of pseries, then it already knows that <=2.7 has pci.0 as a root, pcie.0 otherwise. libvirt has a knowledge what QEMU version has what, right?
It doesn't yet, that's the point :)
We *could* add such knowledge to libvirt[1], but *existing* libvirt versions would still not know about it, which means that upgrading QEMU withough upgrading libvirt will result in failure to create new guests.
But that's okay. In general, we try to promise that: old qemu + old libvirt works new qemu + new libvirt works old qemu + new libvirt works new qemu + old libvirt is untested, and may require a libvirt upgrade In other words, as long as you update your stack from top down (libvirt before qemu), you should be fine. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
On Fri, Nov 18, 2016 at 09:17:22AM +0100, Andrea Bolognani wrote:
On Thu, 2016-11-17 at 13:02 +1100, Alexey Kardashevskiy wrote:
That said, considering that a big part of the PCI address allocation logic is based off whether the specific machine type exposes a legay PCI Root Bus or a PCI Express Root Bus, libvirt will need a way to be able to tell which one is which. Version checks are pretty much out of the question, as they fail as soon as downstream releases enter the picture. A few ways we could deal with the situation: 1) switch to PCI Express on newer machine types, and expose some sort of capability through QMP so that libvirt can know about the switch
[...]
Option 1) would break horribly with existing libvirt versions, and so would Option 2) if we default to using How exactly 1) will break libvirt? Migrating from pseries-2.7 to pseries-2.8 does not work anyway, and machines are allowed to behave different from version to version, what distinct difference will using "pseries-pcie-X.Y" make?
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges.
Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR.
So if you used a new QEMU binary with a libvirt version that doesn't know about the change, new guests would end up using the wrong controllers. Existing guests would not be affected as they would stick with the older machine types, of course.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type.
Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E.
Even if no newer pseries-x.y were to be added after introducing pseries-pcie, you could still easily create guests that use either root bus, so no loss in functionality.
-- Andrea Bolognani / Red Hat / Virtualization
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Wed, 2016-11-23 at 16:00 +1100, David Gibson wrote:
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR.
So, would the PCIe Root Bus in a pseries guest behave differently than the one in a q35 or mach-virt guest? Does it have a different number of slots, do we have to plug different controllers into them, ...? Regardless of how we decide to move forward with the PCIe-enabled pseries machine type, libvirt will have to know about this so it can behave appropriately.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y. Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type. Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E.
Calling the PCIe machine type either pseries-2.8 or pseries-pcie-2.8 would result in the very same amount of work, and in both cases it would be understood that the legacy PCI machine type is no longer going to be updated, but can still be used to run existing guests. -- Andrea Bolognani / Red Hat / Virtualization
On Fri, Nov 25, 2016 at 02:46:21PM +0100, Andrea Bolognani wrote:
On Wed, 2016-11-23 at 16:00 +1100, David Gibson wrote:
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR.
So, would the PCIe Root Bus in a pseries guest behave differently than the one in a q35 or mach-virt guest?
Yes. I had a long discussion with BenH and got a somewhat better idea about this. If only a single host PE (== iommu group) is passed through and there are no emulated devices, the difference isn't too bad: basically on pseries you'll see the subtree that would be below the root complex on q35. But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E). I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
Does it have a different number of slots, do we have to plug different controllers into them, ...?
Regardless of how we decide to move forward with the PCIe-enabled pseries machine type, libvirt will have to know about this so it can behave appropriately.
So there are kind of two extremes of how to address this. There are a variety of options in between, but I suspect they're going to be even more muddled and hideous than the extremes. 1) Give up. You said there's already a flag that says a PCI-E bus is able to accept vanilla-PCI devices. We add a hack flag that says a vanilla-PCI bus is able to accept PCI-E devices. We keep address allocation as it is now - the pseries topology really does resemble vanilla-PCI much better than it does PCI-E. But, we allow PCI-E devices, and PAPR has mechanisms for accessing the extended config space. PCI-E standard hotplug and error reporting will never work, but PAPR provides its own mechanisms for those, so that should be ok. 2) Start exposing the PCI-E heirarchy for pseries guests much more like q35, root complexes and all. It's not clear that PAPR actually *forbids* exposing the root complex, it just doesn't require it and that's not what PowerVM does. But.. there are big questions about whether existing guests will cope with this or not. When you start adding in multiple passed through devices and particularly virtual functions as well, things could get very ugly - we might need to construct multiple emulated virtual root complexes or other messes. In the short to medium term, I'm thinking option (1) seems pretty compelling.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y. Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type. Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E.
Calling the PCIe machine type either pseries-2.8 or pseries-pcie-2.8 would result in the very same amount of work, and in both cases it would be understood that the legacy PCI machine type is no longer going to be updated, but can still be used to run existing guests.
So, I'm not sure if the idea of a new machine type has legs or not, but let's think it through a bit further. Suppose we have a new machine type, let's call it 'papr'. I'm thinking it would be (at least with -nodefaults) basically a super-minimal version of pseries: so each PHB would have to be explicitly created, the VIO bridge would have to be explicitly created, likewise the NVRAM. Not sure about the "devices" which really represent firmware features - the RTC, RNG, hypervisor event source and so forth. Might have some advantages. Then again, it doesn't really solve the specific problem here. It means libvirt (or the user) has to explicitly choose a PCI or PCI-E PHB to put things on, but libvirt's PCI-E address allocation will still be wrong in all probability. Guh. As an aside, here's a RANT. libvirt address allocation. Seriously, W. T. F! libvirt insists on owning address allocation. That's so it can recreate the exact same machine at the far end of a migration. So far so good, except it insists on recording that information in the domain XML in kinda-sorta-but-not-really back end independent form. But the thing is libvirt fundamentally CAN NOT get this right. There are all sorts of possible machine specific address allocation constraints that can exist - from simply which devices are already created by default (for varying values of "default") to complicated constraints depending on details of board wiring. The back end has to know about these - it implements them. The ONLY way libvirt can get this (temporarily) right is by duplicating a huge chunk of the back end's allocation logic, which will inevitably get out of date causing problems just like this. Basically the back end will *always* have better information about how to place devices than libvirt can. So, libvirt should be allowing the back end to do the allocation, then snapshotting that in a back end specific format which can be used for creating migration destinations. But that breaks libvirt's the-domain-XML-is-everything model. In this regard libvirt doesn't just have a design flaw, it has design flaws which breed more design flaws like pestilent cancer. And what's worse the consequences of those design flaws are now making sane design decisions increasingly difficult in adjacent projects like qemu. I'd feel better about this if there seemed to be some recognition of it, and some necessarily long term plan to improve it, but if there is I haven't heard of it. Or at least the closest thing seems to be coming from the qemu side (witness Eduardo's talk at the last KVM forum, and mine at the one before). -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Fri, 2016-12-02 at 15:18 +1100, David Gibson wrote:
But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
It's also somewhat broken if they aren't in the same iommu domain because the way we pass the iommu buid is via the parent node, so a given iommu domain must reside below a common parent and not share it.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
On Fri, Dec 02, 2016 at 04:17:50PM +1100, Benjamin Herrenschmidt wrote:
On Fri, 2016-12-02 at 15:18 +1100, David Gibson wrote:
But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
It's also somewhat broken if they aren't in the same iommu domain because the way we pass the iommu buid is via the parent node, so a given iommu domain must reside below a common parent and not share it.
Uh.. I don't entirely follow you. From the host point of view there are multiple iommu groups (PEs), but from the guest point of view there's only one. On the guest side iommu granularity is always per-vPHB.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Fri, 2016-12-02 at 16:50 +1100, David Gibson wrote:
Uh.. I don't entirely follow you. From the host point of view there are multiple iommu groups (PEs), but from the guest point of view there's only one. On the guest side iommu granularity is always per-vPHB.
Ok so the H_PUT_TCE calls affect all the underlying tables ? Cheers, Ben.
On 03/12/16 08:41, Benjamin Herrenschmidt wrote:
On Fri, 2016-12-02 at 16:50 +1100, David Gibson wrote:
Uh.. I don't entirely follow you. From the host point of view there are multiple iommu groups (PEs), but from the guest point of view there's only one. On the guest side iommu granularity is always per-vPHB.
Ok so the H_PUT_TCE calls affect all the underlying tables ?
Yes, it updates all hardware tables attached to the same guest's LIOBN (which in reality can only be seen on power7 as power8 shares tables between PEs). -- Alexey
On 12/01/2016 11:18 PM, David Gibson wrote:
On Fri, Nov 25, 2016 at 02:46:21PM +0100, Andrea Bolognani wrote:
On Wed, 2016-11-23 at 16:00 +1100, David Gibson wrote:
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges.
Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR. So, would the PCIe Root Bus in a pseries guest behave differently than the one in a q35 or mach-virt guest? Yes. I had a long discussion with BenH and got a somewhat better idea about this.
If only a single host PE (== iommu group) is passed through and there are no emulated devices, the difference isn't too bad: basically on pseries you'll see the subtree that would be below the root complex on q35.
But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
Does it have a different number of slots, do we have to plug different controllers into them, ...?
Regardless of how we decide to move forward with the PCIe-enabled pseries machine type, libvirt will have to know about this so it can behave appropriately. So there are kind of two extremes of how to address this. There are a variety of options in between, but I suspect they're going to be even more muddled and hideous than the extremes.
1) Give up. You said there's already a flag that says a PCI-E bus is able to accept vanilla-PCI devices. We add a hack flag that says a vanilla-PCI bus is able to accept PCI-E devices. We keep address allocation as it is now - the pseries topology really does resemble vanilla-PCI much better than it does PCI-E. But, we allow PCI-E devices, and PAPR has mechanisms for accessing the extended config space. PCI-E standard hotplug and error reporting will never work, but PAPR provides its own mechanisms for those, so that should be ok.
2) Start exposing the PCI-E heirarchy for pseries guests much more like q35, root complexes and all. It's not clear that PAPR actually *forbids* exposing the root complex, it just doesn't require it and that's not what PowerVM does. But.. there are big questions about whether existing guests will cope with this or not. When you start adding in multiple passed through devices and particularly virtual functions as well, things could get very ugly - we might need to construct multiple emulated virtual root complexes or other messes.
In the short to medium term, I'm thinking option (1) seems pretty compelling.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y.
Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type.
Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E. Calling the PCIe machine type either pseries-2.8 or pseries-pcie-2.8 would result in the very same amount of work, and in both cases it would be understood that the legacy PCI machine type is no longer going to be updated, but can still be used to run existing guests. So, I'm not sure if the idea of a new machine type has legs or not, but let's think it through a bit further. Suppose we have a new machine type, let's call it 'papr'. I'm thinking it would be (at least with -nodefaults) basically a super-minimal version of pseries: so each PHB would have to be explicitly created, the VIO bridge would have to be explicitly created, likewise the NVRAM. Not sure about the "devices" which really represent firmware features - the RTC, RNG, hypervisor event source and so forth.
Might have some advantages. Then again, it doesn't really solve the specific problem here. It means libvirt (or the user) has to explicitly choose a PCI or PCI-E PHB to put things on, but libvirt's PCI-E address allocation will still be wrong in all probability.
That's a broad statement. Why? If qemu reports the default devices and characteristics of the devices properly (and libvirt uses that information) there's no reason for it to make the wrong decision.
Guh.
As an aside, here's a RANT.
libvirt address allocation. Seriously, W. T. F!
libvirt insists on owning address allocation. That's so it can recreate the exact same machine at the far end of a migration. So far so good, except it insists on recording that information in the domain XML in kinda-sorta-but-not-really back end independent form.
Explain "kinda-sorta-but-not-really". If there's a deficiency in the model maybe it can be fixed.
But the thing is libvirt fundamentally CAN NOT get this right.
True, but not for the reasons you think. If qemu is able to respond to queries with adequate details about the devices available for a machinetype (and what buses are in place by default), there's no reason that libvirt can't add devices addressed such that all the connections are legal; what libvirt *can't* get right is the policy requested in the next higher layer of management (and ultimately of the user) - does this device need to be hotpluggable? Does the user want to keep all devices on the root complex to avoid extra PCI controllers? And qemu fundamentally CAN NOT get it right either. qemu knows what is possible and what is allowed, but it doesn't know what the user *wants* (beyond "they want device X", which is only 1/2 the story), and has no way of being told what the user wants other than with a PCI address. To back up for a minute, some background info: once a device has been added to a domain, at *any* time in the future (not just during a migration, but forever more until the end of time) that device must always have the same PCI address as it had that first time. In order to guarantee that, libvirt needs to either: a) keep track of the order the devices were added and always put the devices in the same order on the commandline (assuming that qemu guarantees that it actually assigns addresses based on the order of the devices' appearance on the commandline, which has never been stated anywhere as an API guarantee of qemu), or b) remember the address of each device as it is added and specify that on the commandline in the future. libvirt chooses (b). And where is the logical place to store that address? In the config. So we've established that the PCI address of a device needs to be stored in the config. So why does libvirt need to choose it the first time? 1) Because qemu doesn't have (and CAN NOT have) all the information about what are the user's plans for that device: a) It has no idea if the user wants that device to be hotpluggable (on a root-port) or not (on root complex as an integrated device) b) it doesn't know if the user wants the device to be placed on an expander bus so that its NUMA status can be discovered by the guest OS. If there is a choice, there must be a way to make that choice. The way that qemu provides to make the choice is by specifying an address. So libvirt must specify an address in its config. 2) Because qemu is unable/unwilling to automatically add PCIe root ports when necessary, it's *not even possible* (on PCIe machinetypes) for it to place a device on a hotpluggable port without libvirt specifying a PCI address the very first time the device is added (and also adding in a root-port), but libvirt's default policy is that (almost) all devices should be hotpluggable. If we were to follow your recommendation ("libvirt never specifies PCI addresses, but instead allows qemu to assign them"), hotplug on PCIe-based machinetypes would not be possible, though. There have even been mentions that even libvirt is too *low* in the stack to be specifying the PCI address of devices (i.e. that all PCI address decisions should be up to higher level management applications) - I had posted a patch that would allow specifying "hotpluggable='yes|no'" in the XML rather than forcing the call to specify an address, and this was NACKed because it was seen as libvirt dictating policy. (In the end, libvirt *does* dictate a default policy, (it's just that the only way to modify that policy is by manually specifying addresses) - libvirt's default PCI address policy is that (almost) all devices will be assigned an address that makes them hotpluggable, and will not be placed on a non-0 NUMA node. So, in spite of libvirt's effort, in the end we *still* need to expose address configuration to higher level management applications, since they may want to force all devices onto the root complex (e.g. libguestfs, which does it to reduce PCI controller count, and thus startup time) or force certain devices to be on a non-0 NUMA node (e.g. OpenStack when it wants to place a VFIO assigned device on the same NUMA node in the guest as it is in the host). With all of that, I fail to see how it would be at all viable to simply leave PCI address assignment up to qemu.
There are all sorts of possible machine specific address allocation constraints that can exist - from simply which devices are already created by default (for varying values of "default") to complicated constraints depending on details of board wiring. The back end has to know about these - it implements them.
And since qemu knows about them, it should be able to report them. Which is what Eduardo's work is doing. And then libvirt will know about all the constraints in an programmatic manner (rather than the horrible (tedious, error prone) hardcoding of all those details that we've had to suffer with until now).
The ONLY way libvirt can get this (temporarily) right is by duplicating a huge chunk of the back end's allocation logic, which will inevitably get out of date causing problems just like this.
Basically the back end will *always* have better information about how to place devices than libvirt can.
And since no matter how hard qemu might try to come up with a policy for address assignment that will satisfy the needs of 100% of the users 100% of the time, it will fail (because different users have different needs). Because qemu will be unable to properly place all devices all the time, libvirt (and higher level management) will still need to do it. Even in the basic case qemu doesn't provide what libvirt requires as default - that devices be hotpluggable.
So, libvirt should be allowing the back end to do the allocation, then snapshotting that in a back end specific format which can be used for creating migration destinations. But that breaks libvirt's the-domain-XML-is-everything model.
No, that doesn't work because qemu would in many situations place the devices at the wrong address / on the wrong controller, because there are many possible topologies that are legal, and the user may (for perfectly valid reasons) want something different from what qemu would have chosen. (An example of two differing (and valid) policies - libguestfs needs guests to startup as quickly as possible, and thus wants as few PCI controllers as possible (this makes a noticeable difference in Linux boot time), so it wants all devices to be integrated on the root complex. On the other hand, a generic guest in libvirt should make all devices hotpluggable just in case the user wants to unplug them, so by default it tries to place all devices on a pcie-root-port. You can't support both of these if addressing is all left up to qemu)
In this regard libvirt doesn't just have a design flaw, it has design flaws which breed more design flaws like pestilent cancer.
It may make you feel good to say that, but the facts don't back it up. Any project makes design mistakes, but in the specific case you're discussing here, I think you haven't looked from a wide enough viewpoint to see the necessity of what libvirt is doing and why it can't be done by qemu (certainly not all the time anyway).
And what's worse the consequences of those design flaws are now making sane design decisions increasingly difficult in adjacent projects like qemu.
libvirt has always done the best that could be done with the information provided by qemu. The problem isn't that libvirt is creating new problems for qemu out of thin air, it's that qemu is unable to automatically address PCI devices for all possible situations and user policy preferences, so higher levels need to make the decisions about addressing to satisfy their policies (ie what they *want*, eg hotpluggable, integrated on root complex), and qemu hasn't (until Eduardo's patches) been able to provide adequate information about what is *legal* (e.g which devices can be plugged into which model of pci controller, what slots are available on each type of controller, whether those slots are hotpluggable) in a programmatic way, so libvirt has had to hardcode rules about bus-device compatibility and capabilities, slot ranges, etc in order to make proper decisions itself when possible, and to sanity-check decisions about addresses made by higher level management when not. I don't think that's a design flaw. I think that's making the best of a "less than ideal" situation.
I'd feel better about this if there seemed to be some recognition of it, and some necessarily long term plan to improve it, but if there is I haven't heard of it. Or at least the closest thing seems to be coming from the qemu side (witness Eduardo's talk at the last KVM forum, and mine at the one before).
Eduardo's work isn't being done to make up for some mythical design flaw in libvirt. It is being done in order to give libvirt the (previously unavailable) information it needs to do a necessary job, and is being done at least partly at the request of libvirt (we've certainly been demanding some of that stuff for a long time!) The summary is that it's impossible for qemu to correctly decide where to put new devices, especially in a PCIe hierarchy for a few reasons (at least); because of this, libvirt (and higher level management) needs to be able to assign addresses to devices, and in order for us/them to be able to do that properly, qemu needs to provide detailed and accurate information about what buses/controllers/devices are in each machinetype, what controllers/devices are available to add, and what are the legal ways of connecting those devices and controllers.
(Sorry for any duplicates. I sent it from the wrong address the first time) On 12/01/2016 11:18 PM, David Gibson wrote:
On Fri, Nov 25, 2016 at 02:46:21PM +0100, Andrea Bolognani wrote:
On Wed, 2016-11-23 at 16:00 +1100, David Gibson wrote:
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR. So, would the PCIe Root Bus in a pseries guest behave differently than the one in a q35 or mach-virt guest? Yes. I had a long discussion with BenH and got a somewhat better idea about this.
If only a single host PE (== iommu group) is passed through and there are no emulated devices, the difference isn't too bad: basically on pseries you'll see the subtree that would be below the root complex on q35.
But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
Does it have a different number of slots, do we have to plug different controllers into them, ...?
Regardless of how we decide to move forward with the PCIe-enabled pseries machine type, libvirt will have to know about this so it can behave appropriately. So there are kind of two extremes of how to address this. There are a variety of options in between, but I suspect they're going to be even more muddled and hideous than the extremes.
1) Give up. You said there's already a flag that says a PCI-E bus is able to accept vanilla-PCI devices. We add a hack flag that says a vanilla-PCI bus is able to accept PCI-E devices. We keep address allocation as it is now - the pseries topology really does resemble vanilla-PCI much better than it does PCI-E. But, we allow PCI-E devices, and PAPR has mechanisms for accessing the extended config space. PCI-E standard hotplug and error reporting will never work, but PAPR provides its own mechanisms for those, so that should be ok.
2) Start exposing the PCI-E heirarchy for pseries guests much more like q35, root complexes and all. It's not clear that PAPR actually *forbids* exposing the root complex, it just doesn't require it and that's not what PowerVM does. But.. there are big questions about whether existing guests will cope with this or not. When you start adding in multiple passed through devices and particularly virtual functions as well, things could get very ugly - we might need to construct multiple emulated virtual root complexes or other messes.
In the short to medium term, I'm thinking option (1) seems pretty compelling.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y. Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type. Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E. Calling the PCIe machine type either pseries-2.8 or pseries-pcie-2.8 would result in the very same amount of work, and in both cases it would be understood that the legacy PCI machine type is no longer going to be updated, but can still be used to run existing guests. So, I'm not sure if the idea of a new machine type has legs or not, but let's think it through a bit further. Suppose we have a new machine type, let's call it 'papr'. I'm thinking it would be (at least with -nodefaults) basically a super-minimal version of pseries: so each PHB would have to be explicitly created, the VIO bridge would have to be explicitly created, likewise the NVRAM. Not sure about the "devices" which really represent firmware features - the RTC, RNG, hypervisor event source and so forth.
Might have some advantages. Then again, it doesn't really solve the specific problem here. It means libvirt (or the user) has to explicitly choose a PCI or PCI-E PHB to put things on, but libvirt's PCI-E address allocation will still be wrong in all probability.
That's a broad statement. Why? If qemu reports the default devices and characteristics of the devices properly (and libvirt uses that information) there's no reason for it to make the wrong decision.
Guh.
As an aside, here's a RANT.
libvirt address allocation. Seriously, W. T. F!
libvirt insists on owning address allocation. That's so it can recreate the exact same machine at the far end of a migration. So far so good, except it insists on recording that information in the domain XML in kinda-sorta-but-not-really back end independent form.
Explain "kinda-sorta-but-not-really". If there's a deficiency in the model maybe it can be fixed.
But the thing is libvirt fundamentally CAN NOT get this right.
True, but not for the reasons you think. If qemu is able to respond to queries with adequate details about the devices available for a machinetype (and what buses are in place by default), there's no reason that libvirt can't add devices addressed such that all the connections are legal; what libvirt *can't* get right is the policy requested in the next higher layer of management (and ultimately of the user) - does this device need to be hotpluggable? Does the user want to keep all devices on the root complex to avoid extra PCI controllers? And qemu fundamentally CAN NOT get it right either. qemu knows what is possible and what is allowed, but it doesn't know what the user *wants* (beyond "they want device X", which is only 1/2 the story), and has no way of being told what the user wants other than with a PCI address. To back up for a minute, some background info: once a device has been added to a domain, at *any* time in the future (not just during a migration, but forever more until the end of time) that device must always have the same PCI address as it had that first time. In order to guarantee that, libvirt needs to either: a) keep track of the order the devices were added and always put the devices in the same order on the commandline (assuming that qemu guarantees that it actually assigns addresses based on the order of the devices' appearance on the commandline, which has never been stated anywhere as an API guarantee of qemu), or b) remember the address of each device as it is added and specify that on the commandline in the future. libvirt chooses (b). And where is the logical place to store that address? In the config. So we've established that the PCI address of a device needs to be stored in the config. So why does libvirt need to choose it the first time? 1) Because qemu doesn't have (and CAN NOT have) all the information about what are the user's plans for that device: a) It has no idea if the user wants that device to be hotpluggable (on a root-port) or not (on root complex as an integrated device) b) it doesn't know if the user wants the device to be placed on an expander bus so that its NUMA status can be discovered by the guest OS. If there is a choice, there must be a way to make that choice. The way that qemu provides to make the choice is by specifying an address. So libvirt must specify an address in its config. 2) Because qemu is unable/unwilling to automatically add PCIe root ports when necessary, it's *not even possible* (on PCIe machinetypes) for it to place a device on a hotpluggable port without libvirt specifying a PCI address the very first time the device is added (and also adding in a root-port), but libvirt's default policy is that (almost) all devices should be hotpluggable. If we were to follow your recommendation ("libvirt never specifies PCI addresses, but instead allows qemu to assign them"), hotplug on PCIe-based machinetypes would not be possible, though. There have even been mentions that even libvirt is too *low* in the stack to be specifying the PCI address of devices (i.e. that all PCI address decisions should be up to higher level management applications) - I had posted a patch that would allow specifying "hotpluggable='yes|no'" in the XML rather than forcing the call to specify an address, and this was NACKed because it was seen as libvirt dictating policy. (In the end, libvirt *does* dictate a default policy, (it's just that the only way to modify that policy is by manually specifying addresses) - libvirt's default PCI address policy is that (almost) all devices will be assigned an address that makes them hotpluggable, and will not be placed on a non-0 NUMA node. So, in spite of libvirt's effort, in the end we *still* need to expose address configuration to higher level management applications, since they may want to force all devices onto the root complex (e.g. libguestfs, which does it to reduce PCI controller count, and thus startup time) or force certain devices to be on a non-0 NUMA node (e.g. OpenStack when it wants to place a VFIO assigned device on the same NUMA node in the guest as it is in the host). With all of that, I fail to see how it would be at all viable to simply leave PCI address assignment up to qemu.
There are all sorts of possible machine specific address allocation constraints that can exist - from simply which devices are already created by default (for varying values of "default") to complicated constraints depending on details of board wiring. The back end has to know about these - it implements them.
And since qemu knows about them, it should be able to report them. Which is what Eduardo's work is doing. And then libvirt will know about all the constraints in an programmatic manner (rather than the horrible (tedious, error prone) hardcoding of all those details that we've had to suffer with until now).
The ONLY way libvirt can get this (temporarily) right is by duplicating a huge chunk of the back end's allocation logic, which will inevitably get out of date causing problems just like this.
Basically the back end will *always* have better information about how to place devices than libvirt can.
And since no matter how hard qemu might try to come up with a policy for address assignment that will satisfy the needs of 100% of the users 100% of the time, it will fail (because different users have different needs). Because qemu will be unable to properly place all devices all the time, libvirt (and higher level management) will still need to do it. Even in the basic case qemu doesn't provide what libvirt requires as default - that devices be hotpluggable.
So, libvirt should be allowing the back end to do the allocation, then snapshotting that in a back end specific format which can be used for creating migration destinations. But that breaks libvirt's the-domain-XML-is-everything model.
No, that doesn't work because qemu would in many situations place the devices at the wrong address / on the wrong controller, because there are many possible topologies that are legal, and the user may (for perfectly valid reasons) want something different from what qemu would have chosen. (An example of two differing (and valid) policies - libguestfs needs guests to startup as quickly as possible, and thus wants as few PCI controllers as possible (this makes a noticeable difference in Linux boot time), so it wants all devices to be integrated on the root complex. On the other hand, a generic guest in libvirt should make all devices hotpluggable just in case the user wants to unplug them, so by default it tries to place all devices on a pcie-root-port. You can't support both of these if addressing is all left up to qemu)
In this regard libvirt doesn't just have a design flaw, it has design flaws which breed more design flaws like pestilent cancer.
It may make you feel good to say that, but the facts don't back it up. Any project makes design mistakes, but in the specific case you're discussing here, I think you haven't looked from a wide enough viewpoint to see the necessity of what libvirt is doing and why it can't be done by qemu (certainly not all the time anyway).
And what's worse the consequences of those design flaws are now making sane design decisions increasingly difficult in adjacent projects like qemu.
libvirt has always done the best that could be done with the information provided by qemu. The problem isn't that libvirt is creating new problems for qemu out of thin air, it's that qemu is unable to automatically address PCI devices for all possible situations and user policy preferences, so higher levels need to make the decisions about addressing to satisfy their policies (ie what they *want*, eg hotpluggable, integrated on root complex), and qemu hasn't (until Eduardo's patches) been able to provide adequate information about what is *legal* (e.g which devices can be plugged into which model of pci controller, what slots are available on each type of controller, whether those slots are hotpluggable) in a programmatic way, so libvirt has had to hardcode rules about bus-device compatibility and capabilities, slot ranges, etc in order to make proper decisions itself when possible, and to sanity-check decisions about addresses made by higher level management when not. I don't think that's a design flaw. I think that's making the best of a "less than ideal" situation.
I'd feel better about this if there seemed to be some recognition of it, and some necessarily long term plan to improve it, but if there is I haven't heard of it. Or at least the closest thing seems to be coming from the qemu side (witness Eduardo's talk at the last KVM forum, and mine at the one before).
Eduardo's work isn't being done to make up for some mythical design flaw in libvirt. It is being done in order to give libvirt the (previously unavailable) information it needs to do a necessary job, and is being done at least partly at the request of libvirt (we've certainly been demanding some of that stuff for a long time!) The summary is that it's impossible for qemu to correctly decide where to put new devices, especially in a PCIe hierarchy for a few reasons (at least); because of this, libvirt (and higher level management) needs to be able to assign addresses to devices, and in order for us/them to be able to do that properly, qemu needs to provide detailed and accurate information about what buses/controllers/devices are in each machinetype, what controllers/devices are available to add, and what are the legal ways of connecting those devices and controllers.
On Mon, Dec 05, 2016 at 03:54:49PM -0500, Laine Stump wrote:
(Sorry for any duplicates. I sent it from the wrong address the first time)
On 12/01/2016 11:18 PM, David Gibson wrote:
On Fri, Nov 25, 2016 at 02:46:21PM +0100, Andrea Bolognani wrote:
On Wed, 2016-11-23 at 16:00 +1100, David Gibson wrote:
Existing libvirt versions assume that pseries guests have a legacy PCI root bus, and will base their PCI address allocation / PCI topology decisions on that fact: they will, for example, use legacy PCI bridges. Um.. yeah.. trouble is libvirt's PCI-E address allocation probably won't work for spapr PCI-E either, because of the weird PCI-E without root complex presentation we get in PAPR. So, would the PCIe Root Bus in a pseries guest behave differently than the one in a q35 or mach-virt guest? Yes. I had a long discussion with BenH and got a somewhat better idea about this.
If only a single host PE (== iommu group) is passed through and there are no emulated devices, the difference isn't too bad: basically on pseries you'll see the subtree that would be below the root complex on q35.
But if you pass through multiple groups, things get weird. On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt.
Does it have a different number of slots, do we have to plug different controllers into them, ...?
Regardless of how we decide to move forward with the PCIe-enabled pseries machine type, libvirt will have to know about this so it can behave appropriately. So there are kind of two extremes of how to address this. There are a variety of options in between, but I suspect they're going to be even more muddled and hideous than the extremes.
1) Give up. You said there's already a flag that says a PCI-E bus is able to accept vanilla-PCI devices. We add a hack flag that says a vanilla-PCI bus is able to accept PCI-E devices. We keep address allocation as it is now - the pseries topology really does resemble vanilla-PCI much better than it does PCI-E. But, we allow PCI-E devices, and PAPR has mechanisms for accessing the extended config space. PCI-E standard hotplug and error reporting will never work, but PAPR provides its own mechanisms for those, so that should be ok.
2) Start exposing the PCI-E heirarchy for pseries guests much more like q35, root complexes and all. It's not clear that PAPR actually *forbids* exposing the root complex, it just doesn't require it and that's not what PowerVM does. But.. there are big questions about whether existing guests will cope with this or not. When you start adding in multiple passed through devices and particularly virtual functions as well, things could get very ugly - we might need to construct multiple emulated virtual root complexes or other messes.
In the short to medium term, I'm thinking option (1) seems pretty compelling.
I believe after we introduced the very first pseries-pcie-X.Y, we will just stop adding new pseries-X.Y. Isn't i440fx still being updated despite the fact that q35 exists? Granted, there are a lot more differences between those two machine types than just the root bus type. Right, there are heaps of differences between i440fx and q35, and reasons to keep both updated. For pseries we have neither the impetus nor the resources to maintain two different machine type variant, where the only difference is between legacy PCI and weirdly presented PCI-E. Calling the PCIe machine type either pseries-2.8 or pseries-pcie-2.8 would result in the very same amount of work, and in both cases it would be understood that the legacy PCI machine type is no longer going to be updated, but can still be used to run existing guests. So, I'm not sure if the idea of a new machine type has legs or not, but let's think it through a bit further. Suppose we have a new machine type, let's call it 'papr'. I'm thinking it would be (at least with -nodefaults) basically a super-minimal version of pseries: so each PHB would have to be explicitly created, the VIO bridge would have to be explicitly created, likewise the NVRAM. Not sure about the "devices" which really represent firmware features - the RTC, RNG, hypervisor event source and so forth.
Might have some advantages. Then again, it doesn't really solve the specific problem here. It means libvirt (or the user) has to explicitly choose a PCI or PCI-E PHB to put things on, but libvirt's PCI-E address allocation will still be wrong in all probability.
That's a broad statement. Why? If qemu reports the default devices and characteristics of the devices properly (and libvirt uses that information) there's no reason for it to make the wrong decision.
Actually, unlike the comments below, this assertion is neither particularly broad, nor critical of libvirt (it is obliquely critical of PAPR and PowerVM). The point is that the way PCI-E devices are presented under PAPR is different from the way they'd appear on a hardware bus or on virtualized PC or ARM. Specifically we don't present the pseudo-P2P bridges of the root complexes to the guest. Instead (unless you explicitly ask for a bridge) all the devices will appear directly on the root "bus" - obviously this bus is a fiction since PCI-E is point to point in hardware. In short, from the point of view of topology, PAPR PCI-E will look much more like vanilla PCI than PCI-E on an x86. So to the extent that libvirt address allocation differs between PCI and PCI-E, and I gather it does, the PCI-E variant will be wrong for PAPR. That's why for the medium term I advocate option (1) above of treating PAPR PCI-E as if it was a vanilla PCI bus, but with a flag indicating that PCI-E devices can be attached to it.
Guh.
As an aside, here's a RANT.
libvirt address allocation. Seriously, W. T. F!
libvirt insists on owning address allocation. That's so it can recreate the exact same machine at the far end of a migration. So far so good, except it insists on recording that information in the domain XML in kinda-sorta-but-not-really back end independent form.
Explain "kinda-sorta-but-not-really". If there's a deficiency in the model maybe it can be fixed.
My point is that the variety of constraints we can have in the back end is so broad that it's impossible to construct a representation to that level of detail which is truly backend independent.
But the thing is libvirt fundamentally CAN NOT get this right.
True, but not for the reasons you think. If qemu is able to respond to queries with adequate details about the devices available for a machinetype (and what buses are in place by default), there's no reason that libvirt can't add devices addressed such that all the connections are legal;
The problem here is "adequate details". Again, AFAICT the variety of possible constraints given all the vagaries of different platforms is so great that I don't think any format short of code is sufficient to describe it.
what libvirt *can't* get right is the policy requested in the next higher layer of management (and ultimately of the user) - does this device need to be hotpluggable? Does the user want to keep all devices on the root complex to avoid extra PCI controllers?
And qemu fundamentally CAN NOT get it right either. qemu knows what is possible and what is allowed, but it doesn't know what the user *wants* (beyond "they want device X", which is only 1/2 the story), and has no way of being told what the user wants other than with a PCI address.
So, here's a more specific point. You *don't* give PCI addresses to qemu - rather you give device/slot numbers, and which bus/bridge they're attached to. libvirt's use of DDDD:BB:SS.F addresses is a fundamental error, because such addresses aren't meaningful until *after* the bus is enumerated. The bus number is explicitly allocated during enumeration (by guest firmware or OS). You can kind of predict what they'll do, but it requires knowing about every bridge and device on the system, which means it risks breakage every time the backend changes its default setup - and the interpretation of the addresses could be changed by an insertion of a bridge somewhere apparently unrelated in the XML. The domain number is an entirely guest-local concept, except - sorta - on x86 where there's an obvious choice of the index that's used to control the multiplexed access to all the host bridges' config spaces. On platforms which have separate registers for each host bridge, it's entirely possible for libvirt's idea of domain numbers to differ from the guest's which is wonderfully confusing.
To back up for a minute, some background info: once a device has been added to a domain, at *any* time in the future (not just during a migration, but forever more until the end of time) that device must always have the same PCI address as it had that first time. In order to guarantee that, libvirt needs to either:
I dispute the "any time". I'll grant you want to keep the virtual hardware stable for an extended period. But if you ever want to update your backend to newer behaviour (meaning new machine type version, not just new qemu version) then you should revisit address allocation. And, yes, that potentially breaks the guest, but so could the backend changes anyway, just like any hardware update for a physical machine. In most case - assuming it's been set up sensibly - the guest OS will cope with minor hw reshuffles. That is, after all, why RHEL locates filesystems by UUID and NICs by MAC in normal operation.
a) keep track of the order the devices were added and always put the devices in the same order on the commandline (assuming that qemu guarantees that it actually assigns addresses based on the order of the devices' appearance on the commandline, which has never been stated anywhere as an API guarantee of qemu),
or
b) remember the address of each device as it is added and specify that on the commandline in the future. libvirt chooses (b). And where is the logical place to store that address? In the config.
Putting in the config seems like the logical place at first, but it's a mistake: First, the information we need to ensure stable guest hardware is inherently backend specific. On qemu we need to say which bridges and bus ids each device is connected to. Some other system could use PCI addresses (I think that would be a a design flaw in the backend, but if libvirt can't deal with imperfect backends it has failed its remit), for containers the whole question doesn't even make sense. A really simplistic hypervisor could take simply an (ordered) list of plugin devices and always put them into a fixed set of slots. Putting this into the general config forces us to attempt a solution to the insoluble: greating a neutral representation translatable to any backend specific representation. Second, this means that the config now contains both information that was originally entered by the user (or next level up) as what they want for the VM, and also information that libvirt decided once upon a time and is now sticking with. We can no longer separate which bits of information are which. That means when we want to update the backend, we can't tell what we should regenerate. [Meta rant, when it comes to design flaws that breed other design flaws the fact that the domain XML is both a read and write interface is just about the granddaddy of them all. It has two awful consequences: 1) After the VM is first run, we can no longer tell which config constraints were ones actually requested by the user, and which were added by some version of libvirt in the past and are now just there so we make the same decisions as before. That distinction matters. 2) Because the user could be relying on *seeing* something in the XML, not just that whatever it put in there will still be understood by new libvirt versions means we basically can't ever fix errors in the XML design.]
So we've established that the PCI address of a device needs to be stored in the config. So why does libvirt need to choose it the first time?
1) Because qemu doesn't have (and CAN NOT have) all the information about what are the user's plans for that device:
CAN not? Why not? Fundamentally it seems like we need to get the information about what the user wants and what the virtual hardware can do into the same place. At least from the examples you've mentioned so far, pushing "needs to be hotpluggable" and so forth flags down into qemu seems a lot more feasible that pulling up the complex and platform dependent set of constraints into libvirt. And yes, qemu is very far from perfect in this area - the introspection interfaces need to be greatly improved.
a) It has no idea if the user wants that device to be hotpluggable (on a root-port) or not (on root complex as an integrated device)
b) it doesn't know if the user wants the device to be placed on an expander bus so that its NUMA status can be discovered by the guest OS.
So both (a) and (b) are precisely examples of what I'm describing. libvirt is assuming (a) requires hotplug => must be on a root port (b) device bound to NUMA node => use expander bus But (a) is only true for guest platforms using the standard hotplug mechanism. pseries does not. (b) is essentially only true for x86 - the whole concept of the expander bus is a hack to workaround x86's historical inability to have truly independent PCI host bridge. On pseries we can create independent host bridges and assign them each a NUMA node (including the default one). qemu already knew that, libvirt needs to be taught. Just as it seems like libvirt needs to be taught about practically every platform quirk.
If there is a choice, there must be a way to make that choice. The way that qemu provides to make the choice is by specifying an address. So libvirt must specify an address in its config.
Again, no, qemu does not take addresses at all. At least not in the "DDDD:BB:SS.F" sense. Even if it did, DDDD:BB:SS.F addresses have no place in a supposedly back-end independent config.
2) Because qemu is unable/unwilling to automatically add PCIe root ports when necessary, it's *not even possible* (on PCIe machinetypes) for it to place a device on a hotpluggable port without libvirt specifying a PCI address the very first time the device is added (and also adding in a root-port), but libvirt's default policy is that (almost) all devices should be hotpluggable. If we were to follow your recommendation ("libvirt never specifies PCI addresses, but instead allows qemu to assign them"), hotplug on PCIe-based machinetypes would not be possible, though.
Ok, so, yes, I see that a lot of this allocation mess originates because qemu didn't provide sufficient auto-allocation, or introspection, so libvirt had to work around that. The trouble is that those workarounds have now been baked into the libvirt interface - primarily by the fact that the data is integrated back into the primary config rather than stored as adjunct back end specific data. That means it's now insanely difficult to improve the interfaces to qemu, because libvirt XMLs now have all this address information. A few of them are because the user really did request that for some reason, but many are because that was how libvirt tried to implement some more abstract user request. We can now no longer tell the difference, which means we can't translate those abstract requests into a new saner qemu interface.
There have even been mentions that even libvirt is too *low* in the stack to be specifying the PCI address of devices (i.e. that all PCI address decisions should be up to higher level management applications)
Frankly, if the higher level management needs that much control over the PCI topology, going via an abstraction layer like libvirt seems not the right choice. In that case the management needs to understand the guest platform in great detail anyway, so what does libvirt buy you.
- I had posted a patch that would allow specifying "hotpluggable='yes|no'" in the XML rather than forcing the call to specify an address, and this was NACKed because it was seen as libvirt dictating policy.
Huh. Your patch makes perfect sense to me - a step in the right direction - and the reason for NACK seems completely bogus.
(In the end, libvirt *does* dictate a default policy, (it's just that the only way to modify that policy is by manually specifying addresses) - libvirt's default PCI address policy is that (almost) all devices will be assigned an address that makes them hotpluggable, and will not be placed on a non-0 NUMA node.
Yeah, but "address which makes them hotpluggable" depends on guest platform (machine type). As does how you assign them to NUMA nodes.
So, in spite of libvirt's effort, in the end we *still* need to expose address configuration to higher level management applications, since they may want to force all devices onto the root complex (e.g. libguestfs, which does it to reduce PCI controller count, and thus startup time) or force certain devices to be on a non-0 NUMA node (e.g. OpenStack when it wants to place a VFIO assigned device on the same NUMA node in the guest as it is in the host).
So, I'm not necessarily against allowing an explicit address override for special cases (although DDDD:BB:SS.F is the wrong way to encode it). If that breaks due to a backend update, it's reasonable to just punt the error back to the user and get them to fix it - if they didn't understand the guest platform well enough to fix this, they shouldn't have been overriding addresses. But it's not reasonable to punt back to the user when the only thing that got broken by the backend update was information libvirt (maybe not even the current version) decided some time in the past, based on faulty assumptions about the guest platform. But libvirt has no way to distinguish these cases because all the info is folded back into the one XML blob.
With all of that, I fail to see how it would be at all viable to simply leave PCI address assignment up to qemu.
There are all sorts of possible machine specific address allocation constraints that can exist - from simply which devices are already created by default (for varying values of "default") to complicated constraints depending on details of board wiring. The back end has to know about these - it implements them.
And since qemu knows about them, it should be able to report them. Which is what Eduardo's work is doing. And then libvirt will know about all the constraints in an programmatic manner (rather than the horrible (tedious, error prone) hardcoding of all those details that we've had to suffer with until now).
I haven't had a chance to look in detail at Eduardo's latest stuff, but yes it should be an improvement. But I think you vastly underestimate how many different platform quirks we could have which won't be easy to encode.
The ONLY way libvirt can get this (temporarily) right is by duplicating a huge chunk of the back end's allocation logic, which will inevitably get out of date causing problems just like this.
Basically the back end will *always* have better information about how to place devices than libvirt can.
And since no matter how hard qemu might try to come up with a policy for address assignment that will satisfy the needs of 100% of the users 100% of the time, it will fail (because different users have different needs). Because qemu will be unable to properly place all devices all the time, libvirt (and higher level management) will still need to do it. Even in the basic case qemu doesn't provide what libvirt requires as default - that devices be hotpluggable.
So, libvirt should be allowing the back end to do the allocation, then snapshotting that in a back end specific format which can be used for creating migration destinations. But that breaks libvirt's the-domain-XML-is-everything model.
No, that doesn't work because qemu would in many situations place the devices at the wrong address / on the wrong controller, because there are many possible topologies that are legal, and the user may (for perfectly valid reasons) want something different from what qemu would have chosen.
Right, there are certainly reasons for overrides. But at the moment *every* device is overridden, which means when there is something wrong in libvirt's assumptions about how to place things, every device can break, not just overridden ones.
(An example of two differing (and valid) policies - libguestfs needs guests to startup as quickly as possible, and thus wants as few PCI controllers as possible (this makes a noticeable difference in Linux boot time), so it wants all devices to be integrated on the root complex. On the other hand, a generic guest in libvirt should make all devices hotpluggable just in case the user wants to unplug them, so by default it tries to place all devices on a pcie-root-port. You can't support both of these if addressing is all left up to qemu)
With the tools that qemu currently gives you. But now the assumption that address allocation belongs to libvirt means that it's insanely difficult to build those tools without breaking libvirt.
In this regard libvirt doesn't just have a design flaw, it has design flaws which breed more design flaws like pestilent cancer.
It may make you feel good to say that, but the facts don't back it up. Any project makes design mistakes, but in the specific case you're discussing here, I think you haven't looked from a wide enough viewpoint to see the necessity of what libvirt is doing and why it can't be done by qemu (certainly not all the time anyway).
And what's worse the consequences of those design flaws are now making sane design decisions increasingly difficult in adjacent projects like qemu.
libvirt has always done the best that could be done with the information provided by qemu. The problem isn't that libvirt is creating new problems for qemu out of thin air, it's that qemu is unable to automatically address PCI devices for all possible situations and user policy preferences, so higher levels need to make the decisions about addressing to satisfy their policies (ie what they *want*, eg hotpluggable, integrated on root complex), and qemu hasn't (until Eduardo's patches) been able to provide adequate information about what is *legal* (e.g which devices can be plugged into which model of pci controller, what slots are available on each type of controller, whether those slots are hotpluggable) in a programmatic way, so libvirt has had to hardcode rules about bus-device compatibility and capabilities, slot ranges, etc in order to make proper decisions itself when possible, and to sanity-check decisions about addresses made by higher level management when not. I don't think that's a design flaw. I think that's making the best of a "less than ideal" situation.
I'd feel better about this if there seemed to be some recognition of it, and some necessarily long term plan to improve it, but if there is I haven't heard of it. Or at least the closest thing seems to be coming from the qemu side (witness Eduardo's talk at the last KVM forum, and mine at the one before).
Eduardo's work isn't being done to make up for some mythical design flaw in libvirt. It is being done in order to give libvirt the (previously unavailable) information it needs to do a necessary job, and is being done at least partly at the request of libvirt (we've certainly been demanding some of that stuff for a long time!)
Well, I did warn you it was a rant. That said, while there are understandable historical reasons for them design flaws in libvirt which make it really difficult to design other things properly are alas not a myth. Three examples: 1) Read/write XML interface. Makes it nigh-impossible to fix flaws by updating to a more coherent internal representation (with backwards compat glue to parse the old format into the new one). 2) Not separating original user config from libvirt additions / annotations. Discussed repeatedly above. 3) Use of DDDD:BB.SS.F addresses (rather than a reference to bus attachment point + SS.F). You can't use addresses to construct a machine in an address space which only exists once the machine is constructed. And that's just the first three I came up with that have at least a slight connection to the problem at hand.
The summary is that it's impossible for qemu to correctly decide where to put new devices, especially in a PCIe hierarchy for a few reasons (at least); because of this, libvirt (and higher level management) needs to be able to assign addresses to devices, and in order for us/them to be able to do that properly, qemu needs to provide detailed and accurate information about what buses/controllers/devices are in each machinetype, what controllers/devices are available to add, and what are the legal ways of connecting those devices and controllers.
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Fri, 2016-12-02 at 15:18 +1100, David Gibson wrote: > > So, would the PCIe Root Bus in a pseries guest behave > > differently than the one in a q35 or mach-virt guest? > > Yes. I had a long discussion with BenH and got a somewhat better idea > about this. Sorry, but I'm afraid you're going to have to break this down even further for me :( > If only a single host PE (== iommu group) is passed through and there > are no emulated devices, the difference isn't too bad: basically on > pseries you'll see the subtree that would be below the root complex on > q35. > > But if you pass through multiple groups, things get weird. Is the difference between q35 and pseries guests with respect to PCIe only relevant when it comes to assigned devices, or in general? I'm asking this because you seem to focus entirely on assigned devices. > On q35, > you'd generally expect physically separate (different slot) devices to > appear under separate root complexes. This part I don't get at all, so please bear with me. The way I read it you're claiming that eg. a SCSI controller and a network adapter, being physically separate and assigned to separate PCI slots, should have a dedicated PCIe Root Complex each on a q35 guest. That doesn't match with my experience, where you would simply assign them to separate slots of the default PCIe Root Bus (pcie.0), eg. 00:01.0 and 00:02.0. Maybe you're referring to the fact that you might want to create multiple PCIe Root Complexes in order to assign the host devices to separate guest NUMA nodes? How is creating multiple PCIe Root Complexes on q35 using pxb-pcie different than creating multiple PHBs using spapr-pci-host-bridge on pseries? > Whereas on pseries they'll > appear as siblings on a virtual bus (which makes no physical sense for > point-to-point PCI-E). What is the virtual bus in question? Why would it matter that they're siblings? I'm possibly missing the point entirely, but so far it looks to me like there are different configurations you might want to use depending on your goal, and both q35 and pseries give you comparable tools to achieve such configurations. > I suppose we could try treating all devices on pseries as though they > were chipset builtin devices on q35, which will appear on the root > PCI-E bus without root complex. But I suspect that's likely to cause > trouble with hotplug, and it will certainly need different address > allocation from libvirt. PCIe Integrated Endpoint Devices are not hotpluggable on q35, that's why libvirt will follow QEMU's PCIe topology recommendations and place a PCIe Root Port between them; I assume the same could be done for pseries guests as soon as QEMU grows support for generic PCIe Root Ports, something Marcel has already posted patches for. Again, sorry for clearly misunderstanding your explanation, but I'm still not seeing the issue here. I'm sure it's very clear in your mind, but I'm afraid you're going to have to walk me through it :( > > Regardless of how we decide to move forward with the > > PCIe-enabled pseries machine type, libvirt will have to > > know about this so it can behave appropriately. > > So there are kind of two extremes of how to address this. There are a > variety of options in between, but I suspect they're going to be even > more muddled and hideous than the extremes. > > 1) Give up. You said there's already a flag that says a PCI-E bus is > able to accept vanilla-PCI devices. We add a hack flag that says a > vanilla-PCI bus is able to accept PCI-E devices. We keep address > allocation as it is now - the pseries topology really does resemble > vanilla-PCI much better than it does PCI-E. But, we allow PCI-E > devices, and PAPR has mechanisms for accessing the extended config > space. PCI-E standard hotplug and error reporting will never work, > but PAPR provides its own mechanisms for those, so that should be ok. We can definitely special-case pseries guests and take the "anything goes" approach to PCI vs PCIe, but it would certainly be nicer if we could avoid presenting our users the head-scratching situation of PCIe devices being plugged into legacy PCI slots and still showing up as PCIe in the guest. What about virtio devices, which present themselves either as legacy PCI or PCIe depending on the kind of slot they are plugged into? Would they show up as PCIe or legacy PCI on a PCIe-enabled pseries guest? > 2) Start exposing the PCI-E heirarchy for pseries guests much more > like q35, root complexes and all. It's not clear that PAPR actually > *forbids* exposing the root complex, it just doesn't require it and > that's not what PowerVM does. But.. there are big questions about > whether existing guests will cope with this or not. When you start > adding in multiple passed through devices and particularly virtual > functions as well, things could get very ugly - we might need to > construct multiple emulated virtual root complexes or other messes. > > In the short to medium term, I'm thinking option (1) seems pretty > compelling. Is the Root Complex not currently exposed? The Root Bus certainly is, otherwise PCI devices won't work at all, I assume. And I can clearly see a pci.0 bus in the output of 'info qtree' for a pseries guest, and a pci.1 too if I add a spapr-pci-host-bridge. Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead? > So, I'm not sure if the idea of a new machine type has legs or not, > but let's think it through a bit further. Suppose we have a new > machine type, let's call it 'papr'. I'm thinking it would be (at > least with -nodefaults) basically a super-minimal version of pseries: > so each PHB would have to be explicitly created, the VIO bridge would > have to be explicitly created, likewise the NVRAM. Not sure about the > "devices" which really represent firmware features - the RTC, RNG, > hypervisor event source and so forth. > > Might have some advantages. Then again, it doesn't really solve the > specific problem here. It means libvirt (or the user) has to > explicitly choose a PCI or PCI-E PHB to put things on, libvirt would probably add a <controller type='pci' model='pcie-root'/> to the guest XML by default, resulting in a spapr-pcie-host-bridge providing pcie.0 and the same controller / address allocation logic as q35; the user would be able to use <controller type='pci' model='pci-root'/> instead to stick with legacy PCI. This would only matter when using '-nodefaults' anyway, when that flag is not present a PCIe (or legacy PCI) could be created by QEMU to make it more convenient for people that are not using libvirt. Maybe we should have a different model, specific to pseries guests, instead, so that all PHBs would look the same in the guest XML, something like <controller type='pci' model='phb-pcie'/> It would require shuffling libvirt's PCI address allocation code around quite a bit, but it should be doable. And if it makes life easier for our users, then it's worth it. > but libvirt's > PCI-E address allocation will still be wrong in all probability. > > Guh. > As an aside, here's a RANT. [...] Laine already addressed your points extensively, but I'd like to add a few thoughts of my own. * PCI addresses for libvirt guests don't need to be stable only when performing migration, but also to guarantee that no change in guest ABI will happen as a consequence of eg. a simple power cycle. * Even if libvirt left all PCI address assignment to QEMU, we would need a way for users to override QEMU's choices, because one size never fits all and users have all kinds of crazy, yet valid, requirements. So the first time we run QEMU, we would have to take the backend-specific format you suggest, parse it to extract the PCI addresses that have been assigned, and reflect them in the guest XML so that the user can change a bunch of them. Then I guess we could re-encode it in the backend-specific format and pass it to QEMU the next time we run it but, at that point, what's the difference with simply putting the PCI addresses on the command line directly? * It's not just about the addresses, by the way, but also about the controllers - what model is used, how they are plugged together and so on. More stuff that would have to round-trip because users need to be able to take matters into their own hands. * Design mistakes in any software, combined with strict backwards compatibility requirements, make it difficult to make changes in both related components and the software itself, even when the changes would be very beneficial. It can be very frustrating at times, but it's the reality of things and unfortunately there's only so much we can do about it. * Eduardo's work, which you mentioned, is going to be very beneficial in the long run; in the short run, Marcel's PCIe device placement guidelines, a document that has seen contributions from QEMU, OVMF and libvirt developers, have been invaluable to improve libvirt's PCI address allocation logic. So we're already doing better, and more improvements are on the way :) -- Andrea Bolognani / Red Hat / Virtualization
On Tue, Dec 06, 2016 at 06:30:47PM +0100, Andrea Bolognani wrote: > On Fri, 2016-12-02 at 15:18 +1100, David Gibson wrote: > > > So, would the PCIe Root Bus in a pseries guest behave > > > differently than the one in a q35 or mach-virt guest? > > > > Yes. I had a long discussion with BenH and got a somewhat better idea > > about this. > > Sorry, but I'm afraid you're going to have to break this > down even further for me :( > > > If only a single host PE (== iommu group) is passed through and there > > are no emulated devices, the difference isn't too bad: basically on > > pseries you'll see the subtree that would be below the root complex on > > q35. > > > > But if you pass through multiple groups, things get weird. > > Is the difference between q35 and pseries guests with > respect to PCIe only relevant when it comes to assigned > devices, or in general? I'm asking this because you seem to > focus entirely on assigned devices. Well, in a sense that's up to us. The only existing model we have is PowerVM, and PowerVM only does device passthrough, no emulated devices. PAPR doesn't really distinguish one way or the other, but it's written from the perspective of assuming that all PCI devices correspond to physical devices on the host > > On q35, > > you'd generally expect physically separate (different slot) devices to > > appear under separate root complexes. > > This part I don't get at all, so please bear with me. > > The way I read it you're claiming that eg. a SCSI controller > and a network adapter, being physically separate and assigned > to separate PCI slots, should have a dedicated PCIe Root > Complex each on a q35 guest. Right, my understanding was that if the devices were slotted, rather than integrated, each one would sit under a separate root complex, the root complex being a pseudo PCI to PCI bridge. > That doesn't match with my experience, where you would simply > assign them to separate slots of the default PCIe Root Bus > (pcie.0), eg. 00:01.0 and 00:02.0. The qemu default, or the libvirt default? I think this represents treating the devices as though they were integrated devices in the host bridge. I believe on q35 they would not be hotpluggable - but on pseries they would be (because we don't use the standard hot plug controller). > Maybe you're referring to the fact that you might want to > create multiple PCIe Root Complexes in order to assign the > host devices to separate guest NUMA nodes? How is creating > multiple PCIe Root Complexes on q35 using pxb-pcie different > than creating multiple PHBs using spapr-pci-host-bridge on > pseries? Uh.. AIUI the root complex is the PCI to PCI bridge under which PCI-E slots appear. PXB is something different - essentially different host bridges as you say (though with some weird hacks to access config space, which make it dependent on the primary bus in a way which spapr PHBs are not). I'll admit I'm pretty confused myself about the exact distinction between root complex, root port and upstream and downstream ports. > > Whereas on pseries they'll > > appear as siblings on a virtual bus (which makes no physical sense for > > point-to-point PCI-E). > > What is the virtual bus in question? Why would it matter > that they're siblings? On pseries it won't. But my understanding is that libvirt won't create them that way on q35 - instead it will insert the RCs / P2P bridges to allow them to be hotplugged. Inserting that bridge may confuse pseries guests which aren't expecting it. > I'm possibly missing the point entirely, but so far it > looks to me like there are different configurations you > might want to use depending on your goal, and both q35 > and pseries give you comparable tools to achieve such > configurations. > > > I suppose we could try treating all devices on pseries as though they > > were chipset builtin devices on q35, which will appear on the root > > PCI-E bus without root complex. But I suspect that's likely to cause > > trouble with hotplug, and it will certainly need different address > > allocation from libvirt. > > PCIe Integrated Endpoint Devices are not hotpluggable on > q35, that's why libvirt will follow QEMU's PCIe topology > recommendations and place a PCIe Root Port between them; > I assume the same could be done for pseries guests as > soon as QEMU grows support for generic PCIe Root Ports, > something Marcel has already posted patches for. Here you've hit on it. No, we should not do that for pseries, AFAICT. PAPR doesn't really have the concept of integrated endpoint devices, and all devices can be hotplugged via the PAPR mechanisms (and none can via the PCI-E standard hotplug mechanism). > Again, sorry for clearly misunderstanding your explanation, > but I'm still not seeing the issue here. I'm sure it's very > clear in your mind, but I'm afraid you're going to have to > walk me through it :( I wish it were entirely clear in my mind. Like I say I'm still pretty confused by exactly the root complex entails. > > > Regardless of how we decide to move forward with the > > > PCIe-enabled pseries machine type, libvirt will have to > > > know about this so it can behave appropriately. > > > > So there are kind of two extremes of how to address this. There are a > > variety of options in between, but I suspect they're going to be even > > more muddled and hideous than the extremes. > > > > 1) Give up. You said there's already a flag that says a PCI-E bus is > > able to accept vanilla-PCI devices. We add a hack flag that says a > > vanilla-PCI bus is able to accept PCI-E devices. We keep address > > allocation as it is now - the pseries topology really does resemble > > vanilla-PCI much better than it does PCI-E. But, we allow PCI-E > > devices, and PAPR has mechanisms for accessing the extended config > > space. PCI-E standard hotplug and error reporting will never work, > > but PAPR provides its own mechanisms for those, so that should be ok. > > We can definitely special-case pseries guests and take > the "anything goes" approach to PCI vs PCIe, but it would > certainly be nicer if we could avoid presenting our users > the head-scratching situation of PCIe devices being plugged > into legacy PCI slots and still showing up as PCIe in the > guest. > > What about virtio devices, which present themselves either > as legacy PCI or PCIe depending on the kind of slot they > are plugged into? Would they show up as PCIe or legacy PCI > on a PCIe-enabled pseries guest? That we'd have to address on the qemu side with some > > 2) Start exposing the PCI-E heirarchy for pseries guests much more > > like q35, root complexes and all. It's not clear that PAPR actually > > *forbids* exposing the root complex, it just doesn't require it and > > that's not what PowerVM does. But.. there are big questions about > > whether existing guests will cope with this or not. When you start > > adding in multiple passed through devices and particularly virtual > > functions as well, things could get very ugly - we might need to > > construct multiple emulated virtual root complexes or other messes. > > > > In the short to medium term, I'm thinking option (1) seems pretty > > compelling. > > Is the Root Complex not currently exposed? The Root Bus > certainly is, Like I say, I'm fairly confused myself, but I'm pretty sure that Root Complex != Root Bus. The RC sits under the root bus IIRC.. or possibly it consists of the root bus plus something under it as well. Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries. > otherwise PCI devices won't work at all, I > assume. And I can clearly see a pci.0 bus in the output > of 'info qtree' for a pseries guest, and a pci.1 too if > I add a spapr-pci-host-bridge. > > Maybe I just don't quite get the relationship between Root > Complexes and Root Buses, but I guess my question is: what > is preventing us from simply doing whatever a > spapr-pci-host-bridge is doing in order to expose a legacy > PCI Root Bus (pci.*) to the guest, and create a new > spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) > instead? Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use. > > So, I'm not sure if the idea of a new machine type has legs or not, > > but let's think it through a bit further. Suppose we have a new > > machine type, let's call it 'papr'. I'm thinking it would be (at > > least with -nodefaults) basically a super-minimal version of pseries: > > so each PHB would have to be explicitly created, the VIO bridge would > > have to be explicitly created, likewise the NVRAM. Not sure about the > > "devices" which really represent firmware features - the RTC, RNG, > > hypervisor event source and so forth. > > > > Might have some advantages. Then again, it doesn't really solve the > > specific problem here. It means libvirt (or the user) has to > > explicitly choose a PCI or PCI-E PHB to put things on, > > libvirt would probably add a > > <controller type='pci' model='pcie-root'/> > > to the guest XML by default, resulting in a > spapr-pcie-host-bridge providing pcie.0 and the same > controller / address allocation logic as q35; the user > would be able to use > > <controller type='pci' model='pci-root'/> > > instead to stick with legacy PCI. This would only matter > when using '-nodefaults' anyway, when that flag is not > present a PCIe (or legacy PCI) could be created by QEMU > to make it more convenient for people that are not using > libvirt. > > Maybe we should have a different model, specific to > pseries guests, instead, so that all PHBs would look the > same in the guest XML, something like > > <controller type='pci' model='phb-pcie'/> > > It would require shuffling libvirt's PCI address allocation > code around quite a bit, but it should be doable. And if it > makes life easier for our users, then it's worth it. Hrm. So my first inclination would be to stick with the generic names, and map those to creating new pseries host bridges on pseries guests. I would have thought that would be the easier option for users. But I may not have realized all the implications yet. > > but libvirt's > > PCI-E address allocation will still be wrong in all probability. > > > > Guh. > > > As an aside, here's a RANT. > [...] > > Laine already addressed your points extensively, but I'd > like to add a few thoughts of my own. > > * PCI addresses for libvirt guests don't need to be stable > only when performing migration, but also to guarantee > that no change in guest ABI will happen as a consequence > of eg. a simple power cycle. > > * Even if libvirt left all PCI address assignment to QEMU, > we would need a way for users to override QEMU's choices, > because one size never fits all and users have all kinds > of crazy, yet valid, requirements. So the first time we > run QEMU, we would have to take the backend-specific > format you suggest, parse it to extract the PCI addresses > that have been assigned, and reflect them in the guest > XML so that the user can change a bunch of them. Then I > guess we could re-encode it in the backend-specific format > and pass it to QEMU the next time we run it but, at that > point, what's the difference with simply putting the PCI > addresses on the command line directly? > > * It's not just about the addresses, by the way, but also > about the controllers - what model is used, how they are > plugged together and so on. More stuff that would have to > round-trip because users need to be able to take matters > into their own hands. > > * Design mistakes in any software, combined with strict > backwards compatibility requirements, make it difficult > to make changes in both related components and the > software itself, even when the changes would be very > beneficial. It can be very frustrating at times, but > it's the reality of things and unfortunately there's only > so much we can do about it. I think the above I've touched on in my reply to Laine. > * Eduardo's work, which you mentioned, is going to be very > beneficial in the long run; in the short run, Marcel's > PCIe device placement guidelines, a document that has seen > contributions from QEMU, OVMF and libvirt developers, have > been invaluable to improve libvirt's PCI address allocation > logic. So we're already doing better, and more improvements > are on the way :) Right.. so here's the thing, I strongly suspect that Marcel's guidelines will not be correct for pseries. I'm not sure if they'll be definitively wrong, or just different enough from PowerVM that it might confuse guests, but either way. Can you send me a link to that document though, which might help me figure this out. -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
[Added Marcel to CC] On Wed, 2016-12-07 at 15:11 +1100, David Gibson wrote:
Is the difference between q35 and pseries guests with respect to PCIe only relevant when it comes to assigned devices, or in general? I'm asking this because you seem to focus entirely on assigned devices. Well, in a sense that's up to us. The only existing model we have is PowerVM, and PowerVM only does device passthrough, no emulated devices. PAPR doesn't really distinguish one way or the other, but it's written from the perspective of assuming that all PCI devices correspond to physical devices on the host
Okay, that makes sense.
On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes. This part I don't get at all, so please bear with me. The way I read it you're claiming that eg. a SCSI controller and a network adapter, being physically separate and assigned to separate PCI slots, should have a dedicated PCIe Root Complex each on a q35 guest. Right, my understanding was that if the devices were slotted, rather than integrated, each one would sit under a separate root complex, the root complex being a pseudo PCI to PCI bridge.
I assume "slotted" means "plugged into a slot that's not one of those provided by pcie.0" or something along those lines. More on the root complex bit later.
That doesn't match with my experience, where you would simply assign them to separate slots of the default PCIe Root Bus (pcie.0), eg. 00:01.0 and 00:02.0. The qemu default, or the libvirt default?
I'm talking about the libvirt default, which is supposed to follows Marcel's PCIe Guidelines.
I think this represents treating the devices as though they were integrated devices in the host bridge. I believe on q35 they would not be hotpluggable
Yeah, that's indeed not quite what libvirt would do by default: in reality, there would be a ioh3420 between the pcie.0 slots and each device exactly to enable hotplug.
but on pseries they would be (because we don't use the standard hot plug controller).
We can account for that in libvirt and avoid adding the extra ioh3420s (or rather the upcoming generic PCIe Root Ports) for pseries guests.
Maybe you're referring to the fact that you might want to create multiple PCIe Root Complexes in order to assign the host devices to separate guest NUMA nodes? How is creating multiple PCIe Root Complexes on q35 using pxb-pcie different than creating multiple PHBs using spapr-pci-host-bridge on pseries? Uh.. AIUI the root complex is the PCI to PCI bridge under which PCI-E slots appear. PXB is something different - essentially different host bridges as you say (though with some weird hacks to access config space, which make it dependent on the primary bus in a way which spapr PHBs are not). I'll admit I'm pretty confused myself about the exact distinction between root complex, root port and upstream and downstream ports.
I think we both need to get our terminology straight :) I'm sure Marcel will be happy to point us in the right direction. My understanding is that the PCIe Root Complex is the piece of hardware that exposes a PCIe Root Bus (pcie.0 in QEMU); PXBs can be connected to slots in pcie.0 to create more buses that behave, for the most part, like pcie.0 and are mostly useful to bind devices to specific NUMA nodes. Same applies to legacy PCI with the pxb (instead of pxb-pcie) device. In a similar fashion, PHBs are the hardware thingies that expose a PCI Root Bus (pci.0 and so on), the main difference being that they are truly independent: so a q35 guest will always have a "primary" PCIe Root Bus and (optionally) a bunch of "secondary" ones, but the same will not be the case for pseries guests. I don't think the difference is that important though, at least from libvirt's point of view: whether you're creating a pseries guest with two PHBs, or a q35 guest with its built-in PCIe Root Complex and an extra PCIe Expander Bus, you will end up with two "top level" buses that you can plug more devices into. If we had spapr-pcie-host-bridge, we could treat them mostly the same - with caveats such as the one described above, of course.
Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E). What is the virtual bus in question? Why would it matter that they're siblings? On pseries it won't. But my understanding is that libvirt won't create them that way on q35 - instead it will insert the RCs / P2P bridges to allow them to be hotplugged. Inserting that bridge may confuse pseries guests which aren't expecting it.
libvirt will automatically add PCIe Root Ports to make the devices hotpluggable on q35 guests, yes. But, as mentioned above, we can teach it not to.
I'm possibly missing the point entirely, but so far it looks to me like there are different configurations you might want to use depending on your goal, and both q35 and pseries give you comparable tools to achieve such configurations.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex. But I suspect that's likely to cause trouble with hotplug, and it will certainly need different address allocation from libvirt. PCIe Integrated Endpoint Devices are not hotpluggable on q35, that's why libvirt will follow QEMU's PCIe topology recommendations and place a PCIe Root Port between them; I assume the same could be done for pseries guests as soon as QEMU grows support for generic PCIe Root Ports, something Marcel has already posted patches for. Here you've hit on it. No, we should not do that for pseries, AFAICT. PAPR doesn't really have the concept of integrated endpoint devices, and all devices can be hotplugged via the PAPR mechanisms (and none can via the PCI-E standard hotplug mechanism).
Cool, I get it now.
Again, sorry for clearly misunderstanding your explanation, but I'm still not seeing the issue here. I'm sure it's very clear in your mind, but I'm afraid you're going to have to walk me through it :( I wish it were entirely clear in my mind. Like I say I'm still pretty confused by exactly the root complex entails.
Same here, but this back-and-forth is helping! :) [...]
What about virtio devices, which present themselves either as legacy PCI or PCIe depending on the kind of slot they are plugged into? Would they show up as PCIe or legacy PCI on a PCIe-enabled pseries guest? That we'd have to address on the qemu side with some
Unfinished sentence? [...]
Is the Root Complex not currently exposed? The Root Bus certainly is, Like I say, I'm fairly confused myself, but I'm pretty sure that Root Complex != Root Bus. The RC sits under the root bus IIRC.. or possibly it consists of the root bus plus something under it as well. Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries.
Basically the special casing I was mentioning earlier. [...]
Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead? Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
That issue is relevant whether or nor we have different PHB flavors, isn't it? As soon as multiple PHBs are present in a pseries guest, multiple PCI domains will be there as well, and we need to handle that somehow. On q35, on the other hand, I haven't been able to find a way to create extra PCI domains: adding a pxb-pcie certainly didn't work the same as adding an extra spapr-pci-host-bridge in that regard. [...]
Maybe we should have a different model, specific to pseries guests, instead, so that all PHBs would look the same in the guest XML, something like <controller type='pci' model='phb-pcie'/> It would require shuffling libvirt's PCI address allocation code around quite a bit, but it should be doable. And if it makes life easier for our users, then it's worth it. Hrm. So my first inclination would be to stick with the generic names, and map those to creating new pseries host bridges on pseries guests. I would have thought that would be the easier option for users. But I may not have realized all the implications yet.
You're probably right, but I can't immediately see how we would make the user aware of which PHB is which. Maybe we could add some sub-element or extra attribute... Anyway, we should not focus too much on this specific bit at the moment, deciding on a specific XML is mostly bikeshedding :) [...]
* Eduardo's work, which you mentioned, is going to be very beneficial in the long run; in the short run, Marcel's PCIe device placement guidelines, a document that has seen contributions from QEMU, OVMF and libvirt developers, have been invaluable to improve libvirt's PCI address allocation logic. So we're already doing better, and more improvements are on the way :) Right.. so here's the thing, I strongly suspect that Marcel's guidelines will not be correct for pseries. I'm not sure if they'll be definitively wrong, or just different enough from PowerVM that it might confuse guests, but either way.
Those guidelines have been developed with q35/mach-virt in mind[1], so I wouldn't at all be surprised if they didn't apply to pseries guests. And in fact, we just found out that they don't! My point is that we could easily create a similar document for pseries guests, and then libvirt will be able to pick up whatever recommendations we come up with just like it did for q35/mach-virt.
Can you send me a link to that document though, which might help me figure this out.
It's docs/pcie.txt in QEMU's git repository. [1] Even though I now realize that this is not immediately clear by looking at the document itself -- Andrea Bolognani / Red Hat / Virtualization
On 12/07/2016 06:42 PM, Andrea Bolognani wrote:
[Added Marcel to CC]
Hi, Sorry for the late reply.
On Wed, 2016-12-07 at 15:11 +1100, David Gibson wrote:
Is the difference between q35 and pseries guests with respect to PCIe only relevant when it comes to assigned devices, or in general? I'm asking this because you seem to focus entirely on assigned devices.
Well, in a sense that's up to us. The only existing model we have is PowerVM, and PowerVM only does device passthrough, no emulated devices. PAPR doesn't really distinguish one way or the other, but it's written from the perspective of assuming that all PCI devices correspond to physical devices on the host
Okay, that makes sense.
On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes.
This part I don't get at all, so please bear with me.
The way I read it you're claiming that eg. a SCSI controller and a network adapter, being physically separate and assigned to separate PCI slots, should have a dedicated PCIe Root Complex each on a q35 guest.
Not a PCIe Root Complex, but a PCIe Root port.
Right, my understanding was that if the devices were slotted, rather than integrated, each one would sit under a separate root complex, the root complex being a pseudo PCI to PCI bridge.
I assume "slotted" means "plugged into a slot that's not one of those provided by pcie.0" or something along those lines.
More on the root complex bit later.
That doesn't match with my experience, where you would simply assign them to separate slots of the default PCIe Root Bus (pcie.0), eg. 00:01.0 and 00:02.0.
The qemu default, or the libvirt default?
I'm talking about the libvirt default, which is supposed to follows Marcel's PCIe Guidelines.
I think this represents treating the devices as though they were integrated devices in the host bridge. I believe on q35 they would not be hotpluggable
Correct. Please have a look to the new document regarding pcie: docs/pcie.txt and the corresponding presentations.
Yeah, that's indeed not quite what libvirt would do by default: in reality, there would be a ioh3420 between the pcie.0 slots and each device exactly to enable hotplug.
but on pseries they would be (because we don't use the standard hot plug controller).
We can account for that in libvirt and avoid adding the extra ioh3420s (or rather the upcoming generic PCIe Root Ports) for pseries guests.
Maybe you're referring to the fact that you might want to create multiple PCIe Root Complexes in order to assign the host devices to separate guest NUMA nodes? How is creating multiple PCIe Root Complexes on q35 using pxb-pcie different than creating multiple PHBs using spapr-pci-host-bridge on pseries?
Uh.. AIUI the root complex is the PCI to PCI bridge under which PCI-E slots appear. PXB is something different - essentially different host bridges as you say (though with some weird hacks to access config space, which make it dependent on the primary bus in a way which spapr PHBs are not).
I'll admit I'm pretty confused myself about the exact distinction between root complex, root port and upstream and downstream ports.
I think we both need to get our terminology straight :) I'm sure Marcel will be happy to point us in the right direction.
My understanding is that the PCIe Root Complex is the piece of hardware that exposes a PCIe Root Bus (pcie.0 in QEMU);
right
PXBs can be connected to slots in pcie.0 to create more buses that behave, for the most part, like pcie.0 and are mostly useful to bind devices to specific NUMA nodes.
right Same applies
to legacy PCI with the pxb (instead of pxb-pcie) device.
pxb should not be used for PCIe machines, only for legacy PCI ones.
In a similar fashion, PHBs are the hardware thingies that expose a PCI Root Bus (pci.0 and so on), the main difference being that they are truly independent: so a q35 guest will always have a "primary" PCIe Root Bus and (optionally) a bunch of "secondary" ones, but the same will not be the case for pseries guests.
OK
I don't think the difference is that important though, at least from libvirt's point of view: whether you're creating a pseries guest with two PHBs, or a q35 guest with its built-in PCIe Root Complex and an extra PCIe Expander Bus, you will end up with two "top level" buses that you can plug more devices into.
I agree If we had spapr-pcie-host-bridge, we
could treat them mostly the same - with caveats such as the one described above, of course.
Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
What is the virtual bus in question? Why would it matter that they're siblings?
On pseries it won't. But my understanding is that libvirt won't create them that way on q35 - instead it will insert the RCs / P2P bridges to allow them to be hotplugged. Inserting that bridge may confuse pseries guests which aren't expecting it.
libvirt will automatically add PCIe Root Ports to make the devices hotpluggable on q35 guests, yes. But, as mentioned above, we can teach it not to.
I'm possibly missing the point entirely, but so far it looks to me like there are different configurations you might want to use depending on your goal, and both q35 and pseries give you comparable tools to achieve such configurations.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex.
Actually the root PCIe bus is part of a root complex. But I suspect that's likely to cause
trouble with hotplug, and it will certainly need different address allocation from libvirt.
PCIe Integrated Endpoint Devices are not hotpluggable on q35, that's why libvirt will follow QEMU's PCIe topology recommendations and place a PCIe Root Port between them; I assume the same could be done for pseries guests as soon as QEMU grows support for generic PCIe Root Ports, something Marcel has already posted patches for.
Here you've hit on it. No, we should not do that for pseries, AFAICT. PAPR doesn't really have the concept of integrated endpoint devices, and all devices can be hotplugged via the PAPR mechanisms (and none can via the PCI-E standard hotplug mechanism).
This seems to be interfering with the PCIe spec: 1. No PCIe root ports ? those are part of the spec. 2. Only integrated devices ? hotplug is not PCIe native?
Cool, I get it now.
Again, sorry for clearly misunderstanding your explanation, but I'm still not seeing the issue here. I'm sure it's very clear in your mind, but I'm afraid you're going to have to walk me through it :(
I wish it were entirely clear in my mind. Like I say I'm still pretty confused by exactly the root complex entails.
Same here, but this back-and-forth is helping! :)
[...]
What about virtio devices, which present themselves either as legacy PCI or PCIe depending on the kind of slot they are plugged into? Would they show up as PCIe or legacy PCI on a PCIe-enabled pseries guest?
That we'd have to address on the qemu side with some
Unfinished sentence?
[...]
Is the Root Complex not currently exposed? The Root Bus certainly is,
Like I say, I'm fairly confused myself, but I'm pretty sure that Root Complex != Root Bus. The RC sits under the root bus IIRC.. or possibly it consists of the root bus plus something under it as well.
The Root complex includes the PCI bus, some configuration registers if needed, provides access to the configuration space, translates relevant CPU reads/writes to PCI(e) transactions...
Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries.
Basically the special casing I was mentioning earlier.
That looks complicated.. I wish I would no more about the pseries PCIe stuff, does any one know where I can get the info ? (besides 'google it'...)
[...]
Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead?
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
That issue is relevant whether or nor we have different PHB flavors, isn't it? As soon as multiple PHBs are present in a pseries guest, multiple PCI domains will be there as well, and we need to handle that somehow.
On q35, on the other hand, I haven't been able to find a way to create extra PCI domains: adding a pxb-pcie certainly didn't work the same as adding an extra spapr-pci-host-bridge in that regard.
Indeed, all the pxb-pcie devices "share" the same domain.
[...]
Maybe we should have a different model, specific to pseries guests, instead, so that all PHBs would look the same in the guest XML, something like
<controller type='pci' model='phb-pcie'/>
It would require shuffling libvirt's PCI address allocation code around quite a bit, but it should be doable. And if it makes life easier for our users, then it's worth it.
Hrm. So my first inclination would be to stick with the generic names, and map those to creating new pseries host bridges on pseries guests. I would have thought that would be the easier option for users. But I may not have realized all the implications yet.
You're probably right, but I can't immediately see how we would make the user aware of which PHB is which. Maybe we could add some sub-element or extra attribute...
Anyway, we should not focus too much on this specific bit at the moment, deciding on a specific XML is mostly bikeshedding :)
[...]
* Eduardo's work, which you mentioned, is going to be very beneficial in the long run; in the short run, Marcel's PCIe device placement guidelines, a document that has seen contributions from QEMU, OVMF and libvirt developers, have been invaluable to improve libvirt's PCI address allocation logic. So we're already doing better, and more improvements are on the way :)
Right.. so here's the thing, I strongly suspect that Marcel's guidelines will not be correct for pseries.
We should make the document stick for all PCIe archs, if we need to modify it, let's do it. I'm not sure if they'll
be definitively wrong, or just different enough from PowerVM that it might confuse guests, but either way.
I really need to understand how it would confuse the guests, it does not deviate from the PCIe spec, it only adds some restrictions.
Those guidelines have been developed with q35/mach-virt in mind[1], so I wouldn't at all be surprised if they didn't apply to pseries guests. And in fact, we just found out that they don't!
My point is that we could easily create a similar document for pseries guests, and then libvirt will be able to pick up whatever recommendations we come up with just like it did for q35/mach-virt.
Can you send me a link to that document though, which might help me figure this out.
It's docs/pcie.txt in QEMU's git repository.
[1] Even though I now realize that this is not immediately clear by looking at the document itself
I kind of miss the core issue, what is the main problem? Thanks, Marcel
-- Andrea Bolognani / Red Hat / Virtualization
On Tue, 13 Dec 2016 14:25:44 +0200 Marcel Apfelbaum <marcel@redhat.com> wrote:
Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries.
Basically the special casing I was mentioning earlier.
That looks complicated.. I wish I would no more about the pseries PCIe stuff, does any one know where I can get the info ? (besides 'google it'...)
This is described in the LoPAPR specification, available at: https://openpowerfoundation.org/wp-content/uploads/2016/05/LoPAPR_DRAFT_v11_... Cheers. -- Greg
On Tue, 2016-12-13 at 14:25 +0200, Marcel Apfelbaum wrote:
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
That's the thing though, it's *not* bare metal ;-) On bare metal, we use the "powernv" platform for which we haven't yet upstreamed the PCIE support, but there we have real PCIe root ports with all what's exepcted on them. They come on physically different PHB, one root complex per PHB, but they are "proper" PCIe. Hotplug is a bit weird still because it has to go through some FW interactions as the HW doesn't use the PCIe standard slot control bits (and our qemu model doesn't handle it yet) but otherwise it's standard. "pseries" on the other hand is a paravirtualized platform. It's the environment presented to guests by our propritary hypervisor (PowerVM, aka pHyp) and KVM/qemu. I think there's a public spec these days but to cut the story short, it doesn't expose PCIe "properly" for bad historical reasons. It tends to hide the parent, ie, the root port for slots connected directly to a PHB or the entire hierarchy from the root to (and including) the downstream switch port for slots under as switch. So you end up with PCIe devices devoid of a proper "parent" which is a bit weird. Hotplug is implemented using specific firmware interfaces that are identical with pre-PCIe (ie PCI/PCI-X) systems. The FW has interface for supporting 3 kinds of hotplug: - Entire PHBs - P2P Bridges - Individual devices on an existing PHB or P2P bridge In practice these days mostly the first one is used for everything under pHyp, though I think we have a mix of 1 and 3 for KVM/qemu. Cheers, Ben.
On Tue, Dec 13, 2016 at 09:15:37AM -0600, Benjamin Herrenschmidt wrote:
On Tue, 2016-12-13 at 14:25 +0200, Marcel Apfelbaum wrote:
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
That's the thing though, it's *not* bare metal ;-)
On bare metal, we use the "powernv" platform for which we haven't yet upstreamed the PCIE support, but there we have real PCIe root ports with all what's exepcted on them.
They come on physically different PHB, one root complex per PHB, but they are "proper" PCIe. Hotplug is a bit weird still because it has to go through some FW interactions as the HW doesn't use the PCIe standard slot control bits (and our qemu model doesn't handle it yet) but otherwise it's standard.
"pseries" on the other hand is a paravirtualized platform. It's the environment presented to guests by our propritary hypervisor (PowerVM, aka pHyp) and KVM/qemu. I think there's a public spec these days but to cut the story short, it doesn't expose PCIe "properly" for bad historical reasons.
It tends to hide the parent, ie, the root port for slots connected directly to a PHB or the entire hierarchy from the root to (and including) the downstream switch port for slots under as switch.
So you end up with PCIe devices devoid of a proper "parent" which is a bit weird.
Hotplug is implemented using specific firmware interfaces that are identical with pre-PCIe (ie PCI/PCI-X) systems. The FW has interface for supporting 3 kinds of hotplug:
- Entire PHBs - P2P Bridges - Individual devices on an existing PHB or P2P bridge
In practice these days mostly the first one is used for everything under pHyp, though I think we have a mix of 1 and 3 for KVM/qemu.
Alas, no, we only have 3 on KVM/qemu. Well, possibly you could do (2) by plugging a bridge device using (3) then a device onto the bridge. Whole PHB hotplug has never been implemented for qemu/KVM.. but we probably want to. -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On 12/13/2016 05:15 PM, Benjamin Herrenschmidt wrote:
On Tue, 2016-12-13 at 14:25 +0200, Marcel Apfelbaum wrote:
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
That's the thing though, it's *not* bare metal ;-)
On bare metal, we use the "powernv" platform for which we haven't yet upstreamed the PCIE support, but there we have real PCIe root ports with all what's exepcted on them.
They come on physically different PHB, one root complex per PHB, but they are "proper" PCIe. Hotplug is a bit weird still because it has to go through some FW interactions as the HW doesn't use the PCIe standard slot control bits (and our qemu model doesn't handle it yet) but otherwise it's standard.
Thanks for the explanations, so bare metal is standard minus hotplug.
"pseries" on the other hand is a paravirtualized platform. It's the environment presented to guests by our propritary hypervisor (PowerVM, aka pHyp) and KVM/qemu. I think there's a public spec these days but to cut the story short, it doesn't expose PCIe "properly" for bad historical reasons.
It tends to hide the parent, ie, the root port for slots connected directly to a PHB or the entire hierarchy from the root to (and including) the downstream switch port for slots under as switch.
So you end up with PCIe devices devoid of a proper "parent" which is a bit weird.
OK, now is more "clear". I can get back to read the thread from the start :) Thanks, Marcel
Hotplug is implemented using specific firmware interfaces that are identical with pre-PCIe (ie PCI/PCI-X) systems. The FW has interface for supporting 3 kinds of hotplug:
- Entire PHBs - P2P Bridges - Individual devices on an existing PHB or P2P bridge
In practice these days mostly the first one is used for everything under pHyp, though I think we have a mix of 1 and 3 for KVM/qemu.
Cheers, Ben.
On Tue, Dec 13, 2016 at 02:25:44PM +0200, Marcel Apfelbaum wrote:
On 12/07/2016 06:42 PM, Andrea Bolognani wrote:
[Added Marcel to CC]
Hi,
Sorry for the late reply.
On Wed, 2016-12-07 at 15:11 +1100, David Gibson wrote:
Is the difference between q35 and pseries guests with respect to PCIe only relevant when it comes to assigned devices, or in general? I'm asking this because you seem to focus entirely on assigned devices.
Well, in a sense that's up to us. The only existing model we have is PowerVM, and PowerVM only does device passthrough, no emulated devices. PAPR doesn't really distinguish one way or the other, but it's written from the perspective of assuming that all PCI devices correspond to physical devices on the host
Okay, that makes sense.
On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes.
This part I don't get at all, so please bear with me.
The way I read it you're claiming that eg. a SCSI controller and a network adapter, being physically separate and assigned to separate PCI slots, should have a dedicated PCIe Root Complex each on a q35 guest.
Not a PCIe Root Complex, but a PCIe Root port.
Ah, sorry. As I said, I've been pretty confused by all the terminology.
Right, my understanding was that if the devices were slotted, rather than integrated, each one would sit under a separate root complex, the root complex being a pseudo PCI to PCI bridge.
I assume "slotted" means "plugged into a slot that's not one of those provided by pcie.0" or something along those lines.
More on the root complex bit later.
That doesn't match with my experience, where you would simply assign them to separate slots of the default PCIe Root Bus (pcie.0), eg. 00:01.0 and 00:02.0.
The qemu default, or the libvirt default?
I'm talking about the libvirt default, which is supposed to follows Marcel's PCIe Guidelines.
I think this represents treating the devices as though they were integrated devices in the host bridge. I believe on q35 they would not be hotpluggable
Correct. Please have a look to the new document regarding pcie: docs/pcie.txt and the corresponding presentations.
Yeah, that's indeed not quite what libvirt would do by default: in reality, there would be a ioh3420 between the pcie.0 slots and each device exactly to enable hotplug.
but on pseries they would be (because we don't use the standard hot plug controller).
We can account for that in libvirt and avoid adding the extra ioh3420s (or rather the upcoming generic PCIe Root Ports) for pseries guests.
Maybe you're referring to the fact that you might want to create multiple PCIe Root Complexes in order to assign the host devices to separate guest NUMA nodes? How is creating multiple PCIe Root Complexes on q35 using pxb-pcie different than creating multiple PHBs using spapr-pci-host-bridge on pseries?
Uh.. AIUI the root complex is the PCI to PCI bridge under which PCI-E slots appear. PXB is something different - essentially different host bridges as you say (though with some weird hacks to access config space, which make it dependent on the primary bus in a way which spapr PHBs are not).
I'll admit I'm pretty confused myself about the exact distinction between root complex, root port and upstream and downstream ports.
I think we both need to get our terminology straight :) I'm sure Marcel will be happy to point us in the right direction.
My understanding is that the PCIe Root Complex is the piece of hardware that exposes a PCIe Root Bus (pcie.0 in QEMU);
right
Oh.. I wasn't as clear as I'd like to be on what the root complex is. But I thought the root complex did have some guest visible presence in the PCI tree. What you're describing here seems equivalent to what I'd call the PCI Host Bridge (== PHB).
PXBs can be connected to slots in pcie.0 to create more buses that behave, for the most part, like pcie.0 and are mostly useful to bind devices to specific NUMA nodes.
right
Same applies
to legacy PCI with the pxb (instead of pxb-pcie) device.
pxb should not be used for PCIe machines, only for legacy PCI ones.
Noted. And not for pseries at all. Note that because we have a para-virtualized platform (all PCI config access goes via hypercalls) the distinction between PCI and PCI-E is much blurrier than in the x86 case.
In a similar fashion, PHBs are the hardware thingies that expose a PCI Root Bus (pci.0 and so on), the main difference being that they are truly independent: so a q35 guest will always have a "primary" PCIe Root Bus and (optionally) a bunch of "secondary" ones, but the same will not be the case for pseries guests.
OK
I don't think the difference is that important though, at least from libvirt's point of view: whether you're creating a pseries guest with two PHBs, or a q35 guest with its built-in PCIe Root Complex and an extra PCIe Expander Bus, you will end up with two "top level" buses that you can plug more devices into.
I agree
If we had spapr-pcie-host-bridge, we
could treat them mostly the same - with caveats such as the one described above, of course.
Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
What is the virtual bus in question? Why would it matter that they're siblings?
On pseries it won't. But my understanding is that libvirt won't create them that way on q35 - instead it will insert the RCs / P2P bridges to allow them to be hotplugged. Inserting that bridge may confuse pseries guests which aren't expecting it.
libvirt will automatically add PCIe Root Ports to make the devices hotpluggable on q35 guests, yes. But, as mentioned above, we can teach it not to.
I'm possibly missing the point entirely, but so far it looks to me like there are different configurations you might want to use depending on your goal, and both q35 and pseries give you comparable tools to achieve such configurations.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex.
Actually the root PCIe bus is part of a root complex.
So I think what I meant above was "root port". The point is that there won't be the (pseudo) PCI to PCI bridge appearing above the device that there typically would be on q35.
But I suspect that's likely to cause
trouble with hotplug, and it will certainly need different address allocation from libvirt.
PCIe Integrated Endpoint Devices are not hotpluggable on q35, that's why libvirt will follow QEMU's PCIe topology recommendations and place a PCIe Root Port between them; I assume the same could be done for pseries guests as soon as QEMU grows support for generic PCIe Root Ports, something Marcel has already posted patches for.
Here you've hit on it. No, we should not do that for pseries, AFAICT. PAPR doesn't really have the concept of integrated endpoint devices, and all devices can be hotplugged via the PAPR mechanisms (and none can via the PCI-E standard hotplug mechanism).
This seems to be interfering with the PCIe spec: 1. No PCIe root ports ? those are part of the spec.
Yes, I dare say it does interfere with the spec. Nonetheless, there it is.
2. Only integrated devices ? hotplug is not PCIe native?
That's correct. PAPR supplies its own hotplug mechanism, which works for both PCI and PCI-E devices, which is different from the standard PCI-E hotplug mechanism.
Cool, I get it now.
Again, sorry for clearly misunderstanding your explanation, but I'm still not seeing the issue here. I'm sure it's very clear in your mind, but I'm afraid you're going to have to walk me through it :(
I wish it were entirely clear in my mind. Like I say I'm still pretty confused by exactly the root complex entails.
Same here, but this back-and-forth is helping! :)
[...]
What about virtio devices, which present themselves either as legacy PCI or PCIe depending on the kind of slot they are plugged into? Would they show up as PCIe or legacy PCI on a PCIe-enabled pseries guest?
That we'd have to address on the qemu side with some
Unfinished sentence?
[...]
Is the Root Complex not currently exposed? The Root Bus certainly is,
Like I say, I'm fairly confused myself, but I'm pretty sure that Root Complex != Root Bus. The RC sits under the root bus IIRC.. or possibly it consists of the root bus plus something under it as well.
The Root complex includes the PCI bus, some configuration registers if needed, provides access to the configuration space, translates relevant CPU reads/writes to PCI(e) transactions...
Do those configuration registers appear within PCI space, or outside it (e.g. raw MMIO or PIO registers)?
Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries.
Basically the special casing I was mentioning earlier.
That looks complicated.. I wish I would no more about the pseries PCIe stuff, does any one know where I can get the info ? (besides 'google it'...)
Andrea gave a pointer to the PAPR document. Unfortunately how much it covers here I'm not sure about. In particular I'm not sure how much of this is actually PAPR mandated, and how much is just copying PowerVM as the pre-existing PAPR implementation.
[...]
Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead?
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
Which aspect? Having multiple independent host bridges is perfectly reasonable - x86 just doesn't do it well for rather stupid historical reasons. PAPR is quite explicitly a paravirtual platform, you cannot have a bare-metal PAPR machine.
That issue is relevant whether or nor we have different PHB flavors, isn't it? As soon as multiple PHBs are present in a pseries guest, multiple PCI domains will be there as well, and we need to handle that somehow.
On q35, on the other hand, I haven't been able to find a way to create extra PCI domains: adding a pxb-pcie certainly didn't work the same as adding an extra spapr-pci-host-bridge in that regard.
Indeed, all the pxb-pcie devices "share" the same domain.
[...]
Maybe we should have a different model, specific to pseries guests, instead, so that all PHBs would look the same in the guest XML, something like
<controller type='pci' model='phb-pcie'/>
It would require shuffling libvirt's PCI address allocation code around quite a bit, but it should be doable. And if it makes life easier for our users, then it's worth it.
Hrm. So my first inclination would be to stick with the generic names, and map those to creating new pseries host bridges on pseries guests. I would have thought that would be the easier option for users. But I may not have realized all the implications yet.
You're probably right, but I can't immediately see how we would make the user aware of which PHB is which. Maybe we could add some sub-element or extra attribute...
Anyway, we should not focus too much on this specific bit at the moment, deciding on a specific XML is mostly bikeshedding :)
[...]
* Eduardo's work, which you mentioned, is going to be very beneficial in the long run; in the short run, Marcel's PCIe device placement guidelines, a document that has seen contributions from QEMU, OVMF and libvirt developers, have been invaluable to improve libvirt's PCI address allocation logic. So we're already doing better, and more improvements are on the way :)
Right.. so here's the thing, I strongly suspect that Marcel's guidelines will not be correct for pseries.
We should make the document stick for all PCIe archs, if we need to modify it, let's do it.
Yeah, I'm not sure it's possible to cover both x86 and pseries at once. As you noted, it looks rather like PAPR is contradicting the PCI-E spec. Again, one possible option here is to continue to treat pseries as having a vanilla-PCI bus, but with a special flag saying that it's magically able to connect PCI-E devices.
I'm not sure if they'll
be definitively wrong, or just different enough from PowerVM that it might confuse guests, but either way.
I really need to understand how it would confuse the guests, it does not deviate from the PCIe spec, it only adds some restrictions.
Because the guests are written to work with PowerVM, which seems to do something other than the PCIe spec...
Those guidelines have been developed with q35/mach-virt in mind[1], so I wouldn't at all be surprised if they didn't apply to pseries guests. And in fact, we just found out that they don't!
My point is that we could easily create a similar document for pseries guests, and then libvirt will be able to pick up whatever recommendations we come up with just like it did for q35/mach-virt.
Can you send me a link to that document though, which might help me figure this out.
It's docs/pcie.txt in QEMU's git repository.
[1] Even though I now realize that this is not immediately clear by looking at the document itself
I kind of miss the core issue, what is the main problem?
The core problem is that at this stage it's not possible to attach PCIe devices (either emulated or passthrough) to a pseries guest. We need to be able to do that - specifically allowing the guest to access PCIe extended config space. The PAPR virtualized PCI interfaces definitely do allow a guest to access extended config space, but in most other regards they behave more like vanilla-PCI than PCIe. -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On 12/14/2016 04:46 AM, David Gibson wrote:
On Tue, Dec 13, 2016 at 02:25:44PM +0200, Marcel Apfelbaum wrote:
On 12/07/2016 06:42 PM, Andrea Bolognani wrote:
[Added Marcel to CC]
Hi,
Sorry for the late reply.
On Wed, 2016-12-07 at 15:11 +1100, David Gibson wrote:
Is the difference between q35 and pseries guests with respect to PCIe only relevant when it comes to assigned devices, or in general? I'm asking this because you seem to focus entirely on assigned devices.
Well, in a sense that's up to us. The only existing model we have is PowerVM, and PowerVM only does device passthrough, no emulated devices. PAPR doesn't really distinguish one way or the other, but it's written from the perspective of assuming that all PCI devices correspond to physical devices on the host
Okay, that makes sense.
On q35, you'd generally expect physically separate (different slot) devices to appear under separate root complexes.
This part I don't get at all, so please bear with me.
The way I read it you're claiming that eg. a SCSI controller and a network adapter, being physically separate and assigned to separate PCI slots, should have a dedicated PCIe Root Complex each on a q35 guest.
Not a PCIe Root Complex, but a PCIe Root port.
Ah, sorry. As I said, I've been pretty confused by all the terminology.
Right, my understanding was that if the devices were slotted, rather than integrated, each one would sit under a separate root complex, the root complex being a pseudo PCI to PCI bridge.
I assume "slotted" means "plugged into a slot that's not one of those provided by pcie.0" or something along those lines.
More on the root complex bit later.
That doesn't match with my experience, where you would simply assign them to separate slots of the default PCIe Root Bus (pcie.0), eg. 00:01.0 and 00:02.0.
The qemu default, or the libvirt default?
I'm talking about the libvirt default, which is supposed to follows Marcel's PCIe Guidelines.
I think this represents treating the devices as though they were integrated devices in the host bridge. I believe on q35 they would not be hotpluggable
Correct. Please have a look to the new document regarding pcie: docs/pcie.txt and the corresponding presentations.
Yeah, that's indeed not quite what libvirt would do by default: in reality, there would be a ioh3420 between the pcie.0 slots and each device exactly to enable hotplug.
but on pseries they would be (because we don't use the standard hot plug controller).
We can account for that in libvirt and avoid adding the extra ioh3420s (or rather the upcoming generic PCIe Root Ports) for pseries guests.
Maybe you're referring to the fact that you might want to create multiple PCIe Root Complexes in order to assign the host devices to separate guest NUMA nodes? How is creating multiple PCIe Root Complexes on q35 using pxb-pcie different than creating multiple PHBs using spapr-pci-host-bridge on pseries?
Uh.. AIUI the root complex is the PCI to PCI bridge under which PCI-E slots appear. PXB is something different - essentially different host bridges as you say (though with some weird hacks to access config space, which make it dependent on the primary bus in a way which spapr PHBs are not).
I'll admit I'm pretty confused myself about the exact distinction between root complex, root port and upstream and downstream ports.
I think we both need to get our terminology straight :) I'm sure Marcel will be happy to point us in the right direction.
My understanding is that the PCIe Root Complex is the piece of hardware that exposes a PCIe Root Bus (pcie.0 in QEMU);
right
Oh.. I wasn't as clear as I'd like to be on what the root complex is. But I thought the root complex did have some guest visible presence in the PCI tree. What you're describing here seems equivalent to what I'd call the PCI Host Bridge (== PHB).
Yes, a Root Complex is a type of Host Bridge in the sense it bridges between CPU/Memory Controller ant the PCI subsystem.
PXBs can be connected to slots in pcie.0 to create more buses that behave, for the most part, like pcie.0 and are mostly useful to bind devices to specific NUMA nodes.
right
Same applies
to legacy PCI with the pxb (instead of pxb-pcie) device.
pxb should not be used for PCIe machines, only for legacy PCI ones.
Noted. And not for pseries at all. Note that because we have a para-virtualized platform (all PCI config access goes via hypercalls) the distinction between PCI and PCI-E is much blurrier than in the x86 case.
OK
In a similar fashion, PHBs are the hardware thingies that expose a PCI Root Bus (pci.0 and so on), the main difference being that they are truly independent: so a q35 guest will always have a "primary" PCIe Root Bus and (optionally) a bunch of "secondary" ones, but the same will not be the case for pseries guests.
OK
I don't think the difference is that important though, at least from libvirt's point of view: whether you're creating a pseries guest with two PHBs, or a q35 guest with its built-in PCIe Root Complex and an extra PCIe Expander Bus, you will end up with two "top level" buses that you can plug more devices into.
I agree
If we had spapr-pcie-host-bridge, we
could treat them mostly the same - with caveats such as the one described above, of course.
Whereas on pseries they'll appear as siblings on a virtual bus (which makes no physical sense for point-to-point PCI-E).
What is the virtual bus in question? Why would it matter that they're siblings?
On pseries it won't. But my understanding is that libvirt won't create them that way on q35 - instead it will insert the RCs / P2P bridges to allow them to be hotplugged. Inserting that bridge may confuse pseries guests which aren't expecting it.
libvirt will automatically add PCIe Root Ports to make the devices hotpluggable on q35 guests, yes. But, as mentioned above, we can teach it not to.
I'm possibly missing the point entirely, but so far it looks to me like there are different configurations you might want to use depending on your goal, and both q35 and pseries give you comparable tools to achieve such configurations.
I suppose we could try treating all devices on pseries as though they were chipset builtin devices on q35, which will appear on the root PCI-E bus without root complex.
Actually the root PCIe bus is part of a root complex.
So I think what I meant above was "root port".
Yes The point is that
there won't be the (pseudo) PCI to PCI bridge appearing above the device that there typically would be on q35.
Understood, on one hand we have no PCIe Root Ports, on the other hand the devices are not integrated - they can be hot-plugged by platform specific means.
But I suspect that's likely to cause
trouble with hotplug, and it will certainly need different address allocation from libvirt.
PCIe Integrated Endpoint Devices are not hotpluggable on q35, that's why libvirt will follow QEMU's PCIe topology recommendations and place a PCIe Root Port between them; I assume the same could be done for pseries guests as soon as QEMU grows support for generic PCIe Root Ports, something Marcel has already posted patches for.
Here you've hit on it. No, we should not do that for pseries, AFAICT. PAPR doesn't really have the concept of integrated endpoint devices, and all devices can be hotplugged via the PAPR mechanisms (and none can via the PCI-E standard hotplug mechanism).
This seems to be interfering with the PCIe spec: 1. No PCIe root ports ? those are part of the spec.
Yes, I dare say it does interfere with the spec. Nonetheless, there it is.
2. Only integrated devices ? hotplug is not PCIe native?
That's correct. PAPR supplies its own hotplug mechanism, which works for both PCI and PCI-E devices, which is different from the standard PCI-E hotplug mechanism.
Ok the hw is PCIe, but configuration/hot-plug is platform specific.
Cool, I get it now.
Again, sorry for clearly misunderstanding your explanation, but I'm still not seeing the issue here. I'm sure it's very clear in your mind, but I'm afraid you're going to have to walk me through it :(
I wish it were entirely clear in my mind. Like I say I'm still pretty confused by exactly the root complex entails.
Same here, but this back-and-forth is helping! :)
[...]
What about virtio devices, which present themselves either as legacy PCI or PCIe depending on the kind of slot they are plugged into? Would they show up as PCIe or legacy PCI on a PCIe-enabled pseries guest?
That we'd have to address on the qemu side with some
Unfinished sentence?
[...]
Is the Root Complex not currently exposed? The Root Bus certainly is,
Like I say, I'm fairly confused myself, but I'm pretty sure that Root Complex != Root Bus. The RC sits under the root bus IIRC.. or possibly it consists of the root bus plus something under it as well.
The Root complex includes the PCI bus, some configuration registers if needed, provides access to the configuration space, translates relevant CPU reads/writes to PCI(e) transactions...
Do those configuration registers appear within PCI space, or outside it (e.g. raw MMIO or PIO registers)?
Root Complexes use MMIO to expose the PCI configuration space, they call it ECAM (enhanced configuration access mechanism) or MMConfig.
Now... from what Laine was saying it sounds like more of the differences between PCI-E placement and PCI placement may be implemented by libvirt than qemu than I realized. So possibly we do want to make the bus be PCI-E on the qemu side, but have libvirt use the vanilla-PCI placement guidelines rather than PCI-E for pseries.
Basically the special casing I was mentioning earlier.
That looks complicated.. I wish I would no more about the pseries PCIe stuff, does any one know where I can get the info ? (besides 'google it'...)
Andrea gave a pointer to the PAPR document. Unfortunately how much it covers here I'm not sure about. In particular I'm not sure how much of this is actually PAPR mandated, and how much is just copying PowerVM as the pre-existing PAPR implementation.
Understood, I'll have a look, but with low expectations :). Anyway, by now I do have some basic notions of spapr, thanks!
[...]
Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead?
Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use.
This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows...
Which aspect? Having multiple independent host bridges is perfectly reasonable - x86 just doesn't do it well for rather stupid historical reasons.
I agree about the multiple host-bridges, is actually what pxb/pxb-pcie devices (kind of) do. I was talking about having one PCI PHB and another PHB which is PCI Express. I was referring to one system having both PCI and PCIe PHBs.
PAPR is quite explicitly a paravirtual platform, you cannot have a bare-metal PAPR machine.
Understood.
That issue is relevant whether or nor we have different PHB flavors, isn't it? As soon as multiple PHBs are present in a pseries guest, multiple PCI domains will be there as well, and we need to handle that somehow.
On q35, on the other hand, I haven't been able to find a way to create extra PCI domains: adding a pxb-pcie certainly didn't work the same as adding an extra spapr-pci-host-bridge in that regard.
Indeed, all the pxb-pcie devices "share" the same domain.
[...]
Maybe we should have a different model, specific to pseries guests, instead, so that all PHBs would look the same in the guest XML, something like
<controller type='pci' model='phb-pcie'/>
It would require shuffling libvirt's PCI address allocation code around quite a bit, but it should be doable. And if it makes life easier for our users, then it's worth it.
Hrm. So my first inclination would be to stick with the generic names, and map those to creating new pseries host bridges on pseries guests. I would have thought that would be the easier option for users. But I may not have realized all the implications yet.
You're probably right, but I can't immediately see how we would make the user aware of which PHB is which. Maybe we could add some sub-element or extra attribute...
Anyway, we should not focus too much on this specific bit at the moment, deciding on a specific XML is mostly bikeshedding :)
[...]
* Eduardo's work, which you mentioned, is going to be very beneficial in the long run; in the short run, Marcel's PCIe device placement guidelines, a document that has seen contributions from QEMU, OVMF and libvirt developers, have been invaluable to improve libvirt's PCI address allocation logic. So we're already doing better, and more improvements are on the way :)
Right.. so here's the thing, I strongly suspect that Marcel's guidelines will not be correct for pseries.
We should make the document stick for all PCIe archs, if we need to modify it, let's do it.
Yeah, I'm not sure it's possible to cover both x86 and pseries at once. As you noted, it looks rather like PAPR is contradicting the PCI-E spec.
At this point I agree that PAPR PCI-E is not really "by the book", so the PCIe guidelines will not work for PAPR.
Again, one possible option here is to continue to treat pseries as having a vanilla-PCI bus, but with a special flag saying that it's magically able to connect PCI-E devices.
A PCIe bus supporting PCI devices is strange (QEMU allows it ...), but a PCI bus supporting PCIe devices is hard to "swallow". I would say maybe make it a special case of a PCIe bus with different rules. It can derive from the PCIe bus class and override the usual behavior with PAPR specific rule which happen to be similar with the PCI bus rules. Adding Eduardo, he is currently working on a way to properly expose the information on what devices can be plugged on what bus/slot.
I'm not sure if they'll
be definitively wrong, or just different enough from PowerVM that it might confuse guests, but either way.
I really need to understand how it would confuse the guests, it does not deviate from the PCIe spec, it only adds some restrictions.
Because the guests are written to work with PowerVM, which seems to do something other than the PCIe spec...
Those guidelines have been developed with q35/mach-virt in mind[1], so I wouldn't at all be surprised if they didn't apply to pseries guests. And in fact, we just found out that they don't!
My point is that we could easily create a similar document for pseries guests, and then libvirt will be able to pick up whatever recommendations we come up with just like it did for q35/mach-virt.
Can you send me a link to that document though, which might help me figure this out.
It's docs/pcie.txt in QEMU's git repository.
[1] Even though I now realize that this is not immediately clear by looking at the document itself
I kind of miss the core issue, what is the main problem?
The core problem is that at this stage it's not possible to attach PCIe devices (either emulated or passthrough) to a pseries guest. We need to be able to do that - specifically allowing the guest to access PCIe extended config space.
Do we have in QEMU the code to expose the Extended Config Space by other means instead of MMIO (used by x86)?
The PAPR virtualized PCI interfaces definitely do allow a guest to access extended config space, but in most other regards they behave more like vanilla-PCI than PCIe.
So the problem is related to how to expose the information to libvirt? If yes, maybe Eduardo can help. Thanks, Marcel
On Wed, 2016-12-14 at 20:26 +0200, Marcel Apfelbaum wrote:
The Root complex includes the PCI bus, some configuration registers if needed, provides access to the configuration space, translates relevant CPU reads/writes to PCI(e) transactions...
Do those configuration registers appear within PCI space, or outside it (e.g. raw MMIO or PIO registers)?
Root Complexes use MMIO to expose the PCI configuration space, they call it ECAM (enhanced configuration access mechanism) or MMConfig.
Not all of them do. There are plenty of PCIe RCs out there that do config space using different mechanisms such as the good old indirect address/data method, such as ours. IE. Even in real "bare metal" powernv, where the root port is visible to Linux, we still go through firmware for config space to get through to those registers (among other things). My PHB3 implementation (not upstream yet and a bit bitrotted by now) exposed PCIe that way including extended config space and that was working fine. Cheers, Ben.
On Wed, 2016-12-14 at 20:26 +0200, Marcel Apfelbaum wrote:
Maybe I just don't quite get the relationship between Root Complexes and Root Buses, but I guess my question is: what is preventing us from simply doing whatever a spapr-pci-host-bridge is doing in order to expose a legacy PCI Root Bus (pci.*) to the guest, and create a new spapr-pcie-host-bridge that exposes a PCIe Root Bus (pcie.*) instead? Hrm, the suggestion of providing both a vanilla-PCI and PCI-E host bridge came up before. I think one of us spotted a problem with that, but I don't recall what it was now. I guess one is how libvirt would map it's stupid-fake-domain-numbers to which root bus to use. This would be a weird configuration, I never heard of something like that on a bare metal machine, but I never worked on pseries, who knows... Which aspect? Having multiple independent host bridges is perfectly reasonable - x86 just doesn't do it well for rather stupid historical reasons. I agree about the multiple host-bridges, is actually what pxb/pxb-pcie devices (kind of) do. I was talking about having one PCI PHB and another PHB which is PCI Express. I was referring to one system having both PCI and PCIe PHBs.
Sorry this confused you: what I was talking about was just having both a legacy PCI PHB and a PCI Express PHB available in QEMU, not using both for the same guest. The user would pick one or the other and stick with it.
Again, one possible option here is to continue to treat pseries as having a vanilla-PCI bus, but with a special flag saying that it's magically able to connect PCI-E devices. A PCIe bus supporting PCI devices is strange (QEMU allows it ...), but a PCI bus supporting PCIe devices is hard to "swallow".
I agree.
I would say maybe make it a special case of a PCIe bus with different rules. It can derive from the PCIe bus class and override the usual behavior with PAPR specific rule which happen to be similar with the PCI bus rules.
It's pretty clear by now that libvirt will need to have some special handling of pseries guests. Having to choose between "weird legacy PCI" and "weird PCI Express", I think I'd rather go with the latter :) -- Andrea Bolognani / Red Hat / Virtualization
participants (9)
-
Alexey Kardashevskiy -
Andrea Bolognani -
Benjamin Herrenschmidt -
David Gibson -
Eric Blake -
Greg Kurz -
Laine Stump -
Laine Stump -
Marcel Apfelbaum