Re: [PATCH 5/5] cpu: Introduce getHost supoort for ARM
Re-adding libvir-list to the CC line. On Mon, Mar 30, 2020 at 08:20:44PM +0800, Zhenyu Zheng wrote:
Hi, yes, I think we can do that using inline assembly, I can check it out if you think it is a better solution, do you have any suggestions for the features(cpu flags) part? It seems that ARM does not have a location/register that holds all the flags, seems that we have to query alot of different registers to check for features, which could be quite messy.
Perhaps there is a way to record the location/register info in the XML against each feature name, so that the code itself can stay simple and just be driven from the metadata ? I'm not familiar enough with Arm to know how feasiable this is though, so will have to leave that to others to give an opinion.
On Mon, Mar 30, 2020 at 8:01 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
On Mon, Mar 30, 2020 at 07:32:36PM +0800, Zhenyu Zheng wrote:
Hi Daniel,
Thanks for thre review and reply, my first implementation was going to gather data from /proc/cpuinfo, but unlike X86, we can only get this kind of info:
processor : 0 BogoMIPS : 200.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 cpuid CPU implementer : 0x43 CPU architecture: 8 CPU variant : 0x1 CPU part : 0x0a1 CPU revision : 1
so we have to perform some translation to perform human readable information, and I mentioned that 'lscpu' has done that too. So Andrea Bolognani suggested that maybe we can use it directly, to avoid re-implement the translation. Here is the discussion: https://www.redhat.com/archives/libvir-list/2020-March/msg00812.html
On x86 we get majority of info straight from calling the CPUID instruction, not /proc/cpuinfo, and use our XML data files in src/cpu_map to translate things into human readable names. I see you're adding XML data files with names in the earlier patches. Is it possible to add the hex values for the CPU implementer/architecture/variant/part to these XML files so we can directly map them in libvirt, in the same way we do for x86
Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Hi Daniel, Thanks for the review and tips, I've done some learning and finally found the way to read and parse vendor_id, part_id and cpu flags of an ARM CPU directly from registers just like what has been done in X86 drivers. And I've uploaded a V2 patch for these. BR, Zhenyu Zhengh On Mon, Mar 30, 2020 at 8:27 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
Re-adding libvir-list to the CC line.
Hi, yes, I think we can do that using inline assembly, I can check it out if you think it is a better solution, do you have any suggestions for the features(cpu flags) part? It seems
On Mon, Mar 30, 2020 at 08:20:44PM +0800, Zhenyu Zheng wrote: that
ARM does not have a location/register that holds all the flags, seems that we have to query alot of different registers to check for features, which could be quite messy.
Perhaps there is a way to record the location/register info in the XML against each feature name, so that the code itself can stay simple and just be driven from the metadata ?
I'm not familiar enough with Arm to know how feasiable this is though, so will have to leave that to others to give an opinion.
On Mon, Mar 30, 2020 at 8:01 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
On Mon, Mar 30, 2020 at 07:32:36PM +0800, Zhenyu Zheng wrote:
Hi Daniel,
Thanks for thre review and reply, my first implementation was going
gather data from /proc/cpuinfo, but unlike X86, we can only get this kind of info:
processor : 0 BogoMIPS : 200.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 cpuid CPU implementer : 0x43 CPU architecture: 8 CPU variant : 0x1 CPU part : 0x0a1 CPU revision : 1
so we have to perform some translation to perform human readable information, and I mentioned that 'lscpu' has done that too. So Andrea Bolognani suggested that maybe we can use it directly, to avoid re-implement
translation. Here is the discussion: https://www.redhat.com/archives/libvir-list/2020-March/msg00812.html
On x86 we get majority of info straight from calling the CPUID instruction, not /proc/cpuinfo, and use our XML data files in src/cpu_map to
to the translate
things into human readable names. I see you're adding XML data files with names in the earlier patches. Is it possible to add the hex values for the CPU implementer/architecture/variant/part to these XML files so we can directly map them in libvirt, in the same way we do for x86
Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
participants (2)
-
Daniel P. Berrangé -
Zhenyu Zheng