On 23.07.2013 23:20, Jim Fehlig wrote:
One comment below in addition to Konrad's...
Konrad Rzeszutek Wilk wrote:
> On Mon, Jul 22, 2013 at 12:51:05PM +0200, Stefan Bader wrote:
>
>> This fixes the basic setup but there is likely more to do if things
>> like manual CPU hirarchy (nodes, cores, threads) to be working.
>>
>> Cross-posting to xen-devel to make sure I am doing things correctly.
>>
>> -Stefan
>>
>>
>> >From 1ec5e7ea0d3498b9f61b83e8aed87cc3cae106de Mon Sep 17 00:00:00 2001
>> From: Stefan Bader <stefan.bader(a)canonical.com>
>> Date: Fri, 19 Jul 2013 15:20:00 +0200
>> Subject: [PATCH] libxl: Correctly initialize vcpu bitmap
>>
>> The avai_vcpu bitmap has to be allocated before it can be used (using
>>
>
> avail_vcpu ?
>
>
>> the maximum allowed value for that). Then for each available VCPU the
>> bit in the mask has to be set (libxl_bitmap_set takes a bit position
>> as an argument, not the number of bits to set).
>>
>> Without this, I would always only get one VCPU for guests created
>> through libvirt/libxl.
>>
>> Signed-off-by: Stefan Bader <stefan.bader(a)canonical.com>
>>
>
> The libxl calling logic looks Ok to me. So from the libxl perspective
> you can tack on Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk(a)oracle.com>
>
>
>> ---
>> src/libxl/libxl_conf.c | 14 +++++++++++---
>> 1 file changed, 11 insertions(+), 3 deletions(-)
>>
>> diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c
>> index 4a0fba9..7592dd2 100644
>> --- a/src/libxl/libxl_conf.c
>> +++ b/src/libxl/libxl_conf.c
>> @@ -331,7 +331,8 @@ error:
>> }
>>
>> static int
>> -libxlMakeDomBuildInfo(virDomainDefPtr def, libxl_domain_config *d_config)
>> +libxlMakeDomBuildInfo(libxlDriverPrivatePtr driver, virDomainDefPtr def,
>> + libxl_domain_config *d_config)
>> {
>> libxl_domain_build_info *b_info = &d_config->b_info;
>> int hvm = STREQ(def->os.type, "hvm");
>> @@ -343,8 +344,15 @@ libxlMakeDomBuildInfo(virDomainDefPtr def,
libxl_domain_config *d_config)
>> libxl_domain_build_info_init_type(b_info, LIBXL_DOMAIN_TYPE_HVM);
>> else
>> libxl_domain_build_info_init_type(b_info, LIBXL_DOMAIN_TYPE_PV);
>> +
>> b_info->max_vcpus = def->maxvcpus;
>> - libxl_bitmap_set((&b_info->avail_vcpus), def->vcpus);
>> + if (libxl_cpu_bitmap_alloc(driver->ctx, &b_info->avail_vcpus,
>> + def->maxvcpus))
>>
You are using the driver-wide libxl_ctx in libxlDriverPrivate here, but
should be using the per-domain libxl_ctx in libxlDomainObjPrivate
structure IMO.
It looks like libxlBuildDomainConfig, which calls libxlMakeDomBuildInfo,
should have just taken virDomainObjPtr instead of virDomainDefPtr.
libxlBuildDomainConfig would then have access to the per-domain
libxl_ctx, and no longer need the libxlDriverPrivatePtr parameter as well.
So more like attached v2. I am not sure libxlDriverPrivatePtr can really be
dropped (and maybe should not as part of a fix to this issue). Maybe calling
libxl_flask_context_to_sid also should use the per domain context. But at least
libxlMakeVfbList->libxlMakeVfb->virPortAllocatorAcquire sounds a bit like it
might need the driver context.
-Stefan