Messages in this thread | | | Date | Wed, 11 Sep 2013 10:26:59 -0400 | From | Boris Ostrovsky <> | Subject | Re: [Xen-devel] [PATCH v1 0/5] xen/PMU: PMU support for Xen PV guests |
| |
On 09/11/2013 05:33 AM, David Vrabel wrote: > On 10/09/13 16:31, Boris Ostrovsky wrote: >> This is the Linux side of Xen PMU support for PV guests, including dom0. Only >> kernel changes are here, toolstack patch will be provided separately. >> >> Here is description from the hypervisor patch submission that applies to this >> series as well: >> >> This version has following limitations: >> * For accurate profiling of dom0/Xen dom0 VCPUs should be pinned. >> * Hypervisor code is only profiled on processors that have running dom0 VCPUs >> on them. >> * No backtrace support. > These are some pretty significant limitations. Is there a plan for how > to remove them?
I don't have a specific plan (other than do it after this stage is finished) but I do have a rough idea of what would be needed to address these.
Hypervisor changes for all three should be pretty easy.
Linux-wise, for the first one (pinned VCPU) we will probably need to make a change in perf_sample_data. There is a reserved filed there so perhaps we can use it to store PCPU (and maybe domainID). Plus a way to actually write the data, probably a hook or something.
Backtrace support should also not be too bad: we can pass Xen IP stack in the shared data area to dom0 and then again have a hook or something to pass it to perf. (Note that backtracing is not supported for KVM neither so both may benefit here)
The second one is the most difficult: We need to be able somehow to access MSRs (and APIC, I think) on non-dom0 CPUs. Creating interface for such access is not a big deal but integrating it into perf infrastructure will be a challenge. There are other alternatives but they have problems as well.
-boris
| |