[PATCH v2 1/1] drivers/perf: Fix kernel panic due to the invalid mon_ctx pointer

[Will Deacon]

On Thu, Oct 26, 2023 at 06:34:43PM -0500, Shanker Donthineni wrote:
> The return pointer from the resctrl_arch_mon_ctx_alloc_no_wait() function
> is saved in a 32-bit variable 'hwc->idx' which results in the loss of
> the upper 32 bits. This, in turn, triggers a kernel panic when attempting
> to access a corrupted pointer.
> 
> Use 'event->pmu_private' instead of 'hwc->idx' to resolve the issue.
> 
> Another reason contributing to the problem due to continuous 'error irq'
> messages. The resctrl_arch_mon_ctx_alloc_no_wait() function returns a valid
> PTR when it cannot allocate monitor resources, resulting in an incorrect
> MON_SEL configuration. To resolve this issue, return an ERR_PTR when monitor
> allocation fails
> 
> dmesg:
>   mpam: error irq from msc:0 'Monitor_Range', partid:228, pmg: 0, ris: 0
>   mpam: error irq from msc:0 'Monitor_Range', partid:228, pmg: 0, ris: 0
>   mpam: error irq from msc:0 'Monitor_Range', partid:228, pmg: 0, ris: 0
>   ...
>   mpam: error irq from msc:0 'Monitor_Range', partid:228, pmg: 0, ris: 0
> 
> Signed-off-by: Shanker Donthineni <sdonthineni at nvidia.com>
> ---
> Changes in v2:
>  -Use 'event->pmu_private' instead of 'hwc->idx' to keep monitor context
>  -Return ERR_PTR if unable to allocate resource in resctrl_arch_mon_ctx_alloc_no_wait()
> 
>  drivers/perf/resctrl_pmu.c           | 12 ++++++------
>  drivers/platform/mpam/mpam_resctrl.c | 23 ++++++++++++++++-------

I can't see these files upstream or in linux-next, so please don't cc me on
changes to them until they've landed. My inbox is bad enough as it is!

Thanks,

Will