Re: Coverity: kfd_parse_subtype_cache(): Memory - corruptions

[Felix Kuehling]

On 2022-11-04 15:41, coverity-bot wrote:
> Hello!
>
> This is an experimental semi-automated report about issues detected by
> Coverity from a scan of next-20221104 as part of the linux-next scan project:
> https://scan.coverity.com/projects/linux-next-weekly-scan
>
> You're getting this email because you were associated with the identified
> lines of code (noted below) that were touched by commits:
>
>    Fri Dec 8 23:08:59 2017 -0500
>      3a87177eb141 ("drm/amdkfd: Add topology support for dGPUs")
>
> Coverity reported the following:
>
> *** CID 1527133:  Memory - corruptions  (OVERRUN)
> drivers/gpu/drm/amd/amdkfd/kfd_crat.c:1113 in kfd_parse_subtype_cache()
> 1107     			props->cache_size = cache->cache_size;
> 1108     			props->cacheline_size = cache->cache_line_size;
> 1109     			props->cachelines_per_tag = cache->lines_per_tag;
> 1110     			props->cache_assoc = cache->associativity;
> 1111     			props->cache_latency = cache->cache_latency;
> 1112
> vvv     CID 1527133:  Memory - corruptions  (OVERRUN)
> vvv     Overrunning array "cache->sibling_map" of 32 bytes by passing it to a function which accesses it at byte offset 63 using argument "64UL". [Note: The source code implementation of the function has been overridden by a builtin model.]
> 1113     			memcpy(props->sibling_map, cache->sibling_map,
> 1114     					sizeof(props->sibling_map));
> 1115
> 1116     			/* set the sibling_map_size as 32 for CRAT from ACPI */
> 1117     			props->sibling_map_size = CRAT_SIBLINGMAP_SIZE;
> 1118
>
> If this is a false positive, please let us know so we can mark it as
> such, or teach the Coverity rules to be smarter. If not, please make
> sure fixes get into linux-next. :) For patches fixing this, please
> include these lines (but double-check the "Fixes" first):
>
> Reported-by: coverity-bot <keescook+coverity-bot@xxxxxxxxxxxx>
> Addresses-Coverity-ID: 1527133 ("Memory - corruptions")
> Fixes: 3a87177eb141 ("drm/amdkfd: Add topology support for dGPUs")
>
> I'm not sure why this suddenly appeared after 5 years, but the read
> over-run looks legit:


I think this was introduced by a more recent patch that was in fact 
meant to fix an array overrun on HW that is outgrowing the CRAT sibling 
map size:

> commit 0938fbeb6f53fc44bc9b19784dee28496e68ba0c
> Author: Ma Jun <Jun.Ma2@xxxxxxx>
> Date:   Wed Nov 2 15:53:26 2022 +0800
>
>     drm/amdkfd: Fix the warning of array-index-out-of-bounds
>
>     For some GPUs with more CUs, the original sibling_map[32]
>     in struct crat_subtype_cache is not enough
>     to save the cache information when create the VCRAT table,
>     so skip filling the struct crat_subtype_cache info instead
>     fill struct kfd_cache_properties directly to fix this problem.
>
>     Signed-off-by: Ma Jun <Jun.Ma2@xxxxxxx>
>     Reviewed-by: Felix Kuehling <Felix.Kuehling@xxxxxxx>
>     Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
I added Ma Jun to the email.

Regards,
  Felix


> struct crat_subtype_cache {
>          ...
>          uint8_t         sibling_map[CRAT_SIBLINGMAP_SIZE];
>
> #define CRAT_SIBLINGMAP_SIZE    32
>
>
> struct kfd_cache_properties {
>          ...
>          uint8_t                 sibling_map[CACHE_SIBLINGMAP_SIZE];
>
> #define CACHE_SIBLINGMAP_SIZE 64
>
> Thanks for your attention!