Re: [REGRESSION][RFC] memleak on xe load & unload on PTL

[Arunpravin Paneer Selvam <arunpravin.paneerselvam@amd.com>]

Hi Michal,

Please send the patch for the review.

Regards,
Arun.

On 11/18/2025 6:05 PM, Michał Grzelak wrote:
> Hi,
>
> just hit memory leak on xe module load & unload:
>
> unreferenced object 0xffff88811b047d10 (size 16):
>   comm "modprobe", pid 1058, jiffies 4297578480
>   hex dump (first 16 bytes):
>     00 6b 4b 2d 81 88 ff ff 80 7e 4b 2d 81 88 ff ff .kK-.....~K-....
>   backtrace (crc 4f169eaf):
>     kmemleak_alloc+0x4a/0x90
>     __kmalloc_cache_noprof+0x488/0x800
>     drm_buddy_init+0xc2/0x330 [drm_buddy]
>     __xe_ttm_vram_mgr_init+0xc3/0x190 [xe]
>     xe_ttm_stolen_mgr_init+0xf5/0x9d0 [xe]
>     xe_device_probe+0x326/0x9e0 [xe]
>     xe_pci_probe+0x39a/0x610 [xe]
>     local_pci_probe+0x47/0xb0
>     pci_device_probe+0xf3/0x260
>     really_probe+0xf1/0x3c0
>     __driver_probe_device+0x8c/0x180
>     driver_probe_device+0x24/0xd0
>     __driver_attach+0x10f/0x220
>     bus_for_each_dev+0x7f/0xe0
>     driver_attach+0x1e/0x30
>     bus_add_driver+0x151/0x290
>
> Issue was reproduced on PTL & BMG, booted with latest kernel from
> drm-tip. Looks like fault was introduced in commit d4cd665c9
> ("drm/buddy: Separate clear and dirty free block trees"), since 
> reverting it
> makes the leak disappear. Also attached RFC patch, which at first
> glance could fix the issue.
>
> Added xe maintainers and the author to Cc.
>
> BR,
> Michał