On Fri, 6 Mar 2026 11:58:11 +0000 Steven Price <[email protected]> wrote:
> Hi Boris, > > On 05/03/2026 12:43, Boris Brezillon wrote: > > Hello, > > > > This is an attempt at adding a GEM shrinker to panthor so the system > > can finally reclaim GPU memory. > > > > This implementation is losely based on the MSM shrinker (which is why > > I added the MSM maintainers in Cc), and it's relying on the drm_gpuvm > > eviction/validation infrastructure. > > > > I've only done very basic IGT-based [1] and chromium-based (opening > > a lot of tabs on Aquarium until the system starts reclaiming+swaping > > out GPU buffers) testing, but I'm posting this early so I can get > > preliminary feedback on the implementation. If someone knows about > > better tools/ways to test the shrinker, please let me know. > > I did a very basic test with glmark and I can reproduce the below splat: > > [ 290.502999] ------------[ cut here ]------------ > [ 290.504338] refcount_t: underflow; use-after-free. > [ 290.504843] WARNING: lib/refcount.c:28 at > refcount_warn_saturate+0xf4/0x144, CPU#5: kworker/u32:3/75 > [ 290.505715] Modules linked in: panthor drm_gpuvm drm_exec gpu_sched > [ 290.506402] CPU: 5 UID: 0 PID: 75 Comm: kworker/u32:3 Not tainted > 7.0.0-rc1-00176-g608e8196cd63 #202 PREEMPT > [ 290.507323] Hardware name: Radxa ROCK 5B (DT) > [ 290.507733] Workqueue: events_unbound commit_work > [ 290.508185] pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) > [ 290.508835] pc : refcount_warn_saturate+0xf4/0x144 > [ 290.509287] lr : refcount_warn_saturate+0xf4/0x144 > [ 290.509741] sp : ffff800083cb3b80 > [ 290.510056] x29: ffff800083cb3b80 x28: ffff8000821d1e88 x27: > ffff00010fa058a0 > [ 290.510724] x26: 0000000000000000 x25: 0000000000000000 x24: > 00000000ffffffff > [ 290.511398] x23: ffff00010b149000 x22: ffff00010dd3a7c8 x21: > ffff80008226c828 > [ 290.512065] x20: ffff00010dd3a780 x19: ffff00010dd3a780 x18: > 00000000ffffffff > [ 290.512735] x17: 00000000ffffffff x16: ffff800083cb3668 x15: > 0000000000001e00 > [ 290.513403] x14: ffff000102a8f69f x13: ffff8000821fb558 x12: > 000000000000083d > [ 290.514074] x11: 00000000000002bf x10: ffff800082253558 x9 : > ffff8000821fb558 > [ 290.514746] x8 : 00000000ffffefff x7 : ffff800082253558 x6 : > 80000000fffff000 > [ 290.515414] x5 : ffff0001fef31588 x4 : 0000000000000000 x3 : > ffff80017d1e5000 > [ 290.516083] x2 : 0000000000000000 x1 : 0000000000000000 x0 : > ffff000102b61980 > [ 290.516752] Call trace: > [ 290.516987] refcount_warn_saturate+0xf4/0x144 (P) > [ 290.517440] drm_sched_fence_release_scheduled+0xe0/0xe4 [gpu_sched] > [ 290.518046] dma_fence_release+0xb4/0x3cc > [ 290.518429] drm_sched_fence_release_finished+0x94/0xa8 [gpu_sched] > [ 290.519021] dma_fence_release+0xb4/0x3cc > [ 290.519401] dma_fence_array_release+0x94/0x104 > [ 290.519829] dma_fence_release+0xb4/0x3cc > [ 290.520208] drm_atomic_helper_wait_for_fences+0x1a4/0x228 > [ 290.520724] commit_tail+0x38/0x18c > [ 290.521056] commit_work+0x14/0x20 > [ 290.521381] process_one_work+0x208/0x76c > [ 290.521763] worker_thread+0x1c4/0x36c > [ 290.522121] kthread+0x13c/0x148 > [ 290.522430] ret_from_fork+0x10/0x20 > [ 290.522774] irq event stamp: 2167444 > [ 290.523114] hardirqs last enabled at (2167443): [<ffff80008016772c>] > __up_console_sem+0x6c/0x80 > [ 290.523941] hardirqs last disabled at (2167444): [<ffff80008132977c>] > el1_brk64+0x20/0x60 > [ 290.524703] softirqs last enabled at (2167428): [<ffff8000800c94c4>] > handle_softirqs+0x604/0x61c > [ 290.525528] softirqs last disabled at (2167421): [<ffff8000800102d0>] > __do_softirq+0x14/0x20 > [ 290.526320] ---[ end trace 0000000000000000 ]--- > > I haven't yet dug into what's happening. Yikes! I plead guilty for not testing after the rebase...
