Re: [Intel-xe] [PATCH 4/5] drm/xe: Prevent evicting for page tables

[Maarten Lankhorst <maarten.lankhorst@linux.intel.com>]

Hey,

On 2023-05-29 17:02, Thomas Hellström wrote:
>
> On 5/29/23 15:44, Maarten Lankhorst wrote:
>> On 2023-05-26 14:35, Thomas Hellström wrote:
>>> On 5/26/23 14:11, Maarten Lankhorst wrote:
>>>> When creating page tables from xe_exec_ioctl, we may end up freeing
>>>> memory we just validated. To be certain this does not happen, do not
>>>> allow the current reservation to be evicted from the ioctl.
>>>>
>>>> Callchain:
>>>> [  109.008522]  xe_bo_move_notify+0x5c/0xf0 [xe]
>>>> [  109.008548]  xe_bo_move+0x90/0x510 [xe]
>>>> [  109.008573]  ttm_bo_handle_move_mem+0xb7/0x170 [ttm]
>>>> [  109.008581]  ttm_bo_swapout+0x15e/0x360 [ttm]
>>>> [  109.008586]  ttm_device_swapout+0xc2/0x110 [ttm]
>>>> [  109.008592]  ttm_global_swapout+0x47/0xc0 [ttm]
>>>> [  109.008598]  ttm_tt_populate+0x7a/0x130 [ttm]
>>>> [  109.008603]  ttm_bo_handle_move_mem+0x160/0x170 [ttm]
>>>> [  109.008609]  ttm_bo_validate+0xe5/0x1d0 [ttm]
>>>> [  109.008614]  ttm_bo_init_reserved+0xac/0x190 [ttm]
>>>> [  109.008620]  __xe_bo_create_locked+0x153/0x260 [xe]
>>>> [  109.008645]  xe_bo_create_locked_range+0x77/0x360 [xe]
>>>> [  109.008671]  xe_bo_create_pin_map_at+0x33/0x1f0 [xe]
>>>> [  109.008695]  xe_bo_create_pin_map+0x11/0x20 [xe]
>>>> [  109.008721]  xe_pt_create+0x69/0xf0 [xe]
>>>> [  109.008749]  xe_pt_stage_bind_entry+0x208/0x430 [xe]
>>>> [  109.008776]  xe_pt_walk_range+0xe9/0x2a0 [xe]
>>>> [  109.008802]  xe_pt_walk_range+0x223/0x2a0 [xe]
>>>> [  109.008828]  xe_pt_walk_range+0x223/0x2a0 [xe]
>>>> [  109.008853]  __xe_pt_bind_vma+0x28d/0xbd0 [xe]
>>>> [  109.008878]  xe_vm_bind_vma+0xc7/0x2f0 [xe]
>>>> [  109.008904]  xe_vm_rebind+0x72/0x160 [xe]
>>>> [  109.008930]  xe_exec_ioctl+0x22b/0xa70 [xe]
>>>> [  109.008955]  drm_ioctl_kernel+0xb9/0x150 [drm]
>>>> [  109.008972]  drm_ioctl+0x210/0x430 [drm]
>>>> [  109.008988]  __x64_sys_ioctl+0x85/0xb0
>>>> [  109.008990]  do_syscall_64+0x38/0x90
>>>> [  109.008991]  entry_SYSCALL_64_after_hwframe+0x72/0xdc
>>>>
>>>> Original warning:
>>>> [ 5613.149126] WARNING: CPU: 3 PID: 45883 at drivers/gpu/drm/xe/xe_vm.c:504 xe_vm_unlock_dma_resv+0x43/0x50 [xe]
>>>> ...
>>>> [ 5613.226398] RIP: 0010:xe_vm_unlock_dma_resv+0x43/0x50 [xe]
>>>> [ 5613.316098] Call Trace:
>>>> [ 5613.318595]  <TASK>
>>>> [ 5613.320743]  xe_exec_ioctl+0x383/0x8a0 [xe]
>>>> [ 5613.325278]  ? __is_insn_slot_addr+0x8e/0x110
>>>> [ 5613.329719]  ? __is_insn_slot_addr+0x8e/0x110
>>>> [ 5613.334116]  ? kernel_text_address+0x75/0xf0
>>>> [ 5613.338429]  ? __pfx_stack_trace_consume_entry+0x10/0x10
>>>> [ 5613.343778]  ? __kernel_text_address+0x9/0x40
>>>> [ 5613.348181]  ? unwind_get_return_address+0x1a/0x30
>>>> [ 5613.353013]  ? __pfx_stack_trace_consume_entry+0x10/0x10
>>>> [ 5613.358362]  ? arch_stack_walk+0x99/0xf0
>>>> [ 5613.362329]  ? rcu_read_lock_sched_held+0xb/0x70
>>>> [ 5613.366996]  ? lock_acquire+0x287/0x2f0
>>>> [ 5613.370873]  ? rcu_read_lock_sched_held+0xb/0x70
>>>> [ 5613.375530]  ? rcu_read_lock_sched_held+0xb/0x70
>>>> [ 5613.380181]  ? lock_release+0x225/0x2e0
>>>> [ 5613.384059]  ? __pfx_xe_exec_ioctl+0x10/0x10 [xe]
>>>> [ 5613.389092]  drm_ioctl_kernel+0xc0/0x170
>>>> [ 5613.393068]  drm_ioctl+0x1b7/0x490
>>>> [ 5613.396519]  ? __pfx_xe_exec_ioctl+0x10/0x10 [xe]
>>>> [ 5613.401547]  ? lock_release+0x225/0x2e0
>>>> [ 5613.405432]  __x64_sys_ioctl+0x8a/0xb0
>>>> [ 5613.409232]  do_syscall_64+0x37/0x90
>>>>
>>>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
>>>> Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/239
>>> Did you look at passing around the ttm_operation_ctx, or a "allow_res_evict" bool?
>>> In any case would be good to have this fixed asap, so
>>>
>>> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>>>
>> I considered it, but the original callchain was too long. I don't think there is any usecase
>> in which we want to evict from the current context to make room for new pagetables for VM_BIND.
>> Anything locked is most likely used, making room by evicting from current VM (or its bound extobjs)
>> will likely lead to ENOSPC anyway.
>
> Well, I think the use-case where this will cause problems is if we're doing a single VM_BIND on a brand new VRAM BO, and need to evict other VRAM bos from the same VM to make room.
>
> This will then ofc ENOSPC on the next exec, but if we were to introduce a two-pass validation scheme, where we explicitly move suitable BOs with multiple placement options to TT on the first ENOSPC, we could avoid that...

Allowing same-reservation eviction will allow you to evict the BO its VM_BIND page table, leaving no entries to write. :-)

~Maarten