Quoting Bruce Chang (2019-11-13 23:11:04)
> below is the call trace when this issue is hit
>
> <3> [113.316247] BUG: sleeping function called from invalid context at mm/page_alloc.c:4653
> <3> [113.318190] in_atomic(): 1, irqs_disabled(): 0, pid: 678, name: debugfs_test
> <4> [113.319900] no locks held by debugfs_test/678.
> <3> [113.321002] Preemption disabled at:
> <4> [113.321130] [<ffffffffa02506d4>] i915_error_object_create+0x494/0x610 [i915]
> <4> [113.327259] Call Trace:
> <4> [113.327871] dump_stack+0x67/0x9b
> <4> [113.328683] ___might_sleep+0x167/0x250
> <4> [113.329618] __alloc_pages_nodemask+0x26b/0x1110
> <4> [113.330731] ? ___slab_alloc.constprop.34+0x21c/0x380
> <4> [113.331943] ? ___slab_alloc.constprop.34+0x21c/0x380
> <4> [113.333169] ? __slab_alloc.isra.28.constprop.33+0x4d/0x70
> <4> [113.334614] pool_alloc.constprop.19+0x14/0x60 [i915]
> <4> [113.335951] compress_page+0x7c/0x100 [i915]
> <4> [113.337110] i915_error_object_create+0x4bd/0x610 [i915]
> <4> [113.338515] i915_capture_gpu_state+0x384/0x1680 [i915]
> <4> [113.339771] ? __lock_acquire+0x4ac/0x1e90
> <4> [113.340785] ? _raw_spin_lock_irqsave_nested+0x1/0x50
> <4> [113.342127] i915_gpu_info_open+0x44/0x70 [i915]
> <4> [113.343243] full_proxy_open+0x139/0x1b0
> <4> [113.344196] ? open_proxy_open+0xc0/0xc0
> <4> [113.345149] do_dentry_open+0x1ce/0x3a0
> <4> [113.346084] path_openat+0x4c9/0xac0
> <4> [113.346967] do_filp_open+0x96/0x110
> <4> [113.347848] ? __alloc_fd+0xe0/0x1f0
> <4> [113.348736] ? do_sys_open+0x1b8/0x250
> <4> [113.349647] do_sys_open+0x1b8/0x250
> <4> [113.350526] do_syscall_64+0x55/0x1c0
> <4> [113.351418] entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> After the io_mapping_map_atomic_wc/kmap_atomic, the kernel enters atomic context
> but after that, compress_page calls pool_alloc with GFP_KERNEL flag which can
> potentially go to sleep. When the kernel is in atomic context, sleeping is not
> allowed. This is why this bug got triggered.
The last 2 sentences are redundant.
> In order to fix this issue, we either
> 1) not enter into atomic context, i.e., to use non atomic version of
> functions like io_mapping_map_wc/kmap,
> or
> 2) make compress_page run in atomic context.
>
> But it is not a good idea to run slow compression in atomic context, so,
> 1) above is preferred solution which is the implementation of this patch.
Reasonable, though we have and may have to do capture inside atomic
again. (Dropping the atomicity is a recent change that has a surprising
amount of controversy.)
> Signed-off-by: Bruce Chang <yu.bruce.chang@xxxxxxxxx>
> Reviewed-by: Brian Welty <brian.welty@xxxxxxxxx>
> Fixes: 895d8ebeaa924 ("drm/i915: error capture with no ggtt slot")
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
