[Intel-gfx] [PATCH] drm/i915: Mark the userptr invalidate workqueue as WQ_MEM_RECLAIM

Fri Nov 17 11:00:13 UTC 2017

Quoting Matthew Auld (2017-11-17 10:56:54)
> On 14 November 2017 at 17:35, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Commit  21cc6431e0c2 ("drm/i915: Mark the userptr invalidate workqueue
> > as WQ_MEM_RECLAIM") tried to fixup the check_flush_dependency warning
> > for hitting i915_gem_userptr_mn_invalidate_range_start from within the
> > shrinker, but I failed to notice userptr has 2 similarly named
> > workqueues. I marked up i915-userptr-acquire as WQ_MEM_RECLAIM whereas
> > we only wait upon i915-userptr-release from inside the reclaim paths.
> >
> > [62530.869510] workqueue: PF_MEMALLOC task 7983(gem_shrink) is flushing !WQ_MEM_RECLAIM i915-userptr-release:          (null)
> > [62530.869515] ------------[ cut here ]------------
> > [62530.869519] WARNING: CPU: 1 PID: 7983 at kernel/workqueue.c:2434 check_flush_dependency+0x7f/0x110
> > [62530.869519] Modules linked in: pegasus mii ip6table_filter ip6_tables bnep iptable_filter snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic binfmt_misc nls_iso8859_1 intel_rapl x86_pkg_temp_thermal intel_powerclamp coretemp snd_hda_intel snd_hda_codec kvm_intel snd_hda_core snd_hwdep kvm snd_pcm irqbypass snd_seq_midi snd_seq_midi_event snd_rawmidi crct10dif_pclmul crc32_pclmul 8250_dw ghash_clmulni_intel snd_seq pcbc snd_seq_device snd_timer btusb aesni_intel btrtl btbcm aes_x86_64 iwlwifi btintel crypto_simd glue_helper cryptd bluetooth snd intel_cstate input_leds idma64 intel_rapl_perf ecdh_generic serio_raw soundcore cfg80211 wmi_bmof virt_dma intel_lpss_pci intel_lpss acpi_als kfifo_buf industrialio winbond_cir soc_button_array rc_core spidev tpm_crb intel_hid acpi_pad mac_hid sparse_keymap
> > [62530.869546]  parport_pc ppdev lp parport ip_tables x_tables autofs4 hid_generic usbhid i915 i2c_algo_bit prime_numbers drm_kms_helper syscopyarea e1000e sysfillrect sysimgblt fb_sys_fops ahci ptp pps_core libahci drm wmi video i2c_hid hid
> > [62530.869557] CPU: 1 PID: 7983 Comm: gem_shrink Tainted: G     U  W    L  4.14.0-rc8-drm-tip-ww45-commit-1342299+ #1
> > [62530.869558] Hardware name: Intel Corporation CoffeeLake Client Platform/CoffeeLake H DDR4 RVP, BIOS CNLSFWR1.R00.X098.A00.1707301945 07/30/2017
> > [62530.869559] task: ffffa1049dbeec80 task.stack: ffffae7d05c44000
> > [62530.869560] RIP: 0010:check_flush_dependency+0x7f/0x110
> > [62530.869561] RSP: 0018:ffffae7d05c473a0 EFLAGS: 00010286
> > [62530.869562] RAX: 000000000000006e RBX: ffffa1049540f400 RCX: ffffffffa3e55788
> > [62530.869562] RDX: 0000000000000000 RSI: 0000000000000092 RDI: 0000000000000202
> > [62530.869563] RBP: ffffae7d05c473c0 R08: 000000000000006e R09: 000000000038bb0e
> > [62530.869563] R10: 0000000000000000 R11: 000000000000006e R12: ffffa1049dbeec80
> > [62530.869564] R13: 0000000000000000 R14: 0000000000000000 R15: ffffae7d05c473e0
> > [62530.869565] FS:  00007f621b129880(0000) GS:ffffa1050b240000(0000) knlGS:0000000000000000
> > [62530.869566] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [62530.869566] CR2: 00007f6214400000 CR3: 0000000353a17003 CR4: 00000000003606e0
> > [62530.869567] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [62530.869567] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > [62530.869568] Call Trace:
> > [62530.869570]  flush_workqueue+0x115/0x3d0
> > [62530.869573]  ? wake_up_process+0x15/0x20
> > [62530.869596]  i915_gem_userptr_mn_invalidate_range_start+0x12f/0x160 [i915]
> > [62530.869614]  ? i915_gem_userptr_mn_invalidate_range_start+0x12f/0x160 [i915]
> > [62530.869616]  __mmu_notifier_invalidate_range_start+0x55/0x80
> > [62530.869618]  try_to_unmap_one+0x791/0x8b0
> > [62530.869620]  ? call_rwsem_down_read_failed+0x18/0x30
> > [62530.869622]  rmap_walk_anon+0x10b/0x260
> > [62530.869624]  rmap_walk+0x48/0x60
> > [62530.869625]  try_to_unmap+0x93/0xf0
> > [62530.869626]  ? page_remove_rmap+0x2a0/0x2a0
> > [62530.869627]  ? page_not_mapped+0x20/0x20
> > [62530.869629]  ? page_get_anon_vma+0x90/0x90
> > [62530.869630]  ? invalid_mkclean_vma+0x20/0x20
> > [62530.869631]  migrate_pages+0x946/0xaa0
> > [62530.869633]  ? __ClearPageMovable+0x10/0x10
> > [62530.869635]  ? isolate_freepages_block+0x3c0/0x3c0
> > [62530.869636]  compact_zone+0x22f/0x970
> > [62530.869638]  compact_zone_order+0xa3/0xd0
> > [62530.869640]  try_to_compact_pages+0x1a5/0x2a0
> > [62530.869641]  ? try_to_compact_pages+0x1a5/0x2a0
> > [62530.869643]  __alloc_pages_direct_compact+0x50/0x110
> > [62530.869644]  __alloc_pages_slowpath+0x4da/0xf30
> > [62530.869646]  __alloc_pages_nodemask+0x262/0x280
> > [62530.869648]  alloc_pages_vma+0x165/0x1e0
> > [62530.869649]  shmem_alloc_hugepage+0xd0/0x130
> > [62530.869651]  ? __radix_tree_insert+0x45/0x230
> > [62530.869652]  ? __vm_enough_memory+0x29/0x130
> > [62530.869654]  shmem_alloc_and_acct_page+0x10d/0x1e0
> > [62530.869655]  shmem_getpage_gfp+0x426/0xc00
> > [62530.869657]  shmem_fault+0xa0/0x1e0
> > [62530.869659]  ? file_update_time+0x60/0x110
> > [62530.869660]  __do_fault+0x1e/0xc0
> > [62530.869661]  __handle_mm_fault+0xa35/0x1170
> > [62530.869662]  handle_mm_fault+0xcc/0x1c0
> > [62530.869664]  __do_page_fault+0x262/0x4f0
> > [62530.869666]  do_page_fault+0x2e/0xe0
> > [62530.869667]  page_fault+0x22/0x30
> > [62530.869668] RIP: 0033:0x404335
> > [62530.869669] RSP: 002b:00007fff7829e420 EFLAGS: 00010216
> > [62530.869670] RAX: 00007f6210400000 RBX: 0000000000000004 RCX: 0000000000b80000
> > [62530.869670] RDX: 0000000000002e01 RSI: 0000000000008000 RDI: 0000000000000004
> > [62530.869671] RBP: 0000000000000019 R08: 0000000000000002 R09: 0000000000000000
> > [62530.869671] R10: 0000000000000559 R11: 0000000000000246 R12: 0000000008000000
> > [62530.869672] R13: 00000000004042f0 R14: 0000000000000004 R15: 000000000000007e
> > [62530.869673] Code: 00 8b b0 18 05 00 00 48 8d 8b b0 00 00 00 48 8d 90 c0 06 00 00 4d 89 f0 48 c7 c7 40 c0 c8 a3 c6 05 68 c5 e8 00 01 e8 c2 68 04 00 <0f> ff 4d 85 ed 74 18 49 8b 45 20 48 8b 70 08 8b 86 00 01 00 00
> > [62530.869691] ---[ end trace 01e01ad0ff5781f8 ]---
> >
> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103739
> > Fixes: 21cc6431e0c2 ("drm/i915: Mark the userptr invalidate workqueue as WQ_MEM_RECLAIM")
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Michał Winiarski <michal.winiarski at intel.com>
> Reviewed-by: Matthew Auld <matthew.auld at intel.com>

Thank you, another bugzilla bites the dust!
-Chris