drm/i915: Take rpm wakelock for releasing the fence on unbind
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 6 Mar 2017 09:29:16 +0000 (09:29 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Mon, 6 Mar 2017 14:38:18 +0000 (14:38 +0000)
commit181df2d458f3e84eb4809083655fd2fd4ef33642
treef93a99c22f2e151605f26457fe0d7ff00e779ab0
parente1c0c91bdaec0702ebbfdc546926989dd63d19ef
drm/i915: Take rpm wakelock for releasing the fence on unbind

Unbind the vma may happen at any time, outside of the normal GT wakeref.
As such it relies on having a wakeref of its own. However, we can forgo
clearing the register whilst the device is asleep and just mark it as
unused - so that when we do wake up the device, we will clear the unused
fence register (see i915_gem_restore_fences).

[22423.944631] WARNING: CPU: 3 PID: 26178 at drivers/gpu/drm/i915/intel_drv.h:1739 i915_vma_put_fence+0xf3/0x100 [i915]
[22423.946053] RPM wakelock ref not held during HW access
[22423.946056] Modules linked in: vgem(E) i915(E) nls_ascii(E) nls_cp437(E) vfat(E) fat(E) x86_pkg_temp_thermal(E) crct10dif_pclmul(E) crc32_pclmul(E) crc32c_intel(E) ghash_clmulni_intel(E) intel_gtt(E) i2c_algo_bit(E) drm_kms_helper(E) syscopyarea(E) sysfillrect(E) evdev(E) aesni_intel(E) aes_x86_64(E) crypto_simd(E) cryptd(E) glue_helper(E) sysimgblt(E) fb_sys_fops(E) prime_numbers(E) drm(E) efivars(E) mei_me(E) lpc_ich(E) mei(E) mfd_core(E) battery(E) video(E) acpi_pad(E) button(E) tpm_tis(E) tpm_tis_core(E) tpm(E) autofs4(E) i2c_i801(E) thermal(E) fan(E) i2c_designware_platform(E) i2c_designware_core(E)
[22423.946438] CPU: 2 PID: 26178 Comm: gem_concurrent_ Tainted: G            E   4.10.0+ #101
[22423.946513] Hardware name: ��������������������������������� ���������������������������������/���������������������������������, BIOS RYBDWi35.86A.0246.2
[22423.946600] Call Trace:
[22423.946641]  dump_stack+0x68/0x9f
[22423.946703]  __warn+0x107/0x130
[22423.946763]  warn_slowpath_fmt+0xa8/0xe0
[22423.946825]  ? __warn+0x130/0x130
[22423.946868]  ? free_hot_cold_page_list+0x53/0x70
[22423.946942]  ? mark_lock+0xcc/0x7f0
[22423.946997]  ? __lock_is_held+0x84/0x100
[22423.947115]  ? i915_vma_put_fence+0x64/0x100 [i915]
[22423.947224]  i915_vma_put_fence+0xf3/0x100 [i915]
[22423.947335]  i915_vma_unbind+0x4da/0x560 [i915]
[22423.947387]  ? rb_erase+0x812/0x8a0
[22423.947439]  ? kfree+0xa2/0xd0
[22423.947562]  i915_vma_close+0x159/0x180 [i915]
[22423.947674]  intel_ring_free+0x31/0x50 [i915]
[22423.947776]  i915_gem_context_free+0x1ff/0x3d0 [i915]
[22423.947887]  context_close+0x106/0x110 [i915]
[22423.947989]  context_idr_cleanup+0xc/0x10 [i915]
[22423.948041]  idr_for_each+0x14d/0x1d0
[22423.948158]  ? context_close+0x110/0x110 [i915]
[22423.948206]  ? get_from_free_list+0x70/0x70
[22423.948261]  ? __lock_is_held+0x84/0x100
[22423.948325]  ? __mutex_unlock_slowpath+0xd4/0x400
[22423.948448]  i915_gem_context_close+0x4b/0x90 [i915]
[22423.948544]  i915_driver_preclose+0x28/0x50 [i915]
[22423.948620]  drm_release+0x175/0x690 [drm]
[22423.948681]  ? fcntl_setlk+0x5e0/0x5e0
[22423.948746]  __fput+0x17d/0x300
[22423.948807]  ____fput+0x9/0x10
[22423.948859]  task_work_run+0xa7/0xe0
[22423.948924]  do_exit+0x4d2/0x13e0
[22423.948986]  ? mm_update_next_owner+0x320/0x320
[22423.949051]  ? __do_page_fault+0x209/0x5c0
[22423.949110]  ? mark_held_locks+0x23/0xc0
[22423.949166]  ? entry_SYSCALL_64_fastpath+0x5/0xb1
[22423.949232]  do_group_exit+0x93/0x160
[22423.949289]  SyS_exit_group+0x18/0x20
[22423.949350]  entry_SYSCALL_64_fastpath+0x1c/0xb1
[22423.949403] RIP: 0033:0x7f9cc2e154c8
[22423.949484] RSP: 002b:00007ffd7e81b448 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7
[22423.949557] RAX: ffffffffffffffda RBX: ffffffff810ef1f0 RCX: 00007f9cc2e154c8
[22423.949617] RDX: 0000000000000000 RSI: 000000000000003c RDI: 0000000000000000
[22423.949677] RBP: ffff880367e9ff98 R08: 00000000000000e7 R09: ffffffffffffff88
[22423.949741] R10: 00007f9cc1d5c000 R11: 0000000000000246 R12: 00007f9cc30f6c30
[22423.949798] R13: 0000000000000000 R14: 00007f9cc30f6c20 R15: 0000000000000003
[22423.949868]  ? trace_hardirqs_off_caller+0xc0/0x110

v2: Move the rpm check down a layer so that we still perform the
vma/fence update required for the deferred mmio write on resume.
v3: Don't touch i915_gem_object_set_cache_level() and leave the rpm to
the low level routines (such as i915_vma_put_fence).
v4: vma may be null in fence_write, so extract drm_i915_private from
fence->i915

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170306092916.11623-3-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
drivers/gpu/drm/i915/i915_gem_fence_reg.c