drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
Subsystems (2)
PATCH 1/2 drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding mmap_sem is a no-go, so let's try to preallocate it.
Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through msr_create_device.
Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>] apply_workqueue_attrs+0x17/0x50
but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>] i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_nested+0x1b/0x20
i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
i915_gem_userptr_ioctl+0x222/0x2c0 [i915]
drm_ioctl_kernel+0x69/0xb0
drm_ioctl+0x2f9/0x3d0
do_vfs_ioctl+0x94/0x670
SyS_ioctl+0x41/0x70
entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__might_fault+0x68/0x90
_copy_to_user+0x23/0x70
filldir+0xa5/0x120
dcache_readdir+0xf9/0x170
iterate_dir+0x69/0x1a0
SyS_getdents+0xa5/0x140
entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70
handle_create+0xcb/0x1e0
devtmpfsd+0x139/0x180
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
wait_for_common+0x58/0x210
wait_for_completion+0x1d/0x20
devtmpfs_create_node+0x13d/0x160
device_add+0x5eb/0x620
device_create_groups_vargs+0xe0/0xf0
device_create+0x3a/0x40
msr_device_create+0x2b/0x40
cpuhp_invoke_callback+0xa3/0x840
cpuhp_thread_fun+0x7a/0x150
smpboot_thread_fn+0x18a/0x280
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpuhp_issue_call+0x10b/0x170
__cpuhp_setup_state_cpuslocked+0x134/0x2a0
__cpuhp_setup_state+0x46/0x60
page_writeback_init+0x43/0x67
pagecache_init+0x3d/0x42
start_kernel+0x3a8/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_nested+0x1b/0x20
__cpuhp_setup_state_cpuslocked+0x52/0x2a0
__cpuhp_setup_state+0x46/0x60
page_alloc_init+0x28/0x30
start_kernel+0x145/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpus_read_lock+0x3d/0xb0
apply_workqueue_attrs+0x17/0x50
__alloc_workqueue_key+0x1d8/0x4d9
i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915]
i915_gem_userptr_ioctl+0x222/0x2c0 [i915]
drm_ioctl_kernel+0x69/0xb0
drm_ioctl+0x2f9/0x3d0
do_vfs_ioctl+0x94/0x670
SyS_ioctl+0x41/0x70
entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&dev_priv->mm_lock);
lock(&mm->mmap_sem);
lock(&dev_priv->mm_lock); lock(cpu_hotplug_lock.rw_sem);
*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>] i915_gem_userptr_init__mmu_notifier+0x138/0x270 [i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>] i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20
v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939
--- drivers/gpu/drm/i915/i915_gem_userptr.c | 35 +++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 15 deletions(-)
drivers/gpu/drm/i915/i915_gem_userptr.c +20 -15
@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )
@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }
@@ -210,23 +201,37 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn; +struct i915_mmu_notifier *mn; +int err; mn = mm->mn;if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) +return mn; + + err =0 ;down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) { - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn; +if (mm->mn ==NULL ) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } }mutex_unlock (&mm->i915->mm_lock);up_write (&mm->mm->mmap_sem); -return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static int
Signature · 6 lines
2.14.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 2/2 drm/i915: Use rcu instead of stop_machine in set_wedged
stop_machine is not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.
This patch tries to address the locking abuse of stop_machine() from
commit 20e4933c478a1ca694b38fa4ac44d99e659941f5
Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000
drm/i915: Stop the machine as we install the wedged submit_request handler
Chris said parts of the reasons for going with stop_machine() was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, and rcu_read_lock is _real_ fast.
To stay as close as possible to the stop_machine semantics we first update all the submit function pointers to the nop handler, then call synchronize_rcu() to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.
I pondered whether we should annotate engine->submit_request as __rcu and use rcu_assign_pointer and rcu_dereference on it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.
This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>] stop_machine+0x1c/0x40
but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_interruptible_nested+0x1b/0x20
i915_mutex_lock_interruptible+0x51/0x130 [i915]
i915_gem_fault+0x209/0x650 [i915]
__do_fault+0x1e/0x80
__handle_mm_fault+0xa08/0xed0
handle_mm_fault+0x156/0x300
__do_page_fault+0x2c5/0x570
do_page_fault+0x28/0x250
page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__might_fault+0x68/0x90
_copy_to_user+0x23/0x70
filldir+0xa5/0x120
dcache_readdir+0xf9/0x170
iterate_dir+0x69/0x1a0
SyS_getdents+0xa5/0x140
entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70
handle_create+0xcb/0x1e0
devtmpfsd+0x139/0x180
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
wait_for_common+0x58/0x210
wait_for_completion+0x1d/0x20
devtmpfs_create_node+0x13d/0x160
device_add+0x5eb/0x620
device_create_groups_vargs+0xe0/0xf0
device_create+0x3a/0x40
msr_device_create+0x2b/0x40
cpuhp_invoke_callback+0xc9/0xbf0
cpuhp_thread_fun+0x17b/0x240
smpboot_thread_fn+0x18a/0x280
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpuhp_issue_call+0x133/0x1c0
__cpuhp_setup_state_cpuslocked+0x139/0x2a0
__cpuhp_setup_state+0x46/0x60
page_writeback_init+0x43/0x67
pagecache_init+0x3d/0x42
start_kernel+0x3a8/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_nested+0x1b/0x20
__cpuhp_setup_state_cpuslocked+0x53/0x2a0
__cpuhp_setup_state+0x46/0x60
page_alloc_init+0x28/0x30
start_kernel+0x145/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpus_read_lock+0x3d/0xb0
stop_machine+0x1c/0x40
i915_gem_set_wedged+0x1a/0x20 [i915]
i915_reset+0xb9/0x230 [i915]
i915_reset_device+0x1f6/0x260 [i915]
i915_handle_error+0x2d8/0x430 [i915]
hangcheck_declare_hang+0xd3/0xf0 [i915]
i915_hangcheck_elapsed+0x262/0x2d0 [i915]
process_one_work+0x233/0x660
worker_thread+0x4e/0x3b0
kthread+0x152/0x190
ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&dev->struct_mutex);
lock(&mm->mmap_sem);
lock(&dev->struct_mutex); lock(cpu_hotplug_lock.rw_sem);
*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660
#1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660
#2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
v2: Have 1 global synchronize_rcu() barrier across all engines, and improve commit message.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096
--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +12 -19
@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ - engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);
@@ -3041,24 +3033,25 @@ static void engine_set_wedged(struct intel_engine_cs *engine) intel_engine_last_submit(engine)); } -static int __i915_gem_set_wedged_BKL (void *data) +void i915_gem_set_wedged (struct drm_i915_private *i915) { -struct drm_i915_private *i915 = data;struct intel_engine_cs *engine;enum intel_engine_id id;for_each_engine (engine, i915, id) -engine_set_wedged (engine); + engine->submit_request = nop_submit_request; -set_bit (I915_WEDGED , &i915->gpu_error.flags); -wake_up_all (&i915->gpu_error.reset_queue); +/* Make sure no one is running the old callback before we proceed with +* cancelling requests and resetting the completion tracking. Otherwise +* we might submit a request to the hardware which never completes. +*/ +synchronize_rcu (); -return 0 ; -} +for_each_engine (engine, i915, id) +engine_complete_requests (engine); -void i915_gem_set_wedged (struct drm_i915_private *dev_priv) -{ -stop_machine (__i915_gem_set_wedged_BKL, dev_priv,NULL ); +set_bit (I915_WEDGED , &i915->gpu_error.flags); +wake_up_all (&i915->gpu_error.reset_queue); }bool i915_gem_unset_wedged (struct drm_i915_private *i915)
drivers/gpu/drm/i915/i915_gem_request.c +2 -0
@@ -556,7 +556,9 @@ submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)switch (state) {case FENCE_COMPLETE :trace_i915_gem_request_submit (request); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();break ;case FENCE_FREE :
drivers/gpu/drm/i915/selftests/i915_gem_request.c +2 -0
@@ -215,7 +215,9 @@ static int igt_request_rewind(void *arg) }i915_gem_request_get (vip);i915_add_request (vip); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();mutex_unlock (&i915->drm.struct_mutex);
Signature · 6 lines
2.14.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 7 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5
Quoted 8 lines
Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.
I still want a discussion of the reason why keeping the normal path clean and why an alternative is sought, here. That design leads into vv
Quoted 288 lines
To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +12
-19@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ - engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);@@ -3041,24 +3033,25 @@ static void engine_set_wedged(struct intel_engine_cs *engine) intel_engine_last_submit(engine)); } -static int __i915_gem_set_wedged_BKL (void *data) +void i915_gem_set_wedged (struct drm_i915_private *i915) { -struct drm_i915_private *i915 = data;struct intel_engine_cs *engine;enum intel_engine_id id;for_each_engine (engine, i915, id) -engine_set_wedged (engine); + engine->submit_request = nop_submit_request; -set_bit (I915_WEDGED , &i915->gpu_error.flags); -wake_up_all (&i915->gpu_error.reset_queue); +/* Make sure no one is running the old callback before we proceed with +* cancelling requests and resetting the completion tracking. Otherwise +* we might submit a request to the hardware which never completes. +*/ +synchronize_rcu (); -return 0 ; -} +for_each_engine (engine, i915, id) +engine_complete_requests (engine); -void i915_gem_set_wedged (struct drm_i915_private *dev_priv) -{ -stop_machine (__i915_gem_set_wedged_BKL, dev_priv,NULL ); +set_bit (I915_WEDGED , &i915->gpu_error.flags); +wake_up_all (&i915->gpu_error.reset_queue); }bool i915_gem_unset_wedged (struct drm_i915_private *i915)drivers/gpu/drm/i915/i915_gem_request.c +2
-0@@ -556,7 +556,9 @@ submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)switch (state) {case FENCE_COMPLETE :trace_i915_gem_request_submit (request); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();break ;case FENCE_FREE :drivers/gpu/drm/i915/selftests/i915_gem_request.c +2
-0@@ -215,7 +215,9 @@ static int igt_request_rewind(void *arg) }i915_gem_request_get (vip);i915_add_request (vip); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();mutex_unlock (&i915->drm.struct_mutex);Signature · 1 line
2.14.1
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoted 11 lines · Chris Wilson
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 10 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000
Quoted 8 lines
Quoted 5 lines
drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.I still want a discussion of the reason why keeping the normal path clean and why an alternative is sought, here. That design leads into vv
stop_machine() is the least resort when serialization problems cannot be solved otherwise. We try to avoid it where ever we can. While on the call site it looks simple, it's invasive in terms of locking as shown by the lockdep splat and it's imposing latencies and other side effects on all CPUs in the system. So if you don't have a compelling technical reason to use it, then it _is_ the wrong tool.
As Daniel has shown it's not required, so there is no technical reason why stomp_machine() has to be used here.
Thanks,
tglx
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoted 2 lines · Thomas Gleixner
So if you don't have a compelling technical reason to use it, then it _is_ the wrong tool.
This. stop_machine() effectively takes down _all_ CPUs for the duration of your callback. That is something you really should avoid at pretty much any cost.
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
In reply to Thomas GleixnerQuoted 28 lines · Thomas Gleixner
Quoted 11 lines · Chris Wilson
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 10 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000Quoted 8 lines
Quoted 5 lines
drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.I still want a discussion of the reason why keeping the normal path clean and why an alternative is sought, here. That design leads into vv
stop_machine()is the least resort when serialization problems cannot be solved otherwise. We try to avoid it where ever we can. While on the call site it looks simple, it's invasive in terms of locking as shown by the lockdep splat and it's imposing latencies and other side effects on all CPUs in the system. So if you don't have a compelling technical reason to use it, then it _is_ the wrong tool.
Quoted 2 lines
As Daniel has shown it's not required, so there is no technical reason why
stomp_machine()has to be used here.
Well I'm not sure yet whether my fix is actually correct :-)
But imo there's a bunch more reason why stop_machine is uncool, beyond just the "it's a huge shotgun which doesn't play well with anything else" aspect:
- What we actually seem to want is to make sure that all the engine->submit_request have completed, which happen to all run in hardirq context. It's an artifact of stop_machine that it completes all hardirq handlers, but afaiui stop_machine is really just aimed at getting all cpus to execute a specific well know loop (so that your callback can start patching .text and other evil stuff). If we move our callback into a thread that gets preempted, we have a problem.
- As a consequence, no lockdep annotations for the locking we actually want. And since this is for gpu hang recovery (something relatively rare that just _has_ to work) we really need all the support from all the debug tools we can get to catch possible issues.
- Another consequence is that the read side critical sections aren't annotated in the code. That makes it ever so more likely that a redesign moves them out of hardirq context and breaks it all.
- Not relevant here (I think), but stop_machine doesn't remove the need for read-side (compiler) barriers. Not relevant here I think, but in other cases we might still need to sprinkle READ_ONCE all over to make sure gcc doesn't realod and create races that way.
rcu has all these bits covered, is maintained by very smart people, and the overhead is somewhere between 0 and a cacheline access that we touch anyway (preempt_count is also wrangled by our spinlocks in all the callbacks). No way this will ever show up against all the mmio writes the callback does anyway.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
In reply to Daniel VetterQuoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 226 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) {
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +0
-8-/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ - engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);
How are we planning to serialise the intel_engine_init_global_seqno() here with the in-flight nop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.
-Chris
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoted 233 lines · Chris Wilson
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 232 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-7@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +0
-2- engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);How are we planning to serialise the
intel_engine_init_global_seqno()here with the in-flightnop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.
Yeah that part looks indeed fishy. Well the entire "let the nop handler fake-complete requests" logic is something I don't really understand. I guess there's an exclusive relationship between requests handled directly (and cancelled in engine->cancel_request) and requests with external dma_fence dependencies.
But then I'm not really seeing what I'm changing, since even with the stop machine you might end up with a bunch of requests depending upon external fences, which then all complete at roughly the same time and race multiple calls to intel_engine_init_global_seqno with each another.
With the fake submission, do we really need to call intel_engine_init_global_seqno?
So yeah, no idea, but pretty sure I didn't make it worse. -Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoting Daniel Vetter (2017-10-06 15:20:09)
Quoted 249 lines · Chris Wilson
Quoted 233 lines
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 232 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-7@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +0
-2- engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);How are we planning to serialise the
intel_engine_init_global_seqno()here with the in-flightnop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.Yeah that part looks indeed fishy. Well the entire "let the nop handler fake-complete requests" logic is something I don't really understand. I guess there's an exclusive relationship between requests handled directly (and cancelled in
engine->cancel_request) and requests with externaldma_fencedependencies.
Quoted 4 lines
But then I'm not really seeing what I'm changing, since even with the stop machine you might end up with a bunch of requests depending upon external fences, which then all complete at roughly the same time and race multiple calls to
intel_engine_init_global_seqnowith each another.
The stop_machine serialised the update here with the nop_handlers, that's the bit that changes.
Quoted 1 line
With the fake submission, do we really need to call
intel_engine_init_global_seqno?
Yes. Completion is still determined by i915_seqno_passed() comparing the rq against the engine.
You need this
@@ -3246,6 +3246,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)
static void engine_set_wedged(struct intel_engine_cs *engine)
{
+ unsigned long flags;
+
/* We need to be sure that no thread is running the old callback as
* we install the nop handler (otherwise we would submit a request
* to hardware that will never complete). In order to prevent this
@@ -3261,8 +3263,10 @@ static void engine_set_wedged(struct intel_engine_cs *engine)
* (lockless) lookup doesn't try and wait upon the request as we
* reset it.
*/
+ spin_lock_irqsave(&request->engine->timeline->lock, flags);
intel_engine_init_global_seqno(engine,
intel_engine_last_submit(engine));+ spin_unlock_irqrestore(&request->engine->timeline->lock, flags);
}
So that the seqno written is ordered with the same spinlock used inside the nop submission.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoted 284 lines · Chris Wilson
Quoting Daniel Vetter (2017-10-06 15:20:09)
Quoted 249 lines · Chris Wilson
Quoted 233 lines
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 232 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-7@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +0
-2- engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);How are we planning to serialise the
intel_engine_init_global_seqno()here with the in-flightnop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.Yeah that part looks indeed fishy. Well the entire "let the nop handler fake-complete requests" logic is something I don't really understand. I guess there's an exclusive relationship between requests handled directly (and cancelled in
engine->cancel_request) and requests with externaldma_fencedependencies.Quoted 4 lines
But then I'm not really seeing what I'm changing, since even with the stop machine you might end up with a bunch of requests depending upon external fences, which then all complete at roughly the same time and race multiple calls to
intel_engine_init_global_seqnowith each another.The
stop_machineserialised the update here with thenop_handlers, that's the bit that changes.Quoted 1 line
With the fake submission, do we really need to call
intel_engine_init_global_seqno?Yes. Completion is still determined by
i915_seqno_passed()comparing the rq against the engine.You need this
(quoted diff) +4
-0@@ -3246,6 +3246,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request) static void engine_set_wedged(struct intel_engine_cs *engine) { + unsigned long flags; + /* We need to be sure that no thread is running the old callback as * we install the nop handler (otherwise we would submit a request * to hardware that will never complete). In order to prevent this@@ -3261,8 +3263,10 @@ static void engine_set_wedged(struct intel_engine_cs *engine) * (lockless) lookup doesn't try and wait upon the request as we * reset it. */ + spin_lock_irqsave(&request->engine->timeline->lock, flags); intel_engine_init_global_seqno(engine, intel_engine_last_submit(engine)); + spin_unlock_irqrestore(&request->engine->timeline->lock, flags); }
Quoted 2 lines
So that the seqno written is ordered with the same spinlock used inside the nop submission.
Makes sense, I entirely missed the spinlock on Fri evening. Call me blind :-)
All amend the patch.
Thanks, Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
In reply to Daniel VetterQuoting Daniel Vetter (2017-10-06 15:20:09)
Quoted 249 lines · Chris Wilson
Quoted 233 lines
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 232 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-7@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +0
-2- engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);How are we planning to serialise the
intel_engine_init_global_seqno()here with the in-flightnop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.Yeah that part looks indeed fishy. Well the entire "let the nop handler fake-complete requests" logic is something I don't really understand. I guess there's an exclusive relationship between requests handled directly (and cancelled in
engine->cancel_request) and requests with externaldma_fencedependencies.
Quoted 4 lines
But then I'm not really seeing what I'm changing, since even with the stop machine you might end up with a bunch of requests depending upon external fences, which then all complete at roughly the same time and race multiple calls to
intel_engine_init_global_seqnowith each another.
Ugh, there's another issue. If nop_submit_request is executed before cancel_requests, we will consider the execution queue as completed and not in error, i.e. we will not flag those requests with the user visible -EIO.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
Quoted 255 lines · Chris Wilson
Quoting Daniel Vetter (2017-10-06 15:20:09)
Quoted 249 lines · Chris Wilson
Quoted 233 lines
Quoting Daniel Vetter (2017-10-06 10:06:37)
Quoted 232 lines
stop_machineis not really a locking primitive we should use, except when the hw folks tell us the hw is broken and that's the only way to work around it.This patch tries to address the locking abuse of
stop_machine()fromcommit
20e4933c478a1ca694b38fa4ac44d99e659941f5Author: Chris Wilson <[email protected]> Date: Tue Nov 22 14:41:21 2016 +0000drm/i915: Stop the machine as we install the wedged
submit_requesthandlerChris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.To stay as close as possible to the
stop_machinesemantics we first update all the submit function pointers to the nop handler, then callsynchronize_rcu()to make sure no new requests can be submitted. This should give us exactly the huge barrier we want.I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.This should fix the followwing lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc3-CI-CI_DRM_3179+ #1 Tainted: G U
------------------------------------------------------
kworker/3:4/562 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8113d4bc>]stop_machine+0x1c/0x40but task is already holding lock:
(&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>]i915_reset_device+0x1e8/0x260[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev->struct_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_interruptible_nested+0x1b/0x20 i915_mutex_lock_interruptible+0x51/0x130 [i915] i915_gem_fault+0x209/0x650 [i915] __do_fault+0x1e/0x80 __handle_mm_fault+0xa08/0xed0 handle_mm_fault+0x156/0x300 __do_page_fault+0x2c5/0x570 do_page_fault+0x28/0x250 page_fault+0x22/0x30-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xc9/0xbf0 cpuhp_thread_fun+0x17b/0x240 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state-up){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x133/0x1c0 __cpuhp_setup_state_cpuslocked+0x139/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x53/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] i915_handle_error+0x2d8/0x430 [i915] hangcheck_declare_hang+0xd3/0xf0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ret_from_fork+0x27/0x40other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev->struct_mutex Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev->struct_mutex); lock(&mm->mmap_sem); lock(&dev->struct_mutex);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
3 locks held by kworker/3:4/562:
#0: ("events_long"){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #1: ((&(&i915->gpu_error.hangcheck_work)->work)){+.+.}, at: [<ffffffff8109c64a>] process_one_work+0x1aa/0x660 #2: (&dev->struct_mutex){+.+.}, at: [<ffffffffa0136588>] i915_reset_device+0x1e8/0x260 [i915]stack backtrace: CPU: 3 PID: 562 Comm: kworker/3:4 Tainted: G U 4.14.0-rc3-CI-CI_DRM_3179+ #1 Hardware name: /NUC7i5BNB, BIOS BNKBL357.86A.0048.2017.0704.1415 07/04/2017 Workqueue: events_long i915_hangcheck_elapsed [i915] Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 ? irq_work_queue+0x86/0xe0 ? wake_up_klogd+0x53/0x70 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? stop_machine+0x1c/0x40 ? i915_gem_object_truncate+0x50/0x50 [i915] cpus_read_lock+0x3d/0xb0 ? stop_machine+0x1c/0x40 stop_machine+0x1c/0x40 i915_gem_set_wedged+0x1a/0x20 [i915] i915_reset+0xb9/0x230 [i915] i915_reset_device+0x1f6/0x260 [i915] ? gen8_gt_irq_ack+0x170/0x170 [i915] ? work_on_cpu_safe+0x60/0x60 i915_handle_error+0x2d8/0x430 [i915] ? vsnprintf+0xd1/0x4b0 ? scnprintf+0x3a/0x70 hangcheck_declare_hang+0xd3/0xf0 [i915] ? intel_runtime_pm_put+0x56/0xa0 [i915] i915_hangcheck_elapsed+0x262/0x2d0 [i915] process_one_work+0x233/0x660 worker_thread+0x4e/0x3b0 kthread+0x152/0x190 ? process_one_work+0x660/0x660 ? kthread_create_on_node+0x40/0x40 ret_from_fork+0x27/0x40Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hang
Setting dangerous option reset - tainting kernel
i915 0000:00:02.0: Resetting chip after gpu hangv2: Have 1 global
synchronize_rcu()barrier across all engines, and improve commit message.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102886
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103096Cc Chris Wilson <[email protected]>Cc Mika Kuoppala <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Marta Lofstedt <[email protected]>Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-7@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } -static void engine_set_wedged (struct intel_engine_cs *engine) +static void engine_complete_requests (struct intel_engine_cs *engine) { -/* We need to be sure that no thread is running the old callback as -* we install the nop handler (otherwise we would submit a request -* to hardware that will never complete). In order to prevent this -* race, we wait until the machine is idle before making the swap -* (using stop_machine()). -*/ Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c static void nop_submit_request(struct drm_i915_gem_request *request) · line 3026
Quoted 4 lines
drivers/gpu/drm/i915/i915_gem.c +0
-2- engine->submit_request = nop_submit_request; -/* Mark all executing requests as skipped */ engine->cancel_requests (engine);How are we planning to serialise the
intel_engine_init_global_seqno()here with the in-flightnop_submit? With sufficient thrust we will get a stale breadcrumb and an incomplete request.Yeah that part looks indeed fishy. Well the entire "let the nop handler fake-complete requests" logic is something I don't really understand. I guess there's an exclusive relationship between requests handled directly (and cancelled in
engine->cancel_request) and requests with externaldma_fencedependencies.Quoted 4 lines
But then I'm not really seeing what I'm changing, since even with the stop machine you might end up with a bunch of requests depending upon external fences, which then all complete at roughly the same time and race multiple calls to
intel_engine_init_global_seqnowith each another.
Quoted 4 lines
Ugh, there's another issue. If
nop_submit_requestis executed beforecancel_requests, we will consider the execution queue as completed and not in error, i.e. we will not flag those requests with the user visible-EIO.
Hm, that one's a bit a mess. What about the following sequence:
1. We set all submit_request functions to a nop request which does _not_ update the global seqno.
2. synchronize_rcu()
3. ->cancel_requests.
4. We set all submit_request to the current nop submit, i.e. including pushing the global seqno forward to complete them all. Maybe rename that one to nop_submit_complete_requests or something like that.
5. synchronize_rcu()
6. Call the spin-locked global seqno init that's currently in complete requests.
1/3/4/6 would all be wrapped in for_each_engine ofc.
I think that would also clarify a bit why we have to move the global_seqno in the nop_submit_request function.
Definitely needs lots of comments to explain what's going on.
Thoughts?
-Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/2] drm/i915: Use rcu instead of stop_machine in set_wedged
In reply to Daniel VetterQuoted 7 lines · Daniel Vetter
I pondered whether we should annotate
engine->submit_requestas__rcuand usercu_assign_pointerandrcu_dereferenceon it. But the reason behind those is to make sure the compiler/cpu barriers are there for when you have an actual data structure you point at, to make sure all the writes are seen correctly on the read side. But we just have a function pointer, and .text isn't changed, so no need for these barriers and hence no need for annotations.
synchronize_*() provides an smp_mb() on the calling CPU and ensures an smp_mb() on all other CPUs before completion, such that everybody agrees on the state prior to calling syncrhonize_rcu(). So yes, no additional ordering requirements.
Quoted 24 lines
drivers/gpu/drm/i915/i915_gem.c | 31 +++++++++-------------- drivers/gpu/drm/i915/i915_gem_request.c | 2 ++ drivers/gpu/drm/i915/selftests/i915_gem_request.c | 2 ++ 3 files changed, 16 insertions(+), 19 deletions(-)drivers/gpu/drm/i915/i915_gem.c +2
-0@@ -3020,16 +3020,8 @@ static void nop_submit_request(struct drm_i915_gem_request *request)intel_engine_init_global_seqno (request->engine, request->global_seqno); } +static void engine_complete_requests (struct intel_engine_cs *engine) {/* Mark all executing requests as skipped */ engine->cancel_requests (engine);@@ -3041,24 +3033,25 @@ static void engine_set_wedged(struct intel_engine_cs *engine) intel_engine_last_submit(engine)); } +void i915_gem_set_wedged (struct drm_i915_private *i915) {
drivers/gpu/drm/i915/i915_gem.c static void engine_set_wedged(struct intel_engine_cs *engine) · line 3047
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +5
-0struct intel_engine_cs *engine;enum intel_engine_id id;for_each_engine (engine, i915, id) + engine->submit_request = nop_submit_request; +/* Make sure no one is running the old callback before we proceed with +* cancelling requests and resetting the completion tracking. Otherwise +* we might submit a request to the hardware which never completes. +*/
ARGH @ horrid comment style..
http://lkml.iu.edu/hypermail/linux/kernel/1607.1/00627.html
:-)
Quoted 37 lines
drivers/gpu/drm/i915/i915_gem.c +5
-0+synchronize_rcu (); +for_each_engine (engine, i915, id) +engine_complete_requests (engine); +set_bit (I915_WEDGED , &i915->gpu_error.flags); +wake_up_all (&i915->gpu_error.reset_queue); }bool i915_gem_unset_wedged (struct drm_i915_private *i915)drivers/gpu/drm/i915/i915_gem_request.c +2
-0@@ -556,7 +556,9 @@ submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)switch (state) {case FENCE_COMPLETE :trace_i915_gem_request_submit (request); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();break ;case FENCE_FREE :drivers/gpu/drm/i915/selftests/i915_gem_request.c +2
-0@@ -215,7 +215,9 @@ static int igt_request_rewind(void *arg) }i915_gem_request_get (vip);i915_add_request (vip); +rcu_read_lock (); request->engine->submit_request (request); +rcu_read_unlock ();mutex_unlock (&i915->drm.struct_mutex);
Quoted 3 lines · you
Chris said parts of the reasons for going with
stop_machine()was that it's no overhead for the fast-path. But these callbacks use irqsave spinlocks and do a bunch of MMIO, andrcu_read_lockis _real_ fast.
This means that you could simply do synchronize_sched() without the addition of rcu_read_lock()s and still be fine.
Quoting Daniel Vetter (2017-10-06 10:06:36)
Quoted 6 lines
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.
But you haven't mentioned why we might want to preallocate to reduce the mmap/mm_lock coverage anyway.
-Chris
Quoting Daniel Vetter (2017-10-06 10:06:36)
Quoted 179 lines
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through
msr_create_device.Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>]apply_workqueue_attrs+0x17/0x50but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xa3/0x840 cpuhp_thread_fun+0x7a/0x150 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x10b/0x170 __cpuhp_setup_state_cpuslocked+0x134/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x52/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev_priv->mm_lock); lock(&mm->mmap_sem); lock(&dev_priv->mm_lock);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>]i915_gem_userptr_init__mmu_notifier+0x138/0x270[i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
Quoted 10 lines
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939
The difference is that I would have s/Bugzilla/References/. -Chris
Quoted 226 lines · Daniel Vetter
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through
msr_create_device.Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>]apply_workqueue_attrs+0x17/0x50but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xa3/0x840 cpuhp_thread_fun+0x7a/0x150 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x10b/0x170 __cpuhp_setup_state_cpuslocked+0x134/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x52/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev_priv->mm_lock); lock(&mm->mmap_sem); lock(&dev_priv->mm_lock);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>]i915_gem_userptr_init__mmu_notifier+0x138/0x270[i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem_userptr.c | 35 +++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 15 deletions(-)drivers/gpu/drm/i915/i915_gem_userptr.c +0
-10@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }@@ -210,23 +201,37 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn;
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 213
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +5
-0+struct i915_mmu_notifier *mn; +int err; mn = mm->mn;if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) +return mn;
Strictly speaking we don't want to fail just yet, only it we actually needed a new notifier and we failed to create it.
Quoted 20 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +9
-4+ + err =0 ;down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) { - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn; +if (mm->mn ==NULL ) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } }mutex_unlock (&mm->i915->mm_lock);up_write (&mm->mm->mmap_sem);
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 237
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +6
-1-return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static int
Otherwise looks good to me.
I would also put a note in the commit on how working around the locking issue is also beneficial to performance with moving the allocation step outside the mmap_sem.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 228 lines · Tvrtko Ursulin
Quoted 226 lines · Daniel Vetter
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through
msr_create_device.Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>]apply_workqueue_attrs+0x17/0x50but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xa3/0x840 cpuhp_thread_fun+0x7a/0x150 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x10b/0x170 __cpuhp_setup_state_cpuslocked+0x134/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x52/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev_priv->mm_lock); lock(&mm->mmap_sem); lock(&dev_priv->mm_lock);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>]i915_gem_userptr_init__mmu_notifier+0x138/0x270[i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem_userptr.c | 35 +++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 15 deletions(-)drivers/gpu/drm/i915/i915_gem_userptr.c +2
-10@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }@@ -210,23 +201,37 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn; +struct i915_mmu_notifier *mn; +int err; mn = mm->mn;
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 213
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +3
-0if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) +return mn;Strictly speaking we don't want to fail just yet, only it we actually needed a new notifier and we failed to create it.
The check 2 lines above not good enough? It's somewhat racy, but I'm not sure what value we provide by being perfectly correct against low memory. This thread racing against a 2nd one, where the minimal allocation of the 2nd one pushed us perfectly over the oom threshold seems a very unlikely scenario.
Also, small allocations actually never fail :-)
Quoted 32 lines
Quoted 19 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +9
-4+ + err =0 ;down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) { - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn; +if (mm->mn ==NULL ) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } }mutex_unlock (&mm->i915->mm_lock);up_write (&mm->mm->mmap_sem);drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +6
-1-return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static intOtherwise looks good to me.
Quoted 3 lines
I would also put a note in the commit on how working around the locking issue is also beneficial to performance with moving the allocation step outside the
mmap_sem.
Yeah Chris brought that up too, I don't really buy it given how heavy-weight __mmu_notifier_register is. But I can add something like:
"This also has the minor benefit of slightly reducing the critical section where we hold mmap_sem."
r-b with that added to the commit message?
-Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 7 lines · Daniel Vetter
Quoted 6 lines · Tvrtko Ursulin
Quoted 4 lines · Daniel Vetter
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through
msr_create_device.Full lockdep splat:
[snipped lockdep splat]
Quoted 60 lines
Quoted 51 lines
Quoted 51 lines
v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem_userptr.c | 35 +++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 15 deletions(-)drivers/gpu/drm/i915/i915_gem_userptr.c +2
-10@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }@@ -210,23 +201,37 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn; +struct i915_mmu_notifier *mn; +int err; mn = mm->mn;Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 213
Quoted 5 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +3
-0if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) +return mn;Strictly speaking we don't want to fail just yet, only it we actually needed a new notifier and we failed to create it.
Quoted 7 lines
The check 2 lines above not good enough? It's somewhat racy, but I'm not sure what value we provide by being perfectly correct against low memory. This thread racing against a 2nd one, where the minimal allocation of the 2nd one pushed us perfectly over the oom threshold seems a very unlikely scenario.
Also, small allocations actually never fail :-)
Yes, but, we otherwise make each other re-spin for much smaller things than bailout logic being conceptually at the wrong place. So for me I'd like a respin. It's not complicated at all, just move the bailout to to before the __mmu_notifier_register:
...
err = 0;
if (IS_ERR(mn))
err = PTR_ERR(..);
...
if (mana->manah == NULL) { /* ;-D */
/* Protect by mmap_sem...
if (err == 0) {
err = __mmu_notifier_register(..);
...
}
}...
if (mn && !IS_ERR(mn)) {
...free...
}I think.. ?
R-b on this, plus below, unless I got something wrong.
Quoted 37 lines
Quoted 33 lines
Quoted 20 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +9
-4+ + err =0 ;down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) {ed - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn; +if (mm->mn ==NULL ) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } }mutex_unlock (&mm->i915->mm_lock);drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 237
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +6
-1up_write (&mm->mm->mmap_sem); -return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static intOtherwise looks good to me.
Quoted 3 lines
I would also put a note in the commit on how working around the locking issue is also beneficial to performance with moving the allocation step outside the
mmap_sem.
Quoted 7 lines
Yeah Chris brought that up too, I don't really buy it given how heavy-weight
__mmu_notifier_registeris. But I can add something like:"This also has the minor benefit of slightly reducing the critical section where we hold
mmap_sem."r-b with that added to the commit message?
I think for me it is more making it clear that we are working around something where we are not strictly broken by design with a, however meaningless, optimisation at the same time.
Regards,
Tvrtko
Patchwork
CI✗ Fi.CI.BAT: warning for series starting with [1/2] drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
In reply to Daniel Vetter== Series Details ==
== Series Details ==
Series: series starting with [1/2] drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
URL : https://patchwork.freedesktop.org/series/31476/
State : warning
== Summary ==
Series 31476v1 series starting with [1/2] drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
https://patchwork.freedesktop.org/api/1.0/series/31476/revisions/1/mbox/
Test kms_cursor_legacy:
Subgroup basic-busy-flip-before-cursor-legacy:
pass -> DMESG-WARN (fi-bxt-dsi)
Test kms_frontbuffer_tracking:
Subgroup basic:
dmesg-warn -> PASS (fi-bdw-5557u) fdo#102473
Test kms_pipe_crc_basic:
Subgroup suspend-read-crc-pipe-b:
pass -> DMESG-WARN (fi-byt-n2820) fdo#101705fdo#102473 https://bugs.freedesktop.org/show_bug.cgi?id=102473
fdo#101705 https://bugs.freedesktop.org/show_bug.cgi?id=101705
fi-bdw-5557u total:289 pass:268 dwarn:0 dfail:0 fail:0 skip:21 time:456s fi-bdw-gvtdvm total:289 pass:265 dwarn:0 dfail:0 fail:0 skip:24 time:471s fi-blb-e6850 total:289 pass:223 dwarn:1 dfail:0 fail:0 skip:65 time:395s fi-bsw-n3050 total:289 pass:243 dwarn:0 dfail:0 fail:0 skip:46 time:576s fi-bwr-2160 total:289 pass:183 dwarn:0 dfail:0 fail:0 skip:106 time:287s fi-bxt-dsi total:289 pass:258 dwarn:1 dfail:0 fail:0 skip:30 time:526s fi-bxt-j4205 total:289 pass:260 dwarn:0 dfail:0 fail:0 skip:29 time:524s fi-byt-j1900 total:289 pass:253 dwarn:1 dfail:0 fail:0 skip:35 time:543s fi-byt-n2820 total:289 pass:249 dwarn:1 dfail:0 fail:0 skip:39 time:528s fi-cfl-s total:289 pass:256 dwarn:1 dfail:0 fail:0 skip:32 time:567s fi-cnl-y total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:625s fi-elk-e7500 total:289 pass:229 dwarn:0 dfail:0 fail:0 skip:60 time:437s fi-glk-1 total:289 pass:261 dwarn:0 dfail:0 fail:0 skip:28 time:594s fi-hsw-4770 total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:440s fi-hsw-4770r total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:417s fi-ivb-3520m total:289 pass:260 dwarn:0 dfail:0 fail:0 skip:29 time:509s fi-ivb-3770 total:289 pass:260 dwarn:0 dfail:0 fail:0 skip:29 time:475s fi-kbl-7500u total:289 pass:264 dwarn:1 dfail:0 fail:0 skip:24 time:498s fi-kbl-7560u total:289 pass:270 dwarn:0 dfail:0 fail:0 skip:19 time:578s fi-kbl-7567u total:289 pass:265 dwarn:4 dfail:0 fail:0 skip:20 time:493s fi-kbl-r total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:595s fi-pnv-d510 total:289 pass:222 dwarn:1 dfail:0 fail:0 skip:66 time:660s fi-skl-6260u total:289 pass:269 dwarn:0 dfail:0 fail:0 skip:20 time:467s fi-skl-6700hq total:289 pass:263 dwarn:0 dfail:0 fail:0 skip:26 time:660s fi-skl-6700k total:289 pass:265 dwarn:0 dfail:0 fail:0 skip:24 time:536s fi-skl-6770hq total:289 pass:269 dwarn:0 dfail:0 fail:0 skip:20 time:518s fi-skl-gvtdvm total:289 pass:266 dwarn:0 dfail:0 fail:0 skip:23 time:466s fi-snb-2520m total:289 pass:250 dwarn:0 dfail:0 fail:0 skip:39 time:584s fi-snb-2600 total:289 pass:249 dwarn:0 dfail:0 fail:0 skip:40 time:429s
f0ca28b3597bea929c1e58e05eb49a75881d9502 drm-tip: 2017y-10m-06d-11h-12m-22s UTC integration manifest d3d5abd754a3 drm/i915: Use rcu instead of stop_machine in set_wedged0b4e7df35196 drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5926/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 1/1 drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
In reply to Daniel Vetter4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding mmap_sem is a no-go, so let's try to preallocate it.
Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through msr_create_device.
Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>] apply_workqueue_attrs+0x17/0x50
but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>] i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_nested+0x1b/0x20
i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
i915_gem_userptr_ioctl+0x222/0x2c0 [i915]
drm_ioctl_kernel+0x69/0xb0
drm_ioctl+0x2f9/0x3d0
do_vfs_ioctl+0x94/0x670
SyS_ioctl+0x41/0x70
entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__might_fault+0x68/0x90
_copy_to_user+0x23/0x70
filldir+0xa5/0x120
dcache_readdir+0xf9/0x170
iterate_dir+0x69/0x1a0
SyS_getdents+0xa5/0x140
entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70
handle_create+0xcb/0x1e0
devtmpfsd+0x139/0x180
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
wait_for_common+0x58/0x210
wait_for_completion+0x1d/0x20
devtmpfs_create_node+0x13d/0x160
device_add+0x5eb/0x620
device_create_groups_vargs+0xe0/0xf0
device_create+0x3a/0x40
msr_device_create+0x2b/0x40
cpuhp_invoke_callback+0xa3/0x840
cpuhp_thread_fun+0x7a/0x150
smpboot_thread_fn+0x18a/0x280
kthread+0x152/0x190
ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpuhp_issue_call+0x10b/0x170
__cpuhp_setup_state_cpuslocked+0x134/0x2a0
__cpuhp_setup_state+0x46/0x60
page_writeback_init+0x43/0x67
pagecache_init+0x3d/0x42
start_kernel+0x3a8/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
__mutex_lock+0x86/0x9b0
mutex_lock_nested+0x1b/0x20
__cpuhp_setup_state_cpuslocked+0x52/0x2a0
__cpuhp_setup_state+0x46/0x60
page_alloc_init+0x28/0x30
start_kernel+0x145/0x3fc
x86_64_start_reservations+0x2a/0x2c
x86_64_start_kernel+0x6d/0x70
verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840
__lock_acquire+0x1420/0x15e0
lock_acquire+0xb0/0x200
cpus_read_lock+0x3d/0xb0
apply_workqueue_attrs+0x17/0x50
__alloc_workqueue_key+0x1d8/0x4d9
i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915]
i915_gem_userptr_ioctl+0x222/0x2c0 [i915]
drm_ioctl_kernel+0x69/0xb0
drm_ioctl+0x2f9/0x3d0
do_vfs_ioctl+0x94/0x670
SyS_ioctl+0x41/0x70
entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&dev_priv->mm_lock);
lock(&mm->mmap_sem);
lock(&dev_priv->mm_lock); lock(cpu_hotplug_lock.rw_sem);
*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>] i915_gem_userptr_init__mmu_notifier+0x138/0x270 [i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>] i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915]
stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20
Note that this also has the minor benefit of slightly reducing the critical section where we hold mmap_sem.
v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
v4: Repaint in Tvrtko's colors (aka don't report ENOMEM if we race and some other thread managed to not also get an ENOMEM and successfully install the mmu notifier. Note that the kernel guarantees that small allocations succeed, so this never actually happens).
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939
--- drivers/gpu/drm/i915/i915_gem_userptr.c | 38 ++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 15 deletions(-)
drivers/gpu/drm/i915/i915_gem_userptr.c +23 -15
@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )
@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }
@@ -210,23 +201,40 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn; +struct i915_mmu_notifier *mn; +int err =0 ; mn = mm->mn;if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) + err =PTR_ERR (mn); +down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) { - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn; +if (mm->mn ==NULL && !err) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } + }else { +/* someone else raced and successfully installed the mmu +* notifier, we can cancel our own errors */ + err =0 ; }mutex_unlock (&mm->i915->mm_lock);up_write (&mm->mm->mmap_sem); -return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static int
Signature · 6 lines
2.14.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 251 lines · Daniel Vetter
4.14-rc1 gained the fancy new cross-release support in lockdep, which seems to have uncovered a few more rules about what is allowed and isn't.
This one here seems to indicate that allocating a work-queue while holding
mmap_semis a no-go, so let's try to preallocate it.Of course another way to break this chain would be somewhere in the cpu hotplug code, since this isn't the only trace we're finding now which goes through
msr_create_device.Full lockdep splat:
======================================================
WARNING: possible circular locking dependency detected
4.14.0-rc1-CI-CI_DRM_3118+ #1 Tainted: G U
------------------------------------------------------
prime_mmap/1551 is trying to acquire lock:
(cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8109dbb7>]apply_workqueue_attrs+0x17/0x50but task is already holding lock:
(&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #6 (&dev_priv->mm_lock){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 i915_gem_userptr_init__mmu_notifier+0x14a/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #5 (&mm->mmap_sem){++++}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __might_fault+0x68/0x90 _copy_to_user+0x23/0x70 filldir+0xa5/0x120 dcache_readdir+0xf9/0x170 iterate_dir+0x69/0x1a0 SyS_getdents+0xa5/0x140 entry_SYSCALL_64_fastpath+0x1c/0xb1-> #4 (&sb->s_type->i_mutex_key#5){++++}:
down_write+0x3b/0x70 handle_create+0xcb/0x1e0 devtmpfsd+0x139/0x180 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #3 ((complete)&req.done){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 wait_for_common+0x58/0x210 wait_for_completion+0x1d/0x20 devtmpfs_create_node+0x13d/0x160 device_add+0x5eb/0x620 device_create_groups_vargs+0xe0/0xf0 device_create+0x3a/0x40 msr_device_create+0x2b/0x40 cpuhp_invoke_callback+0xa3/0x840 cpuhp_thread_fun+0x7a/0x150 smpboot_thread_fn+0x18a/0x280 kthread+0x152/0x190 ret_from_fork+0x27/0x40-> #2 (cpuhp_state){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpuhp_issue_call+0x10b/0x170 __cpuhp_setup_state_cpuslocked+0x134/0x2a0 __cpuhp_setup_state+0x46/0x60 page_writeback_init+0x43/0x67 pagecache_init+0x3d/0x42 start_kernel+0x3a8/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #1 (cpuhp_state_mutex){+.+.}:
__lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 __mutex_lock+0x86/0x9b0 mutex_lock_nested+0x1b/0x20 __cpuhp_setup_state_cpuslocked+0x52/0x2a0 __cpuhp_setup_state+0x46/0x60 page_alloc_init+0x28/0x30 start_kernel+0x145/0x3fc x86_64_start_reservations+0x2a/0x2c x86_64_start_kernel+0x6d/0x70 verify_cpu+0x0/0xfb-> #0 (cpu_hotplug_lock.rw_sem){++++}:
check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 lock_acquire+0xb0/0x200 cpus_read_lock+0x3d/0xb0 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 do_vfs_ioctl+0x94/0x670 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1other info that might help us debug this:
Chain exists of: cpu_hotplug_lock.rw_sem --> &mm->mmap_sem --> &dev_priv->mm_lock Possible unsafe locking scenario:CPU0 CPU1 ---- ---- lock(&dev_priv->mm_lock); lock(&mm->mmap_sem); lock(&dev_priv->mm_lock);
lock(cpu_hotplug_lock.rw_sem);*** DEADLOCK ***
2 locks held by prime_mmap/1551:
#0: (&mm->mmap_sem){++++}, at: [<ffffffffa01a7b18>]i915_gem_userptr_init__mmu_notifier+0x138/0x270[i915] #1: (&dev_priv->mm_lock){+.+.}, at: [<ffffffffa01a7b2a>]i915_gem_userptr_init__mmu_notifier+0x14a/0x270[i915]stack backtrace: CPU: 4 PID: 1551 Comm: prime_mmap Tainted: G U 4.14.0-rc1-CI-CI_DRM_3118+ #1 Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 Call Trace: dump_stack+0x68/0x9f print_circular_bug+0x235/0x3c0 ? lockdep_init_map_crosslock+0x20/0x20 check_prev_add+0x430/0x840 __lock_acquire+0x1420/0x15e0 ? __lock_acquire+0x1420/0x15e0 ? lockdep_init_map_crosslock+0x20/0x20 lock_acquire+0xb0/0x200 ? apply_workqueue_attrs+0x17/0x50 cpus_read_lock+0x3d/0xb0 ? apply_workqueue_attrs+0x17/0x50 apply_workqueue_attrs+0x17/0x50 __alloc_workqueue_key+0x1d8/0x4d9 ? __lockdep_init_map+0x57/0x1c0 i915_gem_userptr_init__mmu_notifier+0x1fb/0x270 [i915] i915_gem_userptr_ioctl+0x222/0x2c0 [i915] ? i915_gem_userptr_release+0x140/0x140 [i915] drm_ioctl_kernel+0x69/0xb0 drm_ioctl+0x2f9/0x3d0 ? i915_gem_userptr_release+0x140/0x140 [i915] ? __do_page_fault+0x2a4/0x570 do_vfs_ioctl+0x94/0x670 ? entry_SYSCALL_64_fastpath+0x5/0xb1 ? __this_cpu_preempt_check+0x13/0x20 ? trace_hardirqs_on_caller+0xe3/0x1b0 SyS_ioctl+0x41/0x70 entry_SYSCALL_64_fastpath+0x1c/0xb1 RIP: 0033:0x7fbb83c39587 RSP: 002b:00007fff188dc228 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: ffffffff81492963 RCX: 00007fbb83c39587 RDX: 00007fff188dc260 RSI: 00000000c0186473 RDI: 0000000000000003 RBP: ffffc90001487f88 R08: 0000000000000000 R09: 00007fff188dc2ac R10: 00007fbb83efcb58 R11: 0000000000000246 R12: 0000000000000000 R13: 0000000000000003 R14: 00000000c0186473 R15: 00007fff188dc2ac ? __this_cpu_preempt_check+0x13/0x20Note that this also has the minor benefit of slightly reducing the critical section where we hold
mmap_sem.v2: Set ret correctly when we raced with another thread.
v3: Use Chris' diff. Attach the right lockdep splat.
v4: Repaint in Tvrtko's colors (aka don't report ENOMEM if we race and some other thread managed to not also get an ENOMEM and successfully install the mmu notifier. Note that the kernel guarantees that small allocations succeed, so this never actually happens).
Cc: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Joonas Lahtinen <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Sasha Levin <[email protected]>
Cc: Marta Lofstedt <[email protected]>
Cc: Tejun Heo <[email protected]>
References: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3180/shard-hsw3/igt@prime_mmap@test_userptr.html
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102939Signed-off-by Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/i915_gem_userptr.c | 38 ++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 15 deletions(-)drivers/gpu/drm/i915/i915_gem_userptr.c +6
-14@@ -164,7 +164,6 @@ static struct i915_mmu_notifier * i915_mmu_notifier_create(struct mm_struct *mm) {struct i915_mmu_notifier *mn; -int ret; mn =kmalloc (sizeof (*mn),GFP_KERNEL );if (mn ==NULL )@@ -179,14 +178,6 @@ i915_mmu_notifier_create(struct mm_struct *mm)return ERR_PTR (-ENOMEM ); } -/* Protected by mmap_sem (write-lock) */ - ret =__mmu_notifier_register (&mn->mn, mm); -if (ret) { -destroy_workqueue (mn->wq); -kfree (mn); -return ERR_PTR (ret); - } -return mn; }@@ -210,23 +201,40 @@ i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj)static struct i915_mmu_notifier *i915_mmu_notifier_find (struct i915_mm_struct *mm) { -struct i915_mmu_notifier *mn = mm->mn; +struct i915_mmu_notifier *mn; +int err =0 ; mn = mm->mn;if (mn)return mn; + mn =i915_mmu_notifier_create (mm->mm); +if (IS_ERR (mn)) + err =PTR_ERR (mn); +down_write (&mm->mm->mmap_sem);mutex_lock (&mm->i915->mm_lock); -if ((mn = mm->mn) ==NULL ) { - mn =i915_mmu_notifier_create (mm->mm); -if (!IS_ERR (mn)) - mm->mn = mn;
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 226
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +10
-0+if (mm->mn ==NULL && !err) { +/* Protected by mmap_sem (write-lock) */ + err =__mmu_notifier_register (&mn->mn, mm->mm); +if (!err) { +/* Protected by mm_lock */ + mm->mn =fetch_and_zero (&mn); + } + }else { +/* someone else raced and successfully installed the mmu +* notifier, we can cancel our own errors */
Better make sure Linus is not lurking. ;)
Quoted 6 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +1
-0+ err =0 ; }mutex_unlock (&mm->i915->mm_lock);up_write (&mm->mm->mmap_sem);
drivers/gpu/drm/i915/i915_gem_userptr.c i915_gem_userptr_release__mmu_notifier(struct drm_i915_gem_object *obj) · line 240
Quoted 10 lines
drivers/gpu/drm/i915/i915_gem_userptr.c +6
-1-return mn; +if (mn) { +destroy_workqueue (mn->wq); +kfree (mn); + } + +return err ?ERR_PTR (err) : mm->mn; } static int
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI✓ Fi.CI.BAT: success for series starting with drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock (rev2)
In reply to Daniel Vetter== Series Details ==
== Series Details ==
Series: series starting with drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock (rev2)
URL : https://patchwork.freedesktop.org/series/31476/
State : success
== Summary ==
Series 31476v2 series starting with drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
https://patchwork.freedesktop.org/api/1.0/series/31476/revisions/2/mbox/
Test drv_module_reload:
Subgroup basic-reload-inject:
incomplete -> PASS (fi-cfl-s) fdo#103022fdo#103022 https://bugs.freedesktop.org/show_bug.cgi?id=103022
fi-bdw-5557u total:289 pass:268 dwarn:0 dfail:0 fail:0 skip:21 time:455s fi-bdw-gvtdvm total:289 pass:265 dwarn:0 dfail:0 fail:0 skip:24 time:482s fi-blb-e6850 total:289 pass:223 dwarn:1 dfail:0 fail:0 skip:65 time:397s fi-bsw-n3050 total:289 pass:243 dwarn:0 dfail:0 fail:0 skip:46 time:566s fi-bwr-2160 total:289 pass:183 dwarn:0 dfail:0 fail:0 skip:106 time:287s fi-bxt-dsi total:289 pass:259 dwarn:0 dfail:0 fail:0 skip:30 time:526s fi-byt-j1900 total:289 pass:253 dwarn:1 dfail:0 fail:0 skip:35 time:544s fi-byt-n2820 total:289 pass:249 dwarn:1 dfail:0 fail:0 skip:39 time:527s fi-cfl-s total:289 pass:256 dwarn:1 dfail:0 fail:0 skip:32 time:559s fi-cnl-y total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:618s fi-elk-e7500 total:289 pass:229 dwarn:0 dfail:0 fail:0 skip:60 time:430s fi-glk-1 total:289 pass:261 dwarn:0 dfail:0 fail:0 skip:28 time:612s fi-hsw-4770 total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:440s fi-hsw-4770r total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:417s fi-ivb-3520m total:289 pass:260 dwarn:0 dfail:0 fail:0 skip:29 time:504s fi-ivb-3770 total:289 pass:260 dwarn:0 dfail:0 fail:0 skip:29 time:482s fi-kbl-7500u total:289 pass:264 dwarn:1 dfail:0 fail:0 skip:24 time:505s fi-kbl-7560u total:289 pass:270 dwarn:0 dfail:0 fail:0 skip:19 time:584s fi-kbl-7567u total:289 pass:265 dwarn:4 dfail:0 fail:0 skip:20 time:501s fi-kbl-r total:289 pass:262 dwarn:0 dfail:0 fail:0 skip:27 time:597s fi-pnv-d510 total:289 pass:222 dwarn:1 dfail:0 fail:0 skip:66 time:656s fi-skl-6260u total:289 pass:269 dwarn:0 dfail:0 fail:0 skip:20 time:469s fi-skl-6700hq total:289 pass:263 dwarn:0 dfail:0 fail:0 skip:26 time:659s fi-skl-6700k total:289 pass:265 dwarn:0 dfail:0 fail:0 skip:24 time:531s fi-skl-6770hq total:289 pass:269 dwarn:0 dfail:0 fail:0 skip:20 time:516s fi-skl-gvtdvm total:289 pass:266 dwarn:0 dfail:0 fail:0 skip:23 time:476s fi-snb-2520m total:289 pass:250 dwarn:0 dfail:0 fail:0 skip:39 time:582s fi-snb-2600 total:289 pass:249 dwarn:0 dfail:0 fail:0 skip:40 time:432s
fi-bxt-j4205 failed to connect after reboot
cb32cc2ad1c3ccd0803276d5af46c410f5104951 drm-tip: 2017y-10m-06d-15h-01m-44s UTC integration manifest 999c4f026e85 drm/i915: Use rcu instead of stop_machine in set_wedgedcef3c4054a61 drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5935/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI✗ Fi.CI.IGT: warning for series starting with drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock (rev2)
In reply to Daniel Vetter== Series Details ==
== Series Details ==
Series: series starting with drm/i915: Preallocate our mmu notifier workequeu to unbreak cpu hotplug deadlock (rev2)
URL : https://patchwork.freedesktop.org/series/31476/
State : warning
== Summary ==
Test gem_eio:
Subgroup in-flight-contexts:
dmesg-warn -> PASS (shard-hsw) fdo#102886 +4
Test kms_cursor_crc:
Subgroup cursor-64x64-sliding:
pass -> DMESG-WARN (shard-hsw)
Test prime_mmap:
Subgroup test_userptr:
dmesg-warn -> PASS (shard-hsw) fdo#102939fdo#102886 https://bugs.freedesktop.org/show_bug.cgi?id=102886
fdo#102939 https://bugs.freedesktop.org/show_bug.cgi?id=102939
shard-hsw total:2446 pass:1333 dwarn:1 dfail:0 fail:9 skip:1103 time:10142s
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5935/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx