2021-04-16 10:22:47

by Jun Miao

[permalink] [raw]
Subject: [V3] drm/i915/gt: Fix a lockdep warning with interrupts enabled

Don`t simplely disable local interrupt delivery of CPU hardware irq, should race
the region inside signal_irq_work, include intel_breadcrumbs_disarm_irq/intel_breadcrumbs_arm_irq.

RT complains about might sleep inside signal_irq_work() because spin_lock will
be invoked after disabling interrupts.

Thanks to Tvrtko Ursulin for his expert opinions and suggestions.

BUG: sleeping function called from invalid context at kernel/locking/rtmutex.c:969
#0: ffff89c4c00ca970 ((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x1cf/0x6d0
#1: ffffa433c1f53e60 ((work_completion)(&engine->retire_work)){+.+.}-{0:0}, at: process_one_work+0x1cf 0x6d
#2: ffff89c4ccb0a0a8 (kernel_context){+.+.}-{0:0}, at: engine_retire+0x62/0x110 [i915]
#3: ffff89c4cf682300 (wakeref.mutex#3){+.+.}-{0:0}, at: __intel_wakeref_put_last+0x20/0x60 [i915]
#4: ffff89c4ccb08398 (&b->irq_lock){+.+.}-{0:0}, at: intel_breadcrumbs_disarm_irq+0x20/0xd0 [i915]
irq event stamp: 2126
hardirqs last enabled at (2125): [<ffffffffbb134739>] cancel_delayed_work+0xa9/0xc0
hardirqs last disabled at (2126): [<ffffffffc0507fe6>] __intel_breadcrumbs_park+0x76/0x80 [i915]
softirqs last enabled at (0): [<ffffffffbb1099ce>] copy_process+0x63e/0x1630
softirqs last disabled at (0): [<0000000000000000>] 0x0
CPU: 3 PID: 281 Comm: kworker/3:3 Not tainted 5.10.27-rt34-yocto-preempt-rt #1
Hardware name: Intel(R) Client Systems NUC7i5DNKE/NUC7i5DNB, BIOS DNKBLi5v.86A.0064.2019.0523.1933 05/23 2019
Workqueue: events engine_retire [i915]
Call Trace:
show_stack+0x52/0x58
dump_stack+0x7d/0x9f
___might_sleep.cold+0xe3/0xf4
rt_spin_lock+0x3f/0xc0
? intel_breadcrumbs_disarm_irq+0x20/0xd0 [i915]
intel_breadcrumbs_disarm_irq+0x20/0xd0 [i915]
signal_irq_work+0x241/0x660 [i915]
? __this_cpu_preempt_check+0x13/0x20
? lockdep_hardirqs_off+0x106/0x120
__intel_breadcrumbs_park+0x3f/0x80 [i915]
__engine_park+0xbd/0xe0 [i915]
____intel_wakeref_put_last+0x22/0x60 [i915]
__intel_wakeref_put_last+0x50/0x60 [i915]
intel_context_exit_engine+0x5f/0x70 [i915]
i915_request_retire+0x139/0x2d0 [i915]
engine_retire+0xb0/0x110 [i915]
process_one_work+0x26d/0x6d0
worker_thread+0x53/0x330
kthread+0x1b0/0x1d0
? process_one_work+0x6d0/0x6d0
? __kthread_parkme+0xc0/0xc0
ret_from_fork+0x22/0x30

Fixes: 9d5612ca165a ("drm/i915/gt: Defer enabling the breadcrumb interrupt to after submission")
Signed-off-by: Jun Miao <[email protected]>
---
drivers/gpu/drm/i915/gt/intel_breadcrumbs.c | 19 +++++++++++--------
1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
index 34a645d..01eb18e 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -82,13 +82,15 @@ static void __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)

static void intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
{
+ unsigned long flags;
+
if (!b->irq_engine)
return;

- spin_lock(&b->irq_lock);
+ spin_lock_irqsave(&b->irq_lock, flags);
if (!b->irq_armed)
__intel_breadcrumbs_arm_irq(b);
- spin_unlock(&b->irq_lock);
+ spin_unlock_irqrestore(&b->irq_lock, flags);
}

static void __intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
@@ -103,10 +105,12 @@ static void __intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)

static void intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
{
- spin_lock(&b->irq_lock);
+ unsigned long flags;
+
+ spin_lock_irqsave(&b->irq_lock, flags);
if (b->irq_armed)
__intel_breadcrumbs_disarm_irq(b);
- spin_unlock(&b->irq_lock);
+ spin_unlock_irqrestore(&b->irq_lock, flags);
}

static void add_signaling_context(struct intel_breadcrumbs *b,
@@ -200,6 +204,7 @@ static void signal_irq_work(struct irq_work *work)
const ktime_t timestamp = ktime_get();
struct llist_node *signal, *sn;
struct intel_context *ce;
+ unsigned long flags;

signal = NULL;
if (unlikely(!llist_empty(&b->signaled_requests)))
@@ -278,11 +283,11 @@ static void signal_irq_work(struct irq_work *work)
llist_entry(signal, typeof(*rq), signal_node);
struct list_head cb_list;

- spin_lock(&rq->lock);
+ spin_lock_irqsave(&rq->lock, flags);
list_replace(&rq->fence.cb_list, &cb_list);
__dma_fence_signal__timestamp(&rq->fence, timestamp);
__dma_fence_signal__notify(&rq->fence, &cb_list);
- spin_unlock(&rq->lock);
+ spin_unlock_irqrestore(&rq->lock, flags);

i915_request_put(rq);
}
@@ -337,9 +342,7 @@ void __intel_breadcrumbs_park(struct intel_breadcrumbs *b)
/* Kick the work once more to drain the signalers, and disarm the irq */
irq_work_sync(&b->irq_work);
while (READ_ONCE(b->irq_armed) && !atomic_read(&b->active)) {
- local_irq_disable();
signal_irq_work(&b->irq_work);
- local_irq_enable();
cond_resched();
}
}
--
2.7.4