Commit b4d9145b authored by Chris Wilson's avatar Chris Wilson Committed by Joonas Lahtinen

drm/i915: Be wary of data races when reading the active execlists

To implement preempt-to-busy (and so efficient timeslicing and best utilization
of the hardware submission ports) we let the GPU run asynchronously in respect
to the ELSP submission queue. This created challenges in keeping and accessing
the driver state mirroring the asynchronous GPU execution.

The latest occurence of this was spotted by KCSAN:

[ 1413.563200] BUG: KCSAN: data-race in __await_execution+0x217/0x370 [i915]
[ 1413.563221]
[ 1413.563236] race at unknown origin, with read to 0xffff88885bb6c478 of 8 bytes by task 9654 on cpu 1:
[ 1413.563548]  __await_execution+0x217/0x370 [i915]
[ 1413.563891]  i915_request_await_dma_fence+0x4eb/0x6a0 [i915]
[ 1413.564235]  i915_request_await_object+0x421/0x490 [i915]
[ 1413.564577]  i915_gem_do_execbuffer+0x29b7/0x3c40 [i915]
[ 1413.564967]  i915_gem_execbuffer2_ioctl+0x22f/0x5c0 [i915]
[ 1413.564998]  drm_ioctl_kernel+0x156/0x1b0
[ 1413.565022]  drm_ioctl+0x2ff/0x480
[ 1413.565046]  __x64_sys_ioctl+0x87/0xd0
[ 1413.565069]  do_syscall_64+0x4d/0x80
[ 1413.565094]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

To complicate matters, we have to both avoid the read tearing of *active and
avoid any write tearing as perform the pending[] -> inflight[] promotion of the
execlists.

This is because we cannot rely on the memcpy doing u64 aligned copies on all
kernels/platforms and so we opt to open-code it with explicit WRITE_ONCE
annotations to satisfy KCSAN.

v2: When in doubt, write the same comment again.
v3: Expanded commit message.

Fixes: b55230e5 ("drm/i915: Check for awaits on still currently executing requests")
Signed-off-by: default avatarChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: default avatarTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200716142207.13003-1-chris@chris-wilson.co.ukSigned-off-by: default avatarRodrigo Vivi <rodrigo.vivi@intel.com>
[Joonas: Rebased and reordered into drm-intel-gt-next branch]
[Joonas: Added expanded commit message from Tvrtko and Chris]
Signed-off-by: default avatarJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
parent c1793ba8
......@@ -2062,6 +2062,14 @@ static inline void clear_ports(struct i915_request **ports, int count)
memset_p((void **)ports, NULL, count);
}
static inline void
copy_ports(struct i915_request **dst, struct i915_request **src, int count)
{
/* A memcpy_p() would be very useful here! */
while (count--)
WRITE_ONCE(*dst++, *src++); /* avoid write tearing */
}
static void execlists_dequeue(struct intel_engine_cs *engine)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
......@@ -2635,10 +2643,9 @@ static void process_csb(struct intel_engine_cs *engine)
/* switch pending to inflight */
GEM_BUG_ON(!assert_pending_valid(execlists, "promote"));
memcpy(execlists->inflight,
copy_ports(execlists->inflight,
execlists->pending,
execlists_num_ports(execlists) *
sizeof(*execlists->pending));
execlists_num_ports(execlists));
smp_wmb(); /* complete the seqlock */
WRITE_ONCE(execlists->active, execlists->inflight);
......
......@@ -379,17 +379,38 @@ static bool __request_in_flight(const struct i915_request *signal)
* As we know that there are always preemption points between
* requests, we know that only the currently executing request
* may be still active even though we have cleared the flag.
* However, we can't rely on our tracking of ELSP[0] to known
* However, we can't rely on our tracking of ELSP[0] to know
* which request is currently active and so maybe stuck, as
* the tracking maybe an event behind. Instead assume that
* if the context is still inflight, then it is still active
* even if the active flag has been cleared.
*
* To further complicate matters, if there a pending promotion, the HW
* may either perform a context switch to the second inflight execlists,
* or it may switch to the pending set of execlists. In the case of the
* latter, it may send the ACK and we process the event copying the
* pending[] over top of inflight[], _overwriting_ our *active. Since
* this implies the HW is arbitrating and not struck in *active, we do
* not worry about complete accuracy, but we do require no read/write
* tearing of the pointer [the read of the pointer must be valid, even
* as the array is being overwritten, for which we require the writes
* to avoid tearing.]
*
* Note that the read of *execlists->active may race with the promotion
* of execlists->pending[] to execlists->inflight[], overwritting
* the value at *execlists->active. This is fine. The promotion implies
* that we received an ACK from the HW, and so the context is not
* stuck -- if we do not see ourselves in *active, the inflight status
* is valid. If instead we see ourselves being copied into *active,
* we are inflight and may signal the callback.
*/
if (!intel_context_inflight(signal->context))
return false;
rcu_read_lock();
for (port = __engine_active(signal->engine); (rq = *port); port++) {
for (port = __engine_active(signal->engine);
(rq = READ_ONCE(*port)); /* may race with promotion of pending[] */
port++) {
if (rq->context == signal->context) {
inflight = i915_seqno_passed(rq->fence.seqno,
signal->fence.seqno);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment