Messages in this thread Patch in this message | | | From | Oded Gabbay <> | Subject | [PATCH 28/30] habanalabs: fix race between wait and irq | Date | Sat, 22 Jan 2022 21:57:29 +0200 |
| |
There is a race in the user interrupts code, where between checking the target value and adding the new pend to the list, there is a chance the interrupt happened.
In that case, no one will complete the node, and we will get a timeout on it.
Signed-off-by: Oded Gabbay <ogabbay@kernel.org> --- drivers/misc/habanalabs/common/command_submission.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c index 2b27ff826e3c..3d9020825df5 100644 --- a/drivers/misc/habanalabs/common/command_submission.c +++ b/drivers/misc/habanalabs/common/command_submission.c @@ -2901,16 +2901,21 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, pend->cq_kernel_addr = (u64 *) cb->kernel_address + cq_counters_offset; pend->cq_target_value = target_value; + spin_lock_irqsave(&interrupt->wait_list_lock, flags); + /* We check for completion value as interrupt could have been received * before we added the node to the wait list */ if (*pend->cq_kernel_addr >= target_value) { + spin_unlock_irqrestore(&interrupt->wait_list_lock, flags); + *status = HL_WAIT_CS_STATUS_COMPLETED; /* There was no interrupt, we assume the completion is now. */ pend->fence.timestamp = ktime_get(); goto set_timestamp; } else if (!timeout_us) { + spin_unlock_irqrestore(&interrupt->wait_list_lock, flags); *status = HL_WAIT_CS_STATUS_BUSY; pend->fence.timestamp = ktime_get(); goto set_timestamp; @@ -2919,7 +2924,6 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, /* Add pending user interrupt to relevant list for the interrupt * handler to monitor */ - spin_lock_irqsave(&interrupt->wait_list_lock, flags); list_add_tail(&pend->wait_list_node, &interrupt->wait_list_head); spin_unlock_irqrestore(&interrupt->wait_list_lock, flags); -- 2.25.1
| |