On 2019/10/29 0:30, Matthew Wilcox wrote:
> On Mon, Oct 28, 2019 at 05:18:09PM +0800, Xiang Zheng wrote:
>> Commit "7ea7e98fd8d0" suggests that the "pci_lock" is sufficient,
>> and all the callers of pci_wait_cfg() are wrapped with the "pci_lock".
>>
>> However, since the commit "cdcb33f98244" merged, the accesses to
>> the pci_cfg_wait queue are not safe anymore. A "pci_lock" is
>> insufficient and we need to hold an additional queue lock while
>> read/write the wait queue.
>>
>> So let's use the add_wait_queue()/remove_wait_queue() instead of
>> __add_wait_queue()/__remove_wait_queue().
>
> As I said earlier, this reintroduces the deadlock addressed by
> cdcb33f9824429a926b971bf041a6cec238f91ff
>
Thanks Matthew, sorry for that I did not understand the way to reintroduce
the deadlock and sent this patch. If what I think is right, the possible
deadlock may be caused by the condition in which there are three processes:
*Process* *Acquired* *Wait For*
wake_up_all() wq_head->lock pi_lock
snbep_uncore_pci_read_counter() pi_lock pci_lock
pci_wait_cfg() pci_lock wq_head->lock
These processes suffer from the nested locks.:)
But for this problem, what do you think about the solution below:
diff --git a/drivers/pci/access.c b/drivers/pci/access.c
index 2fccb5762c76..09342a74e5ea 100644
--- a/drivers/pci/access.c
+++ b/drivers/pci/access.c
@@ -207,14 +207,14 @@ static noinline void pci_wait_cfg(struct pci_dev *dev)
{
DECLARE_WAITQUEUE(wait, current);
- __add_wait_queue(&pci_cfg_wait, &wait);
do {
set_current_state(TASK_UNINTERRUPTIBLE);
raw_spin_unlock_irq(&pci_lock);
+ add_wait_queue(&pci_cfg_wait, &wait);
schedule();
+ remove_wait_queue(&pci_cfg_wait, &wait);
raw_spin_lock_irq(&pci_lock);
} while (dev->block_cfg_access);
- __remove_wait_queue(&pci_cfg_wait, &wait);
}
/* Returns 0 on success, negative values indicate error. */
> .
>
--
Thanks,
Xiang
