On 7/29/23 11:59, Chengfeng Ye wrote:
> As plx_dma_process_desc() is invoked by both tasklet plx_dma_desc_task()
> under softirq context and plx_dma_tx_status() callback that executed under
> process context, the lock aquicision of &plxdev->ring_lock inside
> plx_dma_process_desc() should disable irq otherwise deadlock could happen
> if the irq preempts the execution of process context code while the lock
> is held in process context on the same CPU.
>
> Possible deadlock scenario:
> plx_dma_tx_status()
> -> plx_dma_process_desc()
> -> spin_lock(&plxdev->ring_lock)
> <tasklet softirq>
> -> plx_dma_desc_task()
> -> plx_dma_process_desc()
> -> spin_lock(&plxdev->ring_lock) (deadlock here)
>
> This flaw was found by an experimental static analysis tool I am developing
> for irq-related deadlock.
>
> The lock was changed from spin_lock_bh() to spin_lock() by a previous patch
> for performance concern but unintentionally brought this potential deadlock
> problem.
>
> This patch reverts back to spin_lock_bh() to fix the deadlock problem.
>
> Fixes: 1d05a0bdb420 ("dmaengine: plx_dma: Move spin_lock_bh() to spin_lock()")
> Signed-off-by: Chengfeng Ye <dg573847474@xxxxxxxxx>
>
Reviewed-by: Logan Gunthorpe <logang@xxxxxxxxxxxx>
Thanks!
Logan
