Fix a potential deadlock bug. Observe that in the mtk-cqdma.c
file, functions like mtk_cqdma_issue_pending() and
mtk_cqdma_free_active_desc() properly acquire the pc lock before the vc
lock when handling pc and vc fields. However, mtk_cqdma_tx_status()
violates this order by first acquiring the vc lock before invoking
mtk_cqdma_find_active_desc(), which subsequently takes the pc lock. This
reversed locking sequence (vc → pc) contradicts the established
pc → vc order and creates deadlock risks.

Fix the issue by moving the vc lock acquisition code from
mtk_cqdma_find_active_desc() to mtk_cqdma_tx_status(). Ensure the pc lock
is acquired before the vc lock in the calling function to maintain correct
locking hierarchy. Note that since mtk_cqdma_find_active_desc() is a
static function with only one caller (mtk_cqdma_tx_status()), this
modification safely eliminates the deadlock possibility without affecting
other components.

This possible bug is found by an experimental static analysis tool
developed by our team. This tool analyzes the locking APIs to extract
function pairs that can be concurrently executed, and then analyzes the
instructions in the paired functions to identify possible concurrency bugs
including deadlocks, data races and atomicity violations.

Fixes: b1f01e48df5a ("dmaengine: mediatek: Add MediaTek Command-Queue DMA controller for MT6765 SoC")
Cc: stable@vger.kernel.org
Signed-off-by: Qiu-ji Chen <chenqiuji666@gmail.com>
---
V2:
Revised the fix approach and updated the description to address the
reduced protection scope of the vc lock in the V1 solution.
---
 drivers/dma/mediatek/mtk-cqdma.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/dma/mediatek/mtk-cqdma.c b/drivers/dma/mediatek/mtk-cqdma.c
index d5ddb4e30e71..e35271ac1eed 100644
--- a/drivers/dma/mediatek/mtk-cqdma.c
+++ b/drivers/dma/mediatek/mtk-cqdma.c
@@ -422,13 +422,10 @@ static struct virt_dma_desc *mtk_cqdma_find_active_desc(struct dma_chan *c,
 	struct virt_dma_desc *vd;
 	unsigned long flags;
 
-	spin_lock_irqsave(&cvc->pc->lock, flags);
 	list_for_each_entry(vd, &cvc->pc->queue, node)
 		if (vd->tx.cookie == cookie) {
-			spin_unlock_irqrestore(&cvc->pc->lock, flags);
 			return vd;
 		}
-	spin_unlock_irqrestore(&cvc->pc->lock, flags);
 
 	list_for_each_entry(vd, &cvc->vc.desc_issued, node)
 		if (vd->tx.cookie == cookie)
@@ -452,9 +449,11 @@ static enum dma_status mtk_cqdma_tx_status(struct dma_chan *c,
 	if (ret == DMA_COMPLETE || !txstate)
 		return ret;
 
+	spin_lock_irqsave(&cvc->pc->lock, flags);
 	spin_lock_irqsave(&cvc->vc.lock, flags);
 	vd = mtk_cqdma_find_active_desc(c, cookie);
 	spin_unlock_irqrestore(&cvc->vc.lock, flags);
+	spin_unlock_irqrestore(&cvc->pc->lock, flags);
 
 	if (vd) {
 		cvd = to_cqdma_vdesc(vd);
-- 
2.34.1

On Thu, 08 May 2025 15:36:33 +0800, Qiu-ji Chen wrote:
> Fix a potential deadlock bug. Observe that in the mtk-cqdma.c
> file, functions like mtk_cqdma_issue_pending() and
> mtk_cqdma_free_active_desc() properly acquire the pc lock before the vc
> lock when handling pc and vc fields. However, mtk_cqdma_tx_status()
> violates this order by first acquiring the vc lock before invoking
> mtk_cqdma_find_active_desc(), which subsequently takes the pc lock. This
> reversed locking sequence (vc → pc) contradicts the established
> pc → vc order and creates deadlock risks.
> 
> [...]

Applied, thanks!

[1/1] dmaengine: mediatek: Fix a possible deadlock error in mtk_cqdma_tx_status()
      commit: 157ae5ffd76a2857ccb4b7ce40bc5a344ca00395

Best regards,
-- 
~Vinod

Il 08/05/25 09:36, Qiu-ji Chen ha scritto:
> Fix a potential deadlock bug. Observe that in the mtk-cqdma.c
> file, functions like mtk_cqdma_issue_pending() and
> mtk_cqdma_free_active_desc() properly acquire the pc lock before the vc
> lock when handling pc and vc fields. However, mtk_cqdma_tx_status()
> violates this order by first acquiring the vc lock before invoking
> mtk_cqdma_find_active_desc(), which subsequently takes the pc lock. This
> reversed locking sequence (vc → pc) contradicts the established
> pc → vc order and creates deadlock risks.
> 
> Fix the issue by moving the vc lock acquisition code from
> mtk_cqdma_find_active_desc() to mtk_cqdma_tx_status(). Ensure the pc lock
> is acquired before the vc lock in the calling function to maintain correct
> locking hierarchy. Note that since mtk_cqdma_find_active_desc() is a
> static function with only one caller (mtk_cqdma_tx_status()), this
> modification safely eliminates the deadlock possibility without affecting
> other components.
> 
> This possible bug is found by an experimental static analysis tool
> developed by our team. This tool analyzes the locking APIs to extract
> function pairs that can be concurrently executed, and then analyzes the
> instructions in the paired functions to identify possible concurrency bugs
> including deadlocks, data races and atomicity violations.
> 
> Fixes: b1f01e48df5a ("dmaengine: mediatek: Add MediaTek Command-Queue DMA controller for MT6765 SoC")
> Cc: stable@vger.kernel.org
> Signed-off-by: Qiu-ji Chen <chenqiuji666@gmail.com>

Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>