Skip to contentSkip to conversation

staging: emxx_udc: Replace GFP_KERNEL with GFP_ATOMIC

PATCH In review Tapasweni Pathak 2 participants 2 messages Feb 20, 2015 – Feb 24, 2015 Staging

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging: emxx_udc: Replace GFP_KERNEL with GFP_ATOMIC

To avoid deadlock, do not call blocking functions with spinlocks held.

Replace GFP_KERNEL with GFP_ATOMIC, as the latter will fail if the pile doesn't have enough free pages but will not sleep and hence deadlock can be avoided.

Found by Coccinelle.

---
Is there any other way this can be fixed as it is better to avoid GFP_ATOMIC?

 drivers/staging/emxx_udc/emxx_udc.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/emxx_udc/emxx_udc.c +1 -1
index 4be646c..6c1de27 100644
@@ -2794,7 +2794,7 @@ static int nbu2ss_ep_queue(
 		if (ep->virt_buf == NULL)
 			ep->virt_buf = (u8 *)dma_alloc_coherent(
 				NULL, PAGE_SIZE,
-				&ep->phys_buf, GFP_KERNEL | GFP_DMA);
+				&ep->phys_buf, GFP_ATOMIC | GFP_DMA);
 		if (ep->epnum > 0)  {
 			if (ep->direct == USB_DIR_IN)
 				memcpy(ep->virt_buf, req->req.buf,

--
1.7.9.5

maintainer
#1 lore raw
Quoted 11 lines · Tapasweni Pathak

To avoid deadlock, do not call blocking functions with spinlocks held.

Replace GFP_KERNEL with GFP_ATOMIC, as the latter will fail if the pile doesn't have enough free pages but will not sleep and hence deadlock can be avoided.

Found by Coccinelle.

---
Is there any other way this can be fixed as it is better to avoid GFP_ATOMIC?

No, this is fine, nice fix.

thanks,

greg k-h