| Message ID | 1470596509-1726-1-git-send-email-robert.jarzmik@free.fr (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
On Sun, Aug 07, 2016 at 09:01:48PM +0200, Robert Jarzmik wrote: > In the case where a descriptor is chained on a running channel, and as > explained in the comment in the code 10 lines above, the success of the > chaining is ensured either if : > - the DMA is still running > - or if the chained transfer is completed > > Unfortunately the transfer completness test was done on the descriptor > to which the transfer was chained, and not the transfer being chained at > the end, ie. hot-chained. > > This corner case is extremely hard to trigger, as usually the DMA chain > is still running, and the first case takes care of returning success of > the hot-chaining. It was seen by hot-chaining several "small transfers" > to a running "big transfer", not in a real-life usecase but by testing > the robustness of the driver. Applied, thanks
diff --git a/drivers/dma/pxa_dma.c b/drivers/dma/pxa_dma.c index e756a30ccba2..64758daa9eee 100644 --- a/drivers/dma/pxa_dma.c +++ b/drivers/dma/pxa_dma.c @@ -635,7 +635,7 @@ static bool pxad_try_hotchain(struct virt_dma_chan *vc, vd_last_issued = list_entry(vc->desc_issued.prev, struct virt_dma_desc, node); pxad_desc_chain(vd_last_issued, vd); - if (is_chan_running(chan) || is_desc_completed(vd_last_issued)) + if (is_chan_running(chan) || is_desc_completed(vd)) return true; }
In the case where a descriptor is chained on a running channel, and as explained in the comment in the code 10 lines above, the success of the chaining is ensured either if : - the DMA is still running - or if the chained transfer is completed Unfortunately the transfer completness test was done on the descriptor to which the transfer was chained, and not the transfer being chained at the end, ie. hot-chained. This corner case is extremely hard to trigger, as usually the DMA chain is still running, and the first case takes care of returning success of the hot-chaining. It was seen by hot-chaining several "small transfers" to a running "big transfer", not in a real-life usecase but by testing the robustness of the driver. Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> --- drivers/dma/pxa_dma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)