* [PATCH] dmaengine: pxa_dma: fix hotchain corner case
@ 2016-08-07 19:01 ` Robert Jarzmik
0 siblings, 0 replies; 6+ messages in thread
From: Robert Jarzmik @ 2016-08-07 19:01 UTC (permalink / raw)
To: Daniel Mack, Haojian Zhuang, Robert Jarzmik, Vinod Koul
Cc: linux-arm-kernel, dmaengine, linux-kernel
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(-)
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;
}
--
2.1.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH] dmaengine: pxa_dma: fix debug message
2016-08-07 19:01 ` Robert Jarzmik
@ 2016-08-07 19:01 ` Robert Jarzmik
-1 siblings, 0 replies; 6+ messages in thread
From: Robert Jarzmik @ 2016-08-07 19:01 UTC (permalink / raw)
To: linux-arm-kernel
In a very tight timeframe, the debug message in the transfer completion
handler can be misleading, as the completion test report can change just
after the message, and the code flow cannot be deduced from the debug
message.
This is just a cleanup to make debugging easier.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
---
drivers/dma/pxa_dma.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/dma/pxa_dma.c b/drivers/dma/pxa_dma.c
index 64758daa9eee..b7414c5c9932 100644
--- a/drivers/dma/pxa_dma.c
+++ b/drivers/dma/pxa_dma.c
@@ -668,6 +668,7 @@ static irqreturn_t pxad_chan_handler(int irq, void *dev_id)
struct virt_dma_desc *vd, *tmp;
unsigned int dcsr;
unsigned long flags;
+ bool vd_completed;
dma_cookie_t last_started = 0;
BUG_ON(!chan);
@@ -678,15 +679,17 @@ static irqreturn_t pxad_chan_handler(int irq, void *dev_id)
spin_lock_irqsave(&chan->vc.lock, flags);
list_for_each_entry_safe(vd, tmp, &chan->vc.desc_issued, node) {
+ vd_completed = is_desc_completed(vd);
dev_dbg(&chan->vc.chan.dev->device,
- "%s(): checking txd %p[%x]: completed=%d\n",
- __func__, vd, vd->tx.cookie, is_desc_completed(vd));
+ "%s(): checking txd %p[%x]: completed=%d dcsr=0x%x\n",
+ __func__, vd, vd->tx.cookie, vd_completed,
+ dcsr);
last_started = vd->tx.cookie;
if (to_pxad_sw_desc(vd)->cyclic) {
vchan_cyclic_callback(vd);
break;
}
- if (is_desc_completed(vd)) {
+ if (vd_completed) {
list_del(&vd->node);
vchan_cookie_complete(vd);
} else {
--
2.1.4
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH] dmaengine: pxa_dma: fix debug message
@ 2016-08-07 19:01 ` Robert Jarzmik
0 siblings, 0 replies; 6+ messages in thread
From: Robert Jarzmik @ 2016-08-07 19:01 UTC (permalink / raw)
To: Daniel Mack, Haojian Zhuang, Robert Jarzmik, Vinod Koul
Cc: linux-arm-kernel, dmaengine, linux-kernel
In a very tight timeframe, the debug message in the transfer completion
handler can be misleading, as the completion test report can change just
after the message, and the code flow cannot be deduced from the debug
message.
This is just a cleanup to make debugging easier.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
---
drivers/dma/pxa_dma.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/dma/pxa_dma.c b/drivers/dma/pxa_dma.c
index 64758daa9eee..b7414c5c9932 100644
--- a/drivers/dma/pxa_dma.c
+++ b/drivers/dma/pxa_dma.c
@@ -668,6 +668,7 @@ static irqreturn_t pxad_chan_handler(int irq, void *dev_id)
struct virt_dma_desc *vd, *tmp;
unsigned int dcsr;
unsigned long flags;
+ bool vd_completed;
dma_cookie_t last_started = 0;
BUG_ON(!chan);
@@ -678,15 +679,17 @@ static irqreturn_t pxad_chan_handler(int irq, void *dev_id)
spin_lock_irqsave(&chan->vc.lock, flags);
list_for_each_entry_safe(vd, tmp, &chan->vc.desc_issued, node) {
+ vd_completed = is_desc_completed(vd);
dev_dbg(&chan->vc.chan.dev->device,
- "%s(): checking txd %p[%x]: completed=%d\n",
- __func__, vd, vd->tx.cookie, is_desc_completed(vd));
+ "%s(): checking txd %p[%x]: completed=%d dcsr=0x%x\n",
+ __func__, vd, vd->tx.cookie, vd_completed,
+ dcsr);
last_started = vd->tx.cookie;
if (to_pxad_sw_desc(vd)->cyclic) {
vchan_cyclic_callback(vd);
break;
}
- if (is_desc_completed(vd)) {
+ if (vd_completed) {
list_del(&vd->node);
vchan_cookie_complete(vd);
} else {
--
2.1.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH] dmaengine: pxa_dma: fix hotchain corner case
2016-08-07 19:01 ` Robert Jarzmik
@ 2016-08-19 6:32 ` Vinod Koul
-1 siblings, 0 replies; 6+ messages in thread
From: Vinod Koul @ 2016-08-19 6:32 UTC (permalink / raw)
To: linux-arm-kernel
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
--
~Vinod
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] dmaengine: pxa_dma: fix hotchain corner case
@ 2016-08-19 6:32 ` Vinod Koul
0 siblings, 0 replies; 6+ messages in thread
From: Vinod Koul @ 2016-08-19 6:32 UTC (permalink / raw)
To: Robert Jarzmik
Cc: Daniel Mack, Haojian Zhuang, linux-arm-kernel, dmaengine,
linux-kernel
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
--
~Vinod
^ permalink raw reply [flat|nested] 6+ messages in thread