Skip to contentSkip to conversation

Solved? - Re: soc-camera: timing out during capture - Re: Testing latest mx3_camera.c

PATCH Applied Agustin Ferrin Pozuelo 5 participants 9 messages May 7, 2009 – May 16, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Holy cow...

Quoted 34 lines · Agustin
Quoted 11 lines · Guennadi Liakhovetski
Quoted 2 lines · Agustin

No, as there is no driver_match_device() in 2.6.29 nor in my patched version. How important is that?

No, sorry, forget it, that's not your problem.

Quoted 3 lines

Meanwhile I noticed that IRQ 176 is being triggered, then discarded as "unhandled" by ipu_idmac, who gives the message "IRQ on active buffer on channel 7, active 0, ready 0, 0, current 0!" below...

Quoted 7 lines

Yes, and this is not good. If you look in drivers/dma/ipu/ipu_idmac.c idmac_interrupt() you'll see, that this message is printed when IDMAC produces an interrupt for a DMA buffer, but at the same time it says, that the buffer, that should have completed is still in use... I've seen a few of such inconsistencies, and up to now always managed to get rid of them in one or another way. But that should not be related to the conversion. Maybe your formats on the sensor and on the SoC do not match, verify that.

Thanks for the tip but I am still out of luck. I enabled DEBUG in ipu_idmac.c just to see that frame start and end are happening more or less when they should:

[...]
camera 0-0: mx3_camera: Submitted cookie 2 DMA 0x86400000
Got SOF IRQ 177 on Channel 7
Got EOF IRQ 178 on Channel 7
dma dma0chan7: ipu_idmac: IDMAC irq 176, buf 0
dma dma0chan7: ipu_idmac: IRQ on active buffer on channel 7, active 0, ready
0, 0, current 0!
Select timeout.
[...]

Quoted 4 lines

I also configured everything to the simplest mode I can have: 8-bit bus, sample falling.

So I am now looking at IDMAC, trying to guess what could be wrong... [snip]

After checking out every single bit in CSI and IDMAC to be correct according to reference and the same I had in the previous/working version...

I thought about the fact that I was only queuing one buffer, and that this might be a corner case as sample code uses a lot of them. And that in the older code that funny things could happen in the handler if we ran out of buffers, though they didn't happen.

So I have queued an extra buffer and voila, got it working.

So thanks again!

However, this could be a bug in ipu_idmac (or some other point), as using a single buffer is very plausible, specially when grabbing huge stills.

--Agustín.

[snip!]

#1 lore raw
Quoted 1 line · Agustin Ferrin Pozuelo

Holy cow...

mu-u-u-u-u-u?:-)

Quoted 9 lines

After checking out every single bit in CSI and IDMAC to be correct according to reference and the same I had in the previous/working version...

I thought about the fact that I was only queuing one buffer, and that this might be a corner case as sample code uses a lot of them. And that in the older code that funny things could happen in the handler if we ran out of buffers, though they didn't happen.

Quoted 7 lines

So I have queued an extra buffer and voila, got it working.

So thanks again!

However, this could be a bug in ipu_idmac (or some other point), as using a single buffer is very plausible, specially when grabbing huge stills.

Great, thanks for testing and debugging! Ok, so, I will have to test this case some time...

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/

#2 lore raw

Grabbing single stills on MX31 - Re: Solved? - Re: soc-camera: timing out during capture - Re: Testing latest mx3_camera.c

Quoted 10 lines · Guennadi Liakhovetski
Quoted 8 lines · Agustin Ferrin Pozuelo

...
I thought about the fact that I was only queuing one buffer, and that this might be a corner case as sample code uses a lot of them. And that in the older code that funny things could happen in the handler if we ran out of buffers, though they didn't happen.

So I have queued an extra buffer and voila, got it working.

Quoted 8 lines
Quoted 5 lines

So thanks again!

However, this could be a bug in ipu_idmac (or some other point), as using a single buffer is very plausible, specially when grabbing huge stills.

Great, thanks for testing and debugging! Ok, so, I will have to test this case some time...

Guennadi,

This workaround (queuing 2 buffers when needing only one) is having the side effect of greatly increasing the time taken.

I did several tests playing with camera vertical blanking and looking at capture times:

Vblank / real / user / sys time:
0 / real 0m 0.90s / user 0m 0.00s / sys 0m 0.34s
1 frame / real 0m 1.04s / user 0m 0.00s / sys 0m 0.34s

2 frames / real    0m 1.18s / user    0m 0.00s / sys     0m 0.33s
2.5 (max)/ real    0m 1.23s / user    0m 0.00s / sys     0m 0.35s

I think the second frame is being captured altogether, and its dma transfer is not allowing any other process to run meanwhile. (VIDIOC_STREAMOFF is being called as soon as the first buffer is ready.)

Do you think it will be too hard to fix?

Regards,
--Agustín.

#3 lore raw

PATCH 1/1 dma: fix ipu_idmac.c to not discard the last queued buffer

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

---

Subject: Re: Grabbing single stills on MX31 - Re: Solved? - Re:
soc-camera: timing out during capture - Re: Testing latest mx3_camera.c

Quoted 11 lines · Agustin
Quoted 9 lines · Guennadi Liakhovetski
Quoted 8 lines · Agustin Ferrin Pozuelo

...
I thought about the fact that I was only queuing one buffer, and that this might be a corner case as sample code uses a lot of them. And that in the older code that funny things could happen in the handler if we ran out of buffers, though they didn't happen.

So I have queued an extra buffer and voila, got it working.

Quoted 8 lines
Quoted 8 lines
Quoted 5 lines

So thanks again!

However, this could be a bug in ipu_idmac (or some other point), as using a single buffer is very plausible, specially when grabbing huge stills.

Great, thanks for testing and debugging! Ok, so, I will have to test this case some time...

Agustin, does this patch fix your problem? Dan, please, pull it as soon as we get a tested-by from Agustin.

Quoted 7 lines

This workaround (queuing 2 buffers when needing only one) is having the side effect of greatly increasing the time taken.

I did several tests playing with camera vertical blanking and looking at capture times:

Vblank / real / user / sys time:

Quoted 8 lines
       0 / real    0m 0.90s / user    0m 0.00s / sys     0m 0.34s
 1 frame / real    0m 1.04s / user    0m 0.00s / sys     0m 0.34s
2 frames / real    0m 1.18s / user    0m 0.00s / sys     0m 0.33s
2.5 (max)/ real    0m 1.23s / user    0m 0.00s / sys     0m 0.35s

I think the second frame is being captured altogether, and its dma transfer is not allowing any other process to run meanwhile. (VIDIOC_STREAMOFF is being called as soon as the first buffer is ready.)

I don't quite understand this. What exactly are you doing above? You submit 2 buffers and change vertical blanking? Where do you change the blanking - in the driver? How exactly?

Quoted 1 line

Do you think it will be too hard to fix?

The blanking issue or the 1-buffer problem?

Thanks
Guennadi

 drivers/dma/ipu/ipu_idmac.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
drivers/dma/ipu/ipu_idmac.c +2 -1
index 3a4deea..9a5bc1a 100644
@@ -1272,7 +1272,8 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 	/* Other interrupts do not interfere with this channel */
 	spin_lock(&ichan->lock);
 	if (unlikely(chan_id != IDMAC_SDC_0 && chan_id != IDMAC_SDC_1 &&
-		     ((curbuf >> chan_id) & 1) == ichan->active_buffer)) {
+		     ((curbuf >> chan_id) & 1) == ichan->active_buffer &&
+		     !list_is_last(ichan->queue.next, &ichan->queue))) {
 		int i = 100;
 
 		/* This doesn't help. See comment in ipu_disable_channel() */
Signature · 1 line
1.6.2.4
#4 lore raw

Re: [PATCH] dma: fix ipu_idmac.c to not discard the last queued buffer

Quoted 6 lines · Guennadi Liakhovetski

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

Reported-by Agustin
Signed-off-by Guennadi Liakhovetski
Signed-off-by Agustin Ferrin Pozuelo
Quoted 24 lines

---

Subject: Re: Grabbing single stills on MX31 - Re: Solved? - Re:
soc-camera: timing out during capture - Re: Testing latest mx3_camera.c

Quoted 17 lines · Agustin
Quoted 15 lines · Guennadi Liakhovetski
Quoted 13 lines · Agustin Ferrin Pozuelo

...
I thought about the fact that I was only queuing one buffer, and that this might be a corner case as sample code uses a lot of them. And that in the older code that funny things could happen in the handler if we ran out of buffers, though they didn't happen.

So I have queued an extra buffer and voila, got it working.

So thanks again!

However, this could be a bug in ipu_idmac (or some other point), as using a single buffer is very plausible, specially when grabbing huge stills.

Quoted 5 lines
Quoted 2 lines
Quoted 2 lines

Great, thanks for testing and debugging! Ok, so, I will have to test this case some time...

Agustin, does this patch fix your problem? Dan, please, pull it as soon as we get a tested-by from Agustin.

Yes it works. And it happens to save 33% of system CPU time in addition to freeing a lot of memory bandwith. Timings after this fix:

  Vblank  /  real        /  user       / sys time:
   [any]  /  real 0.50s  /  user 0.00s / sys 0.22s

(Everything was for a 3888x1944x15bpp still)

Quoted 17 lines
Quoted 8 lines

This workaround (queuing 2 buffers when needing only one) is having the side effect of greatly increasing the time taken.

I did several tests playing with camera vertical blanking and looking at capture times:

Vblank / real / user / sys time:

Quoted 8 lines
       0 / real    0m 0.90s / user    0m 0.00s / sys     0m 0.34s
 1 frame / real    0m 1.04s / user    0m 0.00s / sys     0m 0.34s
2 frames / real    0m 1.18s / user    0m 0.00s / sys     0m 0.33s
2.5 (max)/ real    0m 1.23s / user    0m 0.00s / sys     0m 0.35s

I think the second frame is being captured altogether, and its dma transfer is not allowing any other process to run meanwhile. (VIDIOC_STREAMOFF is being called as soon as the first buffer is ready.)

Quoted 7 lines

I don't quite understand this. What exactly are you doing above? You submit 2 buffers and change vertical blanking? Where do you change the blanking - in the driver? How exactly?

Quoted 1 line

Do you think it will be too hard to fix?

The blanking issue or the 1-buffer problem?

Eh-eh, it was the same!
Thanks a lot!
--Agustín.

Quoted 23 lines

Thanks
Guennadi

drivers/dma/ipu/ipu_idmac.c | 3 ++-

1 files changed, 2 insertions(+), 1 deletions(-)
drivers/dma/ipu/ipu_idmac.c +2 -1
index 3a4deea..9a5bc1a 100644
@@ -1272,7 +1272,8 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
    /* Other interrupts do not interfere with this channel */
    spin_lock(&ichan->lock);
    if (unlikely(chan_id != IDMAC_SDC_0 && chan_id != IDMAC_SDC_1 &&
-             ((curbuf >> chan_id) & 1) == ichan->active_buffer)) {
+             ((curbuf >> chan_id) & 1) == ichan->active_buffer &&
+             !list_is_last(ichan->queue.next, &ichan->queue))) {
        int i = 100;
 
        /* This doesn't help. See comment in ipu_disable_channel() */
Signature · 1 line
1.6.2.4
Applied
#5 lore raw

Re: [PATCH] dma: fix ipu_idmac.c to not discard the last queued buffer

Quoted 11 lines · Agustin
Quoted 6 lines · Guennadi Liakhovetski

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

Reported-by Agustin
Signed-off-by Guennadi Liakhovetski
Signed-off-by Agustin Ferrin Pozuelo

Applied.

#6 lore raw

Re: [PATCH] dma: fix ipu_idmac.c to not discard the last queued buffer

Quoted 8 lines · Dan Williams
Quoted 7 lines · Agustin
Quoted 4 lines · Guennadi Liakhovetski

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Reported-by Agustin
Signed-off-by Guennadi Liakhovetski
Signed-off-by Agustin Ferrin Pozuelo

Applied.

Hopefully with real tags (iow, with email addresses) rather than what's shown above (which is unacceptable.)

#7 lore raw

Re: [PATCH] dma: fix ipu_idmac.c to not discard the last queued buffer

Quoted 12 lines · Russell King - ARM Linux
Quoted 11 lines · Dan Williams
Quoted 10 lines · Agustin
Quoted 6 lines · Guennadi Liakhovetski

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

Reported-by Agustin
Signed-off-by Guennadi Liakhovetski
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Signed-off-by Agustin Ferrin Pozuelo

Applied.

Hopefully with real tags (iow, with email addresses) rather than what's shown above (which is unacceptable.)

Sure, Dan has done it perfectly:

http://git.kernel.org/?p=linux/kernel/git/djbw/async_tx.git;a=commitdiff;h=ad567ffb32f067b30606071eb568cf637fe42185

as it also was in the patch submission

http://marc.info/?l=linux-arm-kernel&m=124212146702853&w=2

and _not_ as it was in the mail that you quote.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/

#8 lore raw

Re: [PATCH] dma: fix ipu_idmac.c to not discard the last queued buffer

Quoted 23 lines · Guennadi Liakhovetski
Quoted 12 lines · Russell King - ARM Linux
Quoted 11 lines · Dan Williams
Quoted 10 lines · Agustin
Quoted 6 lines · Guennadi Liakhovetski

This also fixes the case of a single queued buffer, for example, when taking a single frame snapshot with the mx3_camera driver.

Reported-by Agustin
Signed-off-by Guennadi Liakhovetski
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Signed-off-by Agustin Ferrin Pozuelo

Applied.

Hopefully with real tags (iow, with email addresses) rather than what's shown above (which is unacceptable.)

Sure, Dan has done it perfectly:

Quoted 7 lines

http://git.kernel.org/?p=linux/kernel/git/djbw/async_tx.git;a=commitdiff;h=ad567ffb32f067b30606071eb568cf637fe42185

as it also was in the patch submission

http://marc.info/?l=linux-arm-kernel&m=124212146702853&w=2

and _not_ as it was in the mail that you quote.

Looks like Agustin's mail client is removing the <...> parts from those tags - look at the next message in the thread.