Skip to contentSkip to conversation

commit 271e1b86e691 is breaking DMA uart on SoCFPGA

DISCUSSION Dinh Nguyen 5 participants 10 messages Feb 10, 2016 – Feb 25, 2016

lore
Share & tools
lore
#0 lore raw

Hi Vinod,

It appears that commit 271e1b86e691 "dmaengine: pl330: add quirk for broken no flushp" is breaking uart dma on SoCFPGA. This commit is in linux-next(next-20160210).

Doing a bisect pointed to commit 271e1b86e691, but I had to also revert the following commits (86a8ce7d4103 "dmaengine: pl330: add max burst for dmaengine") and (848e9776fee4 "dmaengine: pl330: support burst mode for dev-to-mem and mem-to-dev transmit") as well because the build would fail if I only revert 271e1b86e691.

The SoCFPGA platform is being tested on Olof's board farm and is failing multi_v7_defconfig build because the PL330 DMA is enabled in this defconfig.

After doing reverts of those 3 commits, the socfpga platform is able boot completely again with multi_v7_defconfig.

Dinh

#1 lore raw
Quoted 12 lines · Dinh Nguyen

Hi Vinod,

It appears that commit 271e1b86e691 "dmaengine: pl330: add quirk for broken no flushp" is breaking uart dma on SoCFPGA. This commit is in linux-next(next-20160210).

Doing a bisect pointed to commit 271e1b86e691, but I had to also revert the following commits (86a8ce7d4103 "dmaengine: pl330: add max burst for dmaengine") and (848e9776fee4 "dmaengine: pl330: support burst mode for dev-to-mem and mem-to-dev transmit") as well because the build would fail if I only revert 271e1b86e691.

Quoted 5 lines

The SoCFPGA platform is being tested on Olof's board farm and is failing multi_v7_defconfig build because the PL330 DMA is enabled in this defconfig.

After doing reverts of those 3 commits, the socfpga platform is able boot completely again with multi_v7_defconfig.

Okay I am removing this topic from next. Also I can drop the series unless someone sends me a fix in couple of days..

Thanks

Signature · 1 line
~Vinod
#2 lore raw

Hi,

Quoted 13 lines · Vinod Koul
Quoted 12 lines · Dinh Nguyen

Hi Vinod,

It appears that commit 271e1b86e691 "dmaengine: pl330: add quirk for broken no flushp" is breaking uart dma on SoCFPGA. This commit is in linux-next(next-20160210).

Doing a bisect pointed to commit 271e1b86e691, but I had to also revert the following commits (86a8ce7d4103 "dmaengine: pl330: add max burst for dmaengine") and (848e9776fee4 "dmaengine: pl330: support burst mode for dev-to-mem and mem-to-dev transmit") as well because the build would fail if I only revert 271e1b86e691.

Quoted 8 lines
Quoted 5 lines

The SoCFPGA platform is being tested on Olof's board farm and is failing multi_v7_defconfig build because the PL330 DMA is enabled in this defconfig.

After doing reverts of those 3 commits, the socfpga platform is able boot completely again with multi_v7_defconfig.

Okay I am removing this topic from next. Also I can drop the series unless someone sends me a fix in couple of days..

I've noticed that these commits also break UART DMA on Samsung Exynos4412 SoCs (I got similar bisection results as Dinh when I've noticed issue with next-20160211 kernel).

Caesar, please keep me on Cc: for potential fix(es) / revised patches (I'll be happy to test them).

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

#3 lore raw

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

Regards,
Alexander.

#9 lore raw
In reply to Bartlomiej Zolnierkiewicz

Hello, here comments from Shawn Lin:

19 февр. 2016 г., в 9:46, Shawn Lin <[email protected]> написал(а):

Seriously review the patch again, commit 271e1b86e691 "dmaengine:
pl330: add quirk for broken no flushp" has not chance to break the platforms which don't add quirk inside the dts stuff. as well as the other works from rockchip.

But refer to Boojin Kim's patch,
dmaengine: pl330: support burst mode for dev-to-mem and mem-to-dev transmit

may some platforms can't support multi-burst? If that's the case, the been-broken driver should limit the maxburst.

2016-02-17 14:12 GMT+03:00 Bartlomiej Zolnierkiewicz <[email protected]>:

Quoted 39 lines

Hi,

Quoted 13 lines · Vinod Koul
Quoted 12 lines · Dinh Nguyen

Hi Vinod,

It appears that commit 271e1b86e691 "dmaengine: pl330: add quirk for broken no flushp" is breaking uart dma on SoCFPGA. This commit is in linux-next(next-20160210).

Doing a bisect pointed to commit 271e1b86e691, but I had to also revert the following commits (86a8ce7d4103 "dmaengine: pl330: add max burst for dmaengine") and (848e9776fee4 "dmaengine: pl330: support burst mode for dev-to-mem and mem-to-dev transmit") as well because the build would fail if I only revert 271e1b86e691.

Quoted 8 lines
Quoted 5 lines

The SoCFPGA platform is being tested on Olof's board farm and is failing multi_v7_defconfig build because the PL330 DMA is enabled in this defconfig.

After doing reverts of those 3 commits, the socfpga platform is able boot completely again with multi_v7_defconfig.

Okay I am removing this topic from next. Also I can drop the series unless someone sends me a fix in couple of days..

I've noticed that these commits also break UART DMA on Samsung Exynos4412 SoCs (I got similar bisection results as Dinh when I've noticed issue with next-20160211 kernel).

Caesar, please keep me on Cc: for potential fix(es) / revised patches (I'll be happy to test them).

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

#4 lore raw
In reply to Alexander Kochetkov

Hi,

Sorry that I couldn't get to this sooner.

Quoted 6 lines · Alexander Kochetkov

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

SoCFPGA is using the 8250_dw.c uart driver.

Quoted 8 lines

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

Quoted 4 lines

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

This case still fails for me.

Quoted 3 lines

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

This case works and the UART is able to operate in DMA mode.

Dinh

#5 lore raw

Hi,

Thanks Dinh & Alexander for the intergration testing.

---

@Dinh,

Can you test it again with the patch[0]?

patch[0]:
https://patchwork.kernel.org/patch/8398511/

-
Caesar

Quoted 40 lines · Dinh Nguyen

Hi,

Sorry that I couldn't get to this sooner.

Quoted 6 lines · Alexander Kochetkov

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

SoCFPGA is using the 8250_dw.c uart driver.

Quoted 8 lines

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

Quoted 4 lines

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

This case still fails for me.

Quoted 3 lines

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

This case works and the UART is able to operate in DMA mode.

Dinh

Signature · 2 lines
Thanks,
Caesar
#8 lore raw
In reply to Alexander Kochetkov

Hi,

Quoted 5 lines · Alexander Kochetkov

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

drivers/tty/serial/samsung.c

Quoted 11 lines

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

This fails (UART DMA doesn't work).

Quoted 2 lines

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

This succeeds (UART DMA works fine).

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

#6 lore raw
In reply to Caesar Wang

Hi,

Quoted 4 lines · Caesar Wang

Hi,

Thanks Dinh & Alexander for the intergration testing.

Quoted 8 lines

---

@Dinh,

Can you test it again with the patch[0]?

patch[0]:
https://patchwork.kernel.org/patch/8398511/

This fixes the problem for me (tested on Samsung Exynos4412 SoC based Trats2 board).

You may add:

to your patch.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

Quoted 36 lines

-
Caesar

Quoted 33 lines · Dinh Nguyen

Hi,

Sorry that I couldn't get to this sooner.

Quoted 6 lines · Alexander Kochetkov

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

SoCFPGA is using the 8250_dw.c uart driver.

Quoted 8 lines

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

Quoted 4 lines

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

This case still fails for me.

Quoted 3 lines

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

This case works and the UART is able to operate in DMA mode.

Dinh

#7 lore raw
Quoted 13 lines · Bartlomiej Zolnierkiewicz

Hi,

Quoted 10 lines · Caesar Wang

Hi,

Thanks Dinh & Alexander for the intergration testing.

---

@Dinh,

Can you test it again with the patch[0]?

Quoted 8 lines
Quoted 2 lines

patch[0]:
https://patchwork.kernel.org/patch/8398511/

This fixes the problem for me (tested on Samsung Exynos4412 SoC based Trats2 board).

You may add:

Thanks Bartlomiej for sharing the test result.:-)

-
Caesar

Quoted 46 lines

to your patch.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

Quoted 36 lines

-
Caesar

Quoted 33 lines · Dinh Nguyen

Hi,

Sorry that I couldn't get to this sooner.

Quoted 6 lines · Alexander Kochetkov

Hello!

Bartlomiej, could you please tell what uart driver is used on Samsung Exynos4412?
Dinh, could you please tell what uart driver is used on SoCFPGA?

SoCFPGA is using the 8250_dw.c uart driver.

Quoted 8 lines

Could you make two test?

1) Checkout tree at commit 848e9776fee4 "dmaengine: pl330: support burst mode for
dev-to-mem and mem-to-dev transmit", make it buildable and see is it
works. In order to make in buildable remove bursts argument from the line:

Quoted 4 lines

if (*bursts == 1)
- return _bursts(pl330, dry_run, buf, pxs, 1);
+ return _bursts(dry_run, buf, pxs, 1);

This case still fails for me.

Quoted 3 lines

2) Checkout next-20160211 kernel tree and set src_maxburst and dst_maxburst to 1 inside UART driver to see is it works?

This case works and the UART is able to operate in DMA mode.

Dinh

Signature · 2 lines
Thanks,
Caesar