serial: 8250: fix regression in 8250 uart driver

Apologies, but my original email bounced for Andy and Heikki, resending.
----

Hi Andy,

I saw that you have discovered that commit ec5a11a91eec ("serial:
8250:
Validate dmaengine rx chan meets requirements") introduced a regression
in the 8250 uart driver. For SoCFPGA platform, I am seeing this error:

[    5.541751] ttyS0 - failed to request DMA

Reverting the commit ec5a11a91eec removes the error.

I saw that you started the discussion, but I didn't see that a patch was
included[1].

The following patch seems to fix the error, but I'm not sure if it's the
same fix that you had in mind.

Thanks,
Dinh

[1] http://marc.info/?l=linux-serial&m=146254187602862&w=2

---------------8<------------

drivers/tty/serial/8250/8250_dma.c +1 -1

index 7f33d1c..847a203 100644
@@ -176,7 +176,7 @@ int serial8250_request_dma(struct uart_8250_port
 *p)
         ret = dma_get_slave_caps(dma->rxchan, &caps);
         if (ret)
                 goto release_rx;
-       if (!caps.cmd_pause || !caps.cmd_terminate ||
+       if ((!caps.cmd_pause || !caps.cmd_terminate) &&

            caps.residue_granularity ==
DMA_RESIDUE_GRANULARITY_DESCRIPTOR) {
                ret = -EINVAL;
                goto release_rx;

Quoted 3 lines

Apologies, but my original email bounced for Andy and Heikki, resending.
----

I suppose this one, i.e. commit dd4e91d538b3 ("dmaengine: slave means at least one of DMA_SLAVE, DMA_CYCLIC"), solves the issue.

Quoted 10 lines · Dinh Nguyen

Quoted 3 lines

Apologies, but my original email bounced for Andy and Heikki, resending.
----

I suppose this one, i.e. commit dd4e91d538b3 ("dmaengine: slave means at
least one of DMA_SLAVE, DMA_CYCLIC"), solves the issue.

Hmm...that commit didn't seem to fix my issue.

In regard to Peter's reply:

"The error is that your dmaengine driver doesn't support the functionality
required by the 8250 driver."

Does this mean that the 8250_dw.c driver can no longer use DMA?

Quoted 5 lines · On Tue, 2016-06-14 at 10:27 -0500, [email protected] wrote:

Quoted 4 lines · Dinh Nguyen

I suppose this one, i.e. commit dd4e91d538b3 ("dmaengine: slave means at
least one of DMA_SLAVE, DMA_CYCLIC"), solves the issue.

Hmm...that commit didn't seem to fix my issue.

[…]

Quoted 1 line

Does this mean that the 8250_dw.c driver can no longer use DMA?

It means that previously the use of DMA engine whatever was used is not correct in your case. There is no issues in UART driver, the issue is in DMA controller driver.

I'm working on a quirk that will allow user to enforce PIO mode for Rx/Tx or both. In that case if DMA controller can't be fixed you still might use DMA for Tx.

Quoted 15 lines · Andy Shevchenko

Quoted 5 lines · On Tue, 2016-06-14 at 10:27 -0500, [email protected] wrote:

Quoted 4 lines · Dinh Nguyen

I suppose this one, i.e. commit dd4e91d538b3 ("dmaengine: slave means at
least one of DMA_SLAVE, DMA_CYCLIC"), solves the issue.

Hmm...that commit didn't seem to fix my issue.

[…]

Quoted 1 line

Does this mean that the 8250_dw.c driver can no longer use DMA?

It means that previously the use of DMA engine whatever was used is not
correct in your case. There is no issues in UART driver, the issue is in
DMA controller driver.

Quoted 4 lines

I'm working on a quirk that will allow user to enforce PIO mode for Rx/Tx or both. In that case if DMA controller can't be fixed you still
might use DMA for Tx.

What's the status of this issue?

Just wondering, as this issue is listed
in my regression reports for 4.7 and it seems nothing much happened for
more than a week now – which is a bad sign as the 4.7 release seems only
a week or two away.

Quoted 13 lines · Andy Shevchenko

Quoted 4 lines · On Tue, 2016-06-14 at 10:27 -0500, [email protected] wrote:

Quoted 3 lines · Dinh Nguyen

Does this mean that the 8250_dw.c driver can no longer use DMA?

It means that previously the use of DMA engine whatever was used is not
correct in your case. There is no issues in UART driver, the issue is in
DMA controller driver.

Quoted 4 lines

I'm working on a quirk that will allow user to enforce PIO mode for Rx/Tx or both. In that case if DMA controller can't be fixed you still
might use DMA for Tx.

What's the status of this issue? Just wondering, as this issue is listed
in my regression reports for 4.7 and it seems nothing much happened for
more than a week now – which is a bad sign as the 4.7 release seems only
a week or two away.

Quoted 19 lines · Thorsten Leemhuis

Quoted 13 lines · Andy Shevchenko

Quoted 4 lines · On Tue, 2016-06-14 at 10:27 -0500, [email protected] wrote:

Quoted 3 lines · Dinh Nguyen

Does this mean that the 8250_dw.c driver can no longer use DMA?

It means that previously the use of DMA engine whatever was used is not
correct in your case. There is no issues in UART driver, the issue is in
DMA controller driver.

Quoted 4 lines

I'm working on a quirk that will allow user to enforce PIO mode for Rx/Tx or both. In that case if DMA controller can't be fixed you still
might use DMA for Tx.

Quoted 7 lines

What's the status of this issue? Just wondering, as this issue is listed
in my regression reports for 4.7 and it seems nothing much happened for
more than a week now – which is a bad sign as the 4.7 release seems only
a week or two away.

I sent a v2 of the series for internal review, same you may found on [1]. If Heikki is okay to that I'll send it here.

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk

I sent a v2 of the series for internal review, same you may found on [1]. If Heikki is okay to that I'll send it here.

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk

Hi Andy,

Quoted 5 lines · Andy Shevchenko

I sent a v2 of the series for internal review, same you may found on [1]. If Heikki is okay to that I'll send it here.

Quoted 2 lines

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk

I tested this branch on SoCFPGA hardware and encountered this error[1].
Doing a bisect led me to this commit[2].

Dinh

--------------------[1]------------------------------

[    5.907714] ttyS0 - failed to request DMA
[    5.921937] VFS: Mounted root (nfs filesystem) on device 0:13. [    5.928732] devtmpfs: mounted
[    5.932504] Freeing unused kernel memory: 1024K (c0900000 - c0a00000)
[    6.004532] Unable to handle kernel NULL pointer dereference at virtual address 0000000c
[    6.012619] pgd = ef670000
[    6.015321] [0000000c] *pgd=2ec0a831, *pte=00000000, *ppte=00000000 [    6.021609] Internal error: Oops: 17 [#1] SMP ARM
[    6.026301] Modules linked in:
[    6.029365] CPU: 1 PID: 1 Comm: init Not tainted 4.8.0-rc3-00063-g6ce52b1 #12
[    6.036476] Hardware name: Altera SOCFPGA
[    6.040476] task: ef4ac000 task.stack: ef4b0000
[    6.045007] PC is at _stop+0x4/0x254
[    6.048579] LR is at pl330_terminate_all+0x4c/0x1e4
[    6.053447] pc : [<c03c2e5c>]    lr : [<c03c30f8>]    psr: 60000093 [    6.053447] sp : ef4b1e88  ip : 00000000  fp : 0000a109 [    6.064890] r10: 00000000  r9 : ef7c7cc0  r8 : eedfaa94 [    6.070099] r7 : eed30410  r6 : eec2be0c  r5 : 60000013  r4 :
eec2be70
[    6.076605] r3 : 00000002  r2 : 00000002  r1 : 00000000  r0 :
00000000
[    6.083113] Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM Segment none
[    6.090318] Control: 10c5387d  Table: 2f67004a  DAC: 00000051 [    6.096046] Process init (pid: 1, stack limit = 0xef4b0218) [    6.101603] Stack: (0xef4b1e88 to 0xef4b2000)
[    6.105953] 1e80:                   eec2be70 60000013 eec2be0c eedfaa94 ef7c7cc0 c03c30f8
[    6.114112] 1ea0: ef4b1ea0 ef4b1ea0 eedfaa00 eed8c02c eec2be20 20000013 c0a66208 c03f2b44
[    6.122272] 1ec0: 00000001 00000000 c0a66208 c0a66208 c0a66324 20000013 c0a66208 eedfaa94
[    6.130432] 1ee0: ef7c7cc0 c03f041c eedfaa00 c0a66208 eedfaa8c c03ea774 eedfaa00 ef7cec00
[    6.138592] 1f00: c0a66208 00000001 eedfaa94 c03eb9a4 ef7cec00 ef1a2930 00000000 ef41bd90
[    6.146751] 1f20: ef1a4440 c03d0790 00080040 00080060 00000000 00000000 00002710 ef7c7cc0
[    6.154911] 1f40: ef1a2930 00000000 ef41bd90 ef1a4440 00000008 ef7c7cc8 0000a109 c021b594
[    6.163071] 1f60: 00000000 00000000 ef4ac3d4 c0a532f8 00000000 ef4ac000 c0107884 ef4b0000
[    6.171231] 1f80: 00000000 c0138a14 ef4b0000 ef4b1fb0 c0107884 00000006 c0107884 c010b274
[    6.179390] 1fa0: 00015500 00000000 bea16c4c c0107714 00000000 00000800 00000045 0000c79c
[    6.187550] 1fc0: 00015500 00000000 bea16c4c 00000006 bea16bc0 00009c59 10000000 0000a109
[    6.195709] 1fe0: 00000000 bea16bb4 0000bf81 b6f6413c 40000010 00000002 00000c84 01000002
[    6.203878] [<c03c2e5c>] (_stop) from [<c03c30f8>] (pl330_terminate_all+0x4c/0x1e4)
[    6.211526] [<c03c30f8>] (pl330_terminate_all) from [<c03f2b44>] (serial8250_release_dma+0x30/0x1c8)
[    6.220640] [<c03f2b44>] (serial8250_release_dma) from [<c03f041c>] (serial8250_do_shutdown+0x64/0x144)
[    6.230019] [<c03f041c>] (serial8250_do_shutdown) from [<c03ea774>] (uart_shutdown+0xd0/0x104)
[    6.238614] [<c03ea774>] (uart_shutdown) from [<c03eb9a4>] (uart_close+0x148/0x298)
[    6.246266] [<c03eb9a4>] (uart_close) from [<c03d0790>] (tty_release+0xf0/0x4d4)
[    6.253658] [<c03d0790>] (tty_release) from [<c021b594>] (__fput+0x80/0x1c8)
[    6.260694] [<c021b594>] (__fput) from [<c0138a14>] (task_work_run+0xb8/0xe8)
[    6.267822] [<c0138a14>] (task_work_run) from [<c010b274>] (do_work_pending+0x7c/0xa4)
[    6.275727] [<c010b274>] (do_work_pending) from [<c0107714>] (slow_work_pending+0xc/0x20)
[    6.283888] Code: e8bd8070 e3a04000 eafffff9 e92d4370 (e590300c) [    6.289963] ---[ end trace e214cd70660516e6 ]---
[    6.294806] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
[    6.294806]
[    6.303932] CPU0: stopping
[    6.306645] CPU: 0 PID: 0 Comm: swapper/0 Tainted: G      D 4.8.0-rc3-00063-g6ce52b1 #12
[    6.315406] Hardware name: Altera SOCFPGA
[    6.319425] [<c010fef0>] (unwind_backtrace) from [<c010b8a4>] (show_stack+0x10/0x14)
[    6.327159] [<c010b8a4>] (show_stack) from [<c037aa18>] (dump_stack+0x84/0x98)
[    6.334372] [<c037aa18>] (dump_stack) from [<c010edb4>] (handle_IPI+0x2a0/0x2bc)
[    6.341755] [<c010edb4>] (handle_IPI) from [<c01014cc>] (gic_handle_irq+0x88/0x8c)
[    6.349310] [<c01014cc>] (gic_handle_irq) from [<c010c38c>] (__irq_svc+0x6c/0x90)
[    6.356775] Exception stack(0xc0a01f50 to 0xc0a01f98)
[    6.361815] 1f40:                                     00000000 ef9cd368 0000153c c01184a0
[    6.369976] 1f60: c0a00000 c0a024a4 c0a023c0 c0a50390 c0a02454 00000001 c0a02454 00000000
[    6.378135] 1f80: 60000013 c0a01fa0 c01081a8 c01081ac 60000013 ffffffff
[    6.384734] [<c010c38c>] (__irq_svc) from [<c01081ac>] (arch_cpu_idle+0x38/0x3c)
[    6.392127] [<c01081ac>] (arch_cpu_idle) from [<c0158cec>] (cpu_startup_entry+0x204/0x25c)
[    6.400379] [<c0158cec>] (cpu_startup_entry) from [<c0900c60>] (start_kernel+0x370/0x37c)
[    6.408542] ---[ end Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
[    6.408542]

--------------------[2]---------------------------------- commit 60a0605738814ca8fa180cdfef20fcc2a32ec1ef
Author: Andy Shevchenko <[email protected]> Date:   Mon Jun 13 12:17:03 2016 +0300

    serial: 8250: allow user to skip using of DMA channels

    Currently DMA mode is enforced if the certain driver has such support. It prevents user to enforce PIO mode for RX, TX, or both.

    The new module parameters skip_rxdma and skip_txdma allow user to choose between auto mode, which is current and default behaviour, disable RX, TX, or both DMA channels without hacking a kernel.

Signed-off-by Andy Shevchenko <[email protected]>

Hi Andy,

Quoted 5 lines · Andy Shevchenko

I sent a v2 of the series for internal review, same you may found on [1]. If Heikki is okay to that I'll send it here.

Quoted 2 lines

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk

I tested this branch on SoCFPGA hardware and encountered this error[1].
Doing a bisect led me to this commit[2].

Quoted 8 lines · Dinh Nguyen

Hi Andy,

Quoted 5 lines · Andy Shevchenko

I sent a v2 of the series for internal review, same you may found on [1]. If Heikki is okay to that I'll send it here.

Quoted 6 lines

Quoted 2 lines

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk

I tested this branch on SoCFPGA hardware and encountered this error[1].
Doing a bisect led me to this commit[2].

Forgot to add you to Cc list, there is new version there.

Quoted 21 lines · Andy Shevchenko

Quoted 18 lines · Dinh Nguyen

Hi Andy,

Quoted 10 lines · Andy Shevchenko

I sent a v2 of the series for internal review, same you may found on
[1]. If Heikki is okay to that I'll send it here.

[1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqr
k

I tested this branch on SoCFPGA hardware and encountered this error[1].
Doing a bisect led me to this commit[2].

Quoted 2 lines

Forgot to add you to Cc list, there is new version there.

I'm no longer getting the previous error, but I am still seeing this:

[    6.025749] ttyS0 - failed to request DMA