spi-atmel.c: problem with PDC on AT91SAM9G20

spi-atmel.c: problem with PDC on AT91SAM9G20

[Igor Plyatov]

Dear Nicolas, Mark, Cyrille and other developers,

please help to manage issue with data corruption by PDC of SPI.


Versions
--------

I use linux4sam kernel 4.9.36+ 
(7e82b52ca2286e9823d2467b64bfe78980b464b7) on the custom AT91SAM9G20 
board (Stamp9G20 SOM from Taskit.de).

The drivers/spi/spi-atmel.c updated by patches from upstream kernel, 
where last commit is 7094576ccdc3acfe1e06a1e2ab547add375baf7f
Author: Cyrille Pitchen <cyrille.pitchen-UWL1GkI3JZL3oGB3hsPCZA@public.gmane.org>
Date:   Fri Jun 23 17:39:16 2017 +0200

     spi: atmel: fix corrupted data issue on SAM9 family SoCs


Issue
-----

Data corruption happens during receiving of SPI data by kernel. It 
happens quite often (once per minute or so) if OS has some light load, 
like periodical writes to SD card in parallel with communication to SPI 
device (DSP).

Corruption of received SPI data by AT91SAM9G20 CPU confirmed with help 
of logical analyzer.
Analyzer is connected to SPI bus right near CPU.
DSP print SPI data to own serial console for debugging purposes.
Logical analyzer show exactly same data which was printed by DSP.
Kernel driver for DSP print SPI data, where some bytes corrupted, into 
kernel logs for debugging purposes.

The DSP and its linux driver has custom protocol to control correctness 
of communication between Linux host and DSP.
Protocol is quite simple:
* Half duplex.
* Main data transceived as 32 bytes long packets with CRC8 as byte #32.
* Each 32 bytes packet ACKed by one byte (error code) sent back in responce.

This protocol allow to control data integrity during communication 
between Linux host and DSP.


spi-atmel.c reports
-------------------

atmel_spi fffc8000.spi: Atmel SPI Controller version 0x199 at 0xfffc8000 
(irq 32)
atmel_spi fffcc000.spi: Atmel SPI Controller version 0x199 at 0xfffcc000 
(irq 33)

And works with use of SPI PDC.


DSP & Logical Analyzer Data
---------------------------

To Linux Host:   01 01 43 F3 00 01 43 E8 00 01 43 EB 00 01 43 E1
To Linux Host:   00 01 43 F4 00 01 43 E1 00 01 43 F0 00 01 43 22
 From Linux Host: CRC error.


Kernel Data
-----------

 From DSP:        01 01 43 F3 00 01 43 E8 00 01 43 EB 00 01 43 E1
 From DSP:        00 01 43 F4 00 01 43 E1 00 01 F0 F0 00 01 43 22
                                                **
To DSP: CRC error.

** Corrupted byte.


The SPI PIO mode works without data corruption, but this is not an 
option, because communication can have critically big lags if OS has 
some load. These lags are not acceptable for our high speed DSP.

Best wishes.
--
Igor Plyatov
--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: spi-atmel.c: problem with PDC on AT91SAM9G20

[Igor Plyatov]

Hello!


> Data corruption happens during receiving of SPI data by kernel. It 
> happens quite often (once per minute or so) if OS has some light load, 
> like periodical writes to SD card in parallel with communication to 
> SPI device (DSP).
... skipped


For those, who has same interest or encountered the same issue as I had...

Notes:
A) My "gs_mgms_dsp" linux driver is the same for linux-2.6.39 and 
linux-4.9.36
    and communicates with DSP through SPI-bus, where data packets are 32 
byte
    long and last byte is CRC8 to check data integrity.
B) CPU is AT91SAM9G20.

I have encoutered SPI data corruption during receiving of data from DSP by
linux-4.9.36 if SPI-bus has big traffic in parallel with big traffic at MMC
interface.

Old linux-2.6.39 does not suffer from such issue, because atmel-mci.c 
driver has
PIO access to MMC interface. While new kernel has changed the 
atmel-mci.c driver
for use of PDC (Peripheral DMA Controller) for access to MMC interface.

Both kernels have Atmel SPI driver where PDC used for SPI data transfers.

SPI data corruption looks like duplication of one of received bytes.
Such bytes surrounded by "**" at log below.

gs_mgms_dsp spi32766.0: CRC error.
gs_mgms_dsp spi32766.0: <- 35 36 37 38 39 3A 3B 3C 3D  3E 3F  40 41 42 43 44
gs_mgms_dsp spi32766.0: <- 45 46 47 48 49 4A 4B 4C 4D *4F 4F* 50 51 52 53 A4
gs_mgms_dsp spi32766.0: -> EIO=0x05
gs_mgms_dsp spi32766.0: CRC error.
gs_mgms_dsp spi32766.0: <- 0F C2 C3 C4 C5 C6 C7 C8 C9  CA CB  CC CD CE CF D0
gs_mgms_dsp spi32766.0: <- D1 D2 D3 D4 D5 D6 D7 D8 D9 *DB DB* DC DD DE DF 02
gs_mgms_dsp spi32766.0: -> EIO=0x05
gs_mgms_dsp spi32766.0: CRC error.
gs_mgms_dsp spi32766.0: <-  03 5A  5B 5C 5D 5E 5F 60 61 62 63 64 65 66 67 68
gs_mgms_dsp spi32766.0: <- *6A 6A* 6B 6C 6D 6E 6F 70 71 72 73 74 75 76 77 25
gs_mgms_dsp spi32766.0: -> EIO=0x05
gs_mgms_dsp spi32766.0: CRC error.
gs_mgms_dsp spi32766.0: <- 15 76 77 78 79 7A  7B 7C  7D 7E 7F 80 81 82 83 84
gs_mgms_dsp spi32766.0: <- 85 86 87 88 89 8A *8C 8C* 8D 8E 8F 90 91 92 93 00
gs_mgms_dsp spi32766.0: -> EIO=0x05
gs_mgms_dsp spi32766.0: CRC error.
gs_mgms_dsp spi32766.0: <- 2A EC ED EE EF F0 F1  F2 F3  F4 F5 F6 F7 F8 F9 FA
gs_mgms_dsp spi32766.0: <- FB FC FD FE FF 00 01 *03 03* 04 05 06 07 08 09 0A
gs_mgms_dsp spi32766.0: -> EIO=0x05

This looks like silent SPI overruns not detected by AT91SAM9G20 HW.

At the end, after some seconds or milliseconds of communication, HW SPI 
overrun
flag detected by the spi-atmel.c driver:
"atmel_spi fffcc000.spi: overrun (0/0 remaining)".

Strictly speaking this is not a regression of spi-atmel.c, but unfortunate
combination with atmel-mmc.c, where PDC used too and I suppose a HW bug in
AT91SAM9G20 CPU.

Please help to answer on questions:
   1) How to modify atmel-mci.c driver to have option for PIO access to MMC
      interface?
   2) Why SPI overrun flag does not asserted each time? I have not found 
such HW bug
      in the Errata for AT91SAM9G20 CPU.
--
To unsubscribe from this list: send the line "unsubscribe linux-spi" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html