Skip to contentSkip to conversation

spi: imx: Increase imx51 ecspi burst length fails on imx6dl and imx8mm

DISCUSSION [email protected] 4 participants 20 messages Nov 18, 2023 – Dec 7, 2023 3 messages not in the archive

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

After upgrade from kernel 6.5.11 to 6.6.1 the spi-devices on my hw colibri-imx6dl and verdin-imx8mm are not working anymore (TPM2 and SPI-SRAM).

Analyzing the problem showed that the 2 commits introduced the problem:

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Reverting the commits solved the problem.

The analyse with the logic-analyser showed a wrong number of transmitted bytes and wrong data.
When I try to send 127 Byte with a incrementing pattern (0x01,0x02,0x03,..), 504 Bytes are sent (0x00,0x00,0x01 0x00,0x00,0x00,0x02, 0x00,0x00,0x00,0x03). We tested with different sizes and patterns, all are not ok. While analysing the configuration and code I was not able to see any obvious mistake.

Has someone else discovered such misbehaviour or has any idea what is wrong? Best Regards
Stefan

#1 lore raw

[CCing people that worked on the relevant commits as well as the regression list, as it should be in the loop for regressions:
https://docs.kernel.org/admin-guide/reporting-regressions.html]

Quoted 21 lines · On 18.11.23 15:25, [email protected] wrote:

After upgrade from kernel 6.5.11 to 6.6.1 the spi-devices on my hw colibri-imx6dl and verdin-imx8mm are not working anymore (TPM2 and SPI-SRAM).

Analyzing the problem showed that the 2 commits introduced the problem:

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Reverting the commits solved the problem.

The analyse with the logic-analyser showed a wrong number of transmitted bytes and wrong data.
When I try to send 127 Byte with a incrementing pattern (0x01,0x02,0x03,..), 504 Bytes are sent (0x00,0x00,0x01 0x00,0x00,0x00,0x02, 0x00,0x00,0x00,0x03). We tested with different sizes and patterns, all are not ok. While analysing the configuration and code I was not able to see any obvious mistake.

Quoted 2 lines

Has someone else discovered such misbehaviour or has any idea what is wrong? Best Regards

Thx for the report!

[TLDR: I'm adding this report to the list of tracked Linux kernel regressions; the text you find below is based on a few templates paragraphs you might have encountered already in similar form. See link in footer if these mails annoy you.]

To ensure the issue doesn't fall through the cracks unnoticed, I'm adding it to regzbot, the Linux kernel regression tracking bot:

#regzbot ^introduced 15a6af94a27
#regzbot title spi: spi-devices on colibri-imx6dl and verdin-imx8mm are not working anymore
#regzbot ignore-activity

This isn't a regression? This issue or a fix for it are already discussed somewhere else? It was fixed already? You want to clarify when the regression started to happen? Or point out I got the title or something else totally wrong? Then just reply and tell me -- ideally while also telling regzbot about it, as explained by the page listed in the footer of this mail.

Developers: When fixing the issue, remember to add 'Link:' tags pointing to the report (the parent of this mail). See page linked in footer for details.

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
That page also explains what to do if mails like this annoy you.

#2 lore raw
Quoted 3 lines · Linux regression tracking (Thorsten Leemhuis)

[CCing people that worked on the relevant commits as well as the regression list, as it should be in the loop for regressions:
https://docs.kernel.org/admin-guide/reporting-regressions.html]

Sorry for the spam, hit "send" before actually adding Stefan and Mark, which now are among the recipients.

Quoted 56 lines · On 18.11.23 15:25, [email protected] wrote:
Quoted 21 lines

After upgrade from kernel 6.5.11 to 6.6.1 the spi-devices on my hw colibri-imx6dl and verdin-imx8mm are not working anymore (TPM2 and SPI-SRAM).

Analyzing the problem showed that the 2 commits introduced the problem:

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Reverting the commits solved the problem.

The analyse with the logic-analyser showed a wrong number of transmitted bytes and wrong data.
When I try to send 127 Byte with a incrementing pattern (0x01,0x02,0x03,..), 504 Bytes are sent (0x00,0x00,0x01 0x00,0x00,0x00,0x02, 0x00,0x00,0x00,0x03). We tested with different sizes and patterns, all are not ok. While analysing the configuration and code I was not able to see any obvious mistake.

Quoted 2 lines

Has someone else discovered such misbehaviour or has any idea what is wrong? Best Regards

Thx for the report!

[TLDR: I'm adding this report to the list of tracked Linux kernel regressions; the text you find below is based on a few templates paragraphs you might have encountered already in similar form. See link in footer if these mails annoy you.]

To ensure the issue doesn't fall through the cracks unnoticed, I'm adding it to regzbot, the Linux kernel regression tracking bot:

#regzbot ^introduced 15a6af94a27
#regzbot title spi: spi-devices on colibri-imx6dl and verdin-imx8mm are not working anymore
#regzbot ignore-activity

This isn't a regression? This issue or a fix for it are already discussed somewhere else? It was fixed already? You want to clarify when the regression started to happen? Or point out I got the title or something else totally wrong? Then just reply and tell me -- ideally while also telling regzbot about it, as explained by the page listed in the footer of this mail.

Developers: When fixing the issue, remember to add 'Link:' tags pointing to the report (the parent of this mail). See page linked in footer for details.

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
That page also explains what to do if mails like this annoy you.

A message that is not in the archive (sent off-list or not yet received): CAB3BuKA+qOY+UhWR-9Ov3qsz3wQr8q8n38MrEMf3FMCthr04yA@mail.gmail.com

#4 lore raw

Hi

I checked the code an printed the used parameters, the ‘bits_per_word’ is set to 8.

Best Regards
Stefan Bigler

Am 2023-11-18T17:54:26.000+01:00 hat Stefan Moring <[email protected]> geschrieben:

Quoted 103 lines

Hi Thorsten,

Looking at the data of your test, it appears as if your ‘bits_per_word’ are set to 32, where you would want 8.

Can you verify this?

Kind regards,

Stefan Moring

Op za 18 nov. 2023 om 16:15 schreef Thorsten Leemhuis < [email protected]>

Quoted 88 lines · Linux regression tracking (Thorsten Leemhuis)
Quoted 4 lines

[CCing people that worked on the relevant commits as well as the
regression list, as it should be in the loop for regressions:
https://docs.kernel.org/admin-guide/reporting-regressions.html]

Sorry for the spam, hit "send" before actually adding Stefan and Mark, which now are among the recipients.

Quoted 78 lines · On 18.11.23 15:25, [email protected] wrote:
Quoted 3 lines

After upgrade from kernel 6.5.11 to 6.6.1 the spi-devices on my hw
colibri-imx6dl and verdin-imx8mm are not working anymore (TPM2 and

SPI-SRAM).

Quoted 72 lines
Quoted 7 lines

Analyzing the problem showed that the 2 commits introduced the problem:

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

Quoted 6 lines

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Reverting the commits solved the problem.

The analyse with the logic-analyser showed a wrong number of

transmitted

Quoted 57 lines
Quoted 3 lines

bytes and wrong data.
When I try to send 127 Byte with a incrementing pattern

(0x01,0x02,0x03,..),

Quoted 52 lines
Quoted 2 lines
 
    504 Bytes are sent (0x00,0x00,0x01 0x00,0x00,0x00,0x02,

0x00,0x00,0x00,0x03).

Quoted 48 lines
Quoted 3 lines

We tested with different sizes and patterns, all are not ok.
While analysing the configuration and code I was not able to see any

obvious

Quoted 43 lines
Quoted 4 lines

mistake.

Has someone else discovered such misbehaviour or has any idea what is

wrong?

Quoted 34 lines
Quoted 2 lines

Best Regards

Thx for the report!

[TLDR: I'm adding this report to the list of tracked Linux kernel regressions; the text you find below is based on a few templates paragraphs you might have encountered already in similar form. See link in footer if these mails annoy you.]

To ensure the issue doesn't fall through the cracks unnoticed, I'm adding it to regzbot, the Linux kernel regression tracking bot:

#regzbot ^introduced 15a6af94a27
#regzbot title spi: spi-devices on colibri-imx6dl and verdin-imx8mm are not working anymore
#regzbot ignore-activity

This isn't a regression? This issue or a fix for it are already discussed somewhere else? It was fixed already? You want to clarify when the regression started to happen? Or point out I got the title or something else totally wrong? Then just reply and tell me -- ideally while also telling regzbot about it, as explained by the page listed in the footer of this mail.

Developers: When fixing the issue, remember to add 'Link:' tags pointing to the report (the parent of this mail). See page linked in footer for details.

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
That page also explains what to do if mails like this annoy you.

A message that is not in the archive (sent off-list or not yet received): CAB3BuKARgJhaVNFsP1FQ+2yLe18QU9H17fHKjc-Sf3izE+MZ1Q@mail.gmail.com

#6 lore raw

Hi Stefan

Thanks for analyzing the problem and sorry for the delay, I had to simplify the code to a minimum so that I can send to you. Now I was able to use spidev.
I have an environment with yocto kirstone 4.0.10.
Load the spi-dma (imx-sdma 302c0000.dma-controller: loaded firmware 4.5), run the spi_imx and the spidev as kenelmodule.

I run the code on a Toradex Verdin Development Board and use the imx8mm-verdin-nonwifi-dev.dts

To add the spidev I patched imx8mm-verdin.dtsi

arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi +9 -0
index 6f0811587142..262500940adc 100644
@@ -209,6 +209,15 @@ &ecspi2 {
        cs-gpios = <&gpio5 13 GPIO_ACTIVE_LOW>;
        pinctrl-names = "default";
        pinctrl-0 = <&pinctrl_ecspi2>;
+
+       spidev@0{
+               compatible = "micron,spi-authenta";
+               reg = <0>;
+               #address-cells = <1>;
+               #size-cells = <0>;
+               spi-max-frequency = <20000000>;
+               status = "okay";
+       };
 };
 

as a spidev test program I used
https://raw.githubusercontent.com/raspberrypi/linux/rpi-3.10.y/Documentation/spi/spidev_test.c

I changed the transmitted data

recipes-spi/spidev/files/spidev_test.c +16 -7
index 16feda9..6056ffd 100644
@@ -39,13 +39,22 @@ static void transfer(int fd)
 {
        int ret;
        uint8_t tx[] = {
-               0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
-               0x40, 0x00, 0x00, 0x00, 0x00, 0x95,
-               0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
-               0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
-               0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
-               0xDE, 0xAD, 0xBE, 0xEF, 0xBA, 0xAD,
-               0xF0, 0x0D,
+               0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+        0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, // 16
+        0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,
+        0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, // 32
+        0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,
+        0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f, 0x20, // 48
+        0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,
+        0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, // 64
+        0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+        0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, // 80
+        0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,
+        0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, // 96
+        0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,
+        0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f, 0x20, // 112
+        0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,
+        0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f,       // 127
        };
        uint8_t rx[ARRAY_SIZE(tx)] = {0, };
 

sending the content showed again the bad data spidev_test --device /dev/spidev1.0 --speed 20000000 --bpw 8

0x00,0x00,0x01,
0x00,0x00,0x00,0x02,
0x00,0x00,0x00,0x03,
0x00,0x00,0x00,0x04,

I you need more information let me know.

Best Regards
Stefan Bigler

Am 2023-11-19T08:52:54.000+01:00 hat Stefan Moring <[email protected]> geschrieben:

Quoted 7 lines

Hi Stefan,

Can you maybe share me your test code? I can try to reproduce it tomorrow.

Kind regards,

Stefan Moring

#7 lore raw
Quoted 26 lines · On Mon, Nov 20, 2023 at 09:27:10AM +0100, [email protected] wrote:

Load the spi-dma (imx-sdma 302c0000.dma-controller: loaded firmware 4.5), run the spi_imx and the spidev as kenelmodule.

I run the code on a Toradex Verdin Development Board and use the imx8mm-verdin-nonwifi-dev.dts

To add the spidev I patched imx8mm-verdin.dtsi

arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi +9 -0
index 6f0811587142..262500940adc 100644
@@ -209,6 +209,15 @@ &ecspi2 {
        cs-gpios = <&gpio5 13 GPIO_ACTIVE_LOW>;
        pinctrl-names = "default";
        pinctrl-0 = <&pinctrl_ecspi2>;
+
+       spidev@0{
+               compatible = "micron,spi-authenta";
+               reg = <0>;
+               #address-cells = <1>;
+               #size-cells = <0>;
+               spi-max-frequency = <20000000>;
+               status = "okay";
+       };
 };
 
Quoted 2 lines

as a spidev test program I used
https://raw.githubusercontent.com/raspberrypi/linux/rpi-3.10.y/Documentation/spi/spidev_test.c

Would you mind doing the following tests looping the SPI interface MISO/MOSI?

```
cd /tmp

dd if=/dev/urandom of=4k-spi-test-data.bin bs=1 count=4k
spidev_test -D "$device" -s 4000000 -i 4k-spi-test-data.bin -o 4k-spi-test-result.bin

dd if=/dev/urandom of=16bytes-spi-test-data.bin bs=1 count=16
spidev_test -D "$device" -s 16000 -i 16bytes-spi-test-data.bin -o 16bytes-spi-test-result.bin
```

with "$device" being your actual spidev device?

Those tests are passing on 6.7.0-rc2 for me.

Francesco

#8 lore raw

Hi Francesco

I run the test and it worked fine. The transmitted data matches with the recived data, no diff.

BUT!!! on the MDIO it is completly different !!!
When I request to send 4kB Data I measure on the line 16384Byte.

The data is again 3Dummy Bytes followed by the databyte.
So the error symetic on send and recive

Regards Stefan

Am 2023-11-20T09:47:48.000+01:00 hat Francesco Dolcini <[email protected]> geschrieben:

Quoted 43 lines · On Mon, Nov 20, 2023 at 09:27:10AM +0100, [email protected] wrote:
Quoted 24 lines

Load the spi-dma (imx-sdma 302c0000.dma-controller: loaded firmware
4.5), run the spi_imx and the spidev as kenelmodule.

I run the code on a Toradex Verdin Development Board and use the imx8mm-verdin-nonwifi-dev.dts

To add the spidev I patched imx8mm-verdin.dtsi
diff --git a/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi b/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi index 6f0811587142..262500940adc 100644
--- a/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi
+++ b/arch/arm64/boot/dts/freescale/imx8mm-verdin.dtsi

 @@ -209,6 +209,15 @@ &ecspi2 {
         cs-gpios = <&gpio5 13 GPIO_ACTIVE_LOW>;
         pinctrl-names = "default";
         pinctrl-0 = <&pinctrl_ecspi2>;
 +
 +       spidev@0{
 +               compatible = "micron,spi-authenta";
 +               reg = <0>;
 +               #address-cells = <1>;
 +               #size-cells = <0>;
 +               spi-max-frequency = <20000000>;
 +               status = "okay";
Quoted 5 lines

+ };
};

as a spidev test program I used
https://raw.githubusercontent.com/raspberrypi/linux/rpi-3.10.y/Documentation/spi/spidev_test.c

Would you mind doing the following tests looping the SPI interface MISO/MOSI?

```
cd /tmp

dd if=/dev/urandom of=4k-spi-test-data.bin bs=1 count=4k
spidev_test -D "$device" -s 4000000 -i 4k-spi-test-data.bin -o 4k-spi-test-result.bin

dd if=/dev/urandom of=16bytes-spi-test-data.bin bs=1 count=16
spidev_test -D "$device" -s 16000 -i 16bytes-spi-test-data.bin -o 16bytes-spi-test-result.bin
```
Quoted 4 lines

with "$device" being your actual spidev device?

Those tests are passing on 6.7.0-rc2 for me.

Francesco

#9 lore raw
Quoted 2 lines · On Mon, Nov 20, 2023 at 10:33:38AM +0100, [email protected] wrote:

I run the test and it worked fine. The transmitted data matches with the recived data, no diff.

ok, thanks for confirming. At least we are on the same page.

Quoted 5 lines

BUT!!! on the MDIO it is completly different !!!
When I request to send 4kB Data I measure on the line 16384Byte.

The data is again 3Dummy Bytes followed by the databyte.
So the error symetic on send and recive

ok, that's annoying. I wonder if the loopback test could be improved to spot such kind of issues, maybe we could verify if the execution time is reasonable given the SPI clock frequency and the payload size ...

Francesco

A message that is not in the archive (sent off-list or not yet received): CAB3BuKDb6uucujD7ac-w4pa1GVNLSQUA4OGE7i074mQSU==WaA@mail.gmail.com

#11 lore raw

Hi Stefan

I added my printing
@@ -655,12 +658,25 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,

        /* Clear BL field and set the right value */
        ctrl &= ~MX51_ECSPI_CTRL_BL_MASK;
+       dev_info(spi_imx->dev, "%s: count=%d, bits_per_word=%d\n",
+            __FUNCTION__, spi_imx->count, spi_imx->bits_per_word);
+
        if (spi_imx->target_mode && is_imx53_ecspi(spi_imx))
                ctrl |= (spi_imx->target_burst * 8 - 1)
                        << MX51_ECSPI_CTRL_BL_OFFSET;

and run the test with the burst enabled
/var/volatile/tmp# spidev_test --device /dev/spidev1.0 --speed 20000000 --bpw 8 -i 127bytes-spi-test-data.bin -o 127bytes-spi-test-result.bin

[  802.729824] spi_imx 30830000.spi: mx51_ecspi_prepare_transfer: count=127, bits_per_word=8

-> transmitted 504 Bytes on the line

then I run it with burst disabled
spidev_test --device /dev/spidev1.0 --speed 20000000 --bpw 8 -i 127bytes-spi-test-data.bin -o 127bytes-spi-test-result.bin

[  316.683807] spi_imx 30830000.spi: mx51_ecspi_prepare_transfer: count=0, bits_per_word=8

-> transmitted 127 Bytes on the line

The count is different
- for burst-enabled 127
- for burst-disable 0

concerning the sdma-firmware I checked on
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/imx/sdma
but the latest I found was 4.5

Regards
Stefan

Am 2023-11-20T16:30:15.000+01:00 hat Stefan Moring <[email protected]> geschrieben:

Quoted 41 lines

Hi Stefan,

I have been trying to reproduce your error on my iMX8M, scoping the SPI lines, but everything works as expected. I don't have the exact same board, but I don't think that should be a problem. One difference I noticed is that my imx-sdma firmware is version 4.6.

Can you verify the values used for the transfer, spi_imx->count and spi_imx->bits_per_word inside the mx51_ecpsi_prepare_transfer() method? Those are the only two things that changed in the commits. Maybe compare them to the working version?

Kind regards,

Stefan Moring

Op ma 20 nov 2023 om 11:18 schreef Francesco Dolcini <[email protected]>:

Quoted 19 lines · On Mon, Nov 20, 2023 at 10:33:38AM +0100, [email protected] wrote:
Quoted 3 lines

I run the test and it worked fine. The transmitted data matches with
the recived data, no diff.

ok, thanks for confirming. At least we are on the same page.

Quoted 5 lines

BUT!!! on the MDIO it is completly different !!!
When I request to send 4kB Data I measure on the line 16384Byte.

The data is again 3Dummy Bytes followed by the databyte.
So the error symetic on send and recive

ok, that's annoying. I wonder if the loopback test could be improved to spot such kind of issues, maybe we could verify if the execution time is reasonable given the SPI clock frequency and the payload size ...

Francesco

#12 lore raw
In reply to a message not in the archive
Quoted 4 lines · Stefan Moring

Can you verify the values used for the transfer,  spi_imx->count and spi_imx-> bits_per_word inside the mx51_ecpsi_prepare_transfer() method? Those are the only two things that changed in the commits. Maybe compare them to the working version?

I would suggest to bisect the issue to the actual commit that introduced the regression, I do not think this was done yet.

Francesco

#13 lore raw
Quoted 8 lines · Francesco Dolcini
Quoted 4 lines · Stefan Moring

Can you verify the values used for the transfer,  spi_imx->count and spi_imx-> bits_per_word inside the mx51_ecpsi_prepare_transfer() method? Those are the only two things that changed in the commits. Maybe compare them to the working version?

I would suggest to bisect the issue to the actual commit that introduced the regression, I do not think this was done yet.

I think it was. To quote

https://lore.kernel.org/all/[email protected]/

"'"
After upgrade from kernel 6.5.11 to 6.6.1 the spi-devices on my hw colibri-imx6dl and verdin-imx8mm are not working anymore (TPM2 and SPI-SRAM).

Analyzing the problem showed that the 2 commits introduced the problem:

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Reverting the commits solved the problem.
"'"

Or am I missing something here?

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
If I did something stupid, please tell me, as explained on that page.

#14 lore raw
Quoted 11 lines · Thorsten Leemhuis
Quoted 8 lines · Francesco Dolcini
Quoted 4 lines · Stefan Moring

Can you verify the values used for the transfer,  spi_imx->count and spi_imx-> bits_per_word inside the mx51_ecpsi_prepare_transfer() method? Those are the only two things that changed in the commits. Maybe compare them to the working version?

I would suggest to bisect the issue to the actual commit that introduced the regression, I do not think this was done yet.

I think it was. To quote

Whoops, you are right.

Quoted 5 lines

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Do we know which one of those two commits introduces this regression?

Francesco

#15 lore raw
Quoted 8 lines · Francesco Dolcini
Quoted 5 lines · Thorsten Leemhuis

spi: Increase imx51 ecspi burst length based on transfer length 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Do we know which one of those two commits introduces this regression?

Whoops again. The second one is a fix for the first one, so my question does not make any sense, forget about this, sorry.

Francesco

#16 lore raw
In reply to Francesco Dolcini

Hi

At least in my test-case the commit is NOT introducing this regression, because the bits_per_word is 8, so the result is the same
spi: imx: Take in account bits per word instead of assuming 8-bits
5f66db08cbd3ca471c66bacb0282902c79db9274

I do not have the latest mx-sdma firmware can you tell me where I get it. On
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/imx/sdma
the latest I found was 4.5

I tried to debug the code but it's hard vor me to understand where the problem could be. I saw then I disable the dma with set

imx51_ecspi_devtype_data {
.has_dmamode = false,
}
the SPI is working fine.

Should I do some more tests, do some loggings..
Please let me know

Regards
Stefan Bigler

Am 2023-11-21T10:16:57.000+01:00 hat Francesco Dolcini <[email protected]> geschrieben:

Quoted 18 lines · Thorsten Leemhuis
Quoted 10 lines · Francesco Dolcini
Quoted 8 lines · Stefan Moring
Quoted 5 lines
    Can you verify the values used for the transfer,  spi_imx->count and spi_imx->

bits_per_word inside the mx51_ecpsi_prepare_transfer() method? Those are the only two things that changed in the commits. Maybe compare them to the working version?

Quoted 4 lines
Quoted 2 lines

I would suggest to bisect the issue to the actual commit that introduced the regression, I do not think this was done yet.

I think it was. To quote

Whoops, you are right.

Quoted 8 lines
Quoted 5 lines

spi: Increase imx51 ecspi burst length based on transfer length
15a6af94a2779d5dfb42ee4bfac858ea8e964a3f

spi: imx: Take in account bits per word instead of assuming 8-bits 5f66db08cbd3ca471c66bacb0282902c79db9274

Do we know which one of those two commits introduces this regression?

Francesco

#17 lore raw

Hi

i did now my test with the latest imx-sdma firmware 4.6 from NXP
https://www.nxp.com/lgfiles/NMG/MAD/YOCTO/firmware-imx-8.18.bin
There is no improvement.
What is the HW that you use? What type of imx8?

Regards Stefan Bigler

Am 2023-11-21T11:34:26.000+01:00 hat <[email protected]> geschrieben:

Quoted 23 lines

Hi

At least in my test-case the commit is NOT introducing this regression, because the bits_per_word is 8, so the result is the same
spi: imx: Take in account bits per word instead of assuming 8-bits
5f66db08cbd3ca471c66bacb0282902c79db9274

I do not have the latest mx-sdma firmware can you tell me where I get it. On
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/imx/sdma
the latest I found was 4.5

I tried to debug the code but it's hard vor me to understand where the problem could be. I saw then I disable the dma with set

imx51_ecspi_devtype_data {
.has_dmamode = false,
}
the SPI is working fine.

Should I do some more tests, do some loggings..
Please let me know

Regards
Stefan Bigler

#18 lore raw
In reply to Francesco Dolcini

Hello,

I'm uncertain whether this is the appropriate place to respond, but I am encountering a similar issue on a phyboard-polis-rdk (imx8mm) with Linux 6.7-rc2. My system uses the imx-sdma firmware version 4.6.

The problem is with the TPM, which fails to function correctly, displaying the error message:
"tpm tpm0: TPM in field failure mode, requires firmware upgrade."

The issue is identical to the one reported by Stefan. It can be resolved by reverting commits 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f and 5f66db08cbd3ca471c66bacb0282902c79db9274, after which the TPM works again.

Additionally, I've conducted the tests which Francesco mentioned:

Quoted 5 lines

```
cd /tmp

dd if=/dev/urandom of=4k-spi-test-data.bin bs=1 count=4k spidev_test -D "$device" -s 4000000 -i 4k-spi-test-data.bin -o 4k-

spi-test-result.bin

Quoted 3 lines

dd if=/dev/urandom of=16bytes-spi-test-data.bin bs=1 count=16 spidev_test -D "$device" -s 16000 -i 16bytes-spi-test-data.bin -o

16bytes-spi-test-result.bin

Quoted 1 line

```

The readback data is correct but on MOSI/MISO there are 16384 bytes.

Please let me know if I can do some more tests.

Best regards,

Benjamin Bigler

#19 lore raw

Hi

I did some debugging and I think the problem is that in this case bits_per_word is 8. So in spi_imx_dma_configure the buswidth is set to DMA_SLAVE_BUSWIDTH_1_BYTE. But in mx51_ecspi_prepare_transfer the BURST_LENGTH is now set to (spi_imx->count * spi_imx->bits_per_word
- 1)
before 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f it was (spi_imx->bits_per_word - 1). Now the spi transmits 4 byte per byte except for the first word. I added the following patch and it worked again

drivers/spi/spi-imx.c +15 -4
index 498e35c8db2c..f514966e2ada 100644
@@ -659,11 +659,22 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
 		ctrl |= (spi_imx->target_burst * 8 - 1)
 			<< MX51_ECSPI_CTRL_BL_OFFSET;
 	else {
-		if (spi_imx->count >= 512)
-			ctrl |= 0xFFF << MX51_ECSPI_CTRL_BL_OFFSET;
-		else
-			ctrl |= (spi_imx->count * spi_imx->bits_per_word - 1)
+		if (spi_imx->usedma)
+			ctrl |= (spi_imx->bits_per_word *
+					 spi_imx_bytes_per_word(
+						 spi_imx->bits_per_word) -
+				 1)
 				<< MX51_ECSPI_CTRL_BL_OFFSET;
+		else {
+			if (spi_imx->count >= MX51_ECSPI_CTRL_MAX_BURST)
+				ctrl |= (MX51_ECSPI_CTRL_MAX_BURST - 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+			else
+				ctrl |= (spi_imx->count *
+						 spi_imx->bits_per_word -
+					 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+		}
 	}
 
 	/* set clock speed */

Best regards,
Benjamin Bigler

#20 lore raw

Hi

Thank you Benjamin for the patch.
I have tested the patch and is works fine.

Best regards
Stefan Bigler

Am 2023-11-26T14:19:56.000+01:00 hat Benjamin Bigler <[email protected]> geschrieben:

Quoted 41 lines

Hi

I did some debugging and I think the problem is that in this case bits_per_word is 8. So in spi_imx_dma_configure the buswidth is set to DMA_SLAVE_BUSWIDTH_1_BYTE. But in mx51_ecspi_prepare_transfer the BURST_LENGTH is now set to (spi_imx->count * spi_imx->bits_per_word
- 1)
before 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f it was (spi_imx->bits_per_word - 1). Now the spi transmits 4 byte per byte except for the first word. I added the following patch and it worked again

drivers/spi/spi-imx.c +15 -4
index 498e35c8db2c..f514966e2ada 100644
@@ -659,11 +659,22 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
 		ctrl |= (spi_imx->target_burst * 8 - 1)
 			<< MX51_ECSPI_CTRL_BL_OFFSET;
 	else {
-		if (spi_imx->count &gt;= 512)
-			ctrl |= 0xFFF << MX51_ECSPI_CTRL_BL_OFFSET;
-		else
-			ctrl |= (spi_imx->count * spi_imx->bits_per_word - 1)
+		if (spi_imx->usedma)
+			ctrl |= (spi_imx->bits_per_word *
+					 spi_imx_bytes_per_word(
+						 spi_imx->bits_per_word) -
+				 1)
 				<< MX51_ECSPI_CTRL_BL_OFFSET;
+		else {
+			if (spi_imx->count &gt;= MX51_ECSPI_CTRL_MAX_BURST)
+				ctrl |= (MX51_ECSPI_CTRL_MAX_BURST - 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+			else
+				ctrl |= (spi_imx->count *
+						 spi_imx->bits_per_word -
+					 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+		}
 	}
 
 	/* set clock speed */
Quoted 1 line

Best regards,

Benjamin Bigler

#21 lore raw
Quoted 3 lines · On 27.11.23 10:09, [email protected] wrote:

Thank you Benjamin for the patch.
I have tested the patch and is works fine.

That's great.

Benjamin, what's the status of the fix for this regression? I tried to find a proper submission for review, but didn't find one. Am I missing something or did that fall through the cracks?

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
If I did something stupid, please tell me, as explained on that page.

#regzbot poke

Quoted 46 lines

Am 2023-11-26T14:19:56.000+01:00 hat Benjamin Bigler <[email protected]> geschrieben:

Quoted 41 lines

Hi

I did some debugging and I think the problem is that in this case bits_per_word is 8. So in spi_imx_dma_configure the buswidth is set to DMA_SLAVE_BUSWIDTH_1_BYTE. But in mx51_ecspi_prepare_transfer the BURST_LENGTH is now set to (spi_imx->count * spi_imx->bits_per_word
- 1)
before 15a6af94a2779d5dfb42ee4bfac858ea8e964a3f it was (spi_imx->bits_per_word - 1). Now the spi transmits 4 byte per byte except for the first word. I added the following patch and it worked again

drivers/spi/spi-imx.c +15 -4
index 498e35c8db2c..f514966e2ada 100644
@@ -659,11 +659,22 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
 		ctrl |= (spi_imx->target_burst * 8 - 1)
 			<< MX51_ECSPI_CTRL_BL_OFFSET;
 	else {
-		if (spi_imx->count &gt;= 512)
-			ctrl |= 0xFFF << MX51_ECSPI_CTRL_BL_OFFSET;
-		else
-			ctrl |= (spi_imx->count * spi_imx->bits_per_word - 1)
+		if (spi_imx->usedma)
+			ctrl |= (spi_imx->bits_per_word *
+					 spi_imx_bytes_per_word(
+						 spi_imx->bits_per_word) -
+				 1)
 				<< MX51_ECSPI_CTRL_BL_OFFSET;
+		else {
+			if (spi_imx->count &gt;= MX51_ECSPI_CTRL_MAX_BURST)
+				ctrl |= (MX51_ECSPI_CTRL_MAX_BURST - 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+			else
+				ctrl |= (spi_imx->count *
+						 spi_imx->bits_per_word -
+					 1)
+					<< MX51_ECSPI_CTRL_BL_OFFSET;
+		}
 	}
 
 	/* set clock speed */
Quoted 1 line

Best regards,

Benjamin Bigler

#22 lore raw
Quoted 10 lines
Quoted 10 lines · On 27.11.23 10:09, [email protected] wrote:
Quoted 3 lines
Quoted 3 lines

Thank you Benjamin for the patch.
I have tested the patch and is works fine.

That's great.

Benjamin, what's the status of the fix for this regression? I tried to find a proper submission for review, but didn't find one. Am I missing something or did that fall through the cracks?

I haven't send a proper patch yet. I will read through the guidelines and submit one on the weekend.

Quoted 10 lines
Quoted 9 lines

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
If I did something stupid, please tell me, as explained on that page.

#regzbot poke

Best regards,
Benjamin Bigler