Skip to contentSkip to conversation

i2c: tegra: Fix i2c-tegra DMA config option processing

PATCH Applied Parker Newman 5 participants 12 messages Aug 3, 2023 – Aug 14, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i2c: tegra: Fix i2c-tegra DMA config option processing

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

---
 drivers/i2c/busses/i2c-tegra.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-tegra.c +1 -1
index bcbbf23aa530..dc6ed3a8d69e 100644
@@ -442,7 +442,7 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
    if (IS_VI(i2c_dev))
        return 0;
 
-   if (!i2c_dev->hw->has_apb_dma) {
+   if (i2c_dev->hw->has_apb_dma) {
        if (!IS_ENABLED(CONFIG_TEGRA20_APB_DMA)) {
            dev_dbg(i2c_dev->dev, "APB DMA support not enabled\n");
            return 0;
Signature · 1 line
2.17.1
maintainer
#1 lore raw

Hi Laxman and/or Dmitry,

Quoted 26 lines · Parker Newman

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

---
 drivers/i2c/busses/i2c-tegra.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-tegra.c +0 -0
index bcbbf23aa530..dc6ed3a8d69e 100644

drivers/i2c/busses/i2c-tegra.c static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev) · line 448

Quoted 8 lines
drivers/i2c/busses/i2c-tegra.c +1 -1
@@ -442,7 +442,7 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
    if (IS_VI(i2c_dev))
        return 0;
 
-   if (!i2c_dev->hw->has_apb_dma) {
+   if (i2c_dev->hw->has_apb_dma) {
        if (!IS_ENABLED(CONFIG_TEGRA20_APB_DMA)) {
            dev_dbg(i2c_dev->dev, "APB DMA support not enabled\n");
            return 0;

Can I have your opinion here, please?

Andi

#2 lore raw

05.08.2023 00:49, Andi Shyti пишет:

Quoted 32 lines

Hi Laxman and/or Dmitry,

Quoted 29 lines · Parker Newman

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

---
 drivers/i2c/busses/i2c-tegra.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-tegra.c +0 -0
index bcbbf23aa530..dc6ed3a8d69e 100644
@@ -442,7 +442,7 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
    if (IS_VI(i2c_dev))
        return 0;
 
Quoted 7 lines

drivers/i2c/busses/i2c-tegra.c static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev) · line 448

Quoted 5 lines
drivers/i2c/busses/i2c-tegra.c +1 -1
-   if (!i2c_dev->hw->has_apb_dma) {
+   if (i2c_dev->hw->has_apb_dma) {
        if (!IS_ENABLED(CONFIG_TEGRA20_APB_DMA)) {
            dev_dbg(i2c_dev->dev, "APB DMA support not enabled\n");
            return 0;

Can I have your opinion here, please?

The patch looks good, thanks Parker for fixing it. I'll be able to test it only sometime later and let you all know if there will be any problem. Previously I haven't noticed any Tegra I2C regressions, maybe we should change that dev_dbg to dev_warn.

#3 lore raw
Quoted 41 lines · Dmitry Osipenko

05.08.2023 00:49, Andi Shyti пишет:

Quoted 32 lines

Hi Laxman and/or Dmitry,

Quoted 29 lines · Parker Newman

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

---
 drivers/i2c/busses/i2c-tegra.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-tegra.c +0 -0
index bcbbf23aa530..dc6ed3a8d69e 100644
@@ -442,7 +442,7 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
    if (IS_VI(i2c_dev))
        return 0;
 
Quoted 7 lines

drivers/i2c/busses/i2c-tegra.c static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev) · line 448

Quoted 5 lines
drivers/i2c/busses/i2c-tegra.c +1 -1
-   if (!i2c_dev->hw->has_apb_dma) {
+   if (i2c_dev->hw->has_apb_dma) {
        if (!IS_ENABLED(CONFIG_TEGRA20_APB_DMA)) {
            dev_dbg(i2c_dev->dev, "APB DMA support not enabled\n");
            return 0;

Can I have your opinion here, please?

Quoted 5 lines

The patch looks good, thanks Parker for fixing it. I'll be able to test it only sometime later and let you all know if there will be any problem. Previously I haven't noticed any Tegra I2C regressions, maybe we should change that dev_dbg to dev_warn.

Hi Dmitry,

You will not notice any issues if both options are set (or not set) as it will fall through and configure the DMA or skip DMA setup as expected.

I only noticed the issue after I enabled CONFIG_TEGRA20_APB_DMA which is a KConfig requirement for the Tegra HS-UART driver and my I2C stopped working.

Which now as I write this I realize is possibly another "bug"... As far as I know the HS UARTs also use GPC DMA on T186 or later? I would need to look into that.

I don't think the print needs to be a warning, not having DMA enabled is a valid option, it just needed the correct CONFIG options to be checked for the DMA type (APB versus GPC).

-Parker

maintainer
#4 lore raw
In reply to Parker Newman

BTW...

Quoted 1 line · Parker Newman

you have a blank line here.

Quoted 10 lines

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

Quoted 6 lines

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

As this is a fix you also need to add

Fixes 48cb6356fae1 ("i2c: tegra: Add GPCDMA support")
Cc Akhil R <[email protected]>
Cc <[email protected]> # v6.1+

Cc'eing Akhil as well for his opinion on this.

Andi

#5 lore raw
Quoted 26 lines

BTW...

Quoted 1 line · Parker Newman

you have a blank line here.

Quoted 11 lines

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but

CONFIG_TEGRA186_GPC_DMA=n

Quoted 5 lines

with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

Quoted 7 lines

As this is a fix you also need to add

Fixes 48cb6356fae1 ("i2c: tegra: Add GPCDMA support")
Cc Akhil R <[email protected]>
Cc <[email protected]> # v6.1+

Cc'eing Akhil as well for his opinion on this.

The fix looks valid to me. Must have been a typo there.

Regards,
Akhil

#6 lore raw
Quoted 32 lines · Akhil R
Quoted 32 lines

BTW...

Quoted 1 line · Parker Newman

you have a blank line here.

Quoted 11 lines

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but

CONFIG_TEGRA186_GPC_DMA=n

Quoted 5 lines

with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

As this is a fix you also need to add

Fixes 48cb6356fae1 ("i2c: tegra: Add GPCDMA support")
Cc Akhil R <[email protected]>
Cc <[email protected]> # v6.1+
Quoted 6 lines
Quoted 1 line

Cc'eing Akhil as well for his opinion on this.

The fix looks valid to me. Must have been a typo there.

Regards,
Akhil

Yes it appears to be a simple typo and if you have both DMA Config options set the bug would get missed.

I am new to the Linux mailing list, should I send a new message to [email protected] or CC them on this one?

Thanks,
Parker

maintainer Acked-by
#7 lore raw

Hi Parker,

Quoted 36 lines · Parker Newman
Quoted 35 lines · Akhil R
Quoted 26 lines

BTW...

Quoted 1 line · Parker Newman

you have a blank line here.

Quoted 11 lines

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

This means if CONFIG_TEGRA20_APB_DMA=y but

CONFIG_TEGRA186_GPC_DMA=n

Quoted 5 lines

with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

Quoted 7 lines

As this is a fix you also need to add

Fixes 48cb6356fae1 ("i2c: tegra: Add GPCDMA support")
Cc Akhil R <[email protected]>
Cc <[email protected]> # v6.1+

Cc'eing Akhil as well for his opinion on this.

The fix looks valid to me. Must have been a typo there.

Quoted 8 lines
Quoted 3 lines

Regards,
Akhil

Yes it appears to be a simple typo and if you have both DMA Config options set the bug would get missed.

I am new to the Linux mailing list, should I send a new message to [email protected] or CC them on this one?

First of all... welcome to the Linux mailing list :)

No need to resend, I can add it... I was hoping for an official ack, but as that's not coming, I will ack it

Thank you Parker and wishing to see more patches from you ;)

Andi

#8 lore raw
In reply to Akhil R
Quoted 3 lines
Quoted 2 lines

Cc'eing Akhil as well for his opinion on this.

The fix looks valid to me. Must have been a typo there.

I'll count this as an ack. Thanks!

Acked-by
#9 lore raw
Quoted 4 lines
Quoted 2 lines
Quoted 1 line

Cc'eing Akhil as well for his opinion on this.

The fix looks valid to me. Must have been a typo there.

I'll count this as an ack. Thanks!

Oh Sorry. I just realized I didn't put an ack.

maintainer
#10 lore raw
In reply to Parker Newman

Hi

Quoted 8 lines · Parker Newman

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

[...]

With the Fixes tag added, applied to i2c/andi-for-current on

https://git.kernel.org/pub/scm/linux/kernel/git/andi.shyti/linux.git

Please note that this patch may still undergo further evaluation and the final decision will be made in collaboration with Wolfram.

Thank you,
Andi

Patches applied
===============
[1/1] i2c: tegra: Fix i2c-tegra DMA config option processing

      commit: fc9a464f3d9a2a361e8bcb960345270a9dc46054
Applied
#11 lore raw
In reply to Parker Newman
Quoted 11 lines · Parker Newman

This patch fixes the Tegra DMA config option processing in the i2c-tegra driver.

Tegra processors prior to Tegra186 used APB DMA for I2C requiring CONFIG_TEGRA20_APB_DMA=y while Tegra186 and later use GPC DMA requiring CONFIG_TEGRA186_GPC_DMA=y.

The check for if the processor uses APB DMA is inverted and so the wrong DMA config options are checked.

Quoted 6 lines

This means if CONFIG_TEGRA20_APB_DMA=y but CONFIG_TEGRA186_GPC_DMA=n with a Tegra186 or later processor the driver will incorrectly think DMA is enabled and attempt to request DMA channels that will never be availible, leaving the driver in a perpetual EPROBE_DEFER state.

Applied to for-current, thanks!