2018-11-26 11:56:01

by Leonard Crestez

[permalink] [raw]
Subject: imx6sx uart failing loopback tests with sdma firmware (ROM works)

Hello,

On a build farm at NXP we found that linux-next now fails on simple
loopback tests on imx6sx-sdb (but not other SOCs AFAICT). I tracked
this down to commit 30fdd51be161
("ARM: imx_v6_v7_defconfig: add CONFIG_FW_LOADER_USER_HELPER")

All that does is enable loading SDMA firmware from rootfs, in theory
this config was always supported so it would not be a new bug.

This doesn't happen with the NXP vendor tree but there is quite a large
delta in uart dma code. In particular upstream dropped IDDMAEN claiming
that this works with both ROM and FW scripts, see commit 905c0decad28
("serial: imx: don't use idle condition detect for DMA transfers")

This doesn't seem to work on 6sx though and I don't know why :(

The issue can be worked-around by forcing RX script from ROM:

diff --git drivers/dma/imx-sdma.c drivers/dma/imx-sdma.c
@@ -894,9 +894,11 @@ static void sdma_get_pc(struct sdma_channel *sdmac,
break;
case IMX_DMATYPE_UART:
- per_2_emi = sdma->script_addrs->uart_2_mcu_addr;
+ //per_2_emi = sdma->script_addrs->uart_2_mcu_addr;
+ per_2_emi = 817;
emi_2_per = sdma->script_addrs->mcu_2_app_addr;
break;
case IMX_DMATYPE_UART_SP:

The above doesn't look like a good fix. I tried to revert 905c0decad28
and resolve conflicts and that also seems to work, pushed here:
https://github.com/cdleonard/linux/commit/0a1757f467e6dc96037ec2e3ea7c88c5a4eb2ceb

You can find test code here (not very interesting):
https://source.codeaurora.org/external/imx/imx-test/tree/test/mxc_uart_test/mxc_uart_test.c?h=imx_4.14.62_1.0.0_beta

Any idea on how to fix this?

--
Regards,
Leonard


2018-11-26 12:05:45

by Lucas Stach

[permalink] [raw]
Subject: Re: imx6sx uart failing loopback tests with sdma firmware (ROM works)

Am Montag, den 26.11.2018, 11:53 +0000 schrieb Leonard Crestez:
> Hello,
>
> On a build farm at NXP we found that linux-next now fails on simple
> loopback tests on imx6sx-sdb (but not other SOCs AFAICT). I tracked
> this down to commit 30fdd51be161
> ("ARM: imx_v6_v7_defconfig: add CONFIG_FW_LOADER_USER_HELPER")
>
> All that does is enable loading SDMA firmware from rootfs, in theory
> this config was always supported so it would not be a new bug.
>
> This doesn't happen with the NXP vendor tree but there is quite a large
> delta in uart dma code. In particular upstream dropped IDDMAEN claiming
> that this works with both ROM and FW scripts, see commit 905c0decad28
> ("serial: imx: don't use idle condition detect for DMA transfers")
>
> This doesn't seem to work on 6sx though and I don't know why :(
>
> The issue can be worked-around by forcing RX script from ROM:
>
> diff --git drivers/dma/imx-sdma.c drivers/dma/imx-sdma.c
> @@ -894,9 +894,11 @@ static void sdma_get_pc(struct sdma_channel *sdmac,
>                 break;
>         case IMX_DMATYPE_UART:
> -               per_2_emi = sdma->script_addrs->uart_2_mcu_addr;
> +               //per_2_emi = sdma->script_addrs->uart_2_mcu_addr;
> +               per_2_emi = 817;
>                 emi_2_per = sdma->script_addrs->mcu_2_app_addr;
>                 break;
>         case IMX_DMATYPE_UART_SP:
>
> The above doesn't look like a good fix. I tried to revert 905c0decad28
> and resolve conflicts and that also seems to work, pushed here:
> https://github.com/cdleonard/linux/commit/0a1757f467e6dc96037ec2e3ea7c88c5a4eb2ceb
>
> You can find test code here (not very interesting):
> https://source.codeaurora.org/external/imx/imx-test/tree/test/mxc_uart_test/mxc_uart_test.c?h=imx_4.14.62_1.0.0_beta
>
> Any idea on how to fix this?

The RAM script is probably busted, as it doesn't provide the same
interface as the ROM script. The assumption of 905c0decad28 was that
the RAM script is an extension to the ROM script, but it seems like
this is not the case and the RAM script actually dropped some crucial
functionality.

See also https://lkml.org/lkml/2018/9/20/522

Please see if you can follow up internally and get a fixed SDMA
firmware pushed out. Fixed probably only means that the UART script
needs to be dropped, as the ROM script is totally fine when doing the
correct setup in the UART driver and the sole purpose of the RAM script
was to work with the broken UART driver.

Regards,
Lucas