mmc: zynq_sdhci: transfer fails with 8-bit data bus

[Michal Simek]

On 11. 05. 20 16:06, Benedikt Grassl wrote:
> Hi,
> 
> with commit 942b5fc0 the 8-bit data bus seems to work fine in U-Boot.
> However, I have one situation where I get transfer errors lateron in
> Linux. At the moment I suspect that this is rather a problem of the
> Linux driver, but I would like to make sure here first.
> My ZynqMP has an eMMC flash connected to SD0 and an SD-card to SD1. When
> I boot the device from SD-card and load a kernel ITB (tested both 4.14
> and 5.4), I immediately get errors when writing to the eMMC flash in
> Linux:
> 
>  print_req_error: I/O error, dev mmcblk0, sector 2048
>  Buffer I/O error on dev mmcblk0p1, logical block 0, lost sync page write
>  mmc0: Got data interrupt 0x00000002 even though no data operation was in progress.
>  mmc0: sdhci: ============ SDHCI REGISTER DUMP ===========
>  mmc0: sdhci: Sys addr:  0x00000400 | Version:  0x00001002
>  mmc0: sdhci: Blk size:  0x00007200 | Blk cnt:  0x00000400
>  mmc0: sdhci: Argument:  0x00000000 | Trn mode: 0x00000033
>  mmc0: sdhci: Present:   0x1ff70000 | Host ctl: 0x0000003d
>  mmc0: sdhci: Power:     0x0000000f | Blk gap:  0x00000080
>  mmc0: sdhci: Wake-up:   0x00000000 | Clock:    0x00000007
>  mmc0: sdhci: Timeout:   0x00000006 | Int stat: 0x00000000
>  mmc0: sdhci: Int enab:  0x02ff000b | Sig enab: 0x02ff000b
>  mmc0: sdhci: AC12 err:  0x00000000 | Slot int: 0x00000000
>  mmc0: sdhci: Caps:      0x75ecc881 | Caps_1:   0x00002007
>  mmc0: sdhci: Cmd:       0x00000c1a | Max curr: 0x00000000
>  mmc0: sdhci: Resp[0]:   0x00000b00 | Resp[1]:  0x36475026
>  mmc0: sdhci: Resp[2]:   0x49533031 | Resp[3]:  0x00000900
>  mmc0: sdhci: Host ctl2: 0x0000008b
>  mmc0: sdhci: ADMA Err:  0x00000000 | ADMA Ptr: 0x0000000000000000
>  mmc0: sdhci: ============================================
> 
> Interestingly, I can work-around this by just typing "mmc info" in the
> U-Boot console. As soon as some interaction with mmc0 happens in U-Boot
> , this problem disappears (i.e. anything that relies on the mmc layer
> to issue transfers to the eMMC flash).
> This did not happen before, when only a 4-bit data bus was supported.
> Also the latest commit e44c2bc on the Xilinx U-boot git repository does
> not solve this problem.

Yes we found that 1.8v dt property requires partial revert of your patch
because logic in u-boot code is problematic and needs to be fixed.

> From my understanding, the Linux driver should be independent of the
> initialization done in U-Boot, right?
> Does anyone have an idea or a comment to this? Did I miss anything?

yes both should be independent. I have no idea what it is wrong in this
case but definitely something to look at.

Thanks,
Michal