[U-Boot] [PATCH] sata: fix reset_sata for dwc_ahsata

[Soeren Moch]

On 26.11.2014 18:38, Nikita Kiryanov wrote:
> Hi Soeren,
> 
> On 11/26/2014 01:40 PM, Soeren Moch wrote:
>> - fix crash when sata device is not initialized
>> - remove disable_sata_clock() since it is not clear which clock for which
>>    device should be disabled here
>> - call disable_sata_clock() for mx6 in preboot_os instead
>>
>> Signed-off-by: Soeren Moch <smoch at web.de>
>> ---
>> Cc: Tom Rini <trini at ti.com>
>> Cc: Nikita Kiryanov <nikita at compulab.co.il>
>> Cc: Simon Glass <sjg at chromium.org>
>> Cc: guillaume.gardet at free.fr
>> Cc: Stefano Babic <sbabic at denx.de>
>> ---
>>   arch/arm/imx-common/cpu.c  |  3 +++
>>   drivers/block/dwc_ahsata.c | 14 ++++++++------
>>   2 files changed, 11 insertions(+), 6 deletions(-)
>>
>> diff --git a/arch/arm/imx-common/cpu.c b/arch/arm/imx-common/cpu.c
>> index b58df7d..28ccd29 100644
>> --- a/arch/arm/imx-common/cpu.c
>> +++ b/arch/arm/imx-common/cpu.c
>> @@ -206,6 +206,9 @@ void arch_preboot_os(void)
>>   {
>>   #if defined(CONFIG_CMD_SATA)
>>       sata_stop();
>> +#if defined(CONFIG_MX6)
>> +    disable_sata_clock();
>> +#endif
> 
> I think a more proper way to do this might be to mirror the
> init sequence completely. The simplest init sequence calls
> setup_sata() somewhere before sata_initialize() is invoked,
> and the distribution of labor is:
> - arch/arm/imx-common/sata.c:setup_sata() Turns on clocks
> - common/cmd_sata.c: sata_initialize() Calls driver init
> 
> The inverse would be:
> - common/cmd_sata.c: sata_stop() Calls driver reset
> - arch/arm/imx-common/sata.c:disable_sata() Turns off clocks
> 
> Thus disable_sata() would need to be defined in
> arch/arm/imx-common/sata.c; it will call disable_sata_clock(),
> and be used in cm_fx6's version of stop_sata().
> 
> It's a little convoluted, but consistent with the existing code
> flow, and it also fixes the MX53 problems because
> arch/arm/imx-common/sata.c is not compiled for this platform.

Nikita,

if I remember correctly, the whole purpose of this patch series was to
get a SSD running, which was not recognized by linux if sata is enabled
on boot. So for me disabling the sata link in u-boot makes perfectly
sense to overcome this issue.
But I cannot imagine that disabling the clock of the sata controller
makes any difference here. But I did not test this (I have no access to
such problematic SSD).
I would prefer to keep the code simple and easy to maintain, I don't see
any advantage in mirroring the init sequence on exit. I would not switch
off the sata controller clock. I'm even not sure that my patch would
work for i.MX6SX, if I further think about it. It is already not easy to
understand all implications.
But this is only my personal opinion, I'm not the maintainer of this
code. My only goal with this patch was to get sata running again, for
boards which enable sata (call setup_sata() in board_init() ), but do
not run 'sata init' in the bootcmd by default - and for other
architectures than mx6.

So how to proceed is a question for the maintainers now, I think.
I see following opportunities:
- implement the clock disabling framework as Nikita has described above
- try to get my patch working with minimal changes of the merged code
  (as we are in -rc2 phase of development),
  and possibly apply additional patches from Nikita on this later
- revert the whole patch series

> 
>>   #endif
>>   #if defined(CONFIG_VIDEO_IPUV3)
>>       /* disable video before launching O/S */
>> diff --git a/drivers/block/dwc_ahsata.c b/drivers/block/dwc_ahsata.c
>> index 9a2b547..9e925d1 100644
>> --- a/drivers/block/dwc_ahsata.c
>> +++ b/drivers/block/dwc_ahsata.c
>> @@ -594,22 +594,24 @@ int init_sata(int dev)
>>
>>   int reset_sata(int dev)
>>   {
>> -    struct ahci_probe_ent *probe_ent =
>> -            (struct ahci_probe_ent *)sata_dev_desc[dev].priv;
>> -    struct sata_host_regs *host_mmio =
>> -            (struct sata_host_regs *)probe_ent->mmio_base;
>> +    struct ahci_probe_ent *probe_ent;
>> +    struct sata_host_regs *host_mmio;
>>
>>       if (dev < 0 || dev > (CONFIG_SYS_SATA_MAX_DEVICE - 1)) {
>>           printf("The sata index %d is out of ranges\n\r", dev);
>>           return -1;
>>       }
>>
>> +    probe_ent = (struct ahci_probe_ent *)sata_dev_desc[dev].priv;
>> +    if (NULL == probe_ent)
>> +        /* not initialized, so nothing to reset */
>> +        return 0;
>> +
>> +    host_mmio = probe_ent->mmio_base;
> 
> The lack of casting generates a warning during compilation:
> drivers/block/dwc_ahsata.c:610:12: warning: assignment makes pointer
> from integer without a cast [enabled by default]
>   host_mmio = probe_ent->mmio_base;
>             ^
> 

Of course you are right here, I accidentally removed the type cast. If
we decide to go on with this patch, I will fix this in a new patch version.

  Soeren

>>       setbits_le32(&host_mmio->ghc, SATA_HOST_GHC_HR);
>>       while (readl(&host_mmio->ghc) & SATA_HOST_GHC_HR)
>>           udelay(100);
>>
>> -    disable_sata_clock();
>> -
>>       return 0;
>>   }
>>
>>
>