Patchwork [1/1] Big endian swap for Marvell 6440 SAS driver

login
register
mail settings
Submitter Cédric Cano
Date June 10, 2011, 1:56 p.m.
Message ID <4DF2228E.7060806@ic.fr>
Download mbox | patch
Permalink /patch/99941/
State Changes Requested
Headers show

Comments

Cédric Cano - June 10, 2011, 1:56 p.m.
Hi,

Marvell SAS driver doesn't work on powerpc architecture due to big 
endian swap. I've tested it with the MV6440 chip.
Here you can find the patch for Linux 2.6.39.1.

Cedric Cano

Signed-off-by: Cedric Cano <ccano@interfaceconcept.com>
---

---
Benjamin Herrenschmidt - June 11, 2011, 10:45 p.m.
On Fri, 2011-06-10 at 15:56 +0200, Cédric Cano wrote:
> Hi,
> 
> Marvell SAS driver doesn't work on powerpc architecture due to big 
> endian swap. I've tested it with the MV6440 chip.
> Here you can find the patch for Linux 2.6.39.1.
> 
> Cedric Cano
> 
> Signed-off-by: Cedric Cano <ccano@interfaceconcept.com>
> ---
> --- drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:04.000000000 +0200
> +++ drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:12.000000000 +0200
> @@ -1143,6 +1143,14 @@
>       MVS_CHIP_DISP->write_port_cfg_addr(mvi, i, PHYR_SATA_SIG0);
>       s[0] = MVS_CHIP_DISP->read_port_cfg_data(mvi, i);
> 
> +    /* swap for big endian devices because of use of these data in bytes */
> +#ifdef __BIG_ENDIAN
> +    s[0] = swab32(s[0]);
> +    s[1] = swab32(s[1]);
> +    s[2] = swab32(s[2]);
> +    s[3] = swab32(s[3]);
> +#endif /* __BIG_ENDIAN */

Hi Cedric !

Thanks for this. CC'ing linux-scsi.

Note that it would probably have been better to use le32_to_cpu, which
avoids the ifdef completely.

Cheers,
Ben.

>       /* Workaround: take some ATAPI devices for ATA */
>       if (((s[1] & 0x00FFFFFF) == 0x00EB1401) && (*(u8 *)&s[3] == 0x01))
>           s[1] = 0x00EB1401 | (*((u8 *)&s[1] + 3) & 0x10);
> 
> ---
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev
Cédric Cano - June 14, 2011, 6:42 a.m.
Hi Ben,

You're alright! I hadn't thought to use le32_to_cpu.
Do you want I make a new patch?

Cedric

Le 12/06/2011 00:45, Benjamin Herrenschmidt a écrit :
> On Fri, 2011-06-10 at 15:56 +0200, Cédric Cano wrote:
>> Hi,
>>
>> Marvell SAS driver doesn't work on powerpc architecture due to big
>> endian swap. I've tested it with the MV6440 chip.
>> Here you can find the patch for Linux 2.6.39.1.
>>
>> Cedric Cano
>>
>> Signed-off-by: Cedric Cano<ccano@interfaceconcept.com>
>> ---
>> --- drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:04.000000000 +0200
>> +++ drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:12.000000000 +0200
>> @@ -1143,6 +1143,14 @@
>>        MVS_CHIP_DISP->write_port_cfg_addr(mvi, i, PHYR_SATA_SIG0);
>>        s[0] = MVS_CHIP_DISP->read_port_cfg_data(mvi, i);
>>
>> +    /* swap for big endian devices because of use of these data in bytes */
>> +#ifdef __BIG_ENDIAN
>> +    s[0] = swab32(s[0]);
>> +    s[1] = swab32(s[1]);
>> +    s[2] = swab32(s[2]);
>> +    s[3] = swab32(s[3]);
>> +#endif /* __BIG_ENDIAN */
> Hi Cedric !
>
> Thanks for this. CC'ing linux-scsi.
>
> Note that it would probably have been better to use le32_to_cpu, which
> avoids the ifdef completely.
>
> Cheers,
> Ben.
>
>>        /* Workaround: take some ATAPI devices for ATA */
>>        if (((s[1]&  0x00FFFFFF) == 0x00EB1401)&&  (*(u8 *)&s[3] == 0x01))
>>            s[1] = 0x00EB1401 | (*((u8 *)&s[1] + 3)&  0x10);
>>
>> ---
>> _______________________________________________
>> Linuxppc-dev mailing list
>> Linuxppc-dev@lists.ozlabs.org
>> https://lists.ozlabs.org/listinfo/linuxppc-dev
>
>
Benjamin Herrenschmidt - June 14, 2011, 7:05 a.m.
On Tue, 2011-06-14 at 08:42 +0200, Cédric Cano wrote:
> Hi Ben,
> 
> You're alright! I hadn't thought to use le32_to_cpu.
> Do you want I make a new patch?

Well, I'm not the maintainer of the subsystem or of the driver, so I'm
not going to be the one accepting/merging it :-)

I'm just reviewing the patch, but yes, I think you should, and you
should send it to the right people, ie, I think at least linux-scsi and
possibly the author/maintainer of the driver itself.

Cheers,
Ben.

> Cedric
> 
> Le 12/06/2011 00:45, Benjamin Herrenschmidt a écrit :
> > On Fri, 2011-06-10 at 15:56 +0200, Cédric Cano wrote:
> >> Hi,
> >>
> >> Marvell SAS driver doesn't work on powerpc architecture due to big
> >> endian swap. I've tested it with the MV6440 chip.
> >> Here you can find the patch for Linux 2.6.39.1.
> >>
> >> Cedric Cano
> >>
> >> Signed-off-by: Cedric Cano<ccano@interfaceconcept.com>
> >> ---
> >> --- drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:04.000000000 +0200
> >> +++ drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:12.000000000 +0200
> >> @@ -1143,6 +1143,14 @@
> >>        MVS_CHIP_DISP->write_port_cfg_addr(mvi, i, PHYR_SATA_SIG0);
> >>        s[0] = MVS_CHIP_DISP->read_port_cfg_data(mvi, i);
> >>
> >> +    /* swap for big endian devices because of use of these data in bytes */
> >> +#ifdef __BIG_ENDIAN
> >> +    s[0] = swab32(s[0]);
> >> +    s[1] = swab32(s[1]);
> >> +    s[2] = swab32(s[2]);
> >> +    s[3] = swab32(s[3]);
> >> +#endif /* __BIG_ENDIAN */
> > Hi Cedric !
> >
> > Thanks for this. CC'ing linux-scsi.
> >
> > Note that it would probably have been better to use le32_to_cpu, which
> > avoids the ifdef completely.
> >
> > Cheers,
> > Ben.
> >
> >>        /* Workaround: take some ATAPI devices for ATA */
> >>        if (((s[1]&  0x00FFFFFF) == 0x00EB1401)&&  (*(u8 *)&s[3] == 0x01))
> >>            s[1] = 0x00EB1401 | (*((u8 *)&s[1] + 3)&  0x10);
> >>
> >> ---
> >> _______________________________________________
> >> Linuxppc-dev mailing list
> >> Linuxppc-dev@lists.ozlabs.org
> >> https://lists.ozlabs.org/listinfo/linuxppc-dev
> >
> >

Patch

--- drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:04.000000000 +0200
+++ drivers/scsi/mvsas/mv_sas.c    2011-06-10 15:50:12.000000000 +0200
@@ -1143,6 +1143,14 @@ 
      MVS_CHIP_DISP->write_port_cfg_addr(mvi, i, PHYR_SATA_SIG0);
      s[0] = MVS_CHIP_DISP->read_port_cfg_data(mvi, i);

+    /* swap for big endian devices because of use of these data in bytes */
+#ifdef __BIG_ENDIAN
+    s[0] = swab32(s[0]);
+    s[1] = swab32(s[1]);
+    s[2] = swab32(s[2]);
+    s[3] = swab32(s[3]);
+#endif /* __BIG_ENDIAN */
+
      /* Workaround: take some ATAPI devices for ATA */
      if (((s[1] & 0x00FFFFFF) == 0x00EB1401) && (*(u8 *)&s[3] == 0x01))
          s[1] = 0x00EB1401 | (*((u8 *)&s[1] + 3) & 0x10);