Re: SiI3124 + SiI3726 pm

Re: SiI3124 + SiI3726 pm

[Tejun Heo]

[-- Attachment #1: Type: text/plain, Size: 466 bytes --]

Jérôme VUARAND wrote:
> I just realised after the mail that the driver was really thinking
> there are 6 ports on the pm, since it's testing ata3.05 and ata4.05.
> These don't exist and generate errors. There's a bug somewhere and
> here is a workaround I made quickly to have correct enumeration
> behaviour (I hope no automatic process parse mails on this mailing
> list for patches, coz this one is not really clean) :

Can you try the attached patch?

-- 
tejun

[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 561 bytes --]

diff --git a/drivers/scsi/libata-pm.c b/drivers/scsi/libata-pm.c
index 002d8b2..7a56ab7 100644
--- a/drivers/scsi/libata-pm.c
+++ b/drivers/scsi/libata-pm.c
@@ -327,6 +327,14 @@ static void ata_pm_quirks(u32 *gscr, int
 		*nr_ports -= 2;
 		*link_flags |= ATA_LFLAG_HRST_TO_RESUME;
 	}
+
+	/* Sil3726 reports one extra port and needs hardreset to
+	 * resume PM link.
+	 */
+	if (vendor == 0x1095 && devid == 0x3726) {
+		*nr_ports -= 1;
+		*link_flags |= ATA_LFLAG_HRST_TO_RESUME;
+	}
 }
 
 static int ata_pm_configure(struct ata_device *dev, int *r_nr_ports,

Re: SiI3124 + SiI3726 pm

[saeed bishara]

Hi,
       I think the 6th port is for the SEMB(Sata Enclosure Management
Bridge) device, this device reportssignature with lbam=0x3c and lbah
0xc3, but the linux classifies it as a disk:

  if (((tf->lbam == 0) && (tf->lbah == 0)) ||
  ((tf->lbam == 0x3c) && (tf->lbah == 0xc3))) {
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
DPRINTK("found ATA device by sig\n");
 return ATA_DEV_ATA;

Can you print the signature of the device on that port?
Anyway, the 0x3C, 0xC3 signature should be classifies as a SEMB device
and not a disk according to the SATA standard.

saeed

On 5/25/06, Tejun Heo <htejun@gmail.com> wrote:
> Jérôme VUARAND wrote:
> > I just realised after the mail that the driver was really thinking
> > there are 6 ports on the pm, since it's testing ata3.05 and ata4.05.
> > These don't exist and generate errors. There's a bug somewhere and
> > here is a workaround I made quickly to have correct enumeration
> > behaviour (I hope no automatic process parse mails on this mailing
> > list for patches, coz this one is not really clean) :
>
> Can you try the attached patch?
>
> --
> tejun
>
>
> diff --git a/drivers/scsi/libata-pm.c b/drivers/scsi/libata-pm.c
> index 002d8b2..7a56ab7 100644
> --- a/drivers/scsi/libata-pm.c
> +++ b/drivers/scsi/libata-pm.c
> @@ -327,6 +327,14 @@ static void ata_pm_quirks(u32 *gscr, int
>                 *nr_ports -= 2;
>                 *link_flags |= ATA_LFLAG_HRST_TO_RESUME;
>         }
> +
> +       /* Sil3726 reports one extra port and needs hardreset to
> +        * resume PM link.
> +        */
> +       if (vendor == 0x1095 && devid == 0x3726) {
> +               *nr_ports -= 1;
> +               *link_flags |= ATA_LFLAG_HRST_TO_RESUME;
> +       }
>  }
>
>  static int ata_pm_configure(struct ata_device *dev, int *r_nr_ports,
>
>
>

Re: SiI3124 + SiI3726 pm

[Tejun Heo]

saeed bishara wrote:
> Hi,
>       I think the 6th port is for the SEMB(Sata Enclosure Management
> Bridge) device, this device reportssignature with lbam=0x3c and lbah
> 0xc3, but the linux classifies it as a disk:
> 
>  if (((tf->lbam == 0) && (tf->lbah == 0)) ||
>  ((tf->lbam == 0x3c) && (tf->lbah == 0xc3))) {
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> DPRINTK("found ATA device by sig\n");
> return ATA_DEV_ATA;
> 
> Can you print the signature of the device on that port?
> Anyway, the 0x3C, 0xC3 signature should be classifies as a SEMB device
> and not a disk according to the SATA standard.

I haven't actually checked that signature but that's probably the same 
class code mix up described for PM in the classification function.  For 
the time being, libata is just ignoring the extra port.  It probably 
should be exported as a SCSI device without HLD attached (/dev/sgX 
only).  I'll try that once hotplug stuff settles.

Thanks.

-- 
tejun

Re: SiI3124 + SiI3726 pm

[Jerome Vuarand]

2006/5/25 Tejun Heo <htejun@gmail.com>:
> saeed bishara wrote:
>>
>> Hi,
>>      I think the 6th port is for the SEMB(Sata Enclosure Management
>> Bridge) device, this device reportssignature with lbam=0x3c and lbah
>> 0xc3, but the linux classifies it as a disk:
>>
>>  if (((tf->lbam == 0) && (tf->lbah == 0)) ||
>>  ((tf->lbam == 0x3c) && (tf->lbah == 0xc3))) {
>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> DPRINTK("found ATA device by sig\n");
>> return ATA_DEV_ATA;
>>
>> Can you print the signature of the device on that port?
>> Anyway, the 0x3C, 0xC3 signature should be classifies as a SEMB device
>> and not a disk according to the SATA standard.
>
> I haven't actually checked that signature but that's probably the same class
> code mix up described for PM in the classification function.  For the time
> being, libata is just ignoring the extra port.  It probably should be
> exported as a SCSI device without HLD attached (/dev/sgX only).  I'll try
> that once hotplug stuff settles.

A few months ago I updated my kernel since it now supports port
multipliers without patch, and the issue of a supplementary port on
the SiI3727 port multiplier rose again. Since I completely forgot
about that 3 year old issue, I started exploring the web again before
remembering that thread.

I found something interesting in the SiI3726 datasheet
(http://www.siliconimage.com/docs/SiI-DS-0121-C1.pdf). You can see on
PDF page 14/41, in the description of the GSCR[2] register, that the 4
least significant bits of the register are labelled "Number of Fan-out
ports". In the description besides, it is explained that the initial
value is 6, and it represents the 5 downstream device ports plus the
upstream host port.

I don't have access to the SATA Port Multiplier specification, but it
would be interesting to check it to verify whether that 4-bit field of
the GSCR[2] register is supposed to contain the number of device ports
(and in that case the problem comes from the SiI3726 firmware) or if
that value is the total number of up+down ports of the multiplier. In
the latter case another mechanism to detect the number of device ports
would be needed (unless multipliers can have several host ports, for
multipath or device sharing, a simple substraction of 1 could be
enough).

In the meantime I'll re-apply the patch Tejun Heo proposed earlier in
that thread. It seems I didn't notice Saeed Bishara request for tests
last time. Tell me if you're still interested in that information, or
if you need any other information to help improve the SiI3726 driver.
I have 6 of these port multipliers (and I suspect some of these are
damaged), and 15 to 20 SATA disks available, so there's plenty of room
for tests.

Re: SiI3124 + SiI3726 pm

[Tejun Heo]

Hello, Jerome.

Jerome Vuarand wrote:
> A few months ago I updated my kernel since it now supports port
> multipliers without patch, and the issue of a supplementary port on
> the SiI3727 port multiplier rose again. Since I completely forgot
> about that 3 year old issue, I started exploring the web again before
> remembering that thread.

Hmm... this is supposed to be worked around by now.  Can you please
post boot log w/o the patch using 2.6.30?

Thanks.

-- 
tejun

Re: SiI3124 + SiI3726 pm

[Jerome Vuarand]

2009/7/28 Tejun Heo <htejun@gmail.com>:
> Jerome Vuarand wrote:
>> A few months ago I updated my kernel since it now supports port
>> multipliers without patch, and the issue of a supplementary port on
>> the SiI3727 port multiplier rose again. Since I completely forgot
>> about that 3 year old issue, I started exploring the web again before
>> remembering that thread.
>
> Hmm... this is supposed to be worked around by now.  Can you please
> post boot log w/o the patch using 2.6.30?

Yes, actually when about to apply the patch, I realized the 6th port
is actually recognized as the PMP SEMB. I guess the datasheet is
somewhat incorrect on that precise page, since it mentions the SEMB
device in other sections. The wiki is somewhat terse about PMP support
(for example it's not very clear what drivers and controller
combination support PMPs), and afaict there is no mention of SEMB
support status (I don't think my PMPs have something connected to the
I²C port of the SiI3726, but that may be helpful to other people).

I moved all my disks to some spare PMP and all my disk detection
problems are gone. I guess some of them are really damaged (also I had
some power issue within my SATA enclosure).

Sorry for the noise :-)