Chris,
Sorry for the long delay.

On Tue, Aug 18, 2009 at 2:54 PM, Chris K[off-list ref] wrote:
...

quoted

Can you confirm the OS is seeing as a 3726?

The later output you posted confirmed this.

quoted

Add a printk to dump SATA_PMP_GSCR_PROD_ID to the console
similar to how it's done in sata_pmp_revalidate_quick().  Or if dev->gscr[]
is already valid, printk dev->gscr[SATA_PMP_GSCR_PROD_ID]. This
assumesyou know how to build/install a kernel from a source tree.
Tell me if that's a bad assumption. :)

How about generous assumption ;-)  Could you elaborate on the steps
for this? I should be able to figure it out with a bit more info on
the process.

The kernel build process is in general steps very simple:
o download the latest stable kernel source from www.kernel.org
o unpacked the source (e.g. tar xzf linux-2.6.*.tar.bz2)
o cd linux-2.6.*
o cp /boot/
o make
o sudo make modules_install
o sudo make install

and check the boot loader (grub likely) knows vmlinuz (not vmlinux) is
installed.
Also need to look if an initrd is also needed. "man mkinitramfs" should help
build the initrd if "make install" isn't automagically doing that.


Then to "dump SATA_PMP_GSCR_PROD_ID", one needs to add a printk to
the libata layer. I've attached a patch to libata_pmp.c as a simple example.
Apply the patch with "patch -p1 < patchname". See
Documentation/SubmittingPatches
on how to a submit/apply patches.

quoted

ISTR some very early 3726 devices reported themselves as 4726 or
something like that.
This can affect which sata_pmp_quirks() are applied.

I noticed this line in the dmesg, does this confirm if it's reporting
the right model properly?
"ata7.15: Port Multiplier 1.1, 0x1095:0x3726 r23, 6 ports, feat 0x1/0x9"

Yes.

quoted

BTW, the full dmesg output attached to the original mail never made it
to linux-ide mailing list...can you attach or post it someplace public?

Yes, you can find my dmesg @ http://pastebin.com/f50585886

Thanks!

Not sure if you saw this part before too - This is the output I get on
the screen during boot:

[59719.989422] ata7: irq_stat 0x01100010, PHY RDY changed
[59719.989477] ata7: SError: { 10B8B }
[59722.781043] ata7: exception Emask 0x10 SAct 0x0 SErr 0x0 action 0xe frozen
[59722.781112] ata7: irq_stat 0x00b40090, PHY RDY changed
[59791.300259] ata7.15: failed to write PMP_FEAT_EN (Emask=0x40)
[59791.300324] ata7: failed to recover PMP after 5 tries, giving up
[59791.300323] ata7: exception Emask 0x3 SAct 0x0 SErr 0x0 action 0x6 frozen t4
[59791.300483] ata7: irq_stat 0x00060002, device error via D2H FIS

The "6 ports" is the clue I was looking for.

[    7.080266] ata7.15: Port Multiplier 1.1, 0x1095:0x3726 r23, 6
ports, feat 0x1/0x9
[    7.080531] ata7.00: hard resetting link
[    7.430307] ata7.00: SATA link up 3.0 Gbps (SStatus 123 SControl 320)
[    7.430311] ata7.01: hard resetting link
[    7.780305] ata7.01: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[    7.780309] ata7.02: hard resetting link
[    8.130305] ata7.02: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[    8.130310] ata7.03: hard resetting link
[    8.480305] ata7.03: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[    8.480310] ata7.04: hard resetting link
[    8.830317] ata7.04: SATA link down (SStatus 0 SControl 320)
[    8.830321] ata7.05: hard resetting link
[    9.180232] ata7.05: failed to read SCR 1 (Emask=0x1)
[    9.180233] ata7.05: failed to read SCR 0 (Emask=0x40)

I don't believe port ata7.05 exists. AFAIK,  sil3736 only has 5 ports (0 to 4).
One of the quirks isn't applying or needs to be added to the kernel.
I don't recall what the psuedo port is for but Tejun or "Sans Digital
TRM4-B" vendor might know.

I think this comment in libata-pmp.c might be relevant:
     /* port 5 is for SEMB device and it doesn't like SRST */

But it looks like port 5 isn't responding at all. You might add a hack to change
"6" to "5" inside the "sil3726 quirks" code chunk.

I would expect the "SEMB" device to respond but perhaps it needs more time
or some other magic that linux doesn't know in order to initialize.

...

LED 1 takes about 2-3 seconds.. the other LEDs are lit very quickly
one after another.

quoted

If CONFIG_PRINTK_TIME=y, the dmesg output should let me figure this out.

This seems to be enabled already with my current kernel.

Yup - it is.

quoted

Can you add debug code to dump how many ports the PMP claims to have?
I'm wondering if a "ghost port" (magic "port" which doesn't have a SATA link but
some magic registers instead) is discovered/reported.

I can sure try! What should I do to enable this?

The console output about "6 ports" already tells us this. But the
patch below will tell us again. :)

hth,
grant