[oi-dev] AMD Hudson SATA

[Jim Klimov]

So, this adventure is going from strange to weird and back.
The Live image with original ahci driver occasionally sees the
HDD in SATA mode, sometimes it doesn't. Reloading the ahci driver
(original or updated - see patch below; I've now also enabled
AHCI_CAP_INIT_PORT_RESET) in the latter case sometimes helps.
It does not seem that the driver tweak does anything (nothing
in dmesg for example).

The installed OS does not see the disk at all and refuses to
load (panics). I tried to sanitize it by importing and exporting
the rpool, and by installing from scratch - both while in SATA
mode - the resulting OS does not boot the same way. With kmdb
pause after panic I see the following relevant messages:

---
Cannot mount pool on /pci at 0,0/pci17aa,5104 at 11/disk at 0,0:a fstype zfs
* The path is correct when I'm booted off the live image
---
Loaded modules: [ scsi_vhci mac uppc mr_sas sd unix mpt zfs krtld
sata apix pmcs genunix specfs pcplusmp cpu.generic mpt_sas ]
* I don't see ahci here. Adding it to /etc/system (forceload: drv/ahci)
and /etc/driver_aliases does not influence the result.
---

As I understand the lookout of things, the miniroot does not
load the ahci module for some reason, and then honestly does
not see the disk attached to ahci controller.

Suggestions welcome.

2012-08-30 14:07, Jim Klimov wrote:
> Hi all,
>
> I am struggling to get AHCI working with my new Thinkpad E335 notebook.
> According to OI Device Driver utility, it includes an "AMD Hudson SATA
> Controller" when it is BIOS-enabled into SATA (AHCI) mode.
>
> However, only twice of a dozen bootups did it see the HDD behind it,
> once was in the shop when I was testing the notebook for compatibility ;)
>
> The system did install and work stably with the controller in
> "Compatibility mode" (seeing the disk as CMDK IDE), but I'd prefer
> SATA working in a stable manner. Apparently, some port-reset hints,
> etc. are in order as for some other controllers in the driver.
>
> I'm trying to give a spin to recent bug report on similar issue,
> but I'd take into account that my box has a different devid
> (mine is pci1022,7804):
> https://www.illumos.org/issues/3099
>
> Do I understand correctly that it is enough to patch ahci.c and build
> the driver and replace one on the laptop?
>
> # gdiff -Nabur usr/src/uts/common/io/sata/adapters/ahci/ahci.c.orig
> usr/src/uts/common/io/sata/adapters/ahci/ahci.c
>
> --- usr/src/uts/common/io/sata/adapters/ahci/ahci.c.orig 2012-04-19
> 02:17:42.110503288 +0400
> +++ usr/src/uts/common/io/sata/adapters/ahci/ahci.c     2012-08-30
> 13:57:09.358521233 +0400
> @@ -4118,6 +4118,31 @@
>                      "is set 0xf and no port multiplier is attached",
> NULL);
>          }
>
> +       /* AMD Hudson D3 AHCI 0x1002 : 0x7801, reportedly similar to
> SB700/750
> +        * See: https://www.illumos.org/issues/3099
> +        * I'll also try 0x7804 as on my box
> +        */
> +       if (ahci_ctlp->ahcictl_venid == 0x1002 &&
> +           ( ahci_ctlp->ahcictl_devid == 0x7801 ||
> +             ahci_ctlp->ahcictl_devid == 0x7804 ) ) {
> +               ahci_ctlp->ahcictl_cap |= AHCI_CAP_COMMU_32BIT_DMA;
> +//             ahci_ctlp->ahcictl_cap |= AHCI_CAP_INIT_PORT_RESET;
> +               ahci_ctlp->ahcictl_cap |= AHCI_CAP_SRST_NO_HOSTPORT;
> +
> +               AHCIDBG(AHCIDBG_INIT, ahci_ctlp,
> +                   "AMD Hudson D3 cannot do 64-bit DMA for communication "
> +                   "memory descriptors though CAP indicates support, "
> +                   "so force it to use 32-bit DMA", NULL);
> +//             AHCIDBG(AHCIDBG_INIT, ahci_ctlp,
> +//                 "AMD Hudson D3 need to do a port reset during
> initialization",
> +//                 NULL);
> +               AHCIDBG(AHCIDBG_INIT, ahci_ctlp,
> +                   "AMD Hudson D3 will get software reset failure if
> pmport "
> +                   "is set 0xf and no port multiplier is attached", NULL);
> +       }
> +
> +
> +
>          /*
>           * Check if capabilities list is supported and if so,
>           * get initial capabilities pointer and clear bits 0,1.
>
>