PCI Radeon RV100 detection hang on sparc64

[Kjetil Oftedal]

Hi,

That looks quite strange. I guess the kernel should map the ROM at the
address OpenBoot/OF assigned to it. ( 10020000 ).

Are pci devices located beneatch pci at 1f,0 not reserving resources
correctly ? (Thus the reuse of addresses when the addresses are
assigned during runtime ? or is there some virtual memory magic
fooling me ?)

Can you provide a dump of lspci -v ?

Regards,
Kjetil Oftedal

On 23/09/2013, mroos at linux.ee <mroos at linux.ee> wrote:
> Resurrecting an old thread...
>
>> On Fri, Sep 7, 2012 at 4:18 AM, David Miller <davem at davemloft.net> wrote:
>> > From: Michel Dänzer <michel at daenzer.net>
>> > Date: Thu, 06 Sep 2012 18:55:51 +0200
>> >
>> >> On Don, 2012-09-06 at 17:41 +0300, Meelis Roos wrote:
>> >>> This is with initialyy unmodified 3.6.0-rc4-00101-g0809095 kernel in
>> >>> Ultra 10 (clean, without my "Video RAM" hack that I talked in other
>> >>> sparclinux posts). When I saw that Sun XVR-100 was detected fine by
>> >>> the
>> >>> kernel, I compiled radeon drm driver with modesetting enabled and
>> >>> tried
>> >>> it:
>> >>>
>> >>> [drm] radeon defaulting to kernel modesetting.
>> >>> [drm] radeon kernel modesetting enabled.
>> >>> PCI: Enabling device: (0000:02:02.0), cmd 82
>> >>> [drm] initializing kernel modesetting (RV100 0x1002:0x5159
>> >>> 0x1002:0x0908).
>> >>> [drm] register mmio base: 0x10000000
>> >>> [drm] register mmio size: 32768
>> >>> [drm:radeon_device_init] *ERROR* Unable to find PCI I/O BAR
>> >>>
>> >>> And here the machine hangs. Debugging printk-s reveal that it does
>> >>> not
>> >>> find any active I/O port resources and then continues into
>> >>> initializing
>> >>> the card. Down in igp_read_bios_from_vram() it successfully ioremaps
>> >>> memory region 0 (vram_base=1ff08000000 and size=40000) and tries to
>> >>> read
>> >>> 2 bytes from there and hangs on reading bios[0].
>> >>
>> >> Is the card initialized by OpenFirmware? E.g., does it display
>> >> anything
>> >> before the kernel boots?
>> >
>> > I think Meelis gave you every single detail you need to know to
>> > analyze and fix this bug.
>> >
>> > He says it crashes on reading bios[0].
>> >
>> > If you look at the function he tells you the crash occurs in, you'll
>> > see clearly that bios is an ioremap()'d pointer.
>> >
>> > You cannot dereference ioremap()'d pointers, you must use the proper
>> > accessors such as readl().
>>
>> we should just skip the igp reading function on not-x86, since IGP on
>> not x86 makes no sense.
>>
>> the other option is to readl out the whole bios from vram instead.
>
> I made several changes.
>
> 1. In igp_read_bios_from_vram():
>
> +#ifndef CONFIG_X86
> +       /* IGP only exists on X86 (32- and 64-bit) */
> +       return false;
> +#endif
>
> This gets me further.
>
> 2. Since "bios" is __iomem pointer, changed its dereferences to
> readb(bios) and readb(bios+1). Maybe relevant, maybe not.
>
> 3. Next it got stuck in radeon_read_bios() trying to do
> kmemdup(bios, size, GFP_KERNEL);
>
> This I changed to mydup for test:
>
> +static void *mydup(uint8_t __iomem *src, size_t len, gfp_t gfp)
> +{
> +       unsigned char *p;
> +       int i;
> +
> +       p = kmalloc(len, gfp);
> +       if (p) {
> +               for (i=0; i<len; i++) {
> +                       p[i] = readb(src+i);
> +               }
> +       }
> +       return p;
> +}
>
> And this copying crashes Adaptec scsi driver in different ways.
>
> Investigating this, I found that it makes BAR 6 of the PCI radeon and
> gets bios=000001ff00000000 assigned. There is also some "Unable to find
> PCI I/O BAR" before - seesm that firmware does not map this ROM...
> To make things worse, the 2 first bytes at 000001ff00000000 actually do
> contain the ROM marker, so it goes on to copying.
>
> [drm] radeon kernel modesetting enabled.
> PCI: Enabling device: (0000:02:02.0), cmd 82
> [drm] initializing kernel modesetting (RV100 0x1002:0x5159 0x1002:0x0908).
> [drm] register mmio base: 0x000001FF10000000
> [drm] register mmio size: 32768
> [drm:radeon_device_init] *ERROR* Unable to find PCI I/O BAR
> [drm] radeon_read_bios 1
> radeon 0000:02:02.0: BAR 6: assigned [mem 0x1ff00000000-0x1ff0001ffff]
> [drm] radeon_read_bios: found ROM signature at 000001ff00000000
> [drm] radeon_read_bios 7: bios=000001ff00000000, size=46592
>
> However, this is where aic7xxx is mapped - /proc/iomem contains this
> (yes, the machine has ffb graphics and 2 ATI magc64 graphics heads too):
>
> 1fc00400000-1fc0040000f : ffb dac
> 1fc00600000-1fc00600983 : ffb fbc
> 1ff00000000-1ffffffffff : /pci at 1f,0
>   1ff00002000-1ff00002fff : aic7xxx
>   1ff000a0000-1ff000bffff : Video RAM area
>   1ff000c0000-1ff000c7fff : Video ROM
>   1ff000f0000-1ff000fffff : System ROM
>   1ff11000000-1ff11ffffff : atyfb
>   1ffc0000000-1ffdfffffff : IOMMU
>   1ffe0000000-1ffe000701f : sunhme
>   1ffe1000000-1ffe1ffffff : atyfb
>   1fff1000000-1fff1001fff : eeprom
>   1fff1200000-1fff120000f : cs4231
>   1fff13062f8-1fff13062ff : su
>   1fff13083f8-1fff13083ff : su
>   1fff1400000-1fff140003f : sab
>   1fff1400040-1fff140007f : sab
>   1fff1702000-1fff170200f : cs4231_pdma
>   1fff1704000-1fff170400f : cs4231_cdma
>   1fff1724000-1fff1724003 : power
>   1fff1726000-1fff1726003 : auxio
>
>
> Where do I go from here?
>
> --
> Meelis Roos (mroos at linux.ee)
> --
> To unsubscribe from this list: send the line "unsubscribe sparclinux" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>