RE: DMAR EHCI failures

Thought someone replied to this,

Basically graphics drivers don't use DMA apis for translation at the
moment, hence when we turn on translation we will get faults since IOMMU
would not be able to translate.

The workaround used to do two things.

1. IF this is dedicated DMAR for igfx, then we just disable the iommu
2. if this is pcie graphics, we provide a 1-1 so we can avoid those
faults.


Sometimes you get those faults for usb devices if the RMRR is not set
correctly by BIOS. An easy workaround is if you disable legacy mode in
bios those should disappear.


Cheers,
ashok raj
- Open Source Technology Center


-----Original Message-----
From: David Brownell [mailto:david-b@xxxxxxxxxxx]
Sent: Saturday, February 09, 2008 6:03 PM
To: Jiri Slaby
Cc: David Brownell; linux-usb@xxxxxxxxxxxxxxx; Alan Stern; Linux Kernel
Mailing List; Raj, Ashok; Li,
Shaohua; Keshavamurthy, Anil S; Dave Airlie
Subject: Re: DMAR EHCI failures

On Monday 04 February 2008, Jiri Slaby wrote:
Hi,

I have this in dmesg:
DMAR:[DMA Write] Request device [00:02.0] fault addr ee1512000
DMAR:[fault reason 05] PTE Write access is not set
PCI-DMA: Intel(R) Virtualization Technology for Directed I/O
DMAR:[DMA Read] Request device [00:1d.7] fault addr 7d5f0000
DMAR:[fault reason 06] PTE Read access is not set
DMAR:[DMA Read] Request device [00:1a.7] fault addr 7d5f1000
DMAR:[fault reason 06] PTE Read access is not set
PCI-GART: No AMD northbridge found.
DMAR:[DMA Read] Request device [00:1a.2] fault addr 7d5f7000
DMAR:[fault reason 06] PTE Read access is not set

CONFIG_DMAR=y
CONFIG_DMAR_GFX_WA=y
CONFIG_DMAR_FLOPPY_WA=y

Without the gfx workaround, there is much more output regarding
00:02.0. Is
there problem with broken hw, bios or drivers?

No idea. Someone who knows the DMA Remapping stuff should have
an answer. Presumably it works with DMAR disabled, yes? If so,
then just don't use DMAR. :)


/sys/firmware/acpi/tables/DMAR:
http://www.fi.muni.cz/~xslaby/sklad/DMAR.bin
dmesg:
http://www.fi.muni.cz/~xslaby/sklad/DMAR.dmesg

# for a in 00:02.0 00:1d.7 00:1a.7 00:1a.2; do lspci -vxxx -s $a;
done
00:02.0 VGA compatible controller: Intel Corporation 82G33/G31
Express
Integrated Graphics Controller (rev 02) (prog-if 00 [VGA controller])
... deletia ...



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Relevant Pages

  • [PATCH 13/21] mm: page fault handler locking
    ... install_page just zaps the entry and reinstalls it). ... But on those architectures whose pte is too big ... But get_user_pages can cause write faults for mappings ... * We enter with non-exclusive mmap_sem (to exclude vma changes, ...
    (Linux-Kernel)
  • [PATCH RFP-V4 11/13] RFP prot support: fix get_user_pages() on VM_MANYPROTS vmas
    ... This is needed since now pte_to_pgprot does heavy changes to the pte, ... * MAY flags. ... servicing faults for write access. ... ptep_establish(vma, address, page_table, entry); ...
    (Linux-Kernel)
  • Re: Proust: in a position than which none
    ... none could be counted upon to involve me in graver consequences at my ... list (doubtless because there was no other class of faults from which ... expect from Proust, even in translation). ... pouvait avoir pour moi, de la part de mes parents, les conséquences les ...
    (alt.usage.english)
  • Re: Populating multiple ptes at fault time
    ... When Linux is operating as a virtual guest, pte population will ... It seems to me there are two cases: major faults, ... we should create mappings for any ...
    (Linux-Kernel)
  • Re: [patch 5/6] mm: merge nopfn into fault
    ... Different kinds of faults, as returned by handle_mm_fault. ... * vma cannot be a COW mapping. ... * We return with mmap_sem still held, but pte unmapped and unlocked. ... return ret; ...
    (Linux-Kernel)