Re: SATA exceptions with 2.6.20-rc5

[Robert Hancock <hancockr@shaw.ca>]

I heard from Larry Walton who was apparently seeing this problem as 
well. He tried my recent "sata_nv: cleanup ADMA error handling v2" patch 
and originally thought it fixed the problem, but it turned out to only 
make it happen less often.

I wouldn't expect that patch to have an effect on this problem. If it 
seems to reduce the frequency that would tend to be further evidence of 
  some kind of timing-related issue where the code change just happens 
to make a difference.

I'll see if I can come up with a debug patch for people having this 
problem to try, which prints out when a flush command is issued and what 
interrupts happen when a flush is pending.

There is one important difference between ADMA and non-ADMA mode for 
non-DMA commands like flushes, which didn't come to mind before: ADMA 
mode uses MMIO registers on the controller whereas non-ADMA mode uses 
legacy IO registers. Posted write flushing is a concern with MMIO 
registers but not with PIO, the libata core is supposed to handle this 
but maybe it doesn't in some case(s). In fact, just looking at 
libata-sff.c there's this comment on the ata_exec_command_mmio function:

  *      FIXME: missing write posting for 400nS delay enforcement

That seems a bit suspicious..

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/