On Wed, Mar 22, 2000 at 02:01:31PM -0800, David Brownell wrote:
> Bob --
>
> So at this point my own leading hypothesis is that this is the "Master
> Abort" case, caused by a bad address in the TD data structures. Which
> for you and me causes UE (controller fails) and for at least one other
> person causes a kernel panic. Could be wrong, but it'd explain more
> than enough to be worth checking out. And I'm tired of trying register
> order tweaks and comparing initialization sequences, anyway!
>
> Still leaves a question about why the NEC and Lucent chips see it, and
> other OHCIs don't (e.g. my ALI one).
>
> So later today (tonight?) I'll probably get friendly with TD data lists
> to check out that hypothesis. Anyone with more recent TD dumping code
> than the 2.2.11 stuff I just noticed, feel free to send me a copy ... :-)
>
> - Dave
Would it make sense to hypothesize [without looking at the code] that
Roman's driver may be forgetting a virt_to_bus call somewhere when filling
in an address for an ED or TD? Some controllers may magically not barf
on this since often the result of a virt_to_bus is similar to ORing the
address with 0xC0000000 and their attempts to access memory over PCI at
0xcbobfood ends up accessing the 0x0bobfood that they meant to access in
the first place either without an error or with an error that they ignore.
(I don't know how PCI is supposed to deal with memory accesses to invalid
addresses but I suspect it could be one of those "iffy" cases on some
hardware that may or may not be implemented 100% right)
[I'm also assuming that virt_to_bus hides any issues of architectures
who's PCI memory accesses get translated through the system's TLB (doesn't
the Alpha do this?) by being an identity macro on those architectures]
> p.s. This is really good info -- many thanks to you and your contact
> at Lucent!! Feels like we're on track to find out what causes
> these problem.
definately! I hope this is the problem. :)
Greg
PS BTW, I'm looking into someones comment that the mailing list
archive stopped archive. I may need to kick it.
--
Gregory P. Smith gnupg/pgp: http://suitenine.com/greg/keys/
C379 1F92 3703 52C9 87C4 BE58 6CDA DB87 105D 9163
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
