LSI problem, endianness & cleanups
benh at kernel.crashing.org
Tue Jul 12 22:46:21 PDT 2011
Hi Alex !
A couple of things I noticed while doing some of the fixes I posted
- There seem to be what could be a HW issue, I would appreciate if you
could double check with your HW guys, at least on rv610 (I haven't been
able to test on other cards).
When MSI is -disabled- (using old style LSIs), my PCI-E bridge detected
an invalid DMA after every interrupt. I am not sure I managed to capture
the address properly, it might be 0 but I need to double check with my
own HW guys.
This doesn't happen when MSI is enabled.
I -looks- like the chip might be trying to shoot an MSI even when MSIs
aren't enabled in the config space, possibly using the (stale) content
of the config space MSI address register. Can you check if your HW folks
know anything about this ?
- There's a lot of horrible duplication of register definitions :-) I
think there's at least 3 copies of CP_RB_CNTL and associated bits. One
thing I'd like to cleanup (not -that- but related) is the way we set the
endian swap bits.
Basically, we currently have #ifdef's that duplicate entire multi-line
statements changing one bit (almost always the same).
I'm thinking about instead defining in a .h something like
#define CP_RB_SWAP CP_RB_SWAP_32BIT
#define CP_RB_SWAP 0
And using CP_RB_SWAP everywhere, removing all the dups (which are bug prone
as always with dups). I'd like to make sure you are ok with that before I
jump and do it, and especially ok with having a _single_ instance of that bit
def for all radeon variants since the bits are in the same place everywhere.
More information about the xorg-driver-ati