On Tue, Jul 10, 2012 at 12:46:21PM +0400, Ilya Yanok wrote:
Hi guys,
On Tue, Jul 10, 2012 at 12:37 PM, Tom Rini trini@ti.com wrote:
What do we want to do about the USB issue (on ARM platforms, with
EHCI,
with >32byte alignment requirements, if dcache isn't build-time disabled, USB is unusable, a change from previous releases), for this release? Are we going to hope the alignment issues can be flushed
out
and fixed well enough before the final release? Should we go with disable the dcache now, continue working on it for the next?
What are the chances that the issue be fixed?
Low ... this is some deep crap that's growing through uboot as whole :-(
And
it's not only USB.
Then I propose 1, 3, 4, 5, 6 from my v4 series (remove dcache_off call from ehci-omap.c as it's wrong, build-time disable DCACHE on USB enabling platforms). Yes, we're papering over bugs for a release, but we (a) know we're doing it and (b) are trying to fix them and (c) can't fix them in time.
I will resend the series this evening. There is no bounce buffering support so unaligned operations (unaligned address passed by user) are broken but otherwise I think everything works.
And I'd really like to make runtime cache disabling work again. Probably we can rename {flush,invalidate}_dcache_range() to __{flush,invalidate}_dcache_range() and call these functions there we really need to do these operations regardless of cache being enabled. Then we could create new {flush,invalidate}_dcache_range() and do if (dcache_enabled()) check in them,,,
We've got a release date now. If you think you can get this done and tested in time, OK. But if we don't have this by -rc2 time we need to put in the big stupid fix instead (since not all distributions come to me with questions about things, I fear some may say USB doesn't work and hold back for who knows how long).