[U-Boot] [PATCH 2/4] net: round up before calling flush_cache

[Marek Vasut]

Dear Stefano Babic,

> On 01/04/2012 15:46, Marek Vasut wrote:
> > Dear Stefano Babic,
> 
> Hi Marek,
> 
> >> If the range passed to flush_cache is not multiple
> >> of ARCH_DMA_MINALIGN, a warning due to mislaignment
> >> is printed.
> >> Detected with fec_mxc, mx35 boards:
> >> 
> >> CACHE: Misaligned operation at range [80800000, 8083c310]
> >> 
> >> Signed-off-by: Stefano Babic <sbabic at denx.de>
> >> CC: Marek Vasut <marex at denx.de>
> >> CC: Joe Hershberger <joe.hershberger at gmail.com>
> >> Cc: Wolfgang Denk <wd at denx.de>
> >> ---
> >> 
> >>  common/cmd_net.c |    2 +-
> >>  1 files changed, 1 insertions(+), 1 deletions(-)
> >> 
> >> diff --git a/common/cmd_net.c b/common/cmd_net.c
> >> index 65f32bc..a500919 100644
> >> --- a/common/cmd_net.c
> >> +++ b/common/cmd_net.c
> >> @@ -256,7 +256,7 @@ static int netboot_common(enum proto_t proto,
> >> cmd_tbl_t *cmdtp, int argc, }
> >> 
> >>  	/* flush cache */
> >> 
> >> -	flush_cache(load_addr, size);
> >> +	flush_cache(load_addr, roundup(size, ARCH_DMA_MINALIGN));
> > 
> > This ain't gonna slide. You might overwrite something, even though this
> > is just loading into memory, right?
> 
> Really we do not reserve space before a network transfer and we pass
> only the start address (load address) where the file is copied. It is
> not known its size before the transfer. So maybe it is not important up
> to further 31 bytes are flushed ;-)
> 
> Anyway, for my understanding: we are calling a function to flush caches.
> This means that if the size is something more as required there will be
> a cache hit misseing, and the SOC will load data again - but without
> corrupt data, right ?

Sure, but you can flush data into the RAM that should have never been flushed to 
the ram in the first place and that might confuse some controller -- or I might 
be overly paranoid.

> 
> > I'm not quite sure how to handle this kind of
> > unaligned access.
> 
> Well, my first goal was to explain which is the cause of the
> misalignment I found last friday testing Anatolij's patch, proofing that
> it is not due to last FEC patches ;-)
> 
> > But adding at least if (unaligned) debug(...); to aid people easily
> > finding these trouble would be nice ;-)
> 
> Not sure - where should be inserted  or what do you exactly mean? size
> is the length of the transfered bytes, and can assueme any value, so it
> is often not aligned.

Well just before the cache-line aligning.

> Best regards,
> Stefano Babic

Best regards,
Marek Vasut