On Sun, Aug 14, 2016 at 07:43:56PM +0000, Karl Beldan wrote:
On Sun, Aug 14, 2016 at 11:47:25AM -0400, Tom Rini wrote:
On Sun, Aug 14, 2016 at 03:03:16PM +0000, Karl Beldan wrote:
check_cache_range() warns that the top boundaries are not properly aligned while flushing and invalidating the buffers and make these operations to fail. ATM the RX bottom boundaries are aligned by design with EMAC_RXBUF_SIZE, properly aligned with ARCH_DMA_MINALIGN, however the top ones are not.
This gets rid of the warnings: CACHE: Misaligned operation at range
Signed-off-by: Karl Beldan karl.beldan+oss@gmail.com
drivers/net/davinci_emac.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/net/davinci_emac.c b/drivers/net/davinci_emac.c index 947bfab..55461b0 100644 --- a/drivers/net/davinci_emac.c +++ b/drivers/net/davinci_emac.c @@ -632,7 +632,8 @@ static int davinci_eth_send_packet (struct eth_device *dev, EMAC_CPPI_EOP_BIT);
flush_dcache_range((unsigned long)packet,
(unsigned long)packet + length);
round_up((unsigned long)packet + length,ARCH_DMA_MINALIGN));It's preferred to use: (unsigned long)packet + ALIGN(length, PKTALIGN)); here instead of ARCH_DMA_MINALIGN.
Hmm, I think your suggestion is buggy. The cache primitives act on [laddr, haddr[, i.e. haddr is excluded, IOW you are missing the tail of the packet (that's why I rounded up).
Joe, we've got some instances using ALIGN(length, PKTALIGN) and others using roundup. And I can't imagine that we shouldn't be using the same thing everywhere here. So, what do you say to Karl's comment? Thanks!
Conceptually I still prefer ARCH_DMA_MINALIGN, also all other code in the base does so.
Being network code, we should use PKTALIGN here and elsewhere.