On Thu, 2015-07-09 at 09:47 +0200, Marcel Ziswiler wrote:
Hi Scott
On Wed, 2015-07-08 at 18:25 -0500, Scott Wood wrote:
As per discussion on v1, this isn't enough to guarantee that the stop address will be aligned. There needs to be a wrapper around memalign() that cache- aligns the size as well.
-Scott
Sorry, that isn't quite clear to me yet. You're saying I would need to handle that in this patch?
Looking through the sources I actually found only one single usage of memalign() which explicitly takes care of this the way you propose:
drivers/usb/host/xhci-mem.c: ptr = memalign(cacheline_size, ALIGN(size, cacheline_size));
I'm just wondering about the 107 other places in U-Boot where memalign() is already used the way I do in this patch.
I suppose there are enough existing ones that it won't hurt to add more -- it's still an improvement over the current situation -- but it'd be good to fix it properly before it gets forgotten.
Wouldn't there need to be an infrastructure solution done to e.g. memalign() itself?
No, just a wrapper:
static inline void *malloc_cache_aligned(size_t size) { return memalign(ARCH_DMA_MINALIGN, ALIGN(size, ARCH_DMA_MINALIGN)); }