Le 04/08/2011 05:45, Hong Xu a écrit :
After DMA transfer, we need to maintain D-Cache coherency. We need to clean cache (write back the dirty lines) and then make the cache invalidate as well(hence CPU will fetch data written by DMA controller from RAM).
Tested on AT91SAM9261EK with Peripheral DMA controller.
Signed-off-by: Hong Xuhong.xu@atmel.com Tested-by: Elen Songelen.song@atmel.com CC: Heiko Schocherhs@denx.de CC: Albert Aribaudalbert.aribaud@free.fr
arch/arm/lib/cache.c | 10 ++++++++++ 1 files changed, 10 insertions(+), 0 deletions(-)
diff --git a/arch/arm/lib/cache.c b/arch/arm/lib/cache.c index 92b61a2..216bde0 100644 --- a/arch/arm/lib/cache.c +++ b/arch/arm/lib/cache.c @@ -53,3 +53,13 @@ void __flush_dcache_all(void) } void flush_dcache_all(void) __attribute__((weak, alias("__flush_dcache_all")));
+void __invalidate_dcache_all(void) +{ +#ifdef CONFIG_ARM926EJS
- asm("mcr p15, 0, %0, c7, c6, 0" : : "r" (0));
+#endif
- return;
+} +void invalidate_dcache_all(void)
- __attribute__((weak, alias("__invalidate_dcache_all")));
NAK in this form.
I don't want invalide function introduced only in full invalidation form, because driver code will start using this, and it is the wrong thing to do as each caller can only tell what it wants invalidated, not what other code wants invalidated; a client doing a full invalidate might (and certainly will, to some extent) trash other data than its own.
So please provide also an invalidate_cache() function similar to the flush_cache() function, with start and size arguments, that callers should use.
full flush and invalidate should only be used by callers that *really* need it (bootm comes to mind).
Amicalement,