Blackfin: fix MPU page permission masks overflow when dealing with async memory
Attempting to use the MPU while doing XIP out of parallel flash hooked up
to the async memory bus would often result in random crashes as the MPU
slowly corrupted memory.
The fallout here is that the async banks gain MPU protection from user
space too. So any accesses have to go through the mmap() interface rather
than just using hardcoded pointers.
Signed-off-by: Barry Song <barry.song@analog.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
diff --git a/arch/blackfin/include/asm/mmu_context.h b/arch/blackfin/include/asm/mmu_context.h
index ae8ef4f..7f363d7 100644
--- a/arch/blackfin/include/asm/mmu_context.h
+++ b/arch/blackfin/include/asm/mmu_context.h
@@ -13,6 +13,7 @@
#include <asm/page.h>
#include <asm/pgalloc.h>
#include <asm/cplbinit.h>
+#include <asm/sections.h>
/* Note: L1 stacks are CPU-private things, so we bluntly disable this
feature in SMP mode, and use the per-CPU scratch SRAM bank only to
@@ -117,9 +118,16 @@
unsigned long flags)
{
unsigned long *mask = mm->context.page_rwx_mask;
- unsigned long page = addr >> 12;
- unsigned long idx = page >> 5;
- unsigned long bit = 1 << (page & 31);
+ unsigned long page;
+ unsigned long idx;
+ unsigned long bit;
+
+ if (unlikely(addr >= ASYNC_BANK0_BASE && addr < ASYNC_BANK3_BASE + ASYNC_BANK3_SIZE))
+ page = (addr - (ASYNC_BANK0_BASE - _ramend)) >> 12;
+ else
+ page = addr >> 12;
+ idx = page >> 5;
+ bit = 1 << (page & 31);
if (flags & VM_READ)
mask[idx] |= bit;