still nfs problems [Was: Linux 2.6.37-rc8]

[James.Bottomley@HansenPartnership.com (James Bottomley)]

On Thu, 2011-01-06 at 12:25 -0600, James Bottomley wrote:
> OK, so thinking about this, it seems that the only danger is actually
> what NFS is doing: reading cache pages via a vmap.  In that case, since
> the requirement is to invalidate the vmap range to prepare for read, we
> could have invalidate_kernel_vmap_range loop over the underlying pages
> and flush them through the kernel alias if the architecture specific
> flag indicates their contents might be dirty.
> 
> The loop adds expense that is probably largely unnecessary to
> invalidate_kernel_vmap_range() but the alternative is adding to the API
> proliferation with something that only flushes the kernel pages if the
> arch specific flag says they're dirty.

This is what I think the arm patch would look like (example only: I
can't compile it).  Is something like this too expensive? the loop can't
be optimised away because of the need to check the pages (and
vmalloc_to_page is a three level page table lookup).

James

---

diff --git a/arch/arm/include/asm/cacheflush.h b/arch/arm/include/asm/cacheflush.h
index 3acd8fa..34469ca 100644
--- a/arch/arm/include/asm/cacheflush.h
+++ b/arch/arm/include/asm/cacheflush.h
@@ -414,8 +414,17 @@ static inline void flush_kernel_vmap_range(void *addr, int size)
 }
 static inline void invalidate_kernel_vmap_range(void *addr, int size)
 {
-	if ((cache_is_vivt() || cache_is_vipt_aliasing()))
-	  __cpuc_flush_dcache_area(addr, (size_t)size);
+	if ((cache_is_vivt() || cache_is_vipt_aliasing())) {
+		void *cursor = addr;
+
+		for ( ; cursor < addr + size; cursor += PAGE_SIZE) {
+			struct page *page = vmalloc_to_page(cursor);
+
+			if (!test_and_set_bit(PG_dcache_clean, &page->flags))
+				__flush_dcache_page(page_mapping(page), page);
+		}
+		__cpuc_flush_dcache_area(addr, (size_t)size);
+	}
 }
 
 #define ARCH_HAS_FLUSH_ANON_PAGE