On Thu, 2009-08-13 at 23:34 +0200, Uwe Kleine-König wrote:
I think it should, since it uses kunmap_high() to undo whatever
kmap_high_get() did. Now, if there'd been a kmap_high_put()... :-)
As to the patch, its not quite right.
it has one, whereas the result of your patch seems to be that it'll
actually create one if its not found.
Something like the below ought to do I guess.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
arch/arm/include/asm/highmem.h | 1 +
arch/arm/mm/dma-mapping.c | 2 +-
mm/highmem.c | 50 ++++++++++++++++++---------------------
3 files changed, 25 insertions(+), 28 deletions(-)
diff --git a/arch/arm/include/asm/highmem.h b/arch/arm/include/asm/highmem.h
index 7f36d00..4d9573b 100644
--- a/arch/arm/include/asm/highmem.h
+++ b/arch/arm/include/asm/highmem.h
@@ -19,6 +19,7 @@ extern pte_t *pkmap_page_table;
extern void *kmap_high(struct page *page);
extern void *kmap_high_get(struct page *page);
+extern void *kmap_high_put(struct page *page);
extern void kunmap_high(struct page *page);
extern void *kmap(struct page *page);
diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 1576176..4a166d9 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -551,7 +551,7 @@ static void dma_cache_maint_contiguous(struct page *page, unsigned long offset,
if (vaddr) {
vaddr += offset;
inner_op(vaddr, vaddr + size);
- kunmap_high(page);
+ kmap_high_put(page);
}
}
diff --git a/mm/highmem.c b/mm/highmem.c
index 66e915a..b2eaefe 100644
--- a/mm/highmem.c
+++ b/mm/highmem.c
@@ -75,26 +75,6 @@ pte_t * pkmap_page_table;
static DECLARE_WAIT_QUEUE_HEAD(pkmap_wait);
-
-/*
- * Most architectures have no use for kmap_high_get(), so let's abstract
- * the disabling of IRQ out of the locking in that case to save on a
- * potential useless overhead.
- */
-#ifdef ARCH_NEEDS_KMAP_HIGH_GET
-#define lock_kmap() spin_lock_irq(&kmap_lock)
-#define unlock_kmap() spin_unlock_irq(&kmap_lock)
-#define lock_kmap_any(flags) spin_lock_irqsave(&kmap_lock, flags)
-#define unlock_kmap_any(flags) spin_unlock_irqrestore(&kmap_lock, flags)
-#else
-#define lock_kmap() spin_lock(&kmap_lock)
-#define unlock_kmap() spin_unlock(&kmap_lock)
-#define lock_kmap_any(flags) \
- do { spin_lock(&kmap_lock); (void)(flags); } while (0)
-#define unlock_kmap_any(flags) \
- do { spin_unlock(&kmap_lock); (void)(flags); } while (0)
-#endif
-
/*
* Try to free a given kmap slot.
*
@@ -361,22 +341,38 @@ EXPORT_SYMBOL(kmap_high);
*
* Returns the page's current virtual memory address, or NULL if no mapping
* exists. When and only when a non null address is returned then a
- * matching call to kunmap_high() is necessary.
+ * matching call to kmap_high_put() is necessary.
*
* This can be called from any context.
*/
void *kmap_high_get(struct page *page)
{
- unsigned long vaddr, flags;
+ unsigned long vaddr;
- lock_kmap_any(flags);
+again:
vaddr = (unsigned long)page_address(page);
if (vaddr) {
- BUG_ON(pkmap_count[PKMAP_NR(vaddr)] < 1);
- pkmap_count[PKMAP_NR(vaddr)]++;
+ atomic_t *counter = &pkmap_count[PKMAP_NR(vaddr)];
+ if (atomic_inc_not_zero(counter)) {
+ unsigned long vaddr2 = (unsigned long)page_address(page);
+
+ if (likely(vaddr == vaddr2))
+ return (void *)vaddr;
+
+ pkmap_put(counter);
+ goto again;
+ }
}
- unlock_kmap_any(flags);
- return (void*) vaddr;
+
+ return NULL;
+}
+
+void kmap_high_put(struct page *page)
+{
+ unsigned long vaddr = (unsigned long)page_address(page);
+
+ BUG_ON(!vaddr);
+ pkmap_put(&pkmap_count[PKMAP_NR(vaddr)]);
}
#endif
--
