From: Bastian Hecht <bhecht@de.adit-jv.com>
Date: Wed, 28 Sep 2016 07:51:57 +0000 (+0100)
Subject: drivers: dma-coherent: Move spinlock in dma_alloc_from_coherent()
X-Git-Url: http://git.lede-project.org./?a=commitdiff_plain;h=dd01c75f1df311793de6ef217c72036552000c9a;p=openwrt%2Fstaging%2Fblogic.git

drivers: dma-coherent: Move spinlock in dma_alloc_from_coherent()

We don't need to hold the spinlock while zeroing the allocated memory.
In case we handle big buffers this is a severe issue as other CPUs might
be spinning half a second or longer.

Signed-off-by: Bastian Hecht <bhecht@de.adit-jv.com>
Signed-off-by: George G. Davis <george_davis@mentor.com>
Signed-off-by: Mark Craske <Mark_Craske@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---

diff --git a/drivers/base/dma-coherent.c b/drivers/base/dma-coherent.c
index 2789f7a95b93..640a7e63c453 100644
--- a/drivers/base/dma-coherent.c
+++ b/drivers/base/dma-coherent.c
@@ -165,6 +165,7 @@ int dma_alloc_from_coherent(struct device *dev, ssize_t size,
 	int order = get_order(size);
 	unsigned long flags;
 	int pageno;
+	int dma_memory_map;
 
 	if (!dev)
 		return 0;
@@ -187,11 +188,12 @@ int dma_alloc_from_coherent(struct device *dev, ssize_t size,
 	 */
 	*dma_handle = mem->device_base + (pageno << PAGE_SHIFT);
 	*ret = mem->virt_base + (pageno << PAGE_SHIFT);
-	if (mem->flags & DMA_MEMORY_MAP)
+	dma_memory_map = (mem->flags & DMA_MEMORY_MAP);
+	spin_unlock_irqrestore(&mem->spinlock, flags);
+	if (dma_memory_map)
 		memset(*ret, 0, size);
 	else
 		memset_io(*ret, 0, size);
-	spin_unlock_irqrestore(&mem->spinlock, flags);
 
 	return 1;