USB: isp1760: Flush the D-cache for the pipe-in transfer buffers
authorCatalin Marinas <catalin.marinas@arm.com>
Tue, 2 Feb 2010 15:31:02 +0000 (15:31 +0000)
committerGreg Kroah-Hartman <gregkh@suse.de>
Tue, 2 Mar 2010 22:54:53 +0000 (14:54 -0800)
When the HDC driver writes the data to the transfer buffers it pollutes
the D-cache (unlike DMA drivers where the device writes the data). If
the corresponding pages get mapped into user space, there are no
additional cache flushing operations performed and this causes random
user space faults on architectures with separate I and D caches
(Harvard) or those with aliasing D-cache.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Matthew Dharm <mdharm-kernel@one-eyed-alien.net>
Cc: Greg KH <greg@kroah.com>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Cc: Oliver Neukum <oliver@neukum.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/usb/host/isp1760-hcd.c

index 27b8f7c..9f01293 100644 (file)
@@ -17,7 +17,9 @@
 #include <linux/debugfs.h>
 #include <linux/uaccess.h>
 #include <linux/io.h>
+#include <linux/mm.h>
 #include <asm/unaligned.h>
+#include <asm/cacheflush.h>
 
 #include "../core/hcd.h"
 #include "isp1760-hcd.h"
@@ -904,6 +906,14 @@ __acquires(priv->lock)
                        status = 0;
        }
 
+       if (usb_pipein(urb->pipe) && usb_pipetype(urb->pipe) != PIPE_CONTROL) {
+               void *ptr;
+               for (ptr = urb->transfer_buffer;
+                    ptr < urb->transfer_buffer + urb->transfer_buffer_length;
+                    ptr += PAGE_SIZE)
+                       flush_dcache_page(virt_to_page(ptr));
+       }
+
        /* complete() can reenter this HCD */
        usb_hcd_unlink_urb_from_ep(priv_to_hcd(priv), urb);
        spin_unlock(&priv->lock);