]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
firewire: fw-ohci: enforce read order for selfID generation
authorStefan Richter <stefanr@s5r6.in-berlin.de>
Sat, 25 Aug 2007 12:08:19 +0000 (14:08 +0200)
committerStefan Richter <stefanr@s5r6.in-berlin.de>
Tue, 16 Oct 2007 22:00:05 +0000 (00:00 +0200)
It seems unlikely, but access to self_id_cpu[0] could at least in theory
be deferred until after the loop over self_id_cpu[1..n] or even after
the subsequent reg_read.  Enforce the desired order by a read barrier.

Also prevent the reg_read from being reordered relative to the for loop.
This isn't necessary if the loop's conditional printk counts as an
implicit barrier, but better make it explicit.

(self_id_cpu[] is a coherent DMA buffer.)

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
drivers/firewire/fw-ohci.c

index e14c1ca7813e2f6d186fd66ce39ddbbb42e1342e..a7947baeee482e33da240d17dc4e63f0d342a8ea 100644 (file)
@@ -30,6 +30,7 @@
 
 #include <asm/uaccess.h>
 #include <asm/semaphore.h>
+#include <asm/system.h>
 
 #include "fw-transaction.h"
 #include "fw-ohci.h"
@@ -926,12 +927,14 @@ static void bus_reset_tasklet(unsigned long data)
 
        self_id_count = (reg_read(ohci, OHCI1394_SelfIDCount) >> 3) & 0x3ff;
        generation = (le32_to_cpu(ohci->self_id_cpu[0]) >> 16) & 0xff;
+       rmb();
 
        for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
                if (ohci->self_id_cpu[i] != ~ohci->self_id_cpu[i + 1])
                        fw_error("inconsistent self IDs\n");
                ohci->self_id_buffer[j] = le32_to_cpu(ohci->self_id_cpu[i]);
        }
+       rmb();
 
        /*
         * Check the consistency of the self IDs we just read.  The