[PATCH] mtd: fix cfi_cmdset_0001 FL_SYNCING race (take 2)

[Alexander Belyakov]

The patch fixes CFI issue with multipartitional devices leading to the set of errors or even deadlock. The problem is CFI FL_SYNCING state race with flash operations (e.g. erase suspend). It is reproduced by running intensive writes on one JFFS2 partition and simultaneously performing mount/unmount cycle on another partition of the same chip.


---
Signed-off-by: Alexander Belyakov <abelyako at googlemail.com>

diff -uNrp a/drivers/mtd/chips/cfi_cmdset_0001.c b/drivers/mtd/chips/cfi_cmdset_0001.c
--- a/drivers/mtd/chips/cfi_cmdset_0001.c	2008-09-08 21:40:20.000000000 +0400
+++ b/drivers/mtd/chips/cfi_cmdset_0001.c	2008-09-19 10:47:34.000000000 +0400
@@ -701,6 +701,10 @@ static int chip_ready (struct map_info *
 	struct cfi_pri_intelext *cfip = cfi->cmdset_priv;
 	unsigned long timeo = jiffies + HZ;
 
+	/* Prevent setting state FL_SYNCING for chip in suspended state. */
+	if (mode == FL_SYNCING && chip->oldstate != FL_READY)
+		goto sleep;
+
 	switch (chip->state) {
 
 	case FL_STATUS:
@@ -806,8 +810,9 @@ static int get_chip(struct map_info *map
 	DECLARE_WAITQUEUE(wait, current);
 
  retry:
-	if (chip->priv && (mode == FL_WRITING || mode == FL_ERASING
-			   || mode == FL_OTP_WRITE || mode == FL_SHUTDOWN)) {
+	if (chip->priv &&
+	    (mode == FL_WRITING || mode == FL_ERASING || mode == FL_OTP_WRITE
+	    || mode == FL_SHUTDOWN) && chip->state != FL_SYNCING) {
 		/*
 		 * OK. We have possibility for contention on the write/erase
 		 * operations which are global to the real chip and not per
@@ -857,6 +862,14 @@ static int get_chip(struct map_info *map
 				return ret;
 			}
 			spin_lock(&shared->lock);
+
+			/* We should not own chip if it is already
+			 * in FL_SYNCING state. Put contender and retry. */
+			if (chip->state == FL_SYNCING) {
+				put_chip(map, contender, contender->start);
+				spin_unlock(contender->mutex);
+				goto retry;
+			}
 			spin_unlock(contender->mutex);
 		}