[PATCH] sbpcd oops on merged requests

From: Paul Gortmaker (p_gortmaker@yahoo.com)
Date: Wed Feb 14 2001 - 04:59:11 EST


Two fixes here - 1st is to make the new max_drives its own separate
module parameter since the driver uses the existing sbpcd module
parameter internally for more than just module parameters <sigh>.

2nd is more important - if you read more than 4 blocks from the
device, you will get a random crash - meaning you could mount a
cd and do "ls" and think it worked if that is all you did. Anything
more gets a rolling oops, EIP pointing off into outer space etc.

Tracked this down to the addition of the generic merge request
handling added to ll_rw_block in 2.3.41 (apparently nobody has
used this driver since then!!!) - it appears these conflict with
the molestation of the request list that sbpcd does from within.

Rather than attack sbpcd with a chainsaw, I went for the minimal
patch approach and simply disabled request merging. Lets face it,
performance is not an issue if you are using a 2x speed CD with
seek times you measure with a wall calendar.

Patch is against 2.4.2-pre3.

Paul.

--- drivers/cdrom/sbpcd.c~ Wed Feb 14 04:02:04 2001
+++ drivers/cdrom/sbpcd.c Wed Feb 14 04:09:55 2001
@@ -323,7 +323,7 @@
  * Andrew J. Kroll <ag784@freenet.buffalo.edu> Wed Jul 26 04:24:10 EDT 2000
  *
  * 4.64 Fix module parameters - were being completely ignored.
- * Can also specify max_drives as 3rd setup int to get rid of
+ * Can also specify max_drives=N as a setup int to get rid of
  * "ghost" drives on crap hardware (aren't they all?) Paul Gortmaker
  *
  * TODO
@@ -339,6 +339,15 @@
  *
  */
 
+/*
+ * Trying to merge requests breaks this driver horribly (as in it goes
+ * boom and apparently has done so since 2.3.41). As it is a legacy
+ * driver for a horribly slow double speed CD on a hideous interface
+ * designed for polled operation, I won't loose any sleep in simply
+ * disallowing merging. Paul G. 02/2001
+ */
+#define DONT_MERGE_REQUESTS
+
 #ifndef SBPCD_ISSUE
 #define SBPCD_ISSUE 1
 #endif SBPCD_ISSUE
@@ -476,7 +485,8 @@
 #else
 static int sbpcd[] = {CDROM_PORT, SBPRO}; /* probe with user's setup only */
 #endif
-MODULE_PARM(sbpcd, "3i");
+MODULE_PARM(sbpcd, "2i");
+MODULE_PARM(max_drives, "i");
 
 #define NUM_PROBE (sizeof(sbpcd) / sizeof(int))
 
@@ -5566,7 +5576,6 @@
 #else
         sbpcd_ioaddr = sbpcd[0];
         sbpro_type = sbpcd[1];
- max_drives = sbpcd[2];
 #endif
         
         CDo_command=sbpcd_ioaddr;
@@ -5661,6 +5670,21 @@
         msg(DBG_SEQ,"found SoundScape interface at %04X.\n", sbpcd_ioaddr);
         return (0);
 }
+
+#ifdef DONT_MERGE_REQUESTS
+static int dont_merge_requests_fn(request_queue_t *q, struct request *req,
+ struct request *next, int max_segments)
+{
+ return 0;
+}
+
+static int dont_bh_merge_fn(request_queue_t *q, struct request *req,
+ struct buffer_head *bh, int max_segments)
+{
+ return 0;
+}
+#endif
+
 /*==========================================================================*/
 /*
  * Test for presence of drive and initialize it.
@@ -5828,6 +5852,11 @@
 #endif MODULE
         }
         blk_init_queue(BLK_DEFAULT_QUEUE(MAJOR_NR), DEVICE_REQUEST);
+#ifdef DONT_MERGE_REQUESTS
+ (BLK_DEFAULT_QUEUE(MAJOR_NR))->back_merge_fn = dont_bh_merge_fn;
+ (BLK_DEFAULT_QUEUE(MAJOR_NR))->front_merge_fn = dont_bh_merge_fn;
+ (BLK_DEFAULT_QUEUE(MAJOR_NR))->merge_requests_fn = dont_merge_requests_fn;
+#endif
         blk_queue_headactive(BLK_DEFAULT_QUEUE(MAJOR_NR), 0);
         read_ahead[MAJOR_NR] = buffers * (CD_FRAMESIZE / 512);
         

_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Thu Feb 15 2001 - 21:00:23 EST