rpms/kernel/devel linux-2.6-kill-blk_attempt_merge.patch, NONE, 1.1.2.1 kernel-2.6.spec, 1.1826.2.7, 1.1826.2.8

fedora-cvs-commits at redhat.com fedora-cvs-commits at redhat.com
Mon Jan 9 16:23:54 UTC 2006 

Author: davej

Update of /cvs/dist/rpms/kernel/devel
In directory cvs.devel.redhat.com:/tmp/cvs-serv11597

Modified Files:
      Tag: private-fc5-test2-branch
	kernel-2.6.spec 
Added Files:
      Tag: private-fc5-test2-branch
	linux-2.6-kill-blk_attempt_merge.patch 
Log Message:
kill blk_attempt_merge which should make things happy again in block layer.



linux-2.6-kill-blk_attempt_merge.patch:
 block/ll_rw_blk.c      |   24 ------------------------
 drivers/ide/ide-cd.c   |   10 ----------
 include/linux/blkdev.h |    1 -
 3 files changed, 35 deletions(-)

--- NEW FILE linux-2.6-kill-blk_attempt_merge.patch ---

On Fri, Jan 06 2006, Dave Jones wrote:
> Looks like some nice slab poison...
> 
> 		Dave
> 
> Oops: 0000 [#1]
> last sysfs file:
> /devices/pci0000:00/0000:00:1d.7/usb1/1-0:1.0/bAlternateSettingModules linked
> in: usb_storage ata_piix libata e1000 ohci1394 ieee1394 uhci_hcdsCPU:    0
> EIP:    0060:[<c01c312e>]    Not tainted VLI
> EFLAGS: 00010002   (2.6.15-1.1819_FC5)
> EIP is at rb_next+0x9/0x22
> eax: 6b6b6b6b   ebx: dfed01f8   ecx: c01bf65a   edx: 6b6b6b6b
> esi: c1991e98   edi: 00000202   ebp: dfed403c   esp: c03c1f14
> ds: 007b   es: 007b   ss: 0068
> Process swapper (pid: 0, threadinfo=c03c1000 task=c0327ba0)
> Stack: c01bf662 c01b5dfa dfed01f8 c01b85dc c040c284 00000004 c1991e98 c023c8ab
>        c1991e98 00000000 c040c284 00000000 c023d3a4 000194d0 00000000 dfed403c
>        c1991e98 c040c284 c1991e98 00000000 c0231f2d 000194d0 00000000 000194d0
> Call Trace:
>  [<c01bf662>] cfq_latter_request+0x8/0x14     [<c01b5dfa>] elv_latter_request+07
> [<c01b85dc>] blk_attempt_remerge+0x1d/0x3c     [<c023c8ab>] cdrom_start_read+0e
> [<c023d3a4>] ide_do_rw_cdrom+0xdd/0x14a     [<c0231f2d>] start_request+0x1b1/01
> [<c023220f>] ide_do_request+0x2a0/0x2fb     [<c023265e>] ide_intr+0xf3/0x11b
>  [<c023c1a7>] cdrom_read_intr+0x0/0x2a1     [<c013632a>] handle_IRQ_event+0x23/c
> [<c01363cd>] __do_IRQ+0x7a/0xcd     [<c01054d0>] do_IRQ+0x5c/0x77
>  =======================
>  [<c0103cea>] common_interrupt+0x1a/0x20     [<c0101120>] mwait_idle+0x1a/0x2e
>  [<c01010a3>] cpu_idle+0x38/0x4d     [<c0390635>] start_kernel+0x17a/0x17c
> Code: 85 c0 74 0b 8b 50 0c 85 d2 74 04 89 d0 eb f5 c3 8b 00 85 c0 74 0b 8b 50 0

The blk_attempt_remerge() is a bad interface to be honest, and it's hard
to get to work reliably because it's done too late. I think the best
option is just to kill it, this will cause problems with other io
schedulers as well.

I've merged this up for 2.6.16-rc inclusion, probably should go to
stabel as well.

---

[PATCH] Kill blk_attempt_remerge()

It's a bad interface, and it's always done too late. Remove it.

Signed-off-by: Jens Axboe <axboe at suse.de>

diff --git a/block/ll_rw_blk.c b/block/ll_rw_blk.c
index 8e13645..c130519 100644
--- a/block/ll_rw_blk.c
+++ b/block/ll_rw_blk.c
@@ -2748,30 +2748,6 @@ static inline int attempt_front_merge(re
 	return 0;
 }
 
-/**
- * blk_attempt_remerge  - attempt to remerge active head with next request
- * @q:    The &request_queue_t belonging to the device
- * @rq:   The head request (usually)
- *
- * Description:
- *    For head-active devices, the queue can easily be unplugged so quickly
- *    that proper merging is not done on the front request. This may hurt
- *    performance greatly for some devices. The block layer cannot safely
- *    do merging on that first request for these queues, but the driver can
- *    call this function and make it happen any way. Only the driver knows
- *    when it is safe to do so.
- **/
-void blk_attempt_remerge(request_queue_t *q, struct request *rq)
-{
-	unsigned long flags;
-
-	spin_lock_irqsave(q->queue_lock, flags);
-	attempt_back_merge(q, rq);
-	spin_unlock_irqrestore(q->queue_lock, flags);
-}
-
-EXPORT_SYMBOL(blk_attempt_remerge);
-
 static void init_request_from_bio(struct request *req, struct bio *bio)
 {
 	req->flags |= REQ_CMD;
 
diff --git a/drivers/ide/ide-cd.c b/drivers/ide/ide-cd.c
index d31117e..e4d55ad 100644
--- a/drivers/ide/ide-cd.c
+++ b/drivers/ide/ide-cd.c
@@ -1332,8 +1332,6 @@ static ide_startstop_t cdrom_start_read 
 	if (cdrom_read_from_buffer(drive))
 		return ide_stopped;
 
-	blk_attempt_remerge(drive->queue, rq);
-
 	/* Clear the local sector buffer. */
 	info->nsectors_buffered = 0;
 
@@ -1874,14 +1872,6 @@ static ide_startstop_t cdrom_start_write
 		return ide_stopped;
 	}
 
-	/*
-	 * for dvd-ram and such media, it's a really big deal to get
-	 * big writes all the time. so scour the queue and attempt to
-	 * remerge requests, often the plugging will not have had time
-	 * to do this properly
-	 */
-	blk_attempt_remerge(drive->queue, rq);
-
 	info->nsectors_buffered = 0;
 
 	/* use dma, if possible. we don't need to check more, since we
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 804cc4e..02a585f 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -595,7 +595,6 @@ extern void generic_make_request(struct 
 extern void blk_put_request(struct request *);
 extern void __blk_put_request(request_queue_t *, struct request *);
 extern void blk_end_sync_rq(struct request *rq, int error);
-extern void blk_attempt_remerge(request_queue_t *, struct request *);
 extern struct request *blk_get_request(request_queue_t *, int, gfp_t);
 extern void blk_insert_request(request_queue_t *, struct request *, int, void *);
 extern void blk_requeue_request(request_queue_t *, struct request *);




Index: kernel-2.6.spec
===================================================================
RCS file: /cvs/dist/rpms/kernel/devel/kernel-2.6.spec,v
retrieving revision 1.1826.2.7
retrieving revision 1.1826.2.8
diff -u -r1.1826.2.7 -r1.1826.2.8
--- kernel-2.6.spec	9 Jan 2006 15:41:31 -0000	1.1826.2.7
+++ kernel-2.6.spec	9 Jan 2006 16:23:52 -0000	1.1826.2.8
@@ -369,6 +369,7 @@
 Patch1830: linux-2.6-w1-hush-debug.patch
 Patch1840: linux-2.6-x86-hp-reboot.patch
 Patch1860: linux-2.6-posix-timers-sched_time-accumulation.patch
+Patch1870: linux-2.6-kill-blk_attempt_merge.patch
 
 # Warn about usage of various obsolete functionality that may go away.
 Patch1900: linux-2.6-obsolete-idescsi-warning.patch
@@ -883,6 +884,8 @@
 %patch1840 -p1
 # Fix posix-cpu-timers sched_time accumulation
 %patch1860 -p1
+# kill blk_attempt_merge
+%patch1870 -p1
 
 # Warn about obsolete functionality usage.
 %patch1900 -p1
@@ -1414,6 +1417,7 @@
 * Mon Jan  9 2006 Dave Jones <davej at redhat.com>
 - Remove vm debug patch that triggers too easily right now.
   (Needs fixing properly post test2).
+- kill blk_attempt_merge() which was horribly broken.
 
 * Mon Jan  9 2006 David Woodhouse <dwmw2 at redhat.com>
 - Fix some usblp problems, add ieee1284_id to sysfs

More information about the fedora-cvs-commits mailing list