Fix soft lockup with iSeries viocd driver
From: Tony Breeds <[email protected]>
Fix soft lockup with iSeries viocd driver, caused by eventually calling end_that_request_first() with nr_bytes 0.
The lockup is triggered by hald, interrogating the device.
---
Implementation based on drivers/ide/ide-io.c:ide_end_request()
viocd.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-)
drivers/cdrom/viocd.c +20 -4
@@ -376,6 +376,22 @@ static int send_request(struct request *return 0 ; } +static void viocd_end_request (struct request *req,int uptodate) +{ +int nsectors = req->hard_cur_sectors; + +/* Make sure it's fully ended */ +if (blk_pc_request (req)) + nsectors = (req->data_len +511 ) >>9 ; +if (!nsectors) + nsectors =1 ; + +if (!end_that_request_first (req, uptodate, nsectors)) { +add_disk_randomness (req->rq_disk); +blkdev_dequeue_request (req); +end_that_request_last (req, uptodate); + } +}static int rwreq;
@@ -385,11 +401,11 @@ static void do_viocd_request(request_quewhile ((rwreq ==0 ) && ((req =elv_next_request (q)) !=NULL )) {if (!blk_fs_request (req)) -end_request (req,0 ); +viocd_end_request (req,0 );else if (send_request (req) <0 ) {printk (VIOCD_KERN_WARNING "unable to send message to OS/400!" ); -end_request (req,0 ); +viocd_end_request (req,0 ); }else rwreq++; }
@@ -601,9 +617,9 @@ return_complete:"with rc %d:0x%04X: %s\n" , req, event->xRc, bevent->sub_result, err->msg); -end_request (req,0 ); +viocd_end_request (req,0 ); }else -end_request (req,1 ); +viocd_end_request (req,1 );/* restart handling of incoming requests */ spin_unlock_irqrestore (&viocd_reqlock, flags);
Yours Tony
linux.conf.au http://linux.conf.au/ || http://lca2008.linux.org.au/
Jan 28 - Feb 02 2008 The Australian Linux Technical Conference!
Quoted 29 lines · Tony Breeds
From: Tony Breeds <[email protected]>
Fix soft lockup with iSeries viocd driver, caused by eventually calling
end_that_request_first()withnr_bytes0.The lockup is triggered by hald, interrogating the device.
Signed-off-by Tony Breeds <[email protected]>---
Implementation based on drivers/ide/ide-io.c:ide_end_request()viocd.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-)Index:
linux-2.6.20-rc5/drivers/cdrom/viocd.c
===================================================================drivers/cdrom/viocd.c +6
-0@@ -376,6 +376,22 @@ static int send_request(struct request *return 0 ; } +static void viocd_end_request (struct request *req,int uptodate) +{ +int nsectors = req->hard_cur_sectors; + +/* Make sure it's fully ended */ +if (blk_pc_request (req))
drivers/cdrom/viocd.c static int send_request(struct request * · line 394
Quoted 10 lines
drivers/cdrom/viocd.c +10
-0+ nsectors = (req->data_len +511 ) >>9 ; +if (!nsectors) + nsectors =1 ; + +if (!end_that_request_first (req, uptodate, nsectors)) { +add_disk_randomness (req->rq_disk); +blkdev_dequeue_request (req); +end_that_request_last (req, uptodate); + } +}
It looks ok (better than end_request()), though it would still be nice to properly end requests. See the recent thread on the lguest block driver.
Signature · 1 line
Jens Axboe
Quoted 3 lines · Jens Axboe
It looks ok (better than
end_request()), though it would still be nice to properly end requests. See the recent thread on the lguest block driver.
Hi Jens,
I read through the lguest thread and came up with the patch
below, I hope I understood correctly. We still need to do the check to make sure we ask for at least one sector to be processed in end_that_request_first(), as the request that triggers the lockup here (an sg_io ioctl in this case), has both req->hard_nr_sectors and req->hard_cur_sectors = 0.
From: Tony Breeds <[email protected]>
Fix soft lockup with iSeries viocd driver, caused by eventually calling end_that_request_first() with nr_bytes 0.
The lockup is triggered by hald, interrogating the device.
--- viocd.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-)
drivers/cdrom/viocd.c +23 -4
@@ -376,6 +376,25 @@ static int send_request(struct request *return 0 ; } +static void viocd_end_request (struct request *req,int uptodate) +{ +int nsectors = req->hard_nr_sectors; + +/* +* Make sure it's fully ended, and ensure that we process +* at least one sector. +*/ +if (blk_pc_request (req)) + nsectors = (req->data_len +511 ) >>9 ; +if (!nsectors) + nsectors =1 ; + +if (end_that_request_first (req, uptodate, nsectors)) +BUG (); +add_disk_randomness (req->rq_disk); +blkdev_dequeue_request (req); +end_that_request_last (req, uptodate); +}static int rwreq;
@@ -385,11 +404,11 @@ static void do_viocd_request(request_quewhile ((rwreq ==0 ) && ((req =elv_next_request (q)) !=NULL )) {if (!blk_fs_request (req)) -end_request (req,0 ); +viocd_end_request (req,0 );else if (send_request (req) <0 ) {printk (VIOCD_KERN_WARNING "unable to send message to OS/400!" ); -end_request (req,0 ); +viocd_end_request (req,0 ); }else rwreq++; }
@@ -601,9 +620,9 @@ return_complete:"with rc %d:0x%04X: %s\n" , req, event->xRc, bevent->sub_result, err->msg); -end_request (req,0 ); +viocd_end_request (req,0 ); }else -end_request (req,1 ); +viocd_end_request (req,1 );/* restart handling of incoming requests */ spin_unlock_irqrestore (&viocd_reqlock, flags);
Yours Tony
linux.conf.au http://linux.conf.au/ || http://lca2008.linux.org.au/
Jan 28 - Feb 02 2008 The Australian Linux Technical Conference!
Quoted 6 lines · Tony Breeds
Quoted 3 lines · Jens Axboe
It looks ok (better than
end_request()), though it would still be nice to properly end requests. See the recent thread on the lguest block driver.
Quoted 7 lines
Hi Jens,
I read through the lguest thread and came up with the patch
below, I hope I understood correctly. We still need to do the check to make sure we ask for at least one sector to be processed inend_that_request_first(), as the request that triggers the lockup here (ansg_ioioctl in this case), has bothreq->hard_nr_sectorsandreq->hard_cur_sectors= 0.
Yep that looks better.
Quoted 8 lines
From: Tony Breeds <[email protected]>
Fix soft lockup with iSeries viocd driver, caused by eventually calling
end_that_request_first()withnr_bytes0.The lockup is triggered by hald, interrogating the device.
Signed-off-by Tony Breeds <[email protected]>
Signature · 1 line
Jens Axboe