Skip to contentSkip to conversation

DVD writing in 2.6.6-rc2

BUG Kenneth Johansson 5 participants 12 messages Apr 27, 2004 – Apr 29, 2004

lore
Share & tools
lore
#0 lore raw

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

I noticed that if I start growisofs and later attach to it with "strace -p" I can make it continue with killing the strace process. just to see it hang in the next poll. But re attaching then killing the strace again a few times and growisofs finally dose a normal exit.

This happens every time.

The only unusual thing with my setup is that I only have the DVD burner on the IDE controller no disks so I guess it's some type of missed interrupt.

I looked at the drivers/ide/pci/via82cxxx.c but that one did not do much apart from setting a few configurations so I guess it's some generic code that dose the real work.

----------VIA BusMastering IDE Configuration----------------

Driver Version:                     3.38
South Bridge:                       VIA vt8233a
Revision:                           ISA 0x0 IDE 0x6
Highest DMA rate:                   UDMA133
BM-DMA base:                        0x9800
PCI clock:                          33.3MHz
Master Read  Cycle IRDY:            0ws
Master Write Cycle IRDY:            0ws
BM IDE Status Register Read Retry:  yes
Max DRDY Pulse Width:               No limit
-----------------------Primary IDE-------Secondary IDE------
Read DMA FIFO flush:          yes                 yes
End Sector FIFO flush:         no                  no
Prefetch Buffer:               no                  no
Post Write Buffer:             no                  no
Enabled:                      yes                 yes
Simplex only:                  no                  no
Cable Type:                   80w                 80w
-------------------drive0----drive1----drive2----drive3-----
Transfer Mode:        PIO       PIO      UDMA      UDMA
Address Setup:      120ns     120ns     120ns     120ns
Cmd Active:         360ns     360ns      90ns      90ns
Cmd Recovery:       210ns     210ns      30ns      30ns
Data Active:        330ns     330ns      90ns      90ns
Data Recovery:      270ns     270ns      30ns      30ns
Cycle Time:         600ns     600ns      60ns      60ns
Transfer Rate:    3.3MB/s   3.3MB/s  33.3MB/s  33.3MB/s
#1 lore raw
Quoted 6 lines · Kenneth Johansson

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

Quoted 6 lines

I noticed that if I start growisofs and later attach to it with "strace -p" I can make it continue with killing the strace process. just to see it hang in the next poll. But re attaching then killing the strace again a few times and growisofs finally dose a normal exit.

This happens every time.

I noted the same thing yesterday with cdrdao, so yours is not an isolated incident. I'll debug it tomorrow.

Signature · 1 line
Jens Axboe
#2 lore raw
Quoted 8 lines · Jens Axboe
Quoted 5 lines

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

I noted the same thing yesterday with cdrdao, so yours is not an isolated incident. I'll debug it tomorrow.

I can confirm this bug for kernel 2.6.3-rc4 as well, so it's not too recent. I found the following on my console after quitting k3b ungracefully, maybe it's useful for narrowing down the problem:

Message from syslogd@gandalf at Tue Apr 27 22:45:53 2004 ... gandalf kernel: MCE: The hardware reports a non fatal, correctable incident occu rred on CPU 0.

Message from syslogd@gandalf at Tue Apr 27 22:45:53 2004 ... gandalf kernel: Bank 2: 940040000000017a

gandalf:/root# Mutex destroy failure: Device or resource busy
gandalf:/root# ICE default IO error handler doing an exit(), pid = 17846, errno = 0
ICE default IO error handler doing an exit(), pid = 17791, errno = 2
#11 lore raw
In reply to Kenneth Johansson

I think this is a growisofs 5.19 problem.
You should download:
http://fy.chalmers.se/~appro/linux/DVD+RW/tools/dvd+rw-tools-5.19-1.4.9.7.tar.gz

Panagiotis Papadakos

Quoted 59 lines · Kenneth Johansson

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

I noticed that if I start growisofs and later attach to it with "strace -p" I can make it continue with killing the strace process. just to see it hang in the next poll. But re attaching then killing the strace again a few times and growisofs finally dose a normal exit.

This happens every time.

The only unusual thing with my setup is that I only have the DVD burner on the IDE controller no disks so I guess it's some type of missed interrupt.

I looked at the drivers/ide/pci/via82cxxx.c but that one did not do much apart from setting a few configurations so I guess it's some generic code that dose the real work.

----------VIA BusMastering IDE Configuration----------------

Driver Version:                     3.38
South Bridge:                       VIA vt8233a
Revision:                           ISA 0x0 IDE 0x6
Highest DMA rate:                   UDMA133
BM-DMA base:                        0x9800
PCI clock:                          33.3MHz
Master Read  Cycle IRDY:            0ws
Master Write Cycle IRDY:            0ws
BM IDE Status Register Read Retry:  yes
Max DRDY Pulse Width:               No limit
-----------------------Primary IDE-------Secondary IDE------
Read DMA FIFO flush:          yes                 yes
End Sector FIFO flush:         no                  no
Prefetch Buffer:               no                  no
Post Write Buffer:             no                  no
Enabled:                      yes                 yes
Simplex only:                  no                  no
Cable Type:                   80w                 80w
-------------------drive0----drive1----drive2----drive3-----
Transfer Mode:        PIO       PIO      UDMA      UDMA
Address Setup:      120ns     120ns     120ns     120ns
Cmd Active:         360ns     360ns      90ns      90ns
Cmd Recovery:       210ns     210ns      30ns      30ns
Data Active:        330ns     330ns      90ns      90ns
Data Recovery:      270ns     270ns      30ns      30ns
Cycle Time:         600ns     600ns      60ns      60ns
Transfer Rate:    3.3MB/s   3.3MB/s  33.3MB/s  33.3MB/s

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

#3 lore raw
In reply to Moritz Muehlenhoff
Quoted 7 lines · Moritz Muehlenhoff
Quoted 6 lines · Jens Axboe
Quoted 5 lines

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

Quoted 5 lines
Quoted 2 lines

I noted the same thing yesterday with cdrdao, so yours is not an isolated incident. I'll debug it tomorrow.

I can confirm this bug for kernel 2.6.3-rc4 as well, so it's not too recent. I found the following on my console after quitting k3b ungracefully, maybe

This is really strange, I haven't been able to locate a kernel problem. Looking at command traces, cdrdao issues two FLUSH_CACHE commands when ending the writes. The last commands are:

0x2a (last real write)
0x35 (first flush cache) (takes about 10 seconds to complete) 0x00 (test unit ready)
0x51 (read disc info, length 4)
0x35 (2nd flush cache) -> never completes

That last sync cache never generates an interrupt, so cdrdao gets stuck forever waiting on it. I cannot even reproduce this with a test case

Signature · 1 line
Jens Axboe
#4 lore raw
Quoted 18 lines · Jens Axboe
Quoted 7 lines · Moritz Muehlenhoff
Quoted 6 lines · Jens Axboe
Quoted 5 lines

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

Quoted 5 lines
Quoted 2 lines

I noted the same thing yesterday with cdrdao, so yours is not an isolated incident. I'll debug it tomorrow.

I can confirm this bug for kernel 2.6.3-rc4 as well, so it's not too recent. I found the following on my console after quitting k3b ungracefully, maybe

This is really strange, I haven't been able to locate a kernel problem. Looking at command traces, cdrdao issues two FLUSH_CACHE commands when ending the writes. The last commands are:

Quoted 8 lines

0x2a (last real write)
0x35 (first flush cache) (takes about 10 seconds to complete) 0x00 (test unit ready)
0x51 (read disc info, length 4)
0x35 (2nd flush cache) -> never completes

That last sync cache never generates an interrupt, so cdrdao gets stuck forever waiting on it. I cannot even reproduce this with a test case

My problem was in growisofs itself. They use poll to wait instead of using nanosleep and totally confusing me. It did not help that strace do not print out the arguments until the syscall return so I never saw the input to poll.

#10 lore raw
In reply to Jens Axboe
Quoted 1 line · Jens Axboe

This is really strange, I haven't been able to locate a kernel problem.

In the case of k3b it turned out to be a bug in growisofs from the dvd+rw-tools package; with the hotfix included in Debian/sid every- things works as expected.
It seems very strange that both triggered a similar bug like this, though.

#5 lore raw
In reply to Kenneth Johansson
Quoted 25 lines · Kenneth Johansson
Quoted 24 lines · Jens Axboe
Quoted 7 lines · Moritz Muehlenhoff
Quoted 6 lines · Jens Axboe
Quoted 5 lines

I have a problem when using growisofs version 5.19.

The problem is that in the very end when gowisofs tries to flush the cache. When stracing the process I can see it sits in a call to poll that never returns.

Quoted 5 lines
Quoted 2 lines

I noted the same thing yesterday with cdrdao, so yours is not an isolated incident. I'll debug it tomorrow.

I can confirm this bug for kernel 2.6.3-rc4 as well, so it's not too recent. I found the following on my console after quitting k3b ungracefully, maybe

This is really strange, I haven't been able to locate a kernel problem. Looking at command traces, cdrdao issues two FLUSH_CACHE commands when ending the writes. The last commands are:

0x2a (last real write)
0x35 (first flush cache) (takes about 10 seconds to complete) 0x00 (test unit ready)
0x51 (read disc info, length 4)
0x35 (2nd flush cache) -> never completes

Quoted 7 lines
Quoted 2 lines

That last sync cache never generates an interrupt, so cdrdao gets stuck forever waiting on it. I cannot even reproduce this with a test case

My problem was in growisofs itself. They use poll to wait instead of using nanosleep and totally confusing me. It did not help that strace do not print out the arguments until the syscall return so I never saw the input to poll.

So it just helped me unearth a different problem :-). There certainly is a bug here, it looks like it's hardware though (see the above description). ide-cd just needs to have it's SYNC_CACHE retries limited, then the kernel should at least recover.

Signature · 1 line
Jens Axboe
#6 lore raw
Quoted 4 lines · Jens Axboe

So it just helped me unearth a different problem :-). There certainly is a bug here, it looks like it's hardware though (see the above description). ide-cd just needs to have it's SYNC_CACHE retries limited, then the kernel should at least recover.

I see cache flushes also go mute and hang forwever. I think ideally we need to hack the IDE layer to have sane timeouts or something perhaps?

--cw

#7 lore raw
Quoted 10 lines · Chris Wedgwood
Quoted 4 lines · Jens Axboe

So it just helped me unearth a different problem :-). There certainly is a bug here, it looks like it's hardware though (see the above description). ide-cd just needs to have it's SYNC_CACHE retries limited, then the kernel should at least recover.

I see cache flushes also go mute and hang forwever. I think ideally we need to hack the IDE layer to have sane timeouts or something perhaps?

Yeah, with 'retries' I mean waiting retries. The command does expire and ide-cd notices it just doesn't put an upper bound on how long it waits. A minute or so should suffice, unless the caller specifies otherwise.

Signature · 1 line
Jens Axboe
#8 lore raw
Quoted 4 lines · Jens Axboe

Yeah, with 'retries' I mean waiting retries. The command does expire and ide-cd notices it just doesn't put an upper bound on how long it waits. A minute or so should suffice, unless the caller specifies otherwise.

Can we poke the IDE drive is it's hung or reset it in these cases? I wonder more in the general case where I've had to reboot because the burner went nuts for some reason...

--cw

#9 lore raw
Quoted 10 lines · Chris Wedgwood
Quoted 4 lines · Jens Axboe

Yeah, with 'retries' I mean waiting retries. The command does expire and ide-cd notices it just doesn't put an upper bound on how long it waits. A minute or so should suffice, unless the caller specifies otherwise.

Can we poke the IDE drive is it's hung or reset it in these cases? I wonder more in the general case where I've had to reboot because the burner went nuts for some reason...

Drive needs to be reset. Something like this should be all that's needed.

===== drivers/ide/ide-cd.c 1.81 vs edited =====

drivers/ide/ide-cd.c +0 -1
@@ -844,7 +844,6 @@
 		case GPCMD_FORMAT_UNIT:
 		case GPCMD_RESERVE_RZONE_TRACK:
 		case GPCMD_CLOSE_TRACK:
-		case GPCMD_FLUSH_CACHE:
 			wait = ATAPI_WAIT_PC;
 			break;
 		default:
Signature · 1 line
Jens Axboe