Skip to contentSkip to conversation

serial port regression caused by "Char: tty_ioctl, use wait_event_interruptible_timeout" patch

DISCUSSION Rick Warner 3 participants 8 messages Feb 5, 2008

lore
Share & tools
lore
#0 lore raw

Hello all,

My company uses a proprietary hardware monitoring solution that utilizes communication over the serial port. A RS232-RS485 converter is plugged into the serial port of the master of a cluster, and cat5 cables are daisy chained between the cards to handle out of band hardware monitoring. The cards themselves speak RS485. We have been using the same software to read the
data from the serial port since the 2.4 kernel days. As of 2.6.23, the
software no longer works.

I narrowed down the problem doing a binary search on git snapshots between .22 and .23, and found the breakage between git6 and git7. Further isolating it found the patch mentioned in the subject to be the cause. I reversed the patch in the .23 source and it now works properly.

Should the code be reverted back as I did, or is there something I should change in our userspace code that reads from the serial port to correct it instead?

Thanks,
Rick

Signature · 4 lines
Richard Warner
Lead Systems Integrator
Microway, Inc
(508)732-5517
#1 lore raw
Quoted 3 lines · Rick Warner

I narrowed down the problem doing a binary search on git snapshots between .22 and .23, and found the breakage between git6 and git7. Further isolating it found the patch mentioned in the subject to be the cause. I reversed the patch in the .23 source and it now works properly.

Should the code be reverted back as I did, or is there something I should change in our userspace code that reads from the serial port to correct it instead?

Instead of reverting the patch can you try modifying this part of the patch:

+ if (wait_event_interruptible_timeout(tty->write_wait,
+ !tty->driver->chars_in_buffer(tty), timeout))
+ return;

by changing it to:

+ if (wait_event_interruptible_timeout(tty->write_wait,
+ !tty->driver->chars_in_buffer(tty), timeout) < 0)
+ return;

It looks like the patch changed the behavior of
tty_wait_until_sent by not calling the driver
specific wait_until_sent if a timeout occurs.

Signature · 2 lines
Paul Fulghum
Microgate Systems, Ltd.
#2 lore raw
Quoted 13 lines · Paul Fulghum

Instead of reverting the patch can you try modifying this part of the patch:

(quoted diff) +3 -0
+	if (wait_event_interruptible_timeout(tty->write_wait,
+			!tty->driver->chars_in_buffer(tty), timeout))
+		return;

by changing it to:

(quoted diff) +3 -0
+	if (wait_event_interruptible_timeout(tty->write_wait,
+			!tty->driver->chars_in_buffer(tty), timeout) < 0)
+		return;
Quoted 3 lines

It looks like the patch changed the behavior of
tty_wait_until_sent by not calling the driver
specific wait_until_sent if a timeout occurs.

I mispoke, the patch changed the behavior by not calling the driver specific wait_until_sent if the condition is true.

Original behavior:
call driver->wait_until_sent() on
timeout or true condition
(skip for signal)

Patch behavior:
call driver->wait_until_sent() only
on timeout (rc == 0)
(skip for signal or true)

By modifying the patch as described above, the original behavior is restored.

Signature · 2 lines
Paul Fulghum
Microgate Systems, Ltd.
#3 lore raw

This modification solved my problem. Will this change go into mainline, or will we need to maintain our own branch of the kernel to keep this working?

Thanks,
Rick

Quoted 34 lines · Paul Fulghum
Quoted 13 lines · Paul Fulghum

Instead of reverting the patch can you try modifying this part of the patch:

(quoted diff) +3 -0
+	if (wait_event_interruptible_timeout(tty->write_wait,
+			!tty->driver->chars_in_buffer(tty), timeout))
+		return;

by changing it to:

(quoted diff) +3 -0
+	if (wait_event_interruptible_timeout(tty->write_wait,
+			!tty->driver->chars_in_buffer(tty), timeout) < 0)
+		return;
Quoted 3 lines

It looks like the patch changed the behavior of
tty_wait_until_sent by not calling the driver
specific wait_until_sent if a timeout occurs.

I mispoke, the patch changed the behavior by not calling the driver specific wait_until_sent if the condition is true.

Original behavior:
call driver->wait_until_sent() on
timeout or true condition
(skip for signal)

Patch behavior:
call driver->wait_until_sent() only
on timeout (rc == 0)
(skip for signal or true)

By modifying the patch as described above, the original behavior is restored.

Signature · 4 lines
Richard Warner
Lead Systems Integrator
Microway, Inc
(508)732-5517
#4 lore raw
Quoted 1 line · Rick Warner

This modification solved my problem. Will this change go into mainline, or will we need to maintain our own branch of the kernel to keep this working?

It should be accepted into mainline as it restores the original behavior. I'll put together a patch submission tomorrow unless Jiri beats me to it.

Thanks,
Paul

Signature · 2 lines
Paul Fulghum
Microgate Systems, Ltd.
#5 lore raw
Quoted 6 lines · Paul Fulghum
Quoted 1 line · Rick Warner

This modification solved my problem. Will this change go into mainline, or will we need to maintain our own branch of the kernel to keep this working?

It should be accepted into mainline as it restores the original behavior. I'll put together a patch submission tomorrow unless Jiri beats me to it.

It should be fixed already as of
git-describe db99247a
v2.6.24-rc6-95-gdb99247

So since 2.6.24-rc7.

Maybe we should consider the fix for 2.6.23 too.

#6 lore raw
Quoted 7 lines · Jiri Slaby

It should be fixed already as of
git-describe db99247a
v2.6.24-rc6-95-gdb99247

So since 2.6.24-rc7.

Maybe we should consider the fix for 2.6.23 too.

Whoops, I missed that.
Problem solved :-)

Thanks,
Paul

#7 lore raw
Quoted 11 lines · Paul Fulghum
Quoted 7 lines · Jiri Slaby

It should be fixed already as of
git-describe db99247a
v2.6.24-rc6-95-gdb99247

So since 2.6.24-rc7.

Maybe we should consider the fix for 2.6.23 too.

Whoops, I missed that.
Problem solved :-)

:) Thank you both for heads up.