We are interested in "Most recent kernel where this bug did not occur"
Please try the latest stable kernel; 2.6.16.20 is available now,
Please try the 2.6.17-rc6-mm2 with option ec_intr=0

With 2.6.16.20 no error messages
With 2.6.17-rc6-mm2 and ec_intr=0 same error messages as without ec_intr=0
More tests?

Please attach acpidump output

Created attachment 8316 [details]
acpidmp output

Here is it (above was the link)

Created attachment 8317 [details]
acpidmp output

Please try the latest stable kernel 2.6.17 - it will give us more info.

As with 2.6.16.20 no error messages with 2.6.17.

So, do you think the bug should be closed?
If NOT, please reopen.
If Yes, and you do want to know why 2.6.17 works, please take a look at 
changelog in which you could find out the individual patch.

For me it is ok to close the bug. I only thought that what is in the -mm kernels
for acpi will go into the main kernel in the next rc round, so I thought this
bug might pop in again. We will see.

Created attachment 8360 [details]
Do not abort method execution if asked to release not acquired mutex

There is no reason to abort method execution if mutex is not acquired. Printing
of  error message should be sufficient.

What is really going on here? In the DSDT, all the releases have a 
corresponding acquire with an infinite wait, so this error should not happen.

It seems that there is an error in our code, that either releases ASL mutex
twice, etc. In either case the correct behavior seems to not abort execution.
Error seems  to vanish in 2.6.17, so we should just apply this patch. 

Alexey confirms that this regression is seen on an HP NX6125,
and that it was not present in 2.6.17 (ACPICA 20060127).

ACPI Error (exmutex-0248): Cannot release Mutex [C112], not acquired [20060623]
ACPI Error (psparse-0537): Method parse/execution failed [\_TZ_.TZ4_._TMP] 
(Node ffff810017d47810), AE_AML_MUTEX_NOT_ACQUIRED

looking at the latest DSDT for the HP NX6125 in bug #5534

it contains the construct where a mutex is acquired in one
method and release in another.  while it isn't he mutex in the
error message, it may be related:

                    Method (C194, 0, NotSerialized)
                    {
                        Acquire (C191, 0xFFFF)
                        Store (0x55, C17C)
                    }

                    Method (C195, 0, NotSerialized)
                    {
                        Store (0xAA, C17C)
                        Release (C191)
                    }

Created attachment 8502 [details]
acoidump

acpidump

Created attachment 8503 [details]
warnings

warnings

This is an additional information about nx6125 system.
The errors are occured on kernel  2.6.18-rc1 and  others git*.

The  simple program tcatacpi and /var/log/warn are attached.
Run  the "tcatacpi"  in parallel  and see the results at least on nx6125 and  
F3200

Created attachment 8505 [details]
2.6.18-rc1 patch to fix acpi_os_get_thread_id()

please try this patch

ACPI: acpi_os_get_thread_id() returns current

While the Linux mutexes and the debug code that
that reference acpi_os_get_thread_id() are happy with 0,
the AML mutexes in exmutex.c expect a unique non-zero
number for each thread - as they track this thread_id
to permit the mutex re-entrancy defined by the ACPI spec.

Verified on '2.6.18-rc1 + patch from comment #17'. AML mutexes work fine, but 
other two error messages still exist - AE_ALREADE_EXISTS and AE_NOT_FOUND,. 
See warnings from comment #15.

 

Created attachment 8507 [details]
namespace fix from upcoming release

Please check if this fix from 20060707 fixes the namespace problem.

Created attachment 8510 [details]
test

Created attachment 8511 [details]
/var/log/warn

Verified on '2.6.18-rc1 + patch #17 + patch #18'.
AE_ALREADE_EXISTS and AE_NOT_FOUND still exist. But there is progress - the
frequency of error messages is very low.
I ran in a loop 'cat /proc/acpi files' concurensly, this test script and
/var/log/warn are attached.

It is possible that the AE_ALREADY_EXISTS problem is fixed in ACPICA version 
20060707, since this is a serialized method that creates namespace objects.

From the release note:
Fixed a problem with Serialized control methods where the semaphore associated 
with the method could be over-signaled after multiple method invocations.

Created attachment 8538 [details]
Patch to move deletion of methods' namespace under its mutex

Here is a patch against 2.6.18-rc1... to move deletion of methods namespace
under mutex of this method. No more errors in dmesg on nx6125

Problem still exists with the last patch.

RE: Patch in comment #23

The interpreter is locked during the execution of TerminateMethod, therefore 
another thread cannot begin execution of a method. I believe that this means 
that the signal/deletion order should not matter.

Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A4] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A5] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A6] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A7] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A8] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0A9] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (dsfield-0441): [C0AA] Namespace 
lookup failure, AE_ALREADY_EXISTS
Jul  7 21:57:14 linux kernel: ACPI Error (psargs-0355): [C0A7] Namespace 
lookup failure, AE_NOT_FOUND

If these do not go away, try the acpi_serialize option.

Any update on this problem please.
Thanks.

As I said already, I am for closing this bug, since quite some time it does not occur anymore.

Len, Alexey, would you agree we should close this bug?