dmsetup fails on latest kernel

dmsetup fails on latest kernel

[Kasatkin, Dmitry]

Hi,

(subject added)

I rebased on the stop of latest mainline tree and now dmsetup fails...
It says: "Failed to create notification semaphore: Function not implemented".
I run strace and saw that dmsetup uses "semget" which returns -ENOSYS...

It works on
122804ecb59493fbb4d31b3ba9ac59faaf45276f

but does not work on
ccb19d263fd1c9e34948e2158c53eacbff369344

Any ideas?

- Dmitry

Re: dmsetup fails on latest kernel

[Kasatkin, Dmitry]

On Thu, Jan 19, 2012 at 3:02 PM, Kasatkin, Dmitry
<dmitry.kasatkin@intel.com> wrote:
> Hi,
>
> (subject added)
>
> I rebased on the stop of latest mainline tree and now dmsetup fails...
> It says: "Failed to create notification semaphore: Function not implemented".
> I run strace and saw that dmsetup uses "semget" which returns -ENOSYS...
>
> It works on
> 122804ecb59493fbb4d31b3ba9ac59faaf45276f
>
> but does not work on
> ccb19d263fd1c9e34948e2158c53eacbff369344
>
> Any ideas?
>
> - Dmitry

Hi,

I have bisected the problem...

It is introduced by this commit:

303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
commit 303395ac3bf3e2cb488435537d416bc840438fcb
Author: H. Peter Anvin <hpa@linux.intel.com>
Date:   Fri Nov 11 16:07:41 2011 -0800

    x86: Generate system call tables and unistd_*.h from tables


It came from:  Merge branch 'x86-syscall-for-linus' of
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Was on 17.01.2012...

I have 32 bit VM and I have SYSVIPC enabled in the configuration...

- Dmitry

Re: dmsetup fails on latest kernel

[Kasatkin, Dmitry]

On Thu, Jan 19, 2012 at 5:56 PM, Kasatkin, Dmitry
<dmitry.kasatkin@intel.com> wrote:
> On Thu, Jan 19, 2012 at 3:02 PM, Kasatkin, Dmitry
> <dmitry.kasatkin@intel.com> wrote:
>> Hi,
>>
>> (subject added)
>>
>> I rebased on the stop of latest mainline tree and now dmsetup fails...
>> It says: "Failed to create notification semaphore: Function not implemented".
>> I run strace and saw that dmsetup uses "semget" which returns -ENOSYS...
>>
>> It works on
>> 122804ecb59493fbb4d31b3ba9ac59faaf45276f
>>
>> but does not work on
>> ccb19d263fd1c9e34948e2158c53eacbff369344
>>
>> Any ideas?
>>
>> - Dmitry
>
> Hi,
>
> I have bisected the problem...
>
> It is introduced by this commit:
>
> 303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
> commit 303395ac3bf3e2cb488435537d416bc840438fcb
> Author: H. Peter Anvin <hpa@linux.intel.com>
> Date:   Fri Nov 11 16:07:41 2011 -0800
>
>    x86: Generate system call tables and unistd_*.h from tables
>
>
> It came from:  Merge branch 'x86-syscall-for-linus' of
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
> Was on 17.01.2012...
>
> I have 32 bit VM and I have SYSVIPC enabled in the configuration...
>
> - Dmitry

It seems "semget" is missing from: arch/x86/syscalls/syscall_32.tbl
But it presents in: arch/x86/syscalls/syscall_64.tbl

Is it a bug then?

- Dmitry

Re: [dm-devel] dmsetup fails on latest kernel

[Bryn M. Reeves]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 01/19/2012 04:10 PM, Kasatkin, Dmitry wrote:
> It seems "semget" is missing from:
> arch/x86/syscalls/syscall_32.tbl But it presents in:
> arch/x86/syscalls/syscall_64.tbl
> 
> Is it a bug then?

Maybe; I noticed that too but it's also missing in the old x86
unistd_32.h.

I thought Linux only provided the ipc(2) call - glibc sources seem to
confirm that (sysdeps/unix/sysv/linux/semget.c).

sys_sem* and friends are present in include/asm-generic/unistd.h so
maybe these values just need to be propagated to syscall_32.tbl?

Actually there's a bunch of SysV IPC calls defined in syscall_64.tbl
that are not present in syscall_32.tbl:

$ egrep 'shm|sem|msg[a-z]' arch/x86/syscalls/syscall_32.tbl
$ egrep 'shm|sem|msg[a-z]' arch/x86/syscalls/syscall_64.tbl
29	64	shmget			sys_shmget
30	64	shmat			sys_shmat
31	64	shmctl			sys_shmctl
64	64	semget			sys_semget
65	64	semop			sys_semop
66	64	semctl			sys_semctl
67	64	shmdt			sys_shmdt
68	64	msgget			sys_msgget
69	64	msgsnd			sys_msgsnd
70	64	msgrcv			sys_msgrcv
71	64	msgctl			sys_msgctl
220	64	semtimedop		sys_semtimedop

But sys_ipc is there:

$ grep sys_ipc arch/x86/syscalls/syscall_32.tbl
117	i386	ipc			sys_ipc				sys32_ipc

Regards,
Bryn.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk8YW7wACgkQ6YSQoMYUY97S0gCfTZOFMENOiVrm5jIprhcYkmDJ
Pu4AnRUkUVwyZkgUG9DAcAfdmwR7kX2N
=Bnwq
-----END PGP SIGNATURE-----

Re: [dm-devel] dmsetup fails on latest kernel

[H. Peter Anvin]

On 01/19/2012 10:06 AM, Bryn M. Reeves wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 01/19/2012 04:10 PM, Kasatkin, Dmitry wrote:
>> It seems "semget" is missing from:
>> arch/x86/syscalls/syscall_32.tbl But it presents in:
>> arch/x86/syscalls/syscall_64.tbl
>>
>> Is it a bug then?
>
> Maybe; I noticed that too but it's also missing in the old x86
> unistd_32.h.
>
> I thought Linux only provided the ipc(2) call - glibc sources seem to
> confirm that (sysdeps/unix/sysv/linux/semget.c).
>
> sys_sem* and friends are present in include/asm-generic/unistd.h so
> maybe these values just need to be propagated to syscall_32.tbl?
>
> Actually there's a bunch of SysV IPC calls defined in syscall_64.tbl
> that are not present in syscall_32.tbl:
>
> $ egrep 'shm|sem|msg[a-z]' arch/x86/syscalls/syscall_32.tbl
> $ egrep 'shm|sem|msg[a-z]' arch/x86/syscalls/syscall_64.tbl
> 29	64	shmget			sys_shmget
> 30	64	shmat			sys_shmat
> 31	64	shmctl			sys_shmctl
> 64	64	semget			sys_semget
> 65	64	semop			sys_semop
> 66	64	semctl			sys_semctl
> 67	64	shmdt			sys_shmdt
> 68	64	msgget			sys_msgget
> 69	64	msgsnd			sys_msgsnd
> 70	64	msgrcv			sys_msgrcv
> 71	64	msgctl			sys_msgctl
> 220	64	semtimedop		sys_semtimedop
>
> But sys_ipc is there:
>
> $ grep sys_ipc arch/x86/syscalls/syscall_32.tbl
> 117	i386	ipc			sys_ipc				sys32_ipc
>

No, the i386 ABI uses several system call multiplexes (sys_ipc, 
sys_socketcall) which are deprecated for new ABIs.  x86-64 and other new 
ABIs use plain system calls.  Thus, this is all as one should expect; 
the right thing to do is to compare against the old 32-bit system call 
table.

	-hpa

Re: dmsetup fails on latest kernel

[Linus Torvalds]

On Thu, Jan 19, 2012 at 7:56 AM, Kasatkin, Dmitry
<dmitry.kasatkin@intel.com> wrote:
> On Thu, Jan 19, 2012 at 3:02 PM, Kasatkin, Dmitry
> <dmitry.kasatkin@intel.com> wrote:
>> Hi,
>>
>> (subject added)
>>
>> I rebased on the stop of latest mainline tree and now dmsetup fails...
>> It says: "Failed to create notification semaphore: Function not implemented".
>> I run strace and saw that dmsetup uses "semget" which returns -ENOSYS...
>>
>> It works on
>> 122804ecb59493fbb4d31b3ba9ac59faaf45276f
>>
>> but does not work on
>> ccb19d263fd1c9e34948e2158c53eacbff369344
>>
>> Any ideas?
>>
>> - Dmitry
>
> Hi,
>
> I have bisected the problem...
>
> It is introduced by this commit:
>
> 303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
> commit 303395ac3bf3e2cb488435537d416bc840438fcb
> Author: H. Peter Anvin <hpa@linux.intel.com>
> Date:   Fri Nov 11 16:07:41 2011 -0800
>
>    x86: Generate system call tables and unistd_*.h from tables

Interesting. I just munged the old pre-merge

  arch/x86/kernel/syscall_table_32.S

and the new

  arch/x86/syscalls/syscall_32.tbl

with some trivial shell scripting, and they are definitely identical.

So the tables look like they match 1:1.

The compat tables match too, except the new list contains the entry
for bdflush (134), but that's a legacy thing that doesn't matter.

I wonder what else could be going on. Wrong number of system calls
check? Peter, any ideas?

                        Linus

Re: dmsetup fails on latest kernel

[H. Peter Anvin]

On 01/19/2012 11:05 AM, Linus Torvalds wrote:
>>
>> Hi,
>>
>> I have bisected the problem...
>>
>> It is introduced by this commit:
>>
>> 303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
>> commit 303395ac3bf3e2cb488435537d416bc840438fcb
>> Author: H. Peter Anvin<hpa@linux.intel.com>
>> Date:   Fri Nov 11 16:07:41 2011 -0800
>>
>>     x86: Generate system call tables and unistd_*.h from tables
>
> Interesting. I just munged the old pre-merge
>
>    arch/x86/kernel/syscall_table_32.S
>
> and the new
>
>    arch/x86/syscalls/syscall_32.tbl
>
> with some trivial shell scripting, and they are definitely identical.
>
> So the tables look like they match 1:1.
>
> The compat tables match too, except the new list contains the entry
> for bdflush (134), but that's a legacy thing that doesn't matter.
>
> I wonder what else could be going on. Wrong number of system calls
> check? Peter, any ideas?
>

Not off the top of my head.  I will try to reproduce this and see what 
the problem is.  Dmitry, what version of binutils and gcc are you using? 
  It could be that there is a toolchain problem.

	-hpa

Re: dmsetup fails on latest kernel

[Kasatkin, Dmitry]

On Thu, Jan 19, 2012 at 9:58 PM, H. Peter Anvin <hpa@linux.intel.com> wrote:
> On 01/19/2012 11:05 AM, Linus Torvalds wrote:
>>>
>>>
>>> Hi,
>>>
>>> I have bisected the problem...
>>>
>>> It is introduced by this commit:
>>>
>>> 303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
>>> commit 303395ac3bf3e2cb488435537d416bc840438fcb
>>> Author: H. Peter Anvin<hpa@linux.intel.com>
>>> Date:   Fri Nov 11 16:07:41 2011 -0800
>>>
>>>    x86: Generate system call tables and unistd_*.h from tables
>>
>>
>> Interesting. I just munged the old pre-merge
>>
>>   arch/x86/kernel/syscall_table_32.S
>>
>> and the new
>>
>>   arch/x86/syscalls/syscall_32.tbl
>>
>> with some trivial shell scripting, and they are definitely identical.
>>
>> So the tables look like they match 1:1.
>>
>> The compat tables match too, except the new list contains the entry
>> for bdflush (134), but that's a legacy thing that doesn't matter.
>>
>> I wonder what else could be going on. Wrong number of system calls
>> check? Peter, any ideas?
>>
>
> Not off the top of my head.  I will try to reproduce this and see what the
> problem is.  Dmitry, what version of binutils and gcc are you using?  It
> could be that there is a toolchain problem.
>

gcc (Ubuntu/Linaro 4.6.1-9ubuntu3) 4.6.1
binutils 2.21.53.20110810-0ubuntu5.1


>        -hpa
>
>

Re: dmsetup fails on latest kernel

[H. Peter Anvin]

Could you please send me:

a) your full strace;
b) arch/x86/kernel/syscall_32.o from your build?

	-hpa

On 01/19/2012 07:56 AM, Kasatkin, Dmitry wrote:
> On Thu, Jan 19, 2012 at 3:02 PM, Kasatkin, Dmitry
> <dmitry.kasatkin@intel.com> wrote:
>> Hi,
>>
>> (subject added)
>>
>> I rebased on the stop of latest mainline tree and now dmsetup fails...
>> It says: "Failed to create notification semaphore: Function not implemented".
>> I run strace and saw that dmsetup uses "semget" which returns -ENOSYS...
>>
>> It works on
>> 122804ecb59493fbb4d31b3ba9ac59faaf45276f
>>
>> but does not work on
>> ccb19d263fd1c9e34948e2158c53eacbff369344
>>
>> Any ideas?
>>
>> - Dmitry
> 
> Hi,
> 
> I have bisected the problem...
> 
> It is introduced by this commit:
> 
> 303395ac3bf3e2cb488435537d416bc840438fcb is the first bad commit
> commit 303395ac3bf3e2cb488435537d416bc840438fcb
> Author: H. Peter Anvin <hpa@linux.intel.com>
> Date:   Fri Nov 11 16:07:41 2011 -0800
> 
>     x86: Generate system call tables and unistd_*.h from tables
> 
> 
> It came from:  Merge branch 'x86-syscall-for-linus' of
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
> Was on 17.01.2012...
> 
> I have 32 bit VM and I have SYSVIPC enabled in the configuration...
> 
> - Dmitry


-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: dmsetup fails on latest kernel

[Linus Torvalds]

On Thu, Jan 19, 2012 at 12:13 PM, H. Peter Anvin <hpa@zytor.com> wrote:
> Could you please send me:
>
> a) your full strace;
> b) arch/x86/kernel/syscall_32.o from your build?

Btw, the strace before-and-after (working vs nonworking) would
probably be useful. They aren't easily comparable (all the random
pointers change depending on allocations etc), but some scripting and
editing out pointer values might make the difference be obvious.

                         Linus

Re: dmsetup fails on latest kernel

[H. Peter Anvin]

On 01/19/2012 12:17 PM, Linus Torvalds wrote:
> On Thu, Jan 19, 2012 at 12:13 PM, H. Peter Anvin <hpa@zytor.com> wrote:
>> Could you please send me:
>>
>> a) your full strace;
>> b) arch/x86/kernel/syscall_32.o from your build?
> 
> Btw, the strace before-and-after (working vs nonworking) would
> probably be useful. They aren't easily comparable (all the random
> pointers change depending on allocations etc), but some scripting and
> editing out pointer values might make the difference be obvious.
> 
>                          Linus

Possible.  I have actually reproduced it and reduced it to a small test
case already, so I hopefully will have it nailed down soon.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: dmsetup fails on latest kernel

[H. Peter Anvin]

[-- Attachment #1: Type: text/plain, Size: 253 bytes --]

Here is a patch which should fix this problem.  I will run a few more
tests and then push it with the rest of the urgent queue to Linus.

	-hpa


-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.


[-- Attachment #2: 0001-x86-syscall-Need-__ARCH_WANT_SYS_IPC-for-32-bits.patch --]
[-- Type: text/x-patch, Size: 1230 bytes --]

>From 497e064320506cd15f7bdd659204b576c80bf24e Mon Sep 17 00:00:00 2001
From: "H. Peter Anvin" <hpa@zytor.com>
Date: Thu, 19 Jan 2012 12:41:25 -0800
Subject: [PATCH] x86, syscall: Need __ARCH_WANT_SYS_IPC for 32 bits

In checkin

  303395ac3bf3 x86: Generate system call tables and unistd_*.h from tables

the feature macros in <asm/unistd.h> were unified between 32 and 64
bits.  Unfortunately 32 bits requires __ARCH_WANT_SYS_IPC and this was
inadvertently dropped.

Reported-by: Dmitry Kasatkin <dmitry.kasatkin@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Link: http://lkml.kernel.org/r/CALLzPKbeXN5gdngo8uYYU8mAow=XhrwBFBhKfG811f37BubQOg@mail.gmail.com
---
 arch/x86/include/asm/unistd.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/unistd.h b/arch/x86/include/asm/unistd.h
index b4a3db7..21f77b8 100644
--- a/arch/x86/include/asm/unistd.h
+++ b/arch/x86/include/asm/unistd.h
@@ -7,6 +7,7 @@
 #  include <asm/unistd_32.h>
 #  define __ARCH_WANT_IPC_PARSE_VERSION
 #  define __ARCH_WANT_STAT64
+#  define __ARCH_WANT_SYS_IPC
 #  define __ARCH_WANT_SYS_OLD_MMAP
 #  define __ARCH_WANT_SYS_OLD_SELECT
 
-- 
1.7.6.5

Re: dmsetup fails on latest kernel

[Kasatkin, Dmitry]

On Thu, Jan 19, 2012 at 11:16 PM, H. Peter Anvin <hpa@zytor.com> wrote:
> Here is a patch which should fix this problem.  I will run a few more
> tests and then push it with the rest of the urgent queue to Linus.
>
>        -hpa

Hi,

Yes. It solves the problem..
Do you still want straces?

- Dmitry

>
>
> --
> H. Peter Anvin, Intel Open Source Technology Center
> I work for Intel.  I don't speak on their behalf.
>

Re: dmsetup fails on latest kernel

[H. Peter Anvin]

On 01/20/2012 05:35 AM, Kasatkin, Dmitry wrote:
> On Thu, Jan 19, 2012 at 11:16 PM, H. Peter Anvin <hpa@zytor.com> wrote:
>> Here is a patch which should fix this problem.  I will run a few more
>> tests and then push it with the rest of the urgent queue to Linus.
>>
>>        -hpa
> 
> Hi,
> 
> Yes. It solves the problem..
> Do you still want straces?
> 

No need, thanks!

	-hpa


-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

[tip:x86/urgent] x86, syscall: Need __ARCH_WANT_SYS_IPC for 32 bits

[tip-bot for H. Peter Anvin]

Commit-ID:  4f2f81a5621de47d42476d0b929be2e0d565df84
Gitweb:     http://git.kernel.org/tip/4f2f81a5621de47d42476d0b929be2e0d565df84
Author:     H. Peter Anvin <hpa@zytor.com>
AuthorDate: Thu, 19 Jan 2012 12:41:25 -0800
Committer:  H. Peter Anvin <hpa@linux.intel.com>
CommitDate: Thu, 19 Jan 2012 12:57:09 -0800

x86, syscall: Need __ARCH_WANT_SYS_IPC for 32 bits

In checkin

  303395ac3bf3 x86: Generate system call tables and unistd_*.h from tables

the feature macros in <asm/unistd.h> were unified between 32 and 64
bits.  Unfortunately 32 bits requires __ARCH_WANT_SYS_IPC and this was
inadvertently dropped.

Reported-by: Dmitry Kasatkin <dmitry.kasatkin@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Link: http://lkml.kernel.org/r/CALLzPKbeXN5gdngo8uYYU8mAow=XhrwBFBhKfG811f37BubQOg@mail.gmail.com
---
 arch/x86/include/asm/unistd.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/unistd.h b/arch/x86/include/asm/unistd.h
index b4a3db7..21f77b8 100644
--- a/arch/x86/include/asm/unistd.h
+++ b/arch/x86/include/asm/unistd.h
@@ -7,6 +7,7 @@
 #  include <asm/unistd_32.h>
 #  define __ARCH_WANT_IPC_PARSE_VERSION
 #  define __ARCH_WANT_STAT64
+#  define __ARCH_WANT_SYS_IPC
 #  define __ARCH_WANT_SYS_OLD_MMAP
 #  define __ARCH_WANT_SYS_OLD_SELECT