MySQL Bugs: #33398: mysys unit test segv in my

Bug #33398	mysys unit test segv in my_atomic-t
Submitted:	20 Dec 2007 12:58	Modified:	15 Apr 2008 23:13
Reporter:	Daniel Fischer	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Tests	Severity:	S7 (Test Cases)
Version:	mysql-6.0 bk	OS:	Solaris (SPARC 64)
Assigned to:		CPU Architecture:	Any

Description:
On Solaris 10 SPARC 64 bit, the mysys unit test is aborted after signal 11 in my_atomic-t:

mysys/bitmap-t......ok
mysys/base64-t......ok
mysys/my_atomic-t...FAILED--Further testing stopped: Signal 11 thrown

Output from manual run:
$ ./my_atomic-t
# N CPUs: 2, atomic ops: mutex
1..7
ok 1 - my_atomic_initialize() returned 0
# Testing my_atomic_add32 with 100 threads, 3000 iterations... 
Bail out! Signal 11 thrown

Ouput from manual run in debugger:
(dbx) run
Running: my_atomic-t 
(process id 2745)
Reading libc_psr.so.1
# N CPUs: 2, atomic ops: mutex
1..7
ok 1 - my_atomic_initialize() returned 0
# Testing my_atomic_add32 with 100 threads, 3000 iterations... 
t@2 (l@2) signal SEGV (no mapping at the fault address) in my_atomic_add32 at 0x10000e718
0x000000010000e718: my_atomic_add32       :     ld       [%o0], %o3
(dbx) where
current thread: t@2
=>[1] my_atomic_add32(0x25b674, 0x792a2471, 0x0, 0x10025b688, 0xffffffff7e6eae40, 0xffffffff7f402400), at 0x10000e718 
  [2] test_atomic_add_handler(0xffffffff7ffff88c, 0x7fffffff, 0x5dc, 0x10025b680, 0x87654321, 0x10025b674), at 0x10000cad8 

How to repeat:
Build mysql-6.0 bk on Solaris 10 SPARC 64-bit and run the mysys unit test, or run my_atomic-t directly.

Setting to verified.

I have just fixed a similar-looking bug in the mysql-maria. Patch for this is to the file include/my_atomic.h:
187c187
< extern int ## S my_atomic_add ## S(Uv_ ## S, U_ ## S);
---
> extern int ## S my_atomic_add ## S(Uv_ ## S U_a, U_ ## S U_v);
190c190
< extern int ## S my_atomic_fas ## S(Uv_ ## S, U_ ## S);
---
> extern int ## S my_atomic_fas ## S(Uv_ ## S U_a, U_ ## S U_v);
193c193
< extern int my_atomic_cas ## S(Uv_ ## S, Uv_ ## S, U_ ## S);
---
> extern int my_atomic_cas ## S(Uv_ ## S U_a, Uv_ ## S U_cmp, U_ ## S U_set);
196c196
< extern int ## S my_atomic_load ## S(Uv_ ## S);
---
> extern int ## S my_atomic_load ## S(Uv_ ## S U_a);
199c199
< extern void my_atomic_store ## S(Uv_ ## S, U_ ## S);
---
> extern void my_atomic_store ## S(Uv_ ## S U_a, U_ ## S U_v);

see also http://lists.mysql.com/commits/42667

Guilhem, this is Maria team code, please take over.

Asked the backup team to apply it to 6.0-backup.

Pushed a small patch on Guilhem's request to mysql-6.0-backup which fixes the my_atomic.h header in preparation for the MyISAM driver release. 

http://lists.mysql.com/commits/43108

Pushed into 6.0.5-alpha

No changelog entry needed, I think. (Fixes problem that doesn't appear in any released version.)