[Bug tapsets/17928] A number of the systemtap.syscall/syscall.exp tests fail on 32-bit arm because size_t sign extended to 64-bit value

["dsmith at redhat dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=17928

--- Comment #9 from David Smith <dsmith at redhat dot com> ---
(In reply to William Cohen from comment #3)
> The changes helped quite a bit.  Before the patches:
> 
> testcase ./systemtap.syscall/syscall.exp completed in 211 seconds
> 
> 		=== systemtap Summary ===
> 
> # of expected passes		85
> # of unexpected failures	28
> # of untested testcases		2
> # of unsupported tests		4
> 
> 
> After the patches the test results are significantly improved:
> 
> testcase ./systemtap.syscall/syscall.exp completed in 217 seconds
> 
> 		=== systemtap Summary ===
> 
> # of expected passes		101
> # of unexpected failures	13
> # of untested testcases		1
> # of unsupported tests		6
> runtest completed at Fri Feb  6 16:23:55 2015
> 
> Comparing the systemtap.sum files there all the no new failures. The
> following tests remain as failure:
> 
> FAIL: 32-bit alarm syscall (can't set timer to time in past) 
> FAIL: 32-bit chmod syscall (chmod arg2 of -1 get odd 01777777777777777777777)
> FAIL: 32-bit dup syscall  (not getting correct args for dup3)
> FAIL: 32-bit fadvise64 syscall (test crashs on first fadvise64)
> FAIL: 32-bit keyctl syscall (returns -EACCES for keyctl syscall)
> FAIL: 32-bit mmap syscall (mlock arg2 != 4096)
> FAIL: 32-bit poll syscall (epoll_wait not mapped to epoll_pwait)
> FAIL: 32-bit recv syscall (doesn't match one recv)
> FAIL: 32-bit recvfrom syscall (recvfrom mishandle 32-bit arg (-1))
> FAIL: 32-bit send syscall (looks similar to recv syscall for send)
> FAIL: 32-bit sendmsg syscall(looks similar to recv syscall for sendmsg)
> FAIL: 32-bit shmat syscall (returning -EINVAL for several shmat/
> FAIL: 32-bit uid syscall (gets wrong value for setuid)
> 
> Will attach the systemtap.log in a moment for details.

I looked through systemtap.log, and have attached a patch that should fix the
following failures:

  keyctl, recv, recvfrom, send, and sendmsg

All but recvfrom are testcase tweaks. On a 32-bit kernel, MAXSTRINGLEN is only
256, which causes us problems in places.

Besides the ones getting arguments incorrectly, the fadvise64 test crash is
odd. You might try compiling and running the testcase by hand as root and see
if it crashes.

-- 
You are receiving this mail because:
You are the assignee for the bug.