Fixes for compiling with g++ 4.3.2

[Mike McCormack]

Hi Hilko,

Which machine are you running the tests on? The tests run successfully on my
Mac Mini (Intel Core Duo), built with gcc version 4.1.2 20061115
(prerelease) (Debian 4.1.1-21).  (from Debian Etch)

The thread test case that you're hitting runs an infinite loop on the client
side (ttclient), to make sure that it the kernel preempts correctly after a
short timeout.  My guess would be that there's a race with the timer setup
or the SIGALRM is not handled correctly.

Perhaps try adding a dprintf in ptrace_address_space_impl::run and see
what's happening to the SIGALRM, if it's coming at all.

I've pushed all your patches onto the public git repository.  I probably
won't be adding any of my own until I get back from holidays at the start of
January.

thanks,

Mike



2008/12/17 Hilko Bengen <bengen@hidden>


> 1. Using CC=gcc-4.2 CXX=g++-4.2
> -------------------------------
>
> In most attempts, the `thread' test leads to a process called
> `ttclient' consuming 100% CPU time; if I kill -9 that process, it
> outputs "failed to get registers".
>
> Sometimes, after I have killed the `ttclient' process during the
> `thread' test, I get an error in the `wine-port' test.
>
> Occasionally, the `thread' test fails with a "1014: debug: 185: wait
> failed" message.
>
> wine-port
> 1018: debug: 115: Expected STATUS_SUCCESS, got c0000033
> process 0010
>        thread 1014
> 1 threads 1 processes left
>
> 2. CC=gcc-4.3 CXX=g++-4.2
> -------------------------
>
> I have been able to observe the same errors with the `thread' and the
> `wine-port' test cases.
>
> 3. CC=gcc-4.2 CXX=g++-4.3
> -------------------------
>
> I have been able to observe the same errors with the `thread' and the
> `wine-port' test cases.
>
> The `file' test (./loader/ntl -q \??\c:\tests\file.exe) eats CPU time
> and rapidly leaks memory every time I call it.
>
> Strange: If I run ./runtest -q file straight from the shell, without
> using make, I don't get this behavior, but a bunch of failed tests.
>
> 1014: debug: 234: failed to query directory 80000006
> 1014: debug: 139: information wrong 00000062
> 1014: debug: 144: FileNameLength wrong 4
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 256: query failed c000000f
> 1014: debug: 139: information wrong 00000001
> 1014: debug: 144: FileNameLength wrong 2
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 276: query failed c000000f
> 1014: debug: 139: information wrong 00000001
> 1014: debug: 144: FileNameLength wrong 2
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 297: query failed c000000f
> 1014: debug: 139: information wrong 00000001
> 1014: debug: 144: FileNameLength wrong 2
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 306: query failed c000000f
> 1014: debug: 139: information wrong 00000001
> 1014: debug: 144: FileNameLength wrong 2
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 311: query failed c000000f
> 1014: debug: 139: information wrong 00000001
> 1014: debug: 144: FileNameLength wrong 2
> 1014: debug: 145: FileName wrong
> 1014: debug: 146: FileAttributes wrong 00000010
> 1014: debug: 30 failed, 117 passed
>
> 4. CC=gcc-4.3 CXX=g++-4.3
> -------------------------
>
> Same as (3).
>
> 5. General
> ----------
>
> Whenever the terminal window is resized while the tests run, I am
> presented with the debugger prompt, apparently because of SIGWINCH
> being caught. This can't be right, can it? Here's a patch.
>
> diff --git a/loader/ptrace_base.cpp b/loader/ptrace_base.cpp
> index f0c017f..59c9146 100644
> --- a/loader/ptrace_base.cpp
> +++ b/loader/ptrace_base.cpp
> @@ -328,6 +328,9 @@ void ptrace_address_space_impl::run( void
> *TebBaseAddress, PCONTEXT ctx, int sin
>                if (WIFSTOPPED(status) && WEXITSTATUS(status) == SIGALRM)
>                        break;
>
> +               if (WIFSTOPPED(status) && WEXITSTATUS(status) == SIGWINCH)
> +                       break;
> +
>                if (WIFSTOPPED(status) && single_step)
>                        break;
>
> Cheers,
> -Hilko
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.infowares.com/archive/ring3k/attachments/20081216/8c39b052/attachment.htm