| Message ID | 1513253436-11438-1-git-send-email-adhemerval.zanella@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | sh: Fix clone exit return code | expand |
Hi Adhemerval! On 12/14/2017 01:10 PM, Adhemerval Zanella wrote: > Since 3f823e87cc (Call exit directly in clone (BZ #21512)) SH clone > implementation fails to set the exit code resulting in the failures: > > FAIL: nptl/tst-align-clone > FAIL: nptl/tst-getpid1 I assume you have run the full testsuite then. Could you also post your results to the glibc wiki for SH? Thanks, Adrian
On Thu, 14 Dec 2017, John Paul Adrian Glaubitz wrote: > Hi Adhemerval! > > On 12/14/2017 01:10 PM, Adhemerval Zanella wrote: > > Since 3f823e87cc (Call exit directly in clone (BZ #21512)) SH clone > > implementation fails to set the exit code resulting in the failures: > > > > FAIL: nptl/tst-align-clone > > FAIL: nptl/tst-getpid1 > I assume you have run the full testsuite then. Could you also post your > results to the glibc wiki for SH? Note that results on the wiki relate to particular releases (well, typically versions during the release freeze) rather than to master. Thus, it would be appropriate to post results on the 2.26 wiki page for 2.26 branch (making clear what version is being tested to avoid confusing anyone using an older branch version and comparing their results), or on the 2.27 wiki page for results from 1 January onwards, but we don't have somewhere to post results for mainline (other than the libc-testresults mailing list - we don't yet have a script for mailing results from a single testsuite run along with information about the system configuration, but having such a way to mail results would be useful).
On 14/12/2017 11:08, John Paul Adrian Glaubitz wrote: > Hi Adhemerval! > > On 12/14/2017 01:10 PM, Adhemerval Zanella wrote: >> Since 3f823e87cc (Call exit directly in clone (BZ #21512)) SH clone >> implementation fails to set the exit code resulting in the failures: >> >> FAIL: nptl/tst-align-clone >> FAIL: nptl/tst-getpid1 > I assume you have run the full testsuite then. Could you also post your > results to the glibc wiki for SH? > > Thanks, > Adrian > I did not run the full testsuite for SH yet, only the nptl tests so far on master (still struggling with system toolchain to make the full testsuite make clean).
On 12/14/2017 04:38 PM, Adhemerval Zanella wrote: > I did not run the full testsuite for SH yet, only the nptl tests so far > on master (still struggling with system toolchain to make the full > testsuite make clean). Ok. I am running "make check" for 2.26 now and will report once it's done. Maybe I should finally set up my other SH evaluation board. Adrian
Also, do people interested in sh think the sh3 support is still useful, or is only sh4 support useful now?
On 12/14/2017 04:48 PM, Joseph Myers wrote: > Also, do people interested in sh think the sh3 support is still useful, or > is only sh4 support useful now? Yes, SH-3 is very useful for the upcoming J-Core J3 CPU [1], so please don't remove support for it. SH-3 and SH-4 are also part of the Jenkins rebootstrap jobs [2]. Adrian > [1] http://j-core.org/roadmap.html > [2] https://jenkins.debian.net/view/rebootstrap/
On Thu, 14 Dec 2017, John Paul Adrian Glaubitz wrote: > On 12/14/2017 04:48 PM, Joseph Myers wrote: > > Also, do people interested in sh think the sh3 support is still useful, or > > is only sh4 support useful now? > > Yes, SH-3 is very useful for the upcoming J-Core J3 CPU [1], so please > don't remove support for it. It would be useful to refactor the sysdeps directories to reduce the need for separate sh3 and sh4 sysdeps (e.g. why shouldn't they share a single setjmp implementation with appropriate conditionals?). My understanding from <https://sourceware.org/ml/libc-alpha/2014-01/msg00388.html> is that SH3 and SH4 have incompatible ABIs, though I don't know what the differences are, or whether there are still differences (e.g. at the function call ABI level) after the kernel sigcontext changes in Linux 4.8 (commit bbe6c77857c38f4acbdc4fc70399515226d1859a) and consequent glibc changes (commit f07820b7d13846431bb1c291414b6286cc5f5cbf). > > [1] http://j-core.org/roadmap.html That refers to SH3 hard float. Does that mean more SH ABI variants are coming (if ABI differences between SH3 and SH4 remain)? (glibc doesn't handle any sort of SH3 hard float at present; most things to do with floating-point registers, exceptions and rounding modes are in sh4 sysdeps directories.)
On 14/12/2017 13:40, John Paul Adrian Glaubitz wrote: > On 12/14/2017 04:38 PM, Adhemerval Zanella wrote: >> I did not run the full testsuite for SH yet, only the nptl tests so far >> on master (still struggling with system toolchain to make the full >> testsuite make clean). > > Ok. I am running "make check" for 2.26 now and will report once it's done. > > Maybe I should finally set up my other SH evaluation board. > > Adrian > Thanks for checking on 2.26. I will commit this patch shortly if no one opposes. Although I haven't check with a full testsuite, the affected testscases, tst-align-clone.c, tst-clone.c, tst-clone2.c, tst-clone3.c, and tst-getpid1, show no regression.
On 12/19/2017 01:31 PM, Adhemerval Zanella wrote: > Thanks for checking on 2.26. I will commit this patch shortly if > no one opposes. Although I haven't check with a full testsuite, > the affected testscases, tst-align-clone.c, tst-clone.c, tst-clone2.c, > tst-clone3.c, and tst-getpid1, show no regression. Sorry, I forgot to post the results. The "make check" actually failed: > https://people.debian.org/~glaubitz/glibc-2.26-SH-check.log I also killed the buildd on tirpitz now to free up more resources for you. Adrian
On 19/12/2017 10:36, John Paul Adrian Glaubitz wrote: > On 12/19/2017 01:31 PM, Adhemerval Zanella wrote: >> Thanks for checking on 2.26. I will commit this patch shortly if >> no one opposes. Although I haven't check with a full testsuite, >> the affected testscases, tst-align-clone.c, tst-clone.c, tst-clone2.c, >> tst-clone3.c, and tst-getpid1, show no regression. > > Sorry, I forgot to post the results. The "make check" actually failed: > >> https://people.debian.org/~glaubitz/glibc-2.26-SH-check.log > > I also killed the buildd on tirpitz now to free up more resources for you. > > Adrian > Thanks, the c++ tests seems to trigger a build failure with system toolchain on tirpitz (although the log you posted did not show much information regarding it).
diff --git a/sysdeps/unix/sysv/linux/sh/clone.S b/sysdeps/unix/sysv/linux/sh/clone.S index b13a64b..d369a82 100644 --- a/sysdeps/unix/sysv/linux/sh/clone.S +++ b/sysdeps/unix/sysv/linux/sh/clone.S @@ -73,6 +73,7 @@ ENTRY(__clone) mov.l @(4,r15), r4 /* we are done, passing the return value through r0 */ + mov r0, r4 mov #+SYS_ify(exit), r3 trapa #0x15 .align 2
Since 3f823e87cc (Call exit directly in clone (BZ #21512)) SH clone implementation fails to set the exit code resulting in the failures: FAIL: nptl/tst-align-clone FAIL: nptl/tst-getpid1 This patch fixes the both testcases. * sysdeps/unix/sysv/linux/sh/clone.S (__clone): Fix exit return code. Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org> --- ChangeLog | 6 ++++++ sysdeps/unix/sysv/linux/sh/clone.S | 1 + 2 files changed, 7 insertions(+)