| Summary: | >=sys-libs/glibc-2.16 signal handling on alpha is broken | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Tobias Klausmann (RETIRED) <klausman> |
| Component: | [OLD] Core system | Assignee: | Gentoo Toolchain Maintainers <toolchain> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | alpha |
| Priority: | Normal | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Bug Depends on: | |||
| Bug Blocks: | 467256 | ||
| Attachments: |
Testing program for gcc test suite
patch |
||
when you dekeyword things in toolchain ebuilds, don't just delete the keyword. a missing keyword means it'll get built (and maybe runtime) tested before ~arch gets added back. if you want to actually indicate support is actively disabled, use -arch instead. (In reply to Tobias Klausmann from comment #0) seems to me that glibc-2.16 fails the same way. can you confirm/deny ? (In reply to SpanKY from comment #2) > (In reply to Tobias Klausmann from comment #0) > > seems to me that glibc-2.16 fails the same way. can you confirm/deny ? Indeed, freshly made chroot with only glibc updated to 2.16.0: # ./a.out RA = 0x120000bfc, CFA = 0x11fd091c0 RA = 0x120000c34, CFA = 0x11fd091c0 RA = 0x120000c58, CFA = 0x11fd091d0 RA = 0x120000c7c, CFA = 0x11fd091d0 RA = 0x200000a29f0, CFA = 0x11fd091e0 RA = (nil), CFA = 0x11fd09468 Aborted bisecting these earlier versions are a pita as glibc-ports is a sep repo :/ Created attachment 363228 [details, diff]
patch
Patch from Richard. Seems to work.
RA = 0x120000c44, CFA = 0x11fc9e800
RA = 0x120000c78, CFA = 0x11fc9e810
RA = 0x200000b49d0, CFA = 0x11fc9e820
RA = 0x120000aac, CFA = 0x11fc9eb58
RA = 0x120000b38, CFA = 0x11fc9eb90
RA = 0x120000b58, CFA = 0x11fc9ec30
RA = 0x1200008e8, CFA = 0x11fc9ec40
RA = 0x2000009ad00, CFA = 0x11fc9ec50
Aborted
I've tested with glibc-2.17, but should be applied to 2.16, 2.17, and 2.18. Please apply.
added to glibc 2.16 & 2.17 http://sources.gentoo.org/gentoo/src/patchsets/glibc/2.16.0/6024_all_alpha-fix-signal-thunk-unwind-info.patch?rev=1.1 http://sources.gentoo.org/gentoo/src/patchsets/glibc/2.17/6024_all_alpha-fix-signal-thunk-unwind-info.patch?rev=1.1 still need to do 2.18 |
Created attachment 355772 [details] Testing program for gcc test suite This was found by Uros Bizjak using the gcc test suite. Testing program is attached. Compile with -fexceptions -fnon-call-exceptions -O2 On broken glibc/alpha, it yields this output: $ ./a.out RA = 0x120000cbc, CFA = 0x11fd9cde0 RA = 0x120000cf4, CFA = 0x11fd9cde0 RA = 0x120000d18, CFA = 0x11fd9cdf0 RA = 0x120000d3c, CFA = 0x11fd9cdf0 RA = 0x2000009cee0, CFA = 0x11fd9ce00 RA = (nil), CFA = 0x11fd9d088 Aborted The (nil) should not be there but rather a proper address. From Uros' report: """ The return address is deduced from RT signal context, but for some reason it is wrong now (nil). The breakage is system wide and is not dependent on gcc or libgcc version. """ I will mask glibc-2.17 on alpha in a moment.