Date: Fri, 28 Sep 2018 20:52:24 +0200 From: Andreas Tobler <andreast@FreeBSD.org> To: Konstantin Belousov <kostikbel@gmail.com> Cc: Michael Tuexen <tuexen@freebsd.org>, src-committers <src-committers@freebsd.org>, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r338930 - head/libexec/rtld-elf/powerpc Message-ID: <89d81c9a-9a93-2c82-93a0-12aa3f8e1a8c@FreeBSD.org> In-Reply-To: <20180927231256.GB5335@kib.kiev.ua> References: <201809251929.w8PJTZl9021331@repo.freebsd.org> <828EF35C-7216-4847-A73D-464111006E52@freebsd.org> <d1442750-1969-d41c-24d8-4bc335426207@FreeBSD.org> <20180927231256.GB5335@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On 28.09.18 01:12, Konstantin Belousov wrote: > On Thu, Sep 27, 2018 at 09:37:31PM +0200, Andreas Tobler wrote: >> On 27.09.18 20:48, Michael Tuexen wrote: >>>> On 25. Sep 2018, at 21:29, Andreas Tobler <andreast@FreeBSD.org> wrote: >>>> >>>> Author: andreast >>>> Date: Tue Sep 25 19:29:35 2018 >>>> New Revision: 338930 >>>> URL: https://svnweb.freebsd.org/changeset/base/338930 >>>> >>>> Log: >>>> Bring the 32-bit powerpc (PowerMac) back to live. The commit 338486 reworked >>>> some TLS bits. This broke operation on the PowerMac. Namely one could not login. >>>> At login the screen/shell was giving back lots of backslashes and the login >>>> shell dumped core. >>>> >>>> The fix to this issue is to revert the powerpc commit from 338486 and to >>>> increase the TLS_TCB_SIZE to 16. >>>> Reverting only did not help, login was possible but userland applications >>>> aborted with strange messages. >>>> >>>> I tested this patch with world/kernel builds and with port upgrades. >>>> Additionally a full gcc8 bootstrap was successfully completed. >>>> >>>> Reviewed by: jhibbits@ >>>> Approved by: re (Glen) >>>> >>>> Modified: >>>> head/libexec/rtld-elf/powerpc/rtld_machdep.h >>>> >>>> Modified: head/libexec/rtld-elf/powerpc/rtld_machdep.h >>>> ============================================================================== >>>> --- head/libexec/rtld-elf/powerpc/rtld_machdep.h Tue Sep 25 18:54:18 2018 (r338929) >>>> +++ head/libexec/rtld-elf/powerpc/rtld_machdep.h Tue Sep 25 19:29:35 2018 (r338930) >>>> @@ -69,12 +69,12 @@ void _rtld_powerpc_pltcall(void); >>>> >>>> #define TLS_TP_OFFSET 0x7000 >>>> #define TLS_DTV_OFFSET 0x8000 >>>> -#define TLS_TCB_SIZE 8 >>>> +#define TLS_TCB_SIZE 16 >>>> >>>> #define round(size, align) \ >>>> (((size) + (align) - 1) & ~((align) - 1)) >>>> #define calculate_first_tls_offset(size, align) \ >>>> - TLS_TCB_SIZE >>>> + round(8, align) >>>> #define calculate_tls_offset(prev_offset, prev_size, size, align) \ >>>> round(prev_offset + prev_size, align) >>>> #define calculate_tls_end(off, size) ((off) + (size)) >>>> >>> After performing a buildworld/buildkernel and running r338956 on a 32-bit powerpc machine (G4 Mac Mini), >>> the following program: >>> >>> #include <pthread.h> >>> #include <stdio.h> >>> #include <string.h> >>> >>> void * >>> f(void * arg) >>> { >>> return (arg); >>> } >>> >>> int >>> main(void) >>> { >>> void *res; >>> pthread_t tid; >>> int err; >>> >>> if ((err = pthread_create(&tid, NULL, f, NULL)) != 0) { >>> fprintf(stderr, "pthread_create: %s\n", strerror(err)); >>> } >>> if ((err = pthread_join(tid, &res)) != 0) { >>> fprintf(stderr, "pthread_join: %s\n", strerror(err)); >>> } >>> return (0); >>> } >>> >>> is killed: >>> >>> tuexen@bsd5:~ % ./test >>> ld-elf.so.1: assert failed: /usr/home/tuexen/head/libexec/rtld-elf/rtld.c:4753 >>> Abort (core dumped) >>> >>> gdb shows: >>> >>> tuexen@bsd5:~ % gdb -c test.core test >>> GNU gdb (GDB) 8.1 [GDB v8.1 for FreeBSD] >>> Copyright (C) 2018 Free Software Foundation, Inc. >>> License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>; >>> This is free software: you are free to change and redistribute it. >>> There is NO WARRANTY, to the extent permitted by law. Type "show copying" >>> and "show warranty" for details. >>> This GDB was configured as "powerpc-portbld-freebsd12.0". >>> Type "show configuration" for configuration details. >>> For bug reporting instructions, please see: >>> <http://www.gnu.org/software/gdb/bugs/>. >>> Find the GDB manual and other documentation resources online at: >>> <http://www.gnu.org/software/gdb/documentation/>. >>> For help, type "help". >>> Type "apropos word" to search for commands related to "word"... >>> Reading symbols from test...done. >>> [New LWP 100112] >>> Core was generated by `./test'. >>> Program terminated with signal SIGABRT, Aborted. >>> #0 0x41829a58 in thr_kill () at thr_kill.S:3 >>> 3 RSYSCALL(thr_kill) >>> (gdb) bt >>> #0 0x41829a58 in thr_kill () at thr_kill.S:3 >>> #1 0x41829378 in __raise (s=6) >>> at /usr/home/tuexen/head/lib/libc/gen/raise.c:52 >>> #2 0x41823b84 in abort () at /usr/home/tuexen/head/lib/libc/stdlib/abort.c:67 >>> #3 0x41813f58 in allocate_tls (objs=0x41843000, oldtcb=<optimized out>, >>> tcbsize=8, tcbalign=<optimized out>) >>> at /usr/home/tuexen/head/libexec/rtld-elf/rtld.c:4753 >>> #4 0x41813fc0 in _rtld_allocate_tls (oldtls=0x0, tcbsize=8, tcbalign=16) >>> at /usr/home/tuexen/head/libexec/rtld-elf/rtld.c:5030 >>> #5 0x41877438 in _tcb_ctor (thread=0x41a8d300, initial=<optimized out>) >>> at /usr/home/tuexen/head/lib/libthr/thread/thr_ctrdtr.c:45 >>> #6 0x41876ebc in _thr_alloc (curthread=0x41a8d000) >>> at /usr/home/tuexen/head/lib/libthr/thread/thr_list.c:172 >>> #7 0x41867118 in _pthread_create (thread=0xffffdc10, attr=0x0, >>> start_routine=0x180073c <f>, arg=0x0) >>> at /usr/home/tuexen/head/lib/libthr/thread/thr_create.c:81 >>> #8 0x01800798 in main () >>> (gdb) quit >>> >>> Any idea what is wrong? >> >> TLS is broken on powerpc ;) >> >> I see the same. >> >> Would you mind reverting my commit and see if it is better/different and >> how different? >> >> For me the commit was an improvement, but I agree this is not enough. >> >> I try to analyze what is going on. Any help is appreciated. > > The cause is quite clear, sizeof(struct tcb) is 8 on power 32bit, see > lib/libthread/arch/powerpc/include/pthread_md.h. I do not think that > you can change this because it is part of the ABI assumed by the > compilers. > Ouch, I think a (big) fail on my side. I'm papering over something else. Michael is happy with TLS_TCB_SIZE == 8. I need to find the cause why portmaster -y --no-confirm -a fails with: --- [andreast@helium] /home/andreast/> portmaster -y --no-confirm -a ===>>> Gathering distinfo list for installed ports a number is expected, got: 0 a number is expected, got: 0 a number is expected, got: 0 a number is expected, got: 0 --- Thanks for making me aware of this fail. Andreas
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?89d81c9a-9a93-2c82-93a0-12aa3f8e1a8c>