From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 14801 invoked by alias); 12 Jul 2019 11:42:57 -0000 Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sourceware.org Received: (qmail 14791 invoked by uid 89); 12 Jul 2019 11:42:57 -0000 Authentication-Results: sourceware.org; auth=none X-Spam-SWARE-Status: No, score=-26.3 required=5.0 tests=AWL,BAYES_00,GIT_PATCH_0,GIT_PATCH_1,GIT_PATCH_2,GIT_PATCH_3,SPF_PASS autolearn=ham version=3.3.1 spammy=RUNTESTFLAGS, runtestflags, target_board, Joel X-HELO: mx1.suse.de Received: from mx2.suse.de (HELO mx1.suse.de) (195.135.220.15) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Fri, 12 Jul 2019 11:42:54 +0000 Received: from relay2.suse.de (unknown [195.135.220.254]) by mx1.suse.de (Postfix) with ESMTP id 75026AD69; Fri, 12 Jul 2019 11:42:52 +0000 (UTC) Subject: Re: [PING][PATCH] Fix amd64->i386 linux syscall restart problem To: Kevin Buettner Cc: gdb-patches@sourceware.org, Pedro Alves References: <20190316221341.021f7c62@f29-4.lan> <975e7120-1b61-2807-5aab-b961e07a80d0@redhat.com> <20190409192916.79fcb539@f29-4.lan> <20190410171606.0e736077@f29-4.lan> <40eb20c2-f6d0-eb16-d6fe-c717138b6559@suse.de> <20190708100001.6f2bc0fc@f29-4.lan> From: Tom de Vries Openpgp: preference=signencrypt Message-ID: <54a90435-e80e-6028-1007-db4423669850@suse.de> Date: Fri, 12 Jul 2019 11:42:00 -0000 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.6.1 MIME-Version: 1.0 In-Reply-To: <20190708100001.6f2bc0fc@f29-4.lan> Content-Type: multipart/mixed; boundary="------------A3D6EAD05DB115320CE7C5BE" X-IsSubscribed: yes X-SW-Source: 2019-07/txt/msg00344.txt.bz2 This is a multi-part message in MIME format. --------------A3D6EAD05DB115320CE7C5BE Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Content-length: 1713 On 08-07-19 19:00, Kevin Buettner wrote: > > On Fri, 21 Jun 2019 08:34:46 +0200 > Tom de Vries wrote: > >> On 21-05-19 14:59, Tom de Vries wrote: >>> On 11-04-19 02:16, Kevin Buettner wrote: >>>> On Wed, 10 Apr 2019 12:42:30 +0100 >>>> Pedro Alves wrote: >>>> >>>>>> Below is a diff showing the new comments. It also includes the >>>>>> changes which wrap the multi-line expressions in parens. >>>>> >>>>> Thanks, that new version of the comment looks great. >>>> >>>> It's in now. Thanks for the review. >>> >>> Hi, >>> >>> the tests fixed by this commit fail on the 8.3 branch (filed as PR24592). >>> >>> The commit applies cleanly on the 8.3 branch, and make the tests pass. >>> >>> OK to backport? >> >> Hi, >> >> [ Ping. ] >> >> I'm probably missing some context here. It seems sofar there are no >> commits to the 8.3 branch. AFAIU, a respin release 8.3.1 is targeted for >> 3 months after 8.3, so around 11th of August. >> >> What kind of fixes are acceptable for the respin? Do they have to be >> regression fixes? Or are functionality fixes like this one also allowed? >> Who's responsibility is it to backport fixes? >> >> Thanks, >> - Tom > > I'm okay with it going into the next point release for 8.3. Hi Kevin, I noticed the follow-up commit e90a813d96 "Fix regression caused by recently added syscall restart code". I reproduced the regression with the 8.3 branch + commit 3f52fdbcb5, and confirmed that the follow-up commit fixes it, so I suppose this one is necessary as well. Since this is the first time I'm pushing something to a gdb release branch, I'm posting here the two pre-commit-formatted patches for review. Thanks, - Tom --------------A3D6EAD05DB115320CE7C5BE Content-Type: text/x-patch; name="0001-Fix-amd64-i386-linux-syscall-restart-problem.patch" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="0001-Fix-amd64-i386-linux-syscall-restart-problem.patch" Content-length: 9845 Fix amd64->i386 linux syscall restart problem [ Backport of master commit 3f52fdbcb5. ] This commit fixes some failures in gdb.base/interrupt.exp when debugging a 32-bit i386 linux inferior from an amd64 host. When running the following test... make check RUNTESTFLAGS="--target_board unix/-m32 interrupt.exp" ... without this commit, I see the following output: FAIL: gdb.base/interrupt.exp: continue (the program exited) FAIL: gdb.base/interrupt.exp: echo data FAIL: gdb.base/interrupt.exp: Send Control-C, second time FAIL: gdb.base/interrupt.exp: signal SIGINT (the program is no longer running) ERROR: Undefined command "". ERROR: GDB process no longer exists === gdb Summary === When the test is run with this commit in place, we see 12 passes instead. This is the desired behavior. Analysis: On Linux, when a syscall is interrupted by a signal, the syscall may return -ERESTARTSYS when a signal occurs. Doing so indicates that the syscall is restartable. Then, depending on settings associated with the signal handler, and after the signal handler is called, the kernel can then either return -EINTR or can cause the syscall to be restarted. In this discussion, we are concerned with the latter case. On i386, the kernel returns this status via the EAX register. When debugging a 32-bit (i386) process from a 64-bit (amd64) GDB, the debugger fetches 64-bit registers even though the process being debugged is 32-bit. Since we're debugging a 32-bit target, only 32 bits are being saved in the register cache. Now, ideally, GDB would save all 64-bits in the regcache and then would be able to restore those same values when it comes time to continue the target. I've looked into doing this, but it's not easy and I don't see many benefits to doing so. One benefit, however, would be that EAX would appear as a negative value for doing syscall restarts. At the moment, GDB is setting the high 32 bits of RAX (and other registers too) to 0. So, when GDB restores EAX just prior to a syscall restart, the high 32 bits of RAX are zeroed, thus making it look like a positive value. For this particular purpose, we need to sign extend EAX so that RAX will appear as a negative value when EAX is set to -ERESTARTSYS. This in turn will cause the signal handling code in the kernel to recognize -ERESTARTSYS which will in turn cause the syscall to be restarted. This commit is based on work by Jan Kratochvil from 2009: https://sourceware.org/ml/gdb-patches/2009-11/msg00592.html Jan's patch had the sign extension code in amd64-nat.c. Several other native targets make use of this code, so it seemed better to move the sign extension code to a linux specific file. I also added similar code to gdbserver. Another approach is to fix the problem in the kernel. Hui Zhu tried to get a fix into the kernel back in 2014, but it was not accepted. Discussion regarding this approach may be found here: https://lore.kernel.org/patchwork/patch/457841/ Even if a fix were to be put into the kernel, we'd still need some kind of fix in GDB in order to support older kernels. Finally, I'll note that Fedora has been carrying a similar patch for at least nine years. Other distributions, including RHEL and CentOS have picked up this change and have been using it too. gdb/ChangeLog: PR gdb/24592 * amd64-linux-nat.c (amd64_linux_collect_native_gregset): New function. (fill_gregset): Call amd64_linux_collect_native_gregset instead of amd64_collect_native_gregset. (amd64_linux_nat_target::store_registers): Likewise. gdb/gdbserver/ChangeLog: PR gdb/24592 * linux-x86-low.c (x86_fill_gregset): Sign extend EAX value when using a 64-bit gdbserver. --- gdb/ChangeLog | 9 ++++++ gdb/amd64-linux-nat.c | 69 +++++++++++++++++++++++++++++++++++++++++-- gdb/gdbserver/ChangeLog | 6 ++++ gdb/gdbserver/linux-x86-low.c | 13 ++++++++ 4 files changed, 95 insertions(+), 2 deletions(-) diff --git a/gdb/ChangeLog b/gdb/ChangeLog index bc8d65abd3..b4781b6114 100644 --- a/gdb/ChangeLog +++ b/gdb/ChangeLog @@ -1,3 +1,12 @@ +2019-04-10 Kevin Buettner + + PR gdb/24592 + * amd64-linux-nat.c (amd64_linux_collect_native_gregset): New + function. + (fill_gregset): Call amd64_linux_collect_native_gregset instead + of amd64_collect_native_gregset. + (amd64_linux_nat_target::store_registers): Likewise. + 2019-05-11 Joel Brobecker * version.in: Set GDB version number to 8.3.0.DATE-git. diff --git a/gdb/amd64-linux-nat.c b/gdb/amd64-linux-nat.c index a0bb105f5a..8d0e8eb35c 100644 --- a/gdb/amd64-linux-nat.c +++ b/gdb/amd64-linux-nat.c @@ -92,6 +92,71 @@ static int amd64_linux_gregset32_reg_offset[] = /* Transfering the general-purpose registers between GDB, inferiors and core files. */ +/* See amd64_collect_native_gregset. This linux specific version handles + issues with negative EAX values not being restored correctly upon syscall + return when debugging 32-bit targets. It has no effect on 64-bit + targets. */ + +static void +amd64_linux_collect_native_gregset (const struct regcache *regcache, + void *gregs, int regnum) +{ + amd64_collect_native_gregset (regcache, gregs, regnum); + + struct gdbarch *gdbarch = regcache->arch (); + if (gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 32) + { + /* Sign extend EAX value to avoid potential syscall restart + problems. + + On Linux, when a syscall is interrupted by a signal, the + (kernel function implementing the) syscall may return + -ERESTARTSYS when a signal occurs. Doing so indicates that + the syscall is restartable. Then, depending on settings + associated with the signal handler, and after the signal + handler is called, the kernel can then either return -EINTR + or it can cause the syscall to be restarted. We are + concerned with the latter case here. + + On (32-bit) i386, the status (-ERESTARTSYS) is placed in the + EAX register. When debugging a 32-bit process from a 64-bit + (amd64) GDB, the debugger fetches 64-bit registers even + though the process being debugged is only 32-bit. The + register cache is only 32 bits wide though; GDB discards the + high 32 bits when placing 64-bit values in the 32-bit + regcache. Normally, this is not a problem since the 32-bit + process should only care about the lower 32-bit portions of + these registers. That said, it can happen that the 64-bit + value being restored will be different from the 64-bit value + that was originally retrieved from the kernel. The one place + (that we know of) where it does matter is in the kernel's + syscall restart code. The kernel's code for restarting a + syscall after a signal expects to see a negative value + (specifically -ERESTARTSYS) in the 64-bit RAX register in + order to correctly cause a syscall to be restarted. + + The call to amd64_collect_native_gregset, above, is setting + the high 32 bits of RAX (and other registers too) to 0. For + syscall restart, we need to sign extend EAX so that RAX will + appear as a negative value when EAX is set to -ERESTARTSYS. + This in turn will cause the signal handling code in the + kernel to recognize -ERESTARTSYS which will in turn cause the + syscall to be restarted. + + The test case gdb.base/interrupt.exp tests for this problem. + Without this sign extension code in place, it'll show + a number of failures when testing against unix/-m32. */ + + if (regnum == -1 || regnum == I386_EAX_REGNUM) + { + void *ptr = ((gdb_byte *) gregs + + amd64_linux_gregset32_reg_offset[I386_EAX_REGNUM]); + + *(int64_t *) ptr = *(int32_t *) ptr; + } + } +} + /* Fill GDB's register cache with the general-purpose register values in *GREGSETP. */ @@ -109,7 +174,7 @@ void fill_gregset (const struct regcache *regcache, elf_gregset_t *gregsetp, int regnum) { - amd64_collect_native_gregset (regcache, gregsetp, regnum); + amd64_linux_collect_native_gregset (regcache, gregsetp, regnum); } /* Transfering floating-point registers between GDB, inferiors and cores. */ @@ -237,7 +302,7 @@ amd64_linux_nat_target::store_registers (struct regcache *regcache, int regnum) if (ptrace (PTRACE_GETREGS, tid, 0, (long) ®s) < 0) perror_with_name (_("Couldn't get registers")); - amd64_collect_native_gregset (regcache, ®s, regnum); + amd64_linux_collect_native_gregset (regcache, ®s, regnum); if (ptrace (PTRACE_SETREGS, tid, 0, (long) ®s) < 0) perror_with_name (_("Couldn't write registers")); diff --git a/gdb/gdbserver/ChangeLog b/gdb/gdbserver/ChangeLog index 6648a1ed53..f9f6abc22e 100644 --- a/gdb/gdbserver/ChangeLog +++ b/gdb/gdbserver/ChangeLog @@ -1,3 +1,9 @@ +2019-04-10 Kevin Buettner + + PR gdb/24592 + * linux-x86-low.c (x86_fill_gregset): Sign extend EAX value + when using a 64-bit gdbserver. + 2019-03-04 Sergio Durigan Junior * configure.srv: Use '$enable_unittest' instead of '$development' diff --git a/gdb/gdbserver/linux-x86-low.c b/gdb/gdbserver/linux-x86-low.c index 029796e361..dd7673126e 100644 --- a/gdb/gdbserver/linux-x86-low.c +++ b/gdb/gdbserver/linux-x86-low.c @@ -338,6 +338,19 @@ x86_fill_gregset (struct regcache *regcache, void *buf) collect_register_by_name (regcache, "orig_eax", ((char *) buf) + ORIG_EAX * REGSIZE); + + /* Sign extend EAX value to avoid potential syscall restart + problems. + + See amd64_linux_collect_native_gregset() in gdb/amd64-linux-nat.c + for a detailed explanation. */ + if (register_size (regcache->tdesc, 0) == 4) + { + void *ptr = ((gdb_byte *) buf + + i386_regmap[find_regno (regcache->tdesc, "eax")]); + + *(int64_t *) ptr = *(int32_t *) ptr; + } } static void --------------A3D6EAD05DB115320CE7C5BE Content-Type: text/x-patch; name="0002-Fix-regression-caused-by-recently-added-syscall-restart-code.patch" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename*0="0002-Fix-regression-caused-by-recently-added-syscall-restart"; filename*1="-code.patch" Content-length: 2822 Fix regression caused by recently added syscall restart code [ Backport of master commit e90a813d96. ] This line of code... *(int64_t *) ptr = *(int32_t *) ptr; ...in linux-x86-low.c is not needed (and does not work correctly) within a 32-bit executable. I added an __x86_64__ ifdef (which is used extensively elsewhere in the file for like purposes) to prevent this code from being included in 32-bit builds. It fixes the following regressions when running on native i686-pc-linux-gnu: FAIL: gdb.server/abspath.exp: continue to main FAIL: gdb.server/connect-without-multi-process.exp: multiprocess=auto: continue to main FAIL: gdb.server/connect-without-multi-process.exp: multiprocess=off: continue to main FAIL: gdb.server/ext-restart.exp: restart: run to main FAIL: gdb.server/ext-restart.exp: run to main FAIL: gdb.server/ext-run.exp: continue to main FAIL: gdb.server/ext-wrapper.exp: print d FAIL: gdb.server/ext-wrapper.exp: restart: print d FAIL: gdb.server/ext-wrapper.exp: restart: run to marker FAIL: gdb.server/ext-wrapper.exp: run to marker FAIL: gdb.server/no-thread-db.exp: continue to breakpoint: after tls assignment FAIL: gdb.server/reconnect-ctrl-c.exp: first: stop with control-c FAIL: gdb.server/reconnect-ctrl-c.exp: second: stop with control-c FAIL: gdb.server/run-without-local-binary.exp: run test program until the end FAIL: gdb.server/server-kill.exp: continue to breakpoint: after server_pid assignment FAIL: gdb.server/server-kill.exp: tstatus FAIL: gdb.server/server-run.exp: continue to main gdb/gdbserver/ChangeLog: PR gdb/24592 * linux-x86-low.c (x86_fill_gregset): Don't compile 64-bit sign extension code on 32-bit builds. --- gdb/gdbserver/ChangeLog | 6 ++++++ gdb/gdbserver/linux-x86-low.c | 2 ++ 2 files changed, 8 insertions(+) diff --git a/gdb/gdbserver/ChangeLog b/gdb/gdbserver/ChangeLog index f9f6abc22e..9968804308 100644 --- a/gdb/gdbserver/ChangeLog +++ b/gdb/gdbserver/ChangeLog @@ -1,3 +1,9 @@ +2019-05-06 Kevin Buettner + + PR gdb/24592 + * linux-x86-low.c (x86_fill_gregset): Don't compile 64-bit + sign extension code on 32-bit builds. + 2019-04-10 Kevin Buettner PR gdb/24592 diff --git a/gdb/gdbserver/linux-x86-low.c b/gdb/gdbserver/linux-x86-low.c index dd7673126e..adddfe7797 100644 --- a/gdb/gdbserver/linux-x86-low.c +++ b/gdb/gdbserver/linux-x86-low.c @@ -339,6 +339,7 @@ x86_fill_gregset (struct regcache *regcache, void *buf) collect_register_by_name (regcache, "orig_eax", ((char *) buf) + ORIG_EAX * REGSIZE); +#ifdef __x86_64__ /* Sign extend EAX value to avoid potential syscall restart problems. @@ -351,6 +352,7 @@ x86_fill_gregset (struct regcache *regcache, void *buf) *(int64_t *) ptr = *(int32_t *) ptr; } +#endif } static void --------------A3D6EAD05DB115320CE7C5BE--