From patchwork Thu Dec 7 10:01:50 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yao Qi X-Patchwork-Id: 24772 Received: (qmail 61308 invoked by alias); 7 Dec 2017 10:01:59 -0000 Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sourceware.org Delivered-To: mailing list gdb-patches@sourceware.org Received: (qmail 61295 invoked by uid 89); 7 Dec 2017 10:01:58 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-26.9 required=5.0 tests=BAYES_00, FREEMAIL_FROM, GIT_PATCH_0, GIT_PATCH_1, GIT_PATCH_2, GIT_PATCH_3, RCVD_IN_DNSWL_NONE, SPF_PASS autolearn=ham version=3.3.2 spammy=2045, 2764, 24517, 2896 X-HELO: mail-wr0-f196.google.com Received: from mail-wr0-f196.google.com (HELO mail-wr0-f196.google.com) (209.85.128.196) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Thu, 07 Dec 2017 10:01:56 +0000 Received: by mail-wr0-f196.google.com with SMTP id h1so6794602wre.12 for ; Thu, 07 Dec 2017 02:01:56 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:subject:date:message-id; bh=NGfAjf2xEllcylRZjXW1P+1C1yJgOdfjKg0jeLpXUvA=; b=QVaE12KdCpCQ23XUgd6ZCR6cs1GE5/Rg+/ot3BMG4+e81jF2rGdUNKqQQl8XWyjB3H w48H/jYKcW5h9FLRDbgyjHU4RWlWkarlJSIT2McZCbqPZJF0fG8JV1PchYa1XD4Miuoo Mce7e/OBUwFp/oIetLs42jRjdhGG80e6RJWJpuFEATAcWhQrPHYmkd6KBxv++1+S2VvN a+8sDe6mFmBzv5pHilmZzBtH+nEO0W/Ul3ZHvFqTARpX4RB7Wwu3MSkBpRy+9NqczFIC p5wO2WkE/hetFQCHYgjeWnH7FPUyU0UFkXSceqvxfyOV31IthWkJEJkpyE++6+uR4j+v JirA== X-Gm-Message-State: AJaThX4NKdaOnCkZisK04tmyyN/kKubqBYhHioI+Df6nUPRCh+RFqQ+q ibUjm4Vw6L4ae14Fn+3CAZw9Sw== X-Google-Smtp-Source: AGs4zMaNzhHD7knZxrWb9XU6IGVJv5ki77oGZkZcDLtgTBV+8CH3unBKvNxaw3SgYH/+R3a4zEDOPw== X-Received: by 10.223.184.47 with SMTP id h44mr21021683wrf.11.1512640913982; Thu, 07 Dec 2017 02:01:53 -0800 (PST) Received: from E107787-LIN.cambridge.arm.com (static.42.136.251.148.clients.your-server.de. [148.251.136.42]) by smtp.gmail.com with ESMTPSA id t19sm5467091wrb.58.2017.12.07.02.01.53 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Thu, 07 Dec 2017 02:01:53 -0800 (PST) From: Yao Qi X-Google-Original-From: Yao Qi To: gdb-patches@sourceware.org Subject: [PATCH] Initialize target description early in IPA Date: Thu, 7 Dec 2017 10:01:50 +0000 Message-Id: <1512640910-26090-1-git-send-email-yao.qi@linaro.org> X-IsSubscribed: yes Target descriptions are allocated lazily, that is fine in GDBserver, but it is not safe to call malloc in gdb_collect in IPA, because we can set a fast tracepoint in malloc, and when the tracepoint is hit, gdb_collect/malloc is called, deadlock or memory corruption may be triggered. #0 0xf7cfc200 in malloc () #1 0xf7efdc07 in operator new(unsigned int) () #2 0xf7ef7636 in allocate_target_description() () #3 0xf7efcbe1 in i386_create_target_description(unsigned long long, bool) () #4 0xf7efb474 in i386_linux_read_description(unsigned long long) () #5 0xf7efb190 in get_ipa_tdesc(int) () #6 0xf7ef9baa in gdb_collect () The fix is to initialize all target descriptions earlier, when the IPA is loaded. In order to guarantee malloc is not called in IPA in gdb_collect, I change the test to set a breakpoint on malloc, if IPA gdb_collect calls malloc, program will hit the breakpoint, and test fail. continue Continuing. Thread 1 "" hit Breakpoint 5, 0xf7cfc200 in malloc () (gdb) FAIL: gdb.trace/ftrace.exp: advance through tracing gdb/gdbserver: 2017-12-07 Yao Qi * linux-aarch64-ipa.c (initialize_low_tracepoint): Call aarch64_linux_read_description. * linux-amd64-ipa.c (idx2mask): New array. (get_ipa_tdesc): Move idx2mask out. (initialize_low_tracepoint): Initialize target descriptions. * linux-i386-ipa.c (idx2mask): New array. (get_ipa_tdesc): Move idx2mask out. (initialize_low_tracepoint): Initialize target descriptions. gdb/testsuite: 2017-12-07 Yao Qi * gdb.trace/ftrace.exp (run_trace_experiment): Set breakpoint on malloc and catch syscall. --- gdb/gdbserver/linux-aarch64-ipa.c | 1 + gdb/gdbserver/linux-amd64-ipa.c | 32 +++++++++++++++++++++----------- gdb/gdbserver/linux-i386-ipa.c | 25 +++++++++++++------------ gdb/testsuite/gdb.trace/ftrace.exp | 33 ++++++++++++++++++++++++++++++--- 4 files changed, 65 insertions(+), 26 deletions(-) diff --git a/gdb/gdbserver/linux-aarch64-ipa.c b/gdb/gdbserver/linux-aarch64-ipa.c index b240100..c39b7ec 100644 --- a/gdb/gdbserver/linux-aarch64-ipa.c +++ b/gdb/gdbserver/linux-aarch64-ipa.c @@ -204,4 +204,5 @@ alloc_jump_pad_buffer (size_t size) void initialize_low_tracepoint (void) { + aarch64_linux_read_description (); } diff --git a/gdb/gdbserver/linux-amd64-ipa.c b/gdb/gdbserver/linux-amd64-ipa.c index 85d0d45..5209dab 100644 --- a/gdb/gdbserver/linux-amd64-ipa.c +++ b/gdb/gdbserver/linux-amd64-ipa.c @@ -169,6 +169,19 @@ supply_static_tracepoint_registers (struct regcache *regcache, #endif /* HAVE_UST */ +#if !defined __ILP32__ +/* Map the tdesc index to xcr0 mask. */ +uint64_t idx2mask[X86_TDESC_LAST] = { + X86_XSTATE_X87_MASK, + X86_XSTATE_SSE_MASK, + X86_XSTATE_AVX_MASK, + X86_XSTATE_MPX_MASK, + X86_XSTATE_AVX_MPX_MASK, + X86_XSTATE_AVX_AVX512_MASK, + X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, +}; +#endif + /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ @@ -194,17 +207,6 @@ get_ipa_tdesc (int idx) break; } #else - /* Map the tdesc index to xcr0 mask. */ - uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, - }; - return amd64_linux_read_description (idx2mask[idx], false); #endif @@ -276,4 +278,12 @@ alloc_jump_pad_buffer (size_t size) void initialize_low_tracepoint (void) { +#if defined __ILP32__ + amd64_linux_read_description (X86_XSTATE_SSE_MASK, true); + amd64_linux_read_description (X86_XSTATE_AVX_MASK, true); + amd64_linux_read_description (X86_XSTATE_AVX_AVX512_MASK, true); +#else + for (auto i = 0; i < X86_TDESC_LAST; i++) + amd64_linux_read_description (idx2mask[i], false); +#endif } diff --git a/gdb/gdbserver/linux-i386-ipa.c b/gdb/gdbserver/linux-i386-ipa.c index 785a63e..8ff91b8 100644 --- a/gdb/gdbserver/linux-i386-ipa.c +++ b/gdb/gdbserver/linux-i386-ipa.c @@ -245,6 +245,17 @@ initialize_fast_tracepoint_trampoline_buffer (void) } } +/* Map the tdesc index to xcr0 mask. */ +uint64_t idx2mask[X86_TDESC_LAST] = { + X86_XSTATE_X87_MASK, + X86_XSTATE_SSE_MASK, + X86_XSTATE_AVX_MASK, + X86_XSTATE_MPX_MASK, + X86_XSTATE_AVX_MPX_MASK, + X86_XSTATE_AVX_AVX512_MASK, + X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, +}; + /* Return target_desc to use for IPA, given the tdesc index passed by gdbserver. */ @@ -256,18 +267,6 @@ get_ipa_tdesc (int idx) internal_error (__FILE__, __LINE__, "unknown ipa tdesc index: %d", idx); } - - /* Map the tdesc index to xcr0 mask. */ - uint64_t idx2mask[X86_TDESC_LAST] = { - X86_XSTATE_X87_MASK, - X86_XSTATE_SSE_MASK, - X86_XSTATE_AVX_MASK, - X86_XSTATE_MPX_MASK, - X86_XSTATE_AVX_MPX_MASK, - X86_XSTATE_AVX_AVX512_MASK, - X86_XSTATE_AVX_MPX_AVX512_PKU_MASK, - }; - return i386_linux_read_description (idx2mask[idx]); } @@ -290,4 +289,6 @@ void initialize_low_tracepoint (void) { initialize_fast_tracepoint_trampoline_buffer (); + for (auto i = 0; i < X86_TDESC_LAST; i++) + i386_linux_read_description (idx2mask[i]); } diff --git a/gdb/testsuite/gdb.trace/ftrace.exp b/gdb/testsuite/gdb.trace/ftrace.exp index b862680..6f24256 100644 --- a/gdb/testsuite/gdb.trace/ftrace.exp +++ b/gdb/testsuite/gdb.trace/ftrace.exp @@ -63,9 +63,36 @@ proc run_trace_experiment {} { gdb_test_no_output "tstart" "start trace experiment" - gdb_test "continue" \ - ".*Breakpoint \[0-9\]+, end .*" \ - "advance through tracing" + # Fast tracepoint can be set in signal handler, so gdb_collect in + # IPA shouldn't call any non-async-signal-safe functions. It is + # impractical to list all non-async-signal-safe functions, and set + # breakpoints on them, so choose malloc only in this test. + gdb_test "b -q malloc" + + # Performance-wise, gdb_collect in IPA shouldn't call any syscall + # in order to keep fast tracepoint fast enough. + global gdb_prompt + set test "catch syscall" + gdb_test_multiple $test $test { + -re "The feature \'catch syscall\' is not supported.*\r\n$gdb_prompt $" { + } + -re ".*$gdb_prompt $" { + pass $test + } + } + + global decimal + set test "advance through tracing" + gdb_test_multiple "continue" $test { + -re "Thread 2 .* hit Catchpoint $decimal \\(call to syscall .*\\).*\r\n$gdb_prompt $" { + # IPA starts a helper thread, which calls accept. Ignore it. + send_gdb "continue\n" + exp_continue + } + -re "Breakpoint \[0-9\]+, end .*$gdb_prompt $" { + pass $test + } + } gdb_test "tstatus" ".*Trace .*" "check on trace status"