LKML Archive on lore.kernel.org
 help / color / Atom feed
From: Andy Lutomirski <luto@amacapital.net>
To: "Jiri Olsa" <jolsa@redhat.com>, "Ingo Molnar" <mingo@redhat.com>,
	"Peter Zijlstra" <peterz@infradead.org>,
	"秦承刚(承刚)" <chenggang.qcg@alibaba-inc.com>
Cc: "Andy Lutomirski" <luto@amacapital.net>,
	"Stephane Eranian" <eranian@google.com>,
	root <chenggang.qin@gmail.com>,
	"Andrew Morton" <akpm@linux-foundation.org>,
	"秦承刚(承刚)" <chenggang.qcg@taobao.com>,
	"Wu Fengguang" <fengguang.wu@intel.com>,
	"Namhyung Kim" <namhyung@gmail.com>,
	"Mike Galbraith" <efault@gmx.de>,
	"Arjan van de Ven" <arjan@linux.intel.com>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	"David Ahern" <dsahern@gmail.com>,
	"Paul Mackerras" <paulus@samba.org>,
	"Yanmin Zhang" <yanmin.zhang@intel.com>
Subject: [PATCH 2/2] x86_64, perf: Improve user regs sampling
Date: Sun,  4 Jan 2015 10:36:20 -0800
Message-ID: <243560c26ff0f739978e2459e203f6515367634d.1420396372.git.luto@amacapital.net> (raw)
In-Reply-To: <cover.1420396372.git.luto@amacapital.net>
In-Reply-To: <cover.1420396372.git.luto@amacapital.net>

Perf reports user regs for kernel-mode samples so that samples can
be backtraced through user code.  The old code was very broken in
syscall context, resulting in useless backtraces.

The new code, in contrast, is still dangerously racy, but it should
at least work most of the time.

Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
 arch/x86/kernel/perf_regs.c | 78 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 76 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/perf_regs.c b/arch/x86/kernel/perf_regs.c
index 3bbbb1a4fb52..781861cc5ee8 100644
--- a/arch/x86/kernel/perf_regs.c
+++ b/arch/x86/kernel/perf_regs.c
@@ -115,7 +115,81 @@ void perf_get_regs_user(struct perf_regs *regs_user,
 			struct pt_regs *regs,
 			struct pt_regs *regs_user_copy)
 {
-	regs_user->regs = task_pt_regs(current);
-	regs_user->abi = perf_reg_abi(current);
+	struct pt_regs *user_regs = task_pt_regs(current);
+
+	/*
+	 * If we're in an NMI that interrupted task_pt_regs setup, then
+	 * we can't sample user regs at all.  This check isn't really
+	 * sufficient, though, as we could be in an NMI inside an interrupt
+	 * that happened during task_pt_regs setup.
+	 */
+	if (regs->sp > (unsigned long)&user_regs->r11 &&
+	    regs->sp <= (unsigned long)(user_regs + 1)) {
+		regs_user->abi = PERF_SAMPLE_REGS_ABI_NONE;
+		regs_user->regs = NULL;
+		return;
+	}
+
+	/*
+	 * RIP, flags, and the argument registers are usually saved.
+	 * orig_ax is probably okay, too.
+	 */
+	regs_user_copy->ip = user_regs->ip;
+	regs_user_copy->cx = user_regs->cx;
+	regs_user_copy->dx = user_regs->dx;
+	regs_user_copy->si = user_regs->si;
+	regs_user_copy->di = user_regs->di;
+	regs_user_copy->r8 = user_regs->r8;
+	regs_user_copy->r9 = user_regs->r9;
+	regs_user_copy->r10 = user_regs->r10;
+	regs_user_copy->r11 = user_regs->r11;
+	regs_user_copy->orig_ax = user_regs->orig_ax;
+	regs_user_copy->flags = user_regs->flags;
+
+	/*
+	 * Don't even try to report the "rest" regs.
+	 */
+	regs_user_copy->bx = -1;
+	regs_user_copy->bp = -1;
+	regs_user_copy->r12 = -1;
+	regs_user_copy->r13 = -1;
+	regs_user_copy->r14 = -1;
+	regs_user_copy->r15 = -1;
+
+	/*
+	 * For this to be at all useful, we need a reasonable guess for
+	 * sp and the ABI.  Be careful: we're in NMI context, and we're
+	 * considering current to be the current task, so we should
+	 * be careful not to look at any other percpu variables that might
+	 * change during context switches.
+	 */
+	if (IS_ENABLED(CONFIG_IA32_EMULATION) &&
+	    task_thread_info(current)->status & TS_COMPAT) {
+		/* Easy case: we're in a compat syscall. */
+		regs_user->abi = PERF_SAMPLE_REGS_ABI_32;
+		regs_user_copy->sp = user_regs->sp;
+		regs_user_copy->cs = user_regs->cs;
+		regs_user_copy->ss = user_regs->ss;
+	} else if (user_regs->orig_ax != -1) {
+		/*
+		 * We're probably in a 64-bit syscall.
+		 * Warning: this code is severely racy.  At least it's better
+		 * than just blindly copying user_regs.
+		 */
+		regs_user->abi = PERF_SAMPLE_REGS_ABI_64;
+		regs_user_copy->sp = this_cpu_read(old_rsp);
+		regs_user_copy->cs = __USER_CS;
+		regs_user_copy->ss = __USER_DS;
+		regs_user_copy->cx = -1;  /* usually contains garbage */
+	} else {
+		/* We're probably in an interrupt or exception. */
+		regs_user->abi = user_64bit_mode(user_regs) ?
+			PERF_SAMPLE_REGS_ABI_64 : PERF_SAMPLE_REGS_ABI_32;
+		regs_user_copy->sp = user_regs->sp;
+		regs_user_copy->cs = user_regs->cs;
+		regs_user_copy->ss = user_regs->ss;
+	}
+
+	regs_user->regs = regs_user_copy;
 }
 #endif /* CONFIG_X86_32 */
-- 
2.1.0


  parent reply index

Thread overview: 23+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-12-23  6:22 [PATCH] perf core: Use KSTK_ESP() instead of pt_regs->sp while output user regs root
2014-12-23  8:30 ` Andy Lutomirski
     [not found] ` <c027bde0-5f4f-441f-8d45-3e7f6f702231@alibaba-inc.com>
2014-12-25 15:48   ` 答复:[PATCH] " Andy Lutomirski
2014-12-25 16:21     ` Andy Lutomirski
2014-12-30 19:03     ` Peter Zijlstra
2014-12-30 23:29       ` Andy Lutomirski
2014-12-31  2:00         ` Andy Lutomirski
2015-01-02 16:11           ` Jan Beulich
2015-01-02 18:03             ` Andy Lutomirski
2015-01-05  8:47               ` Jan Beulich
2015-01-04 16:10       ` Jiri Olsa
2015-01-04 17:18         ` Andy Lutomirski
2015-01-04 17:41           ` Jiri Olsa
2015-01-04 18:36             ` [PATCH 0/2] perf: Improve user regs sampling Andy Lutomirski
2015-01-04 18:36               ` [PATCH 1/2] perf: Move task_pt_regs sampling into arch code Andy Lutomirski
2015-01-05 14:07                 ` Peter Zijlstra
2015-01-05 16:13                   ` Andy Lutomirski
2015-01-05 16:44                     ` Peter Zijlstra
2015-01-05 18:28                       ` Andy Lutomirski
2015-01-09 12:32                 ` [tip:perf/urgent] " tip-bot for Andy Lutomirski
2015-01-04 18:36               ` Andy Lutomirski [this message]
2015-01-09 12:32                 ` [tip:perf/urgent] perf/x86_64: Improve user regs sampling tip-bot for Andy Lutomirski
2015-01-05 10:46               ` [PATCH 0/2] perf: " Jiri Olsa

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=243560c26ff0f739978e2459e203f6515367634d.1420396372.git.luto@amacapital.net \
    --to=luto@amacapital.net \
    --cc=akpm@linux-foundation.org \
    --cc=arjan@linux.intel.com \
    --cc=chenggang.qcg@alibaba-inc.com \
    --cc=chenggang.qcg@taobao.com \
    --cc=chenggang.qin@gmail.com \
    --cc=dsahern@gmail.com \
    --cc=efault@gmx.de \
    --cc=eranian@google.com \
    --cc=fengguang.wu@intel.com \
    --cc=jolsa@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=namhyung@gmail.com \
    --cc=paulus@samba.org \
    --cc=peterz@infradead.org \
    --cc=yanmin.zhang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

LKML Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/lkml/0 lkml/git/0.git
	git clone --mirror https://lore.kernel.org/lkml/1 lkml/git/1.git
	git clone --mirror https://lore.kernel.org/lkml/2 lkml/git/2.git
	git clone --mirror https://lore.kernel.org/lkml/3 lkml/git/3.git
	git clone --mirror https://lore.kernel.org/lkml/4 lkml/git/4.git
	git clone --mirror https://lore.kernel.org/lkml/5 lkml/git/5.git
	git clone --mirror https://lore.kernel.org/lkml/6 lkml/git/6.git
	git clone --mirror https://lore.kernel.org/lkml/7 lkml/git/7.git
	git clone --mirror https://lore.kernel.org/lkml/8 lkml/git/8.git
	git clone --mirror https://lore.kernel.org/lkml/9 lkml/git/9.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 lkml lkml/ https://lore.kernel.org/lkml \
		linux-kernel@vger.kernel.org
	public-inbox-index lkml

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-kernel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git