All of lore.kernel.org
 help / color / mirror / Atom feed
From: Nicholas Piggin <npiggin@gmail.com>
To: qemu-devel@nongnu.org
Cc: "Nicholas Piggin" <npiggin@gmail.com>,
	"Pavel Dovgalyuk" <Pavel.Dovgalyuk@ispras.ru>,
	"Philippe Mathieu-Daudé" <philmd@linaro.org>,
	"Richard Henderson" <richard.henderson@linaro.org>,
	"Alex Bennée" <alex.bennee@linaro.org>,
	"Paolo Bonzini" <pbonzini@redhat.com>,
	"John Snow" <jsnow@redhat.com>, "Cleber Rosa" <crosa@redhat.com>,
	"Wainer dos Santos Moschetta" <wainersm@redhat.com>,
	"Beraldo Leal" <bleal@redhat.com>,
	"Michael Tokarev" <mjt@tls.msk.ru>
Subject: [PATCH v5 16/24] tests/avocado: reverse_debugging.py verify addresses between record and replay
Date: Tue, 19 Mar 2024 01:46:13 +1000	[thread overview]
Message-ID: <20240318154621.2361161-17-npiggin@gmail.com> (raw)
In-Reply-To: <20240318154621.2361161-1-npiggin@gmail.com>

gdb is only attached in the replay phase, and the recorded trace is
single-stepped to build up an address map that is used to then verify
reverse-step and reverse-continue.

This change attaches gdb in the record phase, and builds the address
map by single-stepping the live machine. That address map is then
verified by single-stepping the replay, and then used for the same
reverse debug verification.

This is a stronger verifcation between the record and the replay phases,
as well as providing some testing of gdb operation in record mode.

x86_64 has problems with taking the first step, it does not advance the
icount, so a workaround is added for it.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
 tests/avocado/reverse_debugging.py | 68 +++++++++++++++++++++++++-----
 1 file changed, 58 insertions(+), 10 deletions(-)

diff --git a/tests/avocado/reverse_debugging.py b/tests/avocado/reverse_debugging.py
index e76cf54ad7..8a9578ad5a 100644
--- a/tests/avocado/reverse_debugging.py
+++ b/tests/avocado/reverse_debugging.py
@@ -35,6 +35,10 @@ class ReverseDebugging(LinuxKernelTest):
     STEPS = 10
     endian_is_le = True
 
+    # If first_step_workaround is true, check whether the first step moved
+    # icount, and if not then step again.
+    first_step_workaround = False
+
     def run_vm(self, record, shift, args, replay_path, image_path, port):
         logger = logging.getLogger('replay')
         vm = self.get_vm()
@@ -45,7 +49,7 @@ def run_vm(self, record, shift, args, replay_path, image_path, port):
         else:
             logger.info('replaying the execution...')
             mode = 'replay'
-            vm.add_args('-gdb', 'tcp::%d' % port, '-S')
+        vm.add_args('-gdb', 'tcp::%d' % port, '-S')
         vm.add_args('-icount', 'shift=%s,rr=%s,rrfile=%s,rrsnapshot=init' %
                     (shift, mode, replay_path),
                     '-net', 'none')
@@ -87,6 +91,17 @@ def check_pc(self, g, addr):
         if pc != addr:
             self.fail('Invalid PC (read %x instead of %x)' % (pc, addr))
 
+    @staticmethod
+    def gdb_cont(g):
+        g.cmd(b'c')
+
+    @staticmethod
+    def gdb_cont_nowait(g):
+        # The avocado GDBRemote does not have a good way to disconnect or
+        # continue without waiting for a reply, so open-code our own here.
+        data = g.encode(b'c')
+        g._socket.send(data)
+
     @staticmethod
     def gdb_step(g):
         g.cmd(b's', b'T05thread:01;')
@@ -117,11 +132,40 @@ def reverse_debugging(self, shift=7, args=None):
         replay_path = os.path.join(self.workdir, 'replay.bin')
         port = find_free_port()
 
+        steps = []
+
         # record the log
         vm = self.run_vm(True, shift, args, replay_path, image_path, port)
+        logger.info('connecting to gdbstub')
+        g = gdb.GDBRemote('127.0.0.1', port, False, False)
+        g.connect()
+        r = g.cmd(b'qSupported')
+        if b'qXfer:features:read+' in r:
+            g.cmd(b'qXfer:features:read:target.xml:0,ffb')
+
+        if self.vm_get_icount(vm) != 0:
+            self.fail('icount does not start at zero')
+
+        # save the addresses of the first STEPS instructions executed
+        logger.info('stepping forward')
+        for i in range(self.STEPS):
+            pc = self.get_pc(g)
+            logger.info('saving position %x' % pc)
+            steps.append(pc)
+            self.gdb_step(g)
+            if self.first_step_workaround and i == 0 and self.vm_get_icount(vm) == 0:
+                logger.warn('failed to take first step, stepping again')
+                self.gdb_step(g)
+        if self.vm_get_icount(vm) != self.STEPS:
+            self.fail('icount (%d) does not match number of instructions stepped' % self.vm_get_icount(vm))
+
+        logger.info('continue running')
+        self.gdb_cont_nowait(g)
+
         while self.vm_get_icount(vm) <= self.STEPS:
             pass
         last_icount = self.vm_get_icount(vm)
+        logger.info('shutdown...')
         vm.shutdown()
 
         logger.info("recorded log with %s+ steps" % last_icount)
@@ -139,23 +183,23 @@ def reverse_debugging(self, shift=7, args=None):
         if b'ReverseContinue+' not in r:
             self.fail('Reverse continue is not supported by QEMU')
 
+        # Try single stepping
         logger.info('stepping forward')
-        steps = []
-        # record first instruction addresses
-        for _ in range(self.STEPS):
-            pc = self.get_pc(g)
-            logger.info('saving position %x' % pc)
-            steps.append(pc)
+        for addr in steps:
+            # verify addresses match what initial execution saw
+            self.check_pc(g, addr)
             self.gdb_step(g)
+            logger.info('found position %x' % addr)
 
-        # visit the recorded instruction in reverse order
+        # Try reverse stepping
         logger.info('stepping backward')
         for addr in steps[::-1]:
             self.gdb_bstep(g)
+            # verify addresses match what initial execution saw
             self.check_pc(g, addr)
             logger.info('found position %x' % addr)
 
-        # visit the recorded instruction in forward order
+        # Step forward again
         logger.info('stepping forward')
         for addr in steps:
             self.check_pc(g, addr)
@@ -175,7 +219,7 @@ def reverse_debugging(self, shift=7, args=None):
         # continue - will return after pausing
         # This could stop at the end and get a T02 return, or by
         # re-executing one of the breakpoints and get a T05 return.
-        g.cmd(b'c')
+        self.gdb_cont(g)
         if self.vm_get_icount(vm) == last_icount - 1:
             logger.info('reached the end (icount %s)' % (last_icount - 1))
         else:
@@ -201,6 +245,10 @@ class ReverseDebugging_X86_64(ReverseDebugging):
 
     REG_PC = 0x10
     REG_CS = 0x12
+
+    # The initial step does not change pc on x86 for some reason.
+    first_step_workaround = True
+
     def get_pc(self, g):
         return self.get_reg_le(g, self.REG_PC) \
             + self.get_reg_le(g, self.REG_CS) * 0x10
-- 
2.42.0



  parent reply	other threads:[~2024-03-18 15:48 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-03-18 15:45 [PATCH v5 00/24] replay: fixes and new test cases Nicholas Piggin
2024-03-18 15:45 ` [PATCH v5 01/24] scripts/replay-dump.py: Update to current rr record format Nicholas Piggin
2024-03-18 15:45 ` [PATCH v5 02/24] scripts/replay-dump.py: rejig decoders in event number order Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 03/24] tests/avocado: excercise scripts/replay-dump.py in replay tests Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 04/24] replay: allow runstate shutdown->running when replaying trace Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 05/24] Revert "replay: stop us hanging in rr_wait_io_event" Nicholas Piggin
2024-03-19 17:58   ` Alex Bennée
2024-03-18 15:46 ` [PATCH v5 06/24] tests/avocado: replay_kernel.py add x86-64 q35 machine test Nicholas Piggin
2024-03-19 17:58   ` Alex Bennée
2024-03-18 15:46 ` [PATCH v5 07/24] chardev: set record/replay on the base device of a muxed device Nicholas Piggin
2024-03-19 19:17   ` Alex Bennée
2024-03-18 15:46 ` [PATCH v5 08/24] replay: Fix migration use of clock Nicholas Piggin
2024-03-19 20:40   ` Alex Bennée
2024-03-20  4:48     ` Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 09/24] replay: Fix migration replay_mutex locking Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 10/24] virtio-net: Use replay_schedule_bh_event for bhs that affect machine state Nicholas Piggin
2024-03-19 20:41   ` Alex Bennée
2024-04-05  6:53   ` Pavel Dovgalyuk
2024-03-18 15:46 ` [PATCH v5 11/24] virtio-net: Use virtual time for RSC timers Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 12/24] savevm: Fix load_snapshot error path crash Nicholas Piggin
2024-03-19 20:49   ` Alex Bennée
2024-03-18 15:46 ` [PATCH v5 13/24] tests/avocado: replay_linux.py remove the timeout expected guards Nicholas Piggin
2024-03-19 17:57   ` Alex Bennée
2024-03-20  4:34     ` Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 14/24] tests/avocado: reverse_debugging.py mark aarch64 and pseries as not flaky Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 15/24] tests/avocado: reverse_debugging.py add test for x86-64 q35 machine Nicholas Piggin
2024-03-19 20:50   ` Alex Bennée
2024-03-18 15:46 ` Nicholas Piggin [this message]
2024-03-18 15:46 ` [PATCH v5 17/24] tests/avocado: reverse_debugging.py stop VM before sampling icount Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 18/24] tests/avocado: reverse_debugging reverse-step at the end of the trace Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 19/24] tests/avocado: reverse_debugging.py add snapshot testing Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 20/24] replay: simple auto-snapshot mode for record Nicholas Piggin
2024-04-05  6:56   ` Pavel Dovgalyuk
2024-03-18 15:46 ` [PATCH v5 21/24] tests/avocado: reverse_debugging.py test auto-snapshot mode Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 22/24] target/ppc: fix timebase register reset state Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 23/24] spapr: Fix vpa dispatch count for record-replay Nicholas Piggin
2024-03-18 15:46 ` [PATCH v5 24/24] tests/avocado: replay_linux.py add ppc64 pseries test Nicholas Piggin

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=20240318154621.2361161-17-npiggin@gmail.com \
    --to=npiggin@gmail.com \
    --cc=Pavel.Dovgalyuk@ispras.ru \
    --cc=alex.bennee@linaro.org \
    --cc=bleal@redhat.com \
    --cc=crosa@redhat.com \
    --cc=jsnow@redhat.com \
    --cc=mjt@tls.msk.ru \
    --cc=pbonzini@redhat.com \
    --cc=philmd@linaro.org \
    --cc=qemu-devel@nongnu.org \
    --cc=richard.henderson@linaro.org \
    --cc=wainersm@redhat.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
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.