[PATCH 1/5] gdb/amd-dbgapi: remove unnecessary AMD_DBGAPI_EVENT_KIND_NONE argument

[PATCH 1/5] gdb/amd-dbgapi: remove unnecessary AMD_DBGAPI_EVENT_KIND_NONE argument

[Simon Marchi]

Rely on the default value.

Change-Id: I08c683de005806c5c5d29ed7f9b0c6de81b49a01
---
 gdb/amd-dbgapi-target.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index 8e067b66bd86..c11e834aa408 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -1473,7 +1473,7 @@ amd_dbgapi_target::wait (ptid_t ptid, struct target_waitstatus *ws,
       /* Drain the events for the current inferior from the amd_dbgapi and
 	 preserve the ordering.  */
       auto info = get_amd_dbgapi_inferior_info (current_inferior ());
-      process_event_queue (info->process_id, AMD_DBGAPI_EVENT_KIND_NONE);
+      process_event_queue (info->process_id);
 
       std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
       if (event_ptid == minus_one_ptid)

base-commit: f118937c181f008f78496d299873c879648d2d1a
-- 
2.49.0

[PATCH 2/5] gdb/amd-dbgapi: add assert in require_forward_progress

[Simon Marchi]

I didn't have a problem in this area, but it seems to me that this
pre-condition should always hold.  We should only disable forward
progress requirement if the target says it's ok to do so.  Otherwise, we
could get in a situation where we wait for events from amd-dbgapi, which
will never arrive, because amd-dbgapi didn't actually resume things.

Change-Id: Ifc49f55c7874924b7c47888b8391a07a01d960fc
---
 gdb/amd-dbgapi-target.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index c11e834aa408..8fd8bcf5ae49 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -450,6 +450,12 @@ static void
 require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
 			  bool require)
 {
+  /* If we try to disable forward progress requirement but the target expects
+     resumed threads to be committed to the target, we could wait for events
+     that will never arrive.  */
+  if (!require)
+    gdb_assert (!proc_target->commit_resumed_state);
+
   for (inferior *inf : all_inferiors (proc_target))
     {
       if (ptid != minus_one_ptid && inf->pid != ptid.pid ())
-- 
2.49.0

[PATCH 3/5] gdb/amd-dbgapi: pass amd_dbgapi_inferior_info to process_event_queue

[Simon Marchi]

A following patch will make process_event_queue access a field of
amd_dbgapi_inferior_info.  Prepare for this by making
process_event_queue accept an amd_dbgapi_inferior_info object, instead
of a process id.

Change-Id: I9adc491dd1ff64ff74c40aa7662fffb11bd8332b
---
 gdb/amd-dbgapi-target.c | 25 +++++++++----------------
 1 file changed, 9 insertions(+), 16 deletions(-)

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index 8fd8bcf5ae49..1ec20d1874ec 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -234,7 +234,7 @@ struct amd_dbgapi_inferior_info
 };
 
 static amd_dbgapi_event_id_t process_event_queue
-  (amd_dbgapi_process_id_t process_id,
+  (amd_dbgapi_inferior_info &info,
    amd_dbgapi_event_kind_t until_event_kind = AMD_DBGAPI_EVENT_KIND_NONE);
 
 static const target_info amd_dbgapi_target_info = {
@@ -564,8 +564,7 @@ amd_dbgapi_target_breakpoint::check_status (struct bpstat *bs)
   /* If the action is AMD_DBGAPI_BREAKPOINT_ACTION_HALT, we need to wait until
      a breakpoint resume event for this breakpoint_id is seen.  */
   amd_dbgapi_event_id_t resume_event_id
-    = process_event_queue (info->process_id,
-			   AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
+    = process_event_queue (*info, AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
 
   /* We should always get a breakpoint_resume event after processing all
      events generated by reporting the breakpoint hit.  */
@@ -1337,28 +1336,22 @@ event_kind_str (amd_dbgapi_event_kind_t kind)
   gdb_assert_not_reached ("unhandled amd_dbgapi_event_kind_t value");
 }
 
-/* Drain the dbgapi event queue of a given process_id, or of all processes if
-   process_id is AMD_DBGAPI_PROCESS_NONE.  Stop processing the events if an
-   event of a given kind is requested and `process_id` is not
-   AMD_DBGAPI_PROCESS_NONE.  Wave stop events that are not returned are queued
-   into their inferior's amd_dbgapi_inferior_info pending wave events. */
+/* Drain the dbgapi event queue of a given process_id.  Stop processing the
+   events if an event of a given kind is requested (not AMD_DBGAPI_EVENT_NONE).
+   Wave stop events that are not returned are queued into their inferior's
+   amd_dbgapi_inferior_info pending wave events. */
 
 static amd_dbgapi_event_id_t
-process_event_queue (amd_dbgapi_process_id_t process_id,
+process_event_queue (amd_dbgapi_inferior_info &info,
 		     amd_dbgapi_event_kind_t until_event_kind)
 {
-  /* An event of a given type can only be requested from a single
-     process_id.  */
-  gdb_assert (until_event_kind == AMD_DBGAPI_EVENT_KIND_NONE
-	      || process_id != AMD_DBGAPI_PROCESS_NONE);
-
   while (true)
     {
       amd_dbgapi_event_id_t event_id;
       amd_dbgapi_event_kind_t event_kind;
 
       amd_dbgapi_status_t status
-	= amd_dbgapi_process_next_pending_event (process_id, &event_id,
+	= amd_dbgapi_process_next_pending_event (info.process_id, &event_id,
 						 &event_kind);
 
       if (status != AMD_DBGAPI_STATUS_SUCCESS)
@@ -1479,7 +1472,7 @@ amd_dbgapi_target::wait (ptid_t ptid, struct target_waitstatus *ws,
       /* Drain the events for the current inferior from the amd_dbgapi and
 	 preserve the ordering.  */
       auto info = get_amd_dbgapi_inferior_info (current_inferior ());
-      process_event_queue (info->process_id);
+      process_event_queue (*info);
 
       std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
       if (event_ptid == minus_one_ptid)
-- 
2.49.0

[PATCH 4/5] gdb/amd-dbgapi: factor out require_forward_progress overload to target one inferior

[Simon Marchi]

A following patch will want to call require_forward_progress for a given
inferior.  Extract a new require_forward_progress overload from the
existing require_forward_progress function that targets a specific
inferior.

Change-Id: I54f42b83eb8443d4d91747ffbc86eaeb017f1e49
---
 gdb/amd-dbgapi-target.c | 49 +++++++++++++++++++++++------------------
 1 file changed, 28 insertions(+), 21 deletions(-)

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index 1ec20d1874ec..e86d7f34d0f6 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -443,6 +443,32 @@ async_event_handler_mark ()
   mark_async_event_handler (amd_dbgapi_async_event_handler);
 }
 
+/* Set forward progress requirement to REQUIRE for inferior INFO.  */
+
+static void
+require_forward_progress (amd_dbgapi_inferior_info &info, bool require)
+{
+  /* If we try to disable forward progress requirement but the target expects
+     resumed threads to be committed to the target, we could wait for events
+     that will never arrive.  */
+  if (!require)
+    gdb_assert (!info.inf->process_target ()->commit_resumed_state);
+
+  gdb_assert (info.process_id != AMD_DBGAPI_PROCESS_NONE);
+
+  /* Don't do unnecessary calls to amd-dbgapi to avoid polluting the logs.  */
+  if (info.forward_progress_required == require)
+    return;
+
+  const auto progress
+    = require ? AMD_DBGAPI_PROGRESS_NORMAL : AMD_DBGAPI_PROGRESS_NO_FORWARD;
+  const auto status
+    = amd_dbgapi_process_set_progress (info.process_id, progress);
+  gdb_assert (status == AMD_DBGAPI_STATUS_SUCCESS);
+
+  info.forward_progress_required = require;
+}
+
 /* Set forward progress requirement to REQUIRE for all processes of PROC_TARGET
    matching PTID.  */
 
@@ -450,12 +476,6 @@ static void
 require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
 			  bool require)
 {
-  /* If we try to disable forward progress requirement but the target expects
-     resumed threads to be committed to the target, we could wait for events
-     that will never arrive.  */
-  if (!require)
-    gdb_assert (!proc_target->commit_resumed_state);
-
   for (inferior *inf : all_inferiors (proc_target))
     {
       if (ptid != minus_one_ptid && inf->pid != ptid.pid ())
@@ -463,21 +483,8 @@ require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
 
       amd_dbgapi_inferior_info *info = get_amd_dbgapi_inferior_info (inf);
 
-      if (info->process_id == AMD_DBGAPI_PROCESS_NONE)
-	continue;
-
-      /* Don't do unnecessary calls to amd-dbgapi to avoid polluting the logs.  */
-      if (info->forward_progress_required == require)
-	continue;
-
-      amd_dbgapi_status_t status
-	= amd_dbgapi_process_set_progress
-	    (info->process_id, (require
-				? AMD_DBGAPI_PROGRESS_NORMAL
-				: AMD_DBGAPI_PROGRESS_NO_FORWARD));
-      gdb_assert (status == AMD_DBGAPI_STATUS_SUCCESS);
-
-      info->forward_progress_required = require;
+      if (info->process_id != AMD_DBGAPI_PROCESS_NONE)
+	require_forward_progress (*info, require);
 
       /* If ptid targets a single inferior and we have found it, no need to
 	 continue.  */
-- 
2.49.0

[PATCH 5/5] gdb/amd-dbgapi: disable forward progress requirement in amd_dbgapi_target_breakpoint::check_status

[Simon Marchi]

ROCgdb handles target events very slowly when running a test case like
this, where a breakpoint is preset on HipTest::vectorADD:

    for (int i=0; i < numDevices; ++i) {
      HIPCHECK(hipSetDevice(i));
      hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock), 0, stream[i],
                        static_cast<const int*>(A_d[i]), static_cast<const int*>(B_d[i]), C_d[i], N);
    }

What happens is:

 - A kernel is launched
 - The internal runtime breakpoint is hit during the second
   hipLaunchKernelGGL call, which causes
   amd_dbgapi_target_breakpoint::check_status to be called
 - Meanwhile, all waves of the kernel hit the breakpoint on vectorADD
 - amd_dbgapi_target_breakpoint::check_status calls process_event_queue,
   which pulls the thousand of breakpoint hit events from the kernel
 - As part of handling the breakpoint hit events, we write the PC of the
   waves that stopped to decrement it.  Because the forward progress
   requirement is not disabled, this causes a suspend/resume of the
   queue each time, which is time-consuming.

The stack trace where this all happens is:

    #32 0x00007ffff6b9abda in amd_dbgapi_write_register (wave_id=..., register_id=..., offset=0, value_size=8, value=0x7fffea9fdcc0) at /home/smarchi/src/amd-dbgapi/src/register.cpp:587
    #33 0x00005555588c0bed in amd_dbgapi_target::store_registers (this=0x55555c7b1d20 <the_amd_dbgapi_target>, regcache=0x507000002240, regno=470) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:2504
    #34 0x000055555a5186a1 in target_store_registers (regcache=0x507000002240, regno=470) at /home/smarchi/src/wt/amd/gdb/target.c:3973
    #35 0x0000555559fab831 in regcache::raw_write (this=0x507000002240, regnum=470, src=...) at /home/smarchi/src/wt/amd/gdb/regcache.c:890
    #36 0x0000555559fabd2b in regcache::cooked_write (this=0x507000002240, regnum=470, src=...) at /home/smarchi/src/wt/amd/gdb/regcache.c:915
    #37 0x0000555559fc3ca5 in regcache::cooked_write<unsigned long, void> (this=0x507000002240, regnum=470, val=140737323456768) at /home/smarchi/src/wt/amd/gdb/regcache.c:850
    #38 0x0000555559fab09a in regcache_cooked_write_unsigned (regcache=0x507000002240, regnum=470, val=140737323456768) at /home/smarchi/src/wt/amd/gdb/regcache.c:858
    #39 0x0000555559fb0678 in regcache_write_pc (regcache=0x507000002240, pc=0x7ffff62bd900) at /home/smarchi/src/wt/amd/gdb/regcache.c:1460
    #40 0x00005555588bb37d in process_one_event (event_id=..., event_kind=AMD_DBGAPI_EVENT_KIND_WAVE_STOP) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:1873
    #41 0x00005555588bbf7b in process_event_queue (process_id=..., until_event_kind=AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:2006
    #42 0x00005555588b1aca in amd_dbgapi_target_breakpoint::check_status (this=0x511000140900, bs=0x50600014ed00) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:890
    #43 0x0000555558c50080 in bpstat_stop_status (aspace=0x5070000061b0, bp_addr=0x7fffed0b9ab0, thread=0x518000026c80, ws=..., stop_chain=0x50600014ed00) at /home/smarchi/src/wt/amd/gdb/breakpoint.c:6126
    #44 0x000055555984f4ff in handle_signal_stop (ecs=0x7fffeaa40ef0) at /home/smarchi/src/wt/amd/gdb/infrun.c:7169
    #45 0x000055555984b889 in handle_inferior_event (ecs=0x7fffeaa40ef0) at /home/smarchi/src/wt/amd/gdb/infrun.c:6621
    #46 0x000055555983eab6 in fetch_inferior_event () at /home/smarchi/src/wt/amd/gdb/infrun.c:4750
    #47 0x00005555597caa5f in inferior_event_handler (event_type=INF_REG_EVENT) at /home/smarchi/src/wt/amd/gdb/inf-loop.c:42
    #48 0x00005555588b838e in handle_target_event (client_data=0x0) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:1513

Fix that performance problem by disabling the forward progress
requirement in amd_dbgapi_target_breakpoint::check_status, before
calling process_event_queue, so that we can process all events
efficiently.

Since the same performance problem could theoritically happen any time
process_event_queue is called with forward progress requirement enabled,
add an assert to ensure that forward progress requirement is disabled
when process_event_queue is invoked.  This makes it necessary to add a
require_forward_progress call to amd_dbgapi_finalize_core_attach.  It
looks a bit strange, since core files don't have execution, but it
doesn't hurt.

Add a test that replicates this scenario.  The test launches a kernel
that hits a breakpoint (with an always false condition) repeatedly.
Meanwhile, the host process loads an unloads a code object, causing
check_status to be called.

Bug: SWDEV-482511
Change-Id: Ida86340d679e6bd8462712953458c07ba3fd49ec
---
 gdb/amd-dbgapi-target.c                       |  6 ++
 .../code-object-load-while-breakpoint-hit.cpp | 70 +++++++++++++++++++
 .../code-object-load-while-breakpoint-hit.exp | 68 ++++++++++++++++++
 3 files changed, 144 insertions(+)
 create mode 100644 gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
 create mode 100644 gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index e86d7f34d0f6..44913bbd7532 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -568,6 +568,8 @@ amd_dbgapi_target_breakpoint::check_status (struct bpstat *bs)
   if (action == AMD_DBGAPI_BREAKPOINT_ACTION_RESUME)
     return;
 
+  require_forward_progress (*info, false);
+
   /* If the action is AMD_DBGAPI_BREAKPOINT_ACTION_HALT, we need to wait until
      a breakpoint resume event for this breakpoint_id is seen.  */
   amd_dbgapi_event_id_t resume_event_id
@@ -1352,6 +1354,10 @@ static amd_dbgapi_event_id_t
 process_event_queue (amd_dbgapi_inferior_info &info,
 		     amd_dbgapi_event_kind_t until_event_kind)
 {
+  /* Pulling events with forward progress required may result in bad
+     performance, make sure it is not required.  */
+  gdb_assert (!info.forward_progress_required);
+
   while (true)
     {
       amd_dbgapi_event_id_t event_id;
diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
new file mode 100644
index 000000000000..e97809335569
--- /dev/null
+++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
@@ -0,0 +1,70 @@
+#ifdef DEVICE
+
+#include <hip/hip_runtime.h>
+
+constexpr unsigned int NUM_BREAKPOINT_HITS = 5;
+
+static __device__ void
+break_here ()
+{
+  asm ("s_nop 0");
+}
+
+extern "C" __global__ void
+kernel ()
+{
+  for (int n = 0; n < NUM_BREAKPOINT_HITS; ++n)
+    break_here ();
+}
+
+#else
+
+#include <hip/hip_runtime.h>
+#include <unistd.h>
+
+constexpr unsigned int NUM_ITEMS_PER_BLOCK = 256;
+constexpr unsigned int NUM_BLOCKS = 128;
+constexpr unsigned int NUM_ITEMS = NUM_ITEMS_PER_BLOCK * NUM_BLOCKS;
+constexpr unsigned int NUM_LOAD_UNLOADS = 5;
+
+#define CHECK(cmd)                                                            \
+  {                                                                           \
+    hipError_t error = cmd;                                                   \
+    if (error != hipSuccess)                                                  \
+      {                                                                       \
+	fprintf (stderr, "error: '%s'(%d) at %s:%d\n",                        \
+		 hipGetErrorString (error), error, __FILE__, __LINE__);       \
+	exit (EXIT_FAILURE);                                                  \
+      }                                                                       \
+  }
+
+int
+main (int argc, const char **argv)
+{
+  if (argc != 2)
+    {
+      fprintf (stderr, "Usage: %s <hip_module_path>\n", argv[0]);
+      return 1;
+    }
+
+  const auto module_path = argv[1];
+  hipModule_t module;
+  CHECK (hipModuleLoad (&module, module_path));
+
+  /* Launch the kernel.  */
+  hipFunction_t function;
+  CHECK (hipModuleGetFunction (&function, module, "kernel"));
+  CHECK (hipModuleLaunchKernel (function, NUM_BLOCKS, 1, 1,
+				NUM_ITEMS_PER_BLOCK, 1, 1, 0, nullptr, nullptr,
+				nullptr));
+
+  /* Load and unload the module many times.  */
+  for (int i = 0; i < NUM_LOAD_UNLOADS; ++i)
+    {
+      hipModule_t dummy_module;
+      CHECK (hipModuleLoad (&dummy_module, module_path));
+      CHECK (hipModuleUnload (dummy_module));
+    }
+}
+
+#endif
diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
new file mode 100644
index 000000000000..b8d003f37032
--- /dev/null
+++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
@@ -0,0 +1,68 @@
+# Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+# This file is part of GDB.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# This test verifies what happens when a code object list update happens at the
+# same time as some wave stop events are reported.  It was added following a
+# performance bug fix, where forward progress requirement disabled when
+# pulling events from amd-dbgapi in amd_dbgapi_target_breakpoint::check_status.
+#
+# The test launches a kernel that hits a breakpoint with an always false
+# condition a certain number of times.  Meanwhile, the host loads and unloads
+# a code object in a loop, causing check_status to be called.  The hope is that
+# check_status, when calling process_event_queue, will pull many WAVE_STOP
+# events from the kernel hitting the breakpoint.
+#
+# Without the appropriate fix (of disabling forward progress requirement in
+# check_status), GDB would hit the newly-added assert in process_event_queue,
+# which verifies that forward progress requirement is disabled.  Even without
+# this assert, the test would likely time out (depending on the actual timeout
+# value).
+
+load_lib rocm.exp
+standard_testfile .cpp
+require allow_hipcc_tests
+
+# Build the host executable.
+if { [build_executable "failed to prepare" \
+	  $testfile $srcfile {debug hip}] == -1 } {
+    return -1
+}
+
+set hipmodule_path [standard_output_file ${testfile}.co]
+
+# Build the kernel object file.
+if { [gdb_compile $srcdir/$subdir/$srcfile \
+	$hipmodule_path object \
+	{ debug hip additional_flags=--genco additional_flags=-DDEVICE } ] != "" } {
+    return -1
+}
+
+proc do_test { } {
+    with_rocm_gpu_lock {
+	clean_restart $::binfile
+	gdb_test_no_output "set args $::hipmodule_path" "set args"
+
+	if { ![runto_main] } {
+		return
+	}
+
+	gdb_test "with breakpoint pending on -- break break_here if 0"
+	gdb_continue_to_end "continue to end" "continue" 1
+    }
+}
+
+do_test
-- 
2.49.0

Re: [PATCH 1/5] gdb/amd-dbgapi: remove unnecessary AMD_DBGAPI_EVENT_KIND_NONE argument

[Lancelot SIX]

On Thu, Jun 05, 2025 at 04:16:24PM -0400, Simon Marchi wrote:
> Rely on the default value.

Hi,

This OK.

Approved-By: Lancelot Six <lancelot.six@amd.com> (amdgpu).

Best,
Lancelot.

> 
> Change-Id: I08c683de005806c5c5d29ed7f9b0c6de81b49a01
> ---
>  gdb/amd-dbgapi-target.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index 8e067b66bd86..c11e834aa408 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -1473,7 +1473,7 @@ amd_dbgapi_target::wait (ptid_t ptid, struct target_waitstatus *ws,
>        /* Drain the events for the current inferior from the amd_dbgapi and
>  	 preserve the ordering.  */
>        auto info = get_amd_dbgapi_inferior_info (current_inferior ());
> -      process_event_queue (info->process_id, AMD_DBGAPI_EVENT_KIND_NONE);
> +      process_event_queue (info->process_id);
>  
>        std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
>        if (event_ptid == minus_one_ptid)
> 
> base-commit: f118937c181f008f78496d299873c879648d2d1a
> -- 
> 2.49.0

Re: [PATCH 2/5] gdb/amd-dbgapi: add assert in require_forward_progress

[Lancelot SIX]

On Thu, Jun 05, 2025 at 04:16:25PM -0400, Simon Marchi wrote:
> I didn't have a problem in this area, but it seems to me that this
> pre-condition should always hold.  We should only disable forward
> progress requirement if the target says it's ok to do so.  Otherwise, we
> could get in a situation where we wait for events from amd-dbgapi, which
> will never arrive, because amd-dbgapi didn't actually resume things.
> 

This is OK.

Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)

Best,
Lancelot

> Change-Id: Ifc49f55c7874924b7c47888b8391a07a01d960fc
> ---
>  gdb/amd-dbgapi-target.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index c11e834aa408..8fd8bcf5ae49 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -450,6 +450,12 @@ static void
>  require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
>  			  bool require)
>  {
> +  /* If we try to disable forward progress requirement but the target expects
> +     resumed threads to be committed to the target, we could wait for events
> +     that will never arrive.  */
> +  if (!require)
> +    gdb_assert (!proc_target->commit_resumed_state);
> +
>    for (inferior *inf : all_inferiors (proc_target))
>      {
>        if (ptid != minus_one_ptid && inf->pid != ptid.pid ())
> -- 
> 2.49.0

Re: [PATCH 3/5] gdb/amd-dbgapi: pass amd_dbgapi_inferior_info to process_event_queue

[Lancelot SIX]

On Thu, Jun 05, 2025 at 04:16:26PM -0400, Simon Marchi wrote:
> A following patch will make process_event_queue access a field of
> amd_dbgapi_inferior_info.  Prepare for this by making
> process_event_queue accept an amd_dbgapi_inferior_info object, instead
> of a process id.
> 
> Change-Id: I9adc491dd1ff64ff74c40aa7662fffb11bd8332b
> ---
>  gdb/amd-dbgapi-target.c | 25 +++++++++----------------
>  1 file changed, 9 insertions(+), 16 deletions(-)
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index 8fd8bcf5ae49..1ec20d1874ec 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -234,7 +234,7 @@ struct amd_dbgapi_inferior_info
>  };
>  
>  static amd_dbgapi_event_id_t process_event_queue
> -  (amd_dbgapi_process_id_t process_id,
> +  (amd_dbgapi_inferior_info &info,
>     amd_dbgapi_event_kind_t until_event_kind = AMD_DBGAPI_EVENT_KIND_NONE);
>  
>  static const target_info amd_dbgapi_target_info = {
> @@ -564,8 +564,7 @@ amd_dbgapi_target_breakpoint::check_status (struct bpstat *bs)
>    /* If the action is AMD_DBGAPI_BREAKPOINT_ACTION_HALT, we need to wait until
>       a breakpoint resume event for this breakpoint_id is seen.  */
>    amd_dbgapi_event_id_t resume_event_id
> -    = process_event_queue (info->process_id,
> -			   AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
> +    = process_event_queue (*info, AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
>  
>    /* We should always get a breakpoint_resume event after processing all
>       events generated by reporting the breakpoint hit.  */
> @@ -1337,28 +1336,22 @@ event_kind_str (amd_dbgapi_event_kind_t kind)
>    gdb_assert_not_reached ("unhandled amd_dbgapi_event_kind_t value");
>  }
>  
> -/* Drain the dbgapi event queue of a given process_id, or of all processes if
> -   process_id is AMD_DBGAPI_PROCESS_NONE.  Stop processing the events if an
> -   event of a given kind is requested and `process_id` is not
> -   AMD_DBGAPI_PROCESS_NONE.  Wave stop events that are not returned are queued
> -   into their inferior's amd_dbgapi_inferior_info pending wave events. */
> +/* Drain the dbgapi event queue of a given process_id.  Stop processing the

It reads odd to have process_id here, as we don't pass it as an explicit
parameter anymore.  Could it say "a given inferior" instead?

> +   events if an event of a given kind is requested (not AMD_DBGAPI_EVENT_NONE).
> +   Wave stop events that are not returned are queued into their inferior's
> +   amd_dbgapi_inferior_info pending wave events. */

My first intuition when reading this patch was "hey, the INFO param is
never modified, why not use a const ref".  But eventually ,under
process_one_event, we'll grab a reference to the INFO and modify it to
stash stop events.

Should this patch go one step further, and have the INFO passed to
process_one_event (this one can ensure the event's process_id matches
the INFO's process_id)?  This would also save iterating over all
inferiors to find the one we are looking for.

Best,
Lancelot.

>  
>  static amd_dbgapi_event_id_t
> -process_event_queue (amd_dbgapi_process_id_t process_id,
> +process_event_queue (amd_dbgapi_inferior_info &info,
>  		     amd_dbgapi_event_kind_t until_event_kind)
>  {
> -  /* An event of a given type can only be requested from a single
> -     process_id.  */
> -  gdb_assert (until_event_kind == AMD_DBGAPI_EVENT_KIND_NONE
> -	      || process_id != AMD_DBGAPI_PROCESS_NONE);
> -
>    while (true)
>      {
>        amd_dbgapi_event_id_t event_id;
>        amd_dbgapi_event_kind_t event_kind;
>  
>        amd_dbgapi_status_t status
> -	= amd_dbgapi_process_next_pending_event (process_id, &event_id,
> +	= amd_dbgapi_process_next_pending_event (info.process_id, &event_id,
>  						 &event_kind);
>  
>        if (status != AMD_DBGAPI_STATUS_SUCCESS)
> @@ -1479,7 +1472,7 @@ amd_dbgapi_target::wait (ptid_t ptid, struct target_waitstatus *ws,
>        /* Drain the events for the current inferior from the amd_dbgapi and
>  	 preserve the ordering.  */
>        auto info = get_amd_dbgapi_inferior_info (current_inferior ());
> -      process_event_queue (info->process_id);
> +      process_event_queue (*info);
>  
>        std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
>        if (event_ptid == minus_one_ptid)
> -- 
> 2.49.0

Re: [PATCH 4/5] gdb/amd-dbgapi: factor out require_forward_progress overload to target one inferior

[Lancelot SIX]

On Thu, Jun 05, 2025 at 04:16:27PM -0400, Simon Marchi wrote:
> A following patch will want to call require_forward_progress for a given
> inferior.  Extract a new require_forward_progress overload from the
> existing require_forward_progress function that targets a specific
> inferior.

Hi,

This LGTM, thanks.

Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)

Best,
Lancelot.

> 
> Change-Id: I54f42b83eb8443d4d91747ffbc86eaeb017f1e49
> ---
>  gdb/amd-dbgapi-target.c | 49 +++++++++++++++++++++++------------------
>  1 file changed, 28 insertions(+), 21 deletions(-)
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index 1ec20d1874ec..e86d7f34d0f6 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -443,6 +443,32 @@ async_event_handler_mark ()
>    mark_async_event_handler (amd_dbgapi_async_event_handler);
>  }
>  
> +/* Set forward progress requirement to REQUIRE for inferior INFO.  */
> +
> +static void
> +require_forward_progress (amd_dbgapi_inferior_info &info, bool require)
> +{
> +  /* If we try to disable forward progress requirement but the target expects
> +     resumed threads to be committed to the target, we could wait for events
> +     that will never arrive.  */
> +  if (!require)
> +    gdb_assert (!info.inf->process_target ()->commit_resumed_state);
> +
> +  gdb_assert (info.process_id != AMD_DBGAPI_PROCESS_NONE);
> +
> +  /* Don't do unnecessary calls to amd-dbgapi to avoid polluting the logs.  */
> +  if (info.forward_progress_required == require)
> +    return;
> +
> +  const auto progress
> +    = require ? AMD_DBGAPI_PROGRESS_NORMAL : AMD_DBGAPI_PROGRESS_NO_FORWARD;
> +  const auto status
> +    = amd_dbgapi_process_set_progress (info.process_id, progress);
> +  gdb_assert (status == AMD_DBGAPI_STATUS_SUCCESS);
> +
> +  info.forward_progress_required = require;
> +}
> +
>  /* Set forward progress requirement to REQUIRE for all processes of PROC_TARGET
>     matching PTID.  */
>  
> @@ -450,12 +476,6 @@ static void
>  require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
>  			  bool require)
>  {
> -  /* If we try to disable forward progress requirement but the target expects
> -     resumed threads to be committed to the target, we could wait for events
> -     that will never arrive.  */
> -  if (!require)
> -    gdb_assert (!proc_target->commit_resumed_state);
> -
>    for (inferior *inf : all_inferiors (proc_target))
>      {
>        if (ptid != minus_one_ptid && inf->pid != ptid.pid ())
> @@ -463,21 +483,8 @@ require_forward_progress (ptid_t ptid, process_stratum_target *proc_target,
>  
>        amd_dbgapi_inferior_info *info = get_amd_dbgapi_inferior_info (inf);
>  
> -      if (info->process_id == AMD_DBGAPI_PROCESS_NONE)
> -	continue;
> -
> -      /* Don't do unnecessary calls to amd-dbgapi to avoid polluting the logs.  */
> -      if (info->forward_progress_required == require)
> -	continue;
> -
> -      amd_dbgapi_status_t status
> -	= amd_dbgapi_process_set_progress
> -	    (info->process_id, (require
> -				? AMD_DBGAPI_PROGRESS_NORMAL
> -				: AMD_DBGAPI_PROGRESS_NO_FORWARD));
> -      gdb_assert (status == AMD_DBGAPI_STATUS_SUCCESS);
> -
> -      info->forward_progress_required = require;
> +      if (info->process_id != AMD_DBGAPI_PROCESS_NONE)
> +	require_forward_progress (*info, require);
>  
>        /* If ptid targets a single inferior and we have found it, no need to
>  	 continue.  */
> -- 
> 2.49.0

Re: [PATCH 5/5] gdb/amd-dbgapi: disable forward progress requirement in amd_dbgapi_target_breakpoint::check_status

[Lancelot SIX]

On Thu, Jun 05, 2025 at 04:16:28PM -0400, Simon Marchi wrote:
> ROCgdb handles target events very slowly when running a test case like
> this, where a breakpoint is preset on HipTest::vectorADD:
> 
>     for (int i=0; i < numDevices; ++i) {
>       HIPCHECK(hipSetDevice(i));
>       hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock), 0, stream[i],
>                         static_cast<const int*>(A_d[i]), static_cast<const int*>(B_d[i]), C_d[i], N);
>     }
> 
> What happens is:
> 
>  - A kernel is launched
>  - The internal runtime breakpoint is hit during the second
>    hipLaunchKernelGGL call, which causes
>    amd_dbgapi_target_breakpoint::check_status to be called
>  - Meanwhile, all waves of the kernel hit the breakpoint on vectorADD
>  - amd_dbgapi_target_breakpoint::check_status calls process_event_queue,
>    which pulls the thousand of breakpoint hit events from the kernel
>  - As part of handling the breakpoint hit events, we write the PC of the
>    waves that stopped to decrement it.  Because the forward progress
>    requirement is not disabled, this causes a suspend/resume of the
>    queue each time, which is time-consuming.
> 
> The stack trace where this all happens is:
> 
>     #32 0x00007ffff6b9abda in amd_dbgapi_write_register (wave_id=..., register_id=..., offset=0, value_size=8, value=0x7fffea9fdcc0) at /home/smarchi/src/amd-dbgapi/src/register.cpp:587
>     #33 0x00005555588c0bed in amd_dbgapi_target::store_registers (this=0x55555c7b1d20 <the_amd_dbgapi_target>, regcache=0x507000002240, regno=470) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:2504
>     #34 0x000055555a5186a1 in target_store_registers (regcache=0x507000002240, regno=470) at /home/smarchi/src/wt/amd/gdb/target.c:3973
>     #35 0x0000555559fab831 in regcache::raw_write (this=0x507000002240, regnum=470, src=...) at /home/smarchi/src/wt/amd/gdb/regcache.c:890
>     #36 0x0000555559fabd2b in regcache::cooked_write (this=0x507000002240, regnum=470, src=...) at /home/smarchi/src/wt/amd/gdb/regcache.c:915
>     #37 0x0000555559fc3ca5 in regcache::cooked_write<unsigned long, void> (this=0x507000002240, regnum=470, val=140737323456768) at /home/smarchi/src/wt/amd/gdb/regcache.c:850
>     #38 0x0000555559fab09a in regcache_cooked_write_unsigned (regcache=0x507000002240, regnum=470, val=140737323456768) at /home/smarchi/src/wt/amd/gdb/regcache.c:858
>     #39 0x0000555559fb0678 in regcache_write_pc (regcache=0x507000002240, pc=0x7ffff62bd900) at /home/smarchi/src/wt/amd/gdb/regcache.c:1460
>     #40 0x00005555588bb37d in process_one_event (event_id=..., event_kind=AMD_DBGAPI_EVENT_KIND_WAVE_STOP) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:1873
>     #41 0x00005555588bbf7b in process_event_queue (process_id=..., until_event_kind=AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:2006
>     #42 0x00005555588b1aca in amd_dbgapi_target_breakpoint::check_status (this=0x511000140900, bs=0x50600014ed00) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:890
>     #43 0x0000555558c50080 in bpstat_stop_status (aspace=0x5070000061b0, bp_addr=0x7fffed0b9ab0, thread=0x518000026c80, ws=..., stop_chain=0x50600014ed00) at /home/smarchi/src/wt/amd/gdb/breakpoint.c:6126
>     #44 0x000055555984f4ff in handle_signal_stop (ecs=0x7fffeaa40ef0) at /home/smarchi/src/wt/amd/gdb/infrun.c:7169
>     #45 0x000055555984b889 in handle_inferior_event (ecs=0x7fffeaa40ef0) at /home/smarchi/src/wt/amd/gdb/infrun.c:6621
>     #46 0x000055555983eab6 in fetch_inferior_event () at /home/smarchi/src/wt/amd/gdb/infrun.c:4750
>     #47 0x00005555597caa5f in inferior_event_handler (event_type=INF_REG_EVENT) at /home/smarchi/src/wt/amd/gdb/inf-loop.c:42
>     #48 0x00005555588b838e in handle_target_event (client_data=0x0) at /home/smarchi/src/wt/amd/gdb/amd-dbgapi-target.c:1513
> 
> Fix that performance problem by disabling the forward progress
> requirement in amd_dbgapi_target_breakpoint::check_status, before
> calling process_event_queue, so that we can process all events
> efficiently.
> 
> Since the same performance problem could theoritically happen any time
> process_event_queue is called with forward progress requirement enabled,
> add an assert to ensure that forward progress requirement is disabled
> when process_event_queue is invoked.  This makes it necessary to add a
> require_forward_progress call to amd_dbgapi_finalize_core_attach.  It
> looks a bit strange, since core files don't have execution, but it
> doesn't hurt.
> 
> Add a test that replicates this scenario.  The test launches a kernel
> that hits a breakpoint (with an always false condition) repeatedly.
> Meanwhile, the host process loads an unloads a code object, causing
> check_status to be called.
> 
> Bug: SWDEV-482511
> Change-Id: Ida86340d679e6bd8462712953458c07ba3fd49ec

Hi Simon,

This looks overall good to me.

> ---
>  gdb/amd-dbgapi-target.c                       |  6 ++
>  .../code-object-load-while-breakpoint-hit.cpp | 70 +++++++++++++++++++
>  .../code-object-load-while-breakpoint-hit.exp | 68 ++++++++++++++++++
>  3 files changed, 144 insertions(+)
>  create mode 100644 gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
>  create mode 100644 gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index e86d7f34d0f6..44913bbd7532 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -568,6 +568,8 @@ amd_dbgapi_target_breakpoint::check_status (struct bpstat *bs)
>    if (action == AMD_DBGAPI_BREAKPOINT_ACTION_RESUME)
>      return;
>  
> +  require_forward_progress (*info, false);
> +
>    /* If the action is AMD_DBGAPI_BREAKPOINT_ACTION_HALT, we need to wait until
>       a breakpoint resume event for this breakpoint_id is seen.  */
>    amd_dbgapi_event_id_t resume_event_id
> @@ -1352,6 +1354,10 @@ static amd_dbgapi_event_id_t
>  process_event_queue (amd_dbgapi_inferior_info &info,
>  		     amd_dbgapi_event_kind_t until_event_kind)
>  {
> +  /* Pulling events with forward progress required may result in bad
> +     performance, make sure it is not required.  */
> +  gdb_assert (!info.forward_progress_required);
> +
>    while (true)
>      {
>        amd_dbgapi_event_id_t event_id;
> diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
> new file mode 100644
> index 000000000000..e97809335569
> --- /dev/null
> +++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
> @@ -0,0 +1,70 @@

I think you are missing the copyright header for this file.

> +#ifdef DEVICE
> +
> +#include <hip/hip_runtime.h>
> +
> +constexpr unsigned int NUM_BREAKPOINT_HITS = 5;
> +
> +static __device__ void
> +break_here ()
> +{
> +  asm ("s_nop 0");

I don't expect this inline asssembly is necessary.

> +}
> +
> +extern "C" __global__ void
> +kernel ()
> +{
> +  for (int n = 0; n < NUM_BREAKPOINT_HITS; ++n)
> +    break_here ();
> +}
> +
> +#else
> +
> +#include <hip/hip_runtime.h>
> +#include <unistd.h>
> +
> +constexpr unsigned int NUM_ITEMS_PER_BLOCK = 256;
> +constexpr unsigned int NUM_BLOCKS = 128;
> +constexpr unsigned int NUM_ITEMS = NUM_ITEMS_PER_BLOCK * NUM_BLOCKS;
> +constexpr unsigned int NUM_LOAD_UNLOADS = 5;
> +
> +#define CHECK(cmd)                                                            \
> +  {                                                                           \
> +    hipError_t error = cmd;                                                   \
> +    if (error != hipSuccess)                                                  \
> +      {                                                                       \
> +	fprintf (stderr, "error: '%s'(%d) at %s:%d\n",                        \
> +		 hipGetErrorString (error), error, __FILE__, __LINE__);       \
> +	exit (EXIT_FAILURE);                                                  \
> +      }                                                                       \
> +  }
> +
> +int
> +main (int argc, const char **argv)
> +{
> +  if (argc != 2)
> +    {
> +      fprintf (stderr, "Usage: %s <hip_module_path>\n", argv[0]);
> +      return 1;
> +    }
> +
> +  const auto module_path = argv[1];
> +  hipModule_t module;
> +  CHECK (hipModuleLoad (&module, module_path));
> +
> +  /* Launch the kernel.  */
> +  hipFunction_t function;
> +  CHECK (hipModuleGetFunction (&function, module, "kernel"));
> +  CHECK (hipModuleLaunchKernel (function, NUM_BLOCKS, 1, 1,
> +				NUM_ITEMS_PER_BLOCK, 1, 1, 0, nullptr, nullptr,
> +				nullptr));
> +
> +  /* Load and unload the module many times.  */
> +  for (int i = 0; i < NUM_LOAD_UNLOADS; ++i)
> +    {
> +      hipModule_t dummy_module;
> +      CHECK (hipModuleLoad (&dummy_module, module_path));
> +      CHECK (hipModuleUnload (dummy_module));
> +    }
> +}
> +
> +#endif
> diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
> new file mode 100644
> index 000000000000..b8d003f37032
> --- /dev/null
> +++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
> @@ -0,0 +1,68 @@
> +# Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.

The copyright should be to the FSF, not AMD.

Given the elements above are fixed,

Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)

Best,
Lanecelot.

> +
> +# This file is part of GDB.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +
> +# You should have received a copy of the GNU General Public License
> +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
> +
> +# This test verifies what happens when a code object list update happens at the
> +# same time as some wave stop events are reported.  It was added following a
> +# performance bug fix, where forward progress requirement disabled when
> +# pulling events from amd-dbgapi in amd_dbgapi_target_breakpoint::check_status.
> +#
> +# The test launches a kernel that hits a breakpoint with an always false
> +# condition a certain number of times.  Meanwhile, the host loads and unloads
> +# a code object in a loop, causing check_status to be called.  The hope is that
> +# check_status, when calling process_event_queue, will pull many WAVE_STOP
> +# events from the kernel hitting the breakpoint.
> +#
> +# Without the appropriate fix (of disabling forward progress requirement in
> +# check_status), GDB would hit the newly-added assert in process_event_queue,
> +# which verifies that forward progress requirement is disabled.  Even without
> +# this assert, the test would likely time out (depending on the actual timeout
> +# value).
> +
> +load_lib rocm.exp
> +standard_testfile .cpp
> +require allow_hipcc_tests
> +
> +# Build the host executable.
> +if { [build_executable "failed to prepare" \
> +	  $testfile $srcfile {debug hip}] == -1 } {
> +    return -1
> +}
> +
> +set hipmodule_path [standard_output_file ${testfile}.co]
> +
> +# Build the kernel object file.
> +if { [gdb_compile $srcdir/$subdir/$srcfile \
> +	$hipmodule_path object \
> +	{ debug hip additional_flags=--genco additional_flags=-DDEVICE } ] != "" } {
> +    return -1
> +}
> +
> +proc do_test { } {
> +    with_rocm_gpu_lock {
> +	clean_restart $::binfile
> +	gdb_test_no_output "set args $::hipmodule_path" "set args"
> +
> +	if { ![runto_main] } {
> +		return
> +	}
> +
> +	gdb_test "with breakpoint pending on -- break break_here if 0"
> +	gdb_continue_to_end "continue to end" "continue" 1
> +    }
> +}
> +
> +do_test
> -- 
> 2.49.0

Re: [PATCH 3/5] gdb/amd-dbgapi: pass amd_dbgapi_inferior_info to process_event_queue

[Simon Marchi]

On 6/6/25 7:00 AM, Lancelot SIX wrote:
> On Thu, Jun 05, 2025 at 04:16:26PM -0400, Simon Marchi wrote:
>> A following patch will make process_event_queue access a field of
>> amd_dbgapi_inferior_info.  Prepare for this by making
>> process_event_queue accept an amd_dbgapi_inferior_info object, instead
>> of a process id.
>>
>> Change-Id: I9adc491dd1ff64ff74c40aa7662fffb11bd8332b
>> ---
>>  gdb/amd-dbgapi-target.c | 25 +++++++++----------------
>>  1 file changed, 9 insertions(+), 16 deletions(-)
>>
>> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
>> index 8fd8bcf5ae49..1ec20d1874ec 100644
>> --- a/gdb/amd-dbgapi-target.c
>> +++ b/gdb/amd-dbgapi-target.c
>> @@ -234,7 +234,7 @@ struct amd_dbgapi_inferior_info
>>  };
>>  
>>  static amd_dbgapi_event_id_t process_event_queue
>> -  (amd_dbgapi_process_id_t process_id,
>> +  (amd_dbgapi_inferior_info &info,
>>     amd_dbgapi_event_kind_t until_event_kind = AMD_DBGAPI_EVENT_KIND_NONE);
>>  
>>  static const target_info amd_dbgapi_target_info = {
>> @@ -564,8 +564,7 @@ amd_dbgapi_target_breakpoint::check_status (struct bpstat *bs)
>>    /* If the action is AMD_DBGAPI_BREAKPOINT_ACTION_HALT, we need to wait until
>>       a breakpoint resume event for this breakpoint_id is seen.  */
>>    amd_dbgapi_event_id_t resume_event_id
>> -    = process_event_queue (info->process_id,
>> -			   AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
>> +    = process_event_queue (*info, AMD_DBGAPI_EVENT_KIND_BREAKPOINT_RESUME);
>>  
>>    /* We should always get a breakpoint_resume event after processing all
>>       events generated by reporting the breakpoint hit.  */
>> @@ -1337,28 +1336,22 @@ event_kind_str (amd_dbgapi_event_kind_t kind)
>>    gdb_assert_not_reached ("unhandled amd_dbgapi_event_kind_t value");
>>  }
>>  
>> -/* Drain the dbgapi event queue of a given process_id, or of all processes if
>> -   process_id is AMD_DBGAPI_PROCESS_NONE.  Stop processing the events if an
>> -   event of a given kind is requested and `process_id` is not
>> -   AMD_DBGAPI_PROCESS_NONE.  Wave stop events that are not returned are queued
>> -   into their inferior's amd_dbgapi_inferior_info pending wave events. */
>> +/* Drain the dbgapi event queue of a given process_id.  Stop processing the
> 
> It reads odd to have process_id here, as we don't pass it as an explicit
> parameter anymore.  Could it say "a given inferior" instead?

Yes, makes sense.

> 
>> +   events if an event of a given kind is requested (not AMD_DBGAPI_EVENT_NONE).
>> +   Wave stop events that are not returned are queued into their inferior's
>> +   amd_dbgapi_inferior_info pending wave events. */
> 
> My first intuition when reading this patch was "hey, the INFO param is
> never modified, why not use a const ref".  But eventually ,under
> process_one_event, we'll grab a reference to the INFO and modify it to
> stash stop events.
> 
> Should this patch go one step further, and have the INFO passed to
> process_one_event (this one can ensure the event's process_id matches
> the INFO's process_id)?  This would also save iterating over all
> inferiors to find the one we are looking for.

Oh, yes, that's a good idea.  Or, we could just skip the
AMD_DBGAPI_EVENT_INFO_PROCESS and AMD_DBGAPI_PROCESS_INFO_OS_ID calls
altogether, since we know for which process the event is anyway.  I will
add this as a separate patch just after this one.

Simon

Re: [PATCH 5/5] gdb/amd-dbgapi: disable forward progress requirement in amd_dbgapi_target_breakpoint::check_status

[Simon Marchi]

>> diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
>> new file mode 100644
>> index 000000000000..e97809335569
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.cpp
>> @@ -0,0 +1,70 @@
> 
> I think you are missing the copyright header for this file.

Done.

>> +#ifdef DEVICE
>> +
>> +#include <hip/hip_runtime.h>
>> +
>> +constexpr unsigned int NUM_BREAKPOINT_HITS = 5;
>> +
>> +static __device__ void
>> +break_here ()
>> +{
>> +  asm ("s_nop 0");
> 
> I don't expect this inline asssembly is necessary.

Ok, will remove it.  Not sure why I added it, I might have had some
trouble with empty functions in the past.

>> diff --git a/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
>> new file mode 100644
>> index 000000000000..b8d003f37032
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp
>> @@ -0,0 +1,68 @@
>> +# Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
> 
> The copyright should be to the FSF, not AMD.

Done.

> Given the elements above are fixed,
> 
> Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)

Thanks,

Simon