diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index 40f24b5fc2f..e90628c8183 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -148,7 +148,7 @@ struct amd_dbgapi_inferior_info
 };
 
 static amd_dbgapi_event_id_t process_event_queue
-  (amd_dbgapi_process_id_t process_id = AMD_DBGAPI_PROCESS_NONE,
+  (amd_dbgapi_process_id_t process_id,
    amd_dbgapi_event_kind_t until_event_kind = AMD_DBGAPI_EVENT_KIND_NONE);
 
 static const target_info amd_dbgapi_target_info = {
@@ -1255,8 +1255,10 @@ amd_dbgapi_target::wait (ptid_t ptid, struct target_waitstatus *ws,
   std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
   if (event_ptid == minus_one_ptid)
     {
-      /* Drain the events from the amd_dbgapi and preserve the ordering.  */
-      process_event_queue ();
+      /* 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);
 
       std::tie (event_ptid, gpu_waitstatus) = consume_one_event (ptid.pid ());
       if (event_ptid == minus_one_ptid)
diff --git a/gdb/testsuite/gdb.rocm/multi-inferior-gpu.cpp b/gdb/testsuite/gdb.rocm/multi-inferior-gpu.cpp
new file mode 100644
index 00000000000..ca869233b58
--- /dev/null
+++ b/gdb/testsuite/gdb.rocm/multi-inferior-gpu.cpp
@@ -0,0 +1,113 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2023 Free Software Foundation, Inc.
+
+   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/>.  */
+
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <hip/hip_runtime.h>
+
+#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);                                                 \
+      }                                                                      \
+  }
+
+__global__ void
+kern ()
+{
+  asm ("s_sleep 1");
+}
+
+/* Spawn one child process per detected GPU.  */
+
+static int
+parent (int argc, char **argv)
+{
+  /* Identify how many GPUs we have, and spawn one child for each.  */
+  int num_devices;
+  CHECK (hipGetDeviceCount (&num_devices));
+
+  /* Break here.  */
+
+  for (int i = 0; i < num_devices; i++)
+    {
+      char n[32] = {};
+      snprintf (n, sizeof (n), "%d", i);
+      pid_t pid = fork ();
+      if (pid == -1)
+	{
+	  perror ("Fork failed");
+	  return -1;
+	}
+
+      if (pid == 0)
+	{
+	  /* Exec to force the child to re-initialize the ROCm runtime.  */
+	  if (execl (argv[0], argv[0], n) == -1)
+	    {
+	      perror ("Failed to exec");
+	      return -1;
+	    }
+	}
+    }
+
+  /* Wait for all children.  */
+  while (true)
+    {
+      int ws;
+      pid_t ret = waitpid (-1, &ws, 0);
+      if (ret == -1 && errno == ECHILD)
+	break;
+    }
+
+  /* Last break here.  */
+  return 0;
+}
+
+static int
+child (int argc, char **argv)
+{
+  int dev_number;
+  if (sscanf (argv[1], "%d", &dev_number) != 1)
+    {
+      fprintf (stderr, "Invalid argument \"%s\"\n", argv[1]);
+      return -1;
+    }
+
+  CHECK (hipSetDevice (dev_number));
+  kern<<<1, 1>>> ();
+  hipDeviceSynchronize ();
+  return 0;
+}
+
+/* When called with no argument, identify how many AMDGPU devices are
+   available on the system and spawn one worker process per GPU.  If a
+   command-line argument is provided, it is the index of the GPU to use.  */
+
+int
+main (int argc, char **argv)
+{
+  if (argc <= 1)
+    return parent (argc, argv);
+  else
+    return child (argc, argv);
+}
diff --git a/gdb/testsuite/gdb.rocm/multi-inferior-gpu.exp b/gdb/testsuite/gdb.rocm/multi-inferior-gpu.exp
new file mode 100644
index 00000000000..18b4172ff09
--- /dev/null
+++ b/gdb/testsuite/gdb.rocm/multi-inferior-gpu.exp
@@ -0,0 +1,89 @@
+# Copyright 2023 Free Software Foundation, Inc.
+
+# 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 checks that GDB can debug multiple inferior which uses all
+# the ROCm runtime.
+
+load_lib rocm.exp
+
+standard_testfile .cpp
+
+require allow_hipcc_tests
+require hip_devices_support_debug_multi_process
+
+if {[build_executable "failed to prepare" $testfile $srcfile {debug hip}]} {
+    return
+}
+
+proc do_test {} {
+    clean_restart $::binfile
+    gdb_test_no_output "set non-stop on"
+    gdb_test_no_output "set detach-on-fork off"
+    gdb_test_no_output "set follow-fork parent"
+
+    with_rocm_gpu_lock {
+	gdb_breakpoint [gdb_get_line_number "Break here"]
+	gdb_breakpoint kern allow-pending
+	gdb_breakpoint [gdb_get_line_number "Last break here"]
+
+	# Run until we reach the first breakpoint where we can figure
+	# out how many children will be spawned.
+	gdb_test "run" "hit Breakpoint.*"
+
+	set num_children [get_integer_valueof "num_devices" 0]
+	set bp_to_see $num_children
+	set stopped_gpu_threads [list]
+
+	gdb_test_multiple "continue -a &" "continue to gpu breakpoints" {
+	    -re "Continuing\.\r\n$::gdb_prompt " {
+		pass $gdb_test_name
+	    }
+	}
+
+	gdb_test_multiple "" "wait for gpu stops" {
+	    -re "Thread ($::decimal\.$::decimal)\[^\r\n\]* hit Breakpoint\[^\r\n\]*, kern \(\)\[^\r\n\]*\r\n" {
+		lappend stopped_gpu_threads $expect_out(1,string)
+		incr bp_to_see -1
+		if {$bp_to_see != 0} {
+		    exp_continue
+		} else {
+		    pass $gdb_test_name
+		}
+	    }
+	}
+
+	# Continue all the GPU kernels so all the children processes can reach exit.
+	foreach thread $stopped_gpu_threads {
+	    set infnumber [lindex [split $thread .] 0]
+	    gdb_test "thread $thread" "Switching to thread.*"
+	    gdb_test_multiple "continue $thread" "" {
+		-re "\\\[Inferior $infnumber \[^\n\r\]* exited normally\\]\r\n$::gdb_prompt " {
+		    pass $gdb_test_name
+		}
+	    }
+	}
+
+	gdb_test_multiple "" "reach breakpoint in main" {
+	    -re "hit Breakpoint.*parent" {
+		pass $gdb_test_name
+	    }
+	}
+	# Select main inferior
+	gdb_test "inferior 1" "Switching to inferior 1.*"
+	gdb_continue_to_end "" "continue -a" 1
+    }
+}
+
+do_test