From: Shahab Vahedi <shahab.vahedi@amd.com>
To: <gdb-patches@sourceware.org>
Cc: Shahab Vahedi <list+gdb@vahedi.org>,
Luis Machado <luis.machado.foss@gmail.com>,
Tankut Baris Aktemur <tankutbaris.aktemur@amd.com>,
Lancelot SIX <lancelot.six@amd.com>,
"Simon Marchi" <simon.marchi@efficios.com>,
Shahab Vahedi <shahab.vahedi@amd.com>
Subject: [PUSHED 2/2] gdb.rocm/watchpoint-basic: add XFAILs for known configurations
Date: Mon, 27 Jul 2026 10:58:03 +0200 [thread overview]
Message-ID: <20260727085803.14004-2-shahab.vahedi@amd.com> (raw)
In-Reply-To: <20260727085803.14004-1-shahab.vahedi@amd.com>
Some of the tests in gdb.rom/watchpoint-basic are destined to fail
due to a problem in KFD. This patch marks those tests as such on
configurations that this can happen.
Reviewed-by: Tankut Baris Aktemur <tankutbaris.aktemur@amd.com>
Approved-by: Luis Machado <luis.machado.foss@gmail.com>
---
gdb/testsuite/gdb.rocm/watchpoint-basic.exp | 78 +++++++++++++++++----
1 file changed, 66 insertions(+), 12 deletions(-)
diff --git a/gdb/testsuite/gdb.rocm/watchpoint-basic.exp b/gdb/testsuite/gdb.rocm/watchpoint-basic.exp
index 9626b32d5d2..bc30726515e 100644
--- a/gdb/testsuite/gdb.rocm/watchpoint-basic.exp
+++ b/gdb/testsuite/gdb.rocm/watchpoint-basic.exp
@@ -26,14 +26,25 @@ if {[build_executable "failed to prepare" $testfile $srcfile {debug hip}]} {
}
proc continue_to_watchpoint_hit { old_value new_value test } {
- gdb_test "continue" \
- [multi_line \
- "hit Hardware watchpoint $::decimal:.*" \
- "" \
- "Old value = $old_value" \
- "New value = $new_value" \
- ".*"] \
- $test
+ set hit_re [multi_line \
+ "hit Hardware watchpoint $::decimal:.*" \
+ "" \
+ "Old value = $old_value" \
+ "New value = $new_value" \
+ ".*"]
+
+ set ret false
+ gdb_test_multiple "continue" "$test" {
+ -re -wrap $hit_re {
+ pass "$gdb_test_name"
+ set ret true
+ }
+ -re -wrap ".*$::inferior_exited_re.*" {
+ fail "$gdb_test_name"
+ set ret false
+ }
+ }
+ return $ret
}
# Test inserting a watchpoint on a host variable before the runtime loads, and
@@ -78,6 +89,37 @@ proc_with_prefix test_host_watchpoint_after_runtime_load {} {
}
}
+# On some systems, KFD corrupts the watchpoint configurations
+# during the first dispatch's queue mapping. This leads to
+# watchpoints not being triggered, if they were set before the
+# first dispatch. As a result, those test scenarios run to the
+# end and exit prematurely.
+
+gdb_caching_proc target_has_xfail {} {
+ set xfail_arches {gfx1201}
+
+ set targets [find_amdgpu_devices]
+ if {[llength $targets] == 0} {
+ # Can't determine GPU type, don't set up xfail. The test will probably
+ # not run correctly anyway.
+ return 0
+ }
+
+ # The test will run on GPU-0, so it should be the first of the list.
+ set target [lindex $targets 0]
+
+ return [expr {[lsearch -exact $xfail_arches $target] != -1}]
+}
+
+# This is used when dealing with the first watchpoints that are
+# set before the first dispatch.
+
+proc maybe_xfail {} {
+ if {[target_has_xfail]} {
+ setup_xfail "*-*-*" "watchpoint configuration corruption in kfd."
+ }
+}
+
# Test inserting a watchpoint before the kernel is launched, then hitting
# it when the kernel runs.
@@ -93,7 +135,10 @@ proc_with_prefix test_watchpoint_before_kernel {} {
"Hardware watchpoint $::decimal: .*" \
"set watchpoint on *ptr1"
- continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"
+ maybe_xfail
+ if {![continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"]} {
+ return
+ }
continue_to_watchpoint_hit 10 30 "continue to watchpoint hit 2"
continue_to_watchpoint_hit 30 40 "continue to watchpoint hit 3"
continue_to_watchpoint_hit 40 60 "continue to watchpoint hit 4"
@@ -145,7 +190,10 @@ proc_with_prefix test_remove_watchpoint_inside_kernel {} {
"Hardware watchpoint $::decimal: .*" \
"set watchpoint"
- continue_to_watchpoint_hit 0 10 "continue to watchpoint hit"
+ maybe_xfail
+ if {![continue_to_watchpoint_hit 0 10 "continue to watchpoint hit"]} {
+ return
+ }
gdb_test "with confirm off -- delete" "" "delete all breakpoints"
gdb_continue_to_end "" continue 1
@@ -170,7 +218,10 @@ proc_with_prefix test_multiple_watchpoints {} {
"Hardware watchpoint $::decimal: .*" \
"set watchpoint on *ptr2"
- continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"
+ maybe_xfail
+ if {![continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"]} {
+ return
+ }
continue_to_watchpoint_hit 0 100 "continue to watchpoint hit 2"
continue_to_watchpoint_hit 10 30 "continue to watchpoint hit 3"
continue_to_watchpoint_hit 100 300 "continue to watchpoint hit 4"
@@ -210,7 +261,10 @@ proc_with_prefix test_disable_enable_watchpoint {} {
}
}
- continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"
+ maybe_xfail
+ if {![continue_to_watchpoint_hit 0 10 "continue to watchpoint hit 1"]} {
+ return
+ }
gdb_test_no_output "disable $wp1_num" "disable watchpoint"
--
2.55.0
prev parent reply other threads:[~2026-07-27 9:00 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-07-27 8:58 [PUSHED 1/2] gdb.rocm/watchpoint-basic: use gdb_continue_to_end Shahab Vahedi
2026-07-27 8:58 ` Shahab Vahedi [this message]
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=20260727085803.14004-2-shahab.vahedi@amd.com \
--to=shahab.vahedi@amd.com \
--cc=gdb-patches@sourceware.org \
--cc=lancelot.six@amd.com \
--cc=list+gdb@vahedi.org \
--cc=luis.machado.foss@gmail.com \
--cc=simon.marchi@efficios.com \
--cc=tankutbaris.aktemur@amd.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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox