[PATCH v2] gdb: fix "frame function" issue when call is last instruction

[PATCH v2] gdb: fix "frame function" issue when call is last instruction

[Sahil Siddiq]

Currently, the "frame function" fails when the last
instruction is a call. In such a case, the $rip
register points to an address that lies outside the
frame.

Using "get_frame_address_in_block" instead of
"get_frame_pc" resolves this issue.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30929
---
Changes since v1:
* frame-selection-last-instr-call.exp: Modify test to fix regressions on arm/aarch64

Patch v1: https://sourceware.org/pipermail/gdb-patches/2024-June/210251.html

 gdb/stack.c                                   |   4 +-
 .../frame-selection-last-instr-call.c         |  28 ++++
 .../frame-selection-last-instr-call.exp       | 130 ++++++++++++++++++
 3 files changed, 160 insertions(+), 2 deletions(-)
 create mode 100644 gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
 create mode 100644 gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp

diff --git a/gdb/stack.c b/gdb/stack.c
index b36193be2f..8249866468 100644
--- a/gdb/stack.c
+++ b/gdb/stack.c
@@ -2860,8 +2860,8 @@ find_frame_for_function (const char *function_name)
   do
     {
       for (size_t i = 0; (i < sals.size () && !found); i++)
-	found = (get_frame_pc (frame) >= func_bounds[i].low
-		 && get_frame_pc (frame) < func_bounds[i].high);
+	found = (get_frame_address_in_block (frame) >= func_bounds[i].low
+		 && get_frame_address_in_block (frame) < func_bounds[i].high);
       if (!found)
 	{
 	  level = 1;
diff --git a/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
new file mode 100644
index 0000000000..4f056332af
--- /dev/null
+++ b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
@@ -0,0 +1,28 @@
+/* Copyright 2024 Free Software Foundation, Inc.
+
+   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/>.  */
+
+void
+frame_1 (void)
+{
+  __builtin_abort();
+}
+
+int
+main (void)
+{
+  frame_1 ();
+}
diff --git a/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
new file mode 100644
index 0000000000..5ba52cc069
--- /dev/null
+++ b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
@@ -0,0 +1,130 @@
+# Copyright 2024 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 tests GDB's frame selection as used by the 'frame',
+# 'select-frame', and 'info frame' commands in the corner case
+# when the last instruction in a frame is a call.
+
+standard_testfile
+
+if {[prepare_for_testing "failed to prepare" $testfile $srcfile {debug}]} {
+    return -1
+}
+
+runto_main
+gdb_breakpoint abort
+gdb_continue_to_breakpoint abort
+
+gdb_test "bt" "#0  $hex in abort .*#1  $hex in frame_1 .*#2  $hex in main.*" "backtrace at breakpoint"
+
+# Perform "info frame" to extract the frame's address.
+proc get_frame_address { {testname ""} } {
+    global hex gdb_prompt
+
+    set frame_address "unknown"
+    set testname "get_frame_address: ${testname}"
+    gdb_test_multiple "info frame" $testname {
+	-re ", frame at ($hex):\r\n.*\r\n$gdb_prompt $" {
+	    set frame_address $expect_out(1,string)
+	    pass $testname
+	}
+    }
+
+    return $frame_address
+}
+
+# Check that the current frame is at stack depth LEVEL, at frame
+# address ADDRESS, and is in FUNCTION.
+proc check_frame { level address function } {
+    global hex gdb_prompt
+
+    set re [multi_line \
+		"Stack level ${level}, frame at ($address):" \
+		".* = $hex in ${function}( \(\[^\r\n\]*\))*; saved .* = $hex" \
+		".*\r\n$gdb_prompt $" ]
+
+    set testname "check frame level ${level}"
+    gdb_test_multiple "info frame" $testname {
+	-re $re {
+	    pass $testname
+	}
+    }
+}
+
+# Select frame using level, but relying on this being the default
+# action, so "frame 0" performs "frame level 0".
+gdb_test "frame 0" "#0  $hex in abort.*"
+set frame_0_address [ get_frame_address "frame 0" ]
+gdb_test "frame 1" "#1  $hex in frame_1.*"
+set frame_1_address [ get_frame_address "frame 1" ]
+gdb_test "frame 2" "#2  $hex in main.*"
+set frame_2_address [ get_frame_address "frame 2" ]
+
+# Select frame using 'level' specification.
+gdb_test "frame level 0" "#0  $hex in abort.*"
+gdb_test "frame level 1" "#1  $hex in frame_1.*"
+gdb_test "frame level 2" "#2  $hex in main.*"
+
+# Select frame by address.
+gdb_test "frame address ${frame_0_address}" "#0  $hex in abort.*" \
+    "select frame 0 by address"
+gdb_test "frame address ${frame_1_address}" "#1  $hex in frame_1.*" \
+    "select frame 1 by address"
+gdb_test "frame address ${frame_2_address}" "#2  $hex in main.*" \
+    "select frame 2 by address"
+
+# Select frame by function.
+gdb_test "frame function abort" "#0  $hex in abort.*"
+gdb_test "frame function frame_1" "#1  $hex in frame_1.*"
+gdb_test "frame function main" "#2  $hex in main.*"
+
+with_test_prefix "select-frame, no keyword" {
+    gdb_test_no_output "select-frame 0"
+    check_frame "0" "${frame_0_address}" "abort"
+    gdb_test_no_output "select-frame 1"
+    check_frame "1" "${frame_1_address}" "frame_1"
+    gdb_test_no_output "select-frame 2"
+    check_frame "2" "${frame_2_address}" "main"
+}
+
+with_test_prefix "select-frame, keyword=level" {
+    gdb_test_no_output "select-frame level 0"
+    check_frame "0" "${frame_0_address}" "abort"
+    gdb_test_no_output "select-frame level 1"
+    check_frame "1" "${frame_1_address}" "frame_1"
+    gdb_test_no_output "select-frame level 2"
+    check_frame "2" "${frame_2_address}" "main"
+}
+
+with_test_prefix "select-frame, keyword=address" {
+    gdb_test_no_output "select-frame address ${frame_0_address}" \
+	"select frame 0 by address"
+    check_frame "0" "${frame_0_address}" "abort"
+    gdb_test_no_output "select-frame address ${frame_1_address}" \
+	"select frame 1 by address"
+    check_frame "1" "${frame_1_address}" "frame_1"
+    gdb_test_no_output "select-frame address ${frame_2_address}" \
+	"select frame 2 by address"
+    check_frame "2" "${frame_2_address}" "main"
+}
+
+with_test_prefix "select-frame, keyword=function" {
+    gdb_test_no_output "select-frame function abort"
+    check_frame "0" "${frame_0_address}" "abort"
+    gdb_test_no_output "select-frame function frame_1"
+    check_frame "1" "${frame_1_address}" "frame_1"
+    gdb_test_no_output "select-frame function main"
+    check_frame "2" "${frame_2_address}" "main"
+}
-- 
2.45.2

Re: [PATCH v2] gdb: fix "frame function" issue when call is last instruction

[Guinevere Larsen]

On 7/1/24 4:53 PM, Sahil Siddiq wrote:
> Currently, the "frame function" fails when the last
> instruction is a call. In such a case, the $rip
> register points to an address that lies outside the
> frame.
Hello!

Thanks for working on this patch. I tested this patch and I can confirm 
it does fix the issue you linked, but I have a few inline comments.

I would prefer if you wrote "the last instruction in a frame is a call" 
or something similar. I ask this because when I read the commit message, 
I thought "last instruction" was referring to last instruction executed, 
and couldn't understand  why I couldn't reproduce without __builtin_abort.

>
> Using "get_frame_address_in_block" instead of
> "get_frame_pc" resolves this issue.
>
> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30929
> ---
> Changes since v1:
> * frame-selection-last-instr-call.exp: Modify test to fix regressions on arm/aarch64
>
> Patch v1: https://sourceware.org/pipermail/gdb-patches/2024-June/210251.html

And responding to your v1 comment about why you created a new test file, 
you mention creating a new C file for this test. Have you tried 
extending the original frame-selection.c file with a new function that 
calls __builtin_abort instead? It seems to me that it should work, and 
you should be able to extend the existing frame-selection.exp file, but 
I might be missing something.

I would prefer if we just extended the test, instead of added a new one, 
but if I indeed missed something and it isn't as easy to extend it as I 
think, we can have the new test :)

>
>   gdb/stack.c                                   |   4 +-
>   .../frame-selection-last-instr-call.c         |  28 ++++
>   .../frame-selection-last-instr-call.exp       | 130 ++++++++++++++++++
>   3 files changed, 160 insertions(+), 2 deletions(-)
>   create mode 100644 gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
>   create mode 100644 gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
>
> diff --git a/gdb/stack.c b/gdb/stack.c
> index b36193be2f..8249866468 100644
> --- a/gdb/stack.c
> +++ b/gdb/stack.c
> @@ -2860,8 +2860,8 @@ find_frame_for_function (const char *function_name)
>     do
>       {
>         for (size_t i = 0; (i < sals.size () && !found); i++)
> -	found = (get_frame_pc (frame) >= func_bounds[i].low
> -		 && get_frame_pc (frame) < func_bounds[i].high);
> +	found = (get_frame_address_in_block (frame) >= func_bounds[i].low
> +		 && get_frame_address_in_block (frame) < func_bounds[i].high);
>         if (!found)
>   	{
>   	  level = 1;
> diff --git a/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
> new file mode 100644
> index 0000000000..4f056332af
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.c
> @@ -0,0 +1,28 @@
> +/* Copyright 2024 Free Software Foundation, Inc.
> +
> +   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/>.  */
> +
> +void
> +frame_1 (void)
> +{
> +  __builtin_abort();
> +}
> +
> +int
> +main (void)
> +{
> +  frame_1 ();
> +}
> diff --git a/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
> new file mode 100644
> index 0000000000..5ba52cc069
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
> @@ -0,0 +1,130 @@
> +# Copyright 2024 Free Software Foundation, Inc.
If we do keep this test, the copyright year should be kept as 2018-2024, 
since the file is very close to an existing file.
> +
> +# 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 tests GDB's frame selection as used by the 'frame',
> +# 'select-frame', and 'info frame' commands in the corner case
> +# when the last instruction in a frame is a call.
> +
> +standard_testfile
> +
> +if {[prepare_for_testing "failed to prepare" $testfile $srcfile {debug}]} {
> +    return -1
> +}
> +
> +runto_main
> +gdb_breakpoint abort
> +gdb_continue_to_breakpoint abort
> +
> +gdb_test "bt" "#0  $hex in abort .*#1  $hex in frame_1 .*#2  $hex in main.*" "backtrace at breakpoint"

Similarly, if we keep this test, I would prefer if this backtrace test 
used the "multi_line". That's because ".*" could eat whole lines, and 
make the test pass when it shouldn't. Something like:

gdb_test "bt" [multi_line "#0 $hex in abort \[^\\r\\n\]*" "#1 $hex in 
frame_1\[^\\r\\n\]*" "#2  $hex in main \[^\\r\\n\]*"] "backtrace at 
breakpoint"

(with the appropriate line breaks, also I haven't tested the regexes, 
they probably have some error in them).

-- 
Cheers,
Guinevere Larsen
She/Her/Hers

> +
> +# Perform "info frame" to extract the frame's address.
> +proc get_frame_address { {testname ""} } {
> +    global hex gdb_prompt
> +
> +    set frame_address "unknown"
> +    set testname "get_frame_address: ${testname}"
> +    gdb_test_multiple "info frame" $testname {
> +	-re ", frame at ($hex):\r\n.*\r\n$gdb_prompt $" {
> +	    set frame_address $expect_out(1,string)
> +	    pass $testname
> +	}
> +    }
> +
> +    return $frame_address
> +}
> +
> +# Check that the current frame is at stack depth LEVEL, at frame
> +# address ADDRESS, and is in FUNCTION.
> +proc check_frame { level address function } {
> +    global hex gdb_prompt
> +
> +    set re [multi_line \
> +		"Stack level ${level}, frame at ($address):" \
> +		".* = $hex in ${function}( \(\[^\r\n\]*\))*; saved .* = $hex" \
> +		".*\r\n$gdb_prompt $" ]
> +
> +    set testname "check frame level ${level}"
> +    gdb_test_multiple "info frame" $testname {
> +	-re $re {
> +	    pass $testname
> +	}
> +    }
> +}
> +
> +# Select frame using level, but relying on this being the default
> +# action, so "frame 0" performs "frame level 0".
> +gdb_test "frame 0" "#0  $hex in abort.*"
> +set frame_0_address [ get_frame_address "frame 0" ]
> +gdb_test "frame 1" "#1  $hex in frame_1.*"
> +set frame_1_address [ get_frame_address "frame 1" ]
> +gdb_test "frame 2" "#2  $hex in main.*"
> +set frame_2_address [ get_frame_address "frame 2" ]
> +
> +# Select frame using 'level' specification.
> +gdb_test "frame level 0" "#0  $hex in abort.*"
> +gdb_test "frame level 1" "#1  $hex in frame_1.*"
> +gdb_test "frame level 2" "#2  $hex in main.*"
> +
> +# Select frame by address.
> +gdb_test "frame address ${frame_0_address}" "#0  $hex in abort.*" \
> +    "select frame 0 by address"
> +gdb_test "frame address ${frame_1_address}" "#1  $hex in frame_1.*" \
> +    "select frame 1 by address"
> +gdb_test "frame address ${frame_2_address}" "#2  $hex in main.*" \
> +    "select frame 2 by address"
> +
> +# Select frame by function.
> +gdb_test "frame function abort" "#0  $hex in abort.*"
> +gdb_test "frame function frame_1" "#1  $hex in frame_1.*"
> +gdb_test "frame function main" "#2  $hex in main.*"
> +
> +with_test_prefix "select-frame, no keyword" {
> +    gdb_test_no_output "select-frame 0"
> +    check_frame "0" "${frame_0_address}" "abort"
> +    gdb_test_no_output "select-frame 1"
> +    check_frame "1" "${frame_1_address}" "frame_1"
> +    gdb_test_no_output "select-frame 2"
> +    check_frame "2" "${frame_2_address}" "main"
> +}
> +
> +with_test_prefix "select-frame, keyword=level" {
> +    gdb_test_no_output "select-frame level 0"
> +    check_frame "0" "${frame_0_address}" "abort"
> +    gdb_test_no_output "select-frame level 1"
> +    check_frame "1" "${frame_1_address}" "frame_1"
> +    gdb_test_no_output "select-frame level 2"
> +    check_frame "2" "${frame_2_address}" "main"
> +}
> +
> +with_test_prefix "select-frame, keyword=address" {
> +    gdb_test_no_output "select-frame address ${frame_0_address}" \
> +	"select frame 0 by address"
> +    check_frame "0" "${frame_0_address}" "abort"
> +    gdb_test_no_output "select-frame address ${frame_1_address}" \
> +	"select frame 1 by address"
> +    check_frame "1" "${frame_1_address}" "frame_1"
> +    gdb_test_no_output "select-frame address ${frame_2_address}" \
> +	"select frame 2 by address"
> +    check_frame "2" "${frame_2_address}" "main"
> +}
> +
> +with_test_prefix "select-frame, keyword=function" {
> +    gdb_test_no_output "select-frame function abort"
> +    check_frame "0" "${frame_0_address}" "abort"
> +    gdb_test_no_output "select-frame function frame_1"
> +    check_frame "1" "${frame_1_address}" "frame_1"
> +    gdb_test_no_output "select-frame function main"
> +    check_frame "2" "${frame_2_address}" "main"
> +}

Re: [PATCH v2] gdb: fix "frame function" issue when call is last instruction

[Sahil]

Hi,

Thank you for the review.

On Wednesday, July 3, 2024 7:54:43 PM GMT+5:30 Guinevere Larsen wrote:
> [...]
> I would prefer if you wrote "the last instruction in a frame is a call"
> or something similar. I ask this because when I read the commit message,
> I thought "last instruction" was referring to last instruction executed,
> and couldn't understand  why I couldn't reproduce without __builtin_abort.
> 

Sorry about that. I should have been clearer. I'll modify the commit message.

> > Using "get_frame_address_in_block" instead of
> > "get_frame_pc" resolves this issue.
> > 
> > Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30929
> > ---
> > Changes since v1:
> > * frame-selection-last-instr-call.exp: Modify test to fix regressions on
> > arm/aarch64
> > 
> > Patch v1:
> > https://sourceware.org/pipermail/gdb-patches/2024-June/210251.html
> And responding to your v1 comment about why you created a new test file,
> you mention creating a new C file for this test. Have you tried
> extending the original frame-selection.c file with a new function that
> calls __builtin_abort instead?

I tried extending frame_selection.c initially but I was uncertain about
adding a function that calls __builtin_abort since "main" would never
return. I thought the last statement in main - "return i + j" - also plays
a role in the test.

After taking a look at the test again, I don't think that should cause
a hindrance. I'll experiment with this.

> It seems to me that it should work, and
> you should be able to extend the existing frame-selection.exp file, but
> I might be missing something.
> 
> I would prefer if we just extended the test, instead of added a new one,
> but if I indeed missed something and it isn't as easy to extend it as I
> think, we can have the new test :)
> 

I don't think extending frame-selection.exp should be difficult. I was mainly
unsure about frame-selection.c.

> [...]
> > diff --git a/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
> > b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp new file
> > mode 100644
> > index 0000000000..5ba52cc069
> > --- /dev/null
> > +++ b/gdb/testsuite/gdb.base/frame-selection-last-instr-call.exp
> > @@ -0,0 +1,130 @@
> > +# Copyright 2024 Free Software Foundation, Inc.
> 
> If we do keep this test, the copyright year should be kept as 2018-2024,
> since the file is very close to an existing file.

Ok, understood.

> [...]
> > +runto_main
> > +gdb_breakpoint abort
> > +gdb_continue_to_breakpoint abort
> > +
> > +gdb_test "bt" "#0  $hex in abort .*#1  $hex in frame_1 .*#2  $hex in
> > main.*" "backtrace at breakpoint"
> Similarly, if we keep this test, I would prefer if this backtrace test
> used the "multi_line". That's because ".*" could eat whole lines, and
> make the test pass when it shouldn't. Something like:
> 
> gdb_test "bt" [multi_line "#0 $hex in abort \[^\\r\\n\]*" "#1 $hex in
> frame_1\[^\\r\\n\]*" "#2  $hex in main \[^\\r\\n\]*"] "backtrace at
> breakpoint"
> 
> (with the appropriate line breaks, also I haven't tested the regexes,
> they probably have some error in them).
> 

Got it. I'll modify this.

Thanks,
Sahil