| Message ID | 20240420213307.976401-3-thiago.bauermann@linaro.org |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Add testcase for libc memory operations | expand |
On Sat, 20 Apr 2024 18:33:07 -0300 Thiago Jung Bauermann <thiago.bauermann@linaro.org> wrote: > diff --git a/gdb/testsuite/gdb.base/memops-watchpoint.exp b/gdb/testsuite/gdb.base/memops-watchpoint.exp > new file mode 100644 > index 000000000000..6fc84eb469c4 > --- /dev/null > +++ b/gdb/testsuite/gdb.base/memops-watchpoint.exp > @@ -0,0 +1,83 @@ > +# 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/>. > + > +# Test a binary that uses standard libc memory operation functions. They are > +# frequently optimized with specialized instructions, so make sure GDB behaves > +# correctly in their presence. > + > +# It's not possible to check in which libc function the watchpoint triggers > +# without its debug info. > +require libc_has_debug_info I'm wondering about the need for this requirement. When I comment it out and run it on a machine without libc debuginfo, I do see 3 FAILs, but it seems to me that those could be turned into PASSes by changing the regular expressions for the "continue until..." tests. E.g. for the first one, with libc debuginfo, I see: continue Continuing. Hardware watchpoint 2: -location a[31] Old value = 101 'e' New value = 0 '\000' __memset_avx2_unaligned () at ../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:146 146 VMOVU %VMM(0), (%rdi) (gdb) PASS: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits But, without libc debuginfo, the watchpoint still works: continue Continuing. Hardware watchpoint 2: -location a[31] Old value = 101 'e' New value = 0 '\000' 0x00007ffff7e3553a in __memset_avx2_unaligned () from /lib64/libc.so.6 (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits As stated earlier, this could be turned into a PASS by tweaking the RE. In both cases, we know that it's in a "memset" function. (The presence of minimal symbols provides GDB with this information.) Kevin
Thank you for your review! Kevin Buettner <kevinb@redhat.com> writes: > On Sat, 20 Apr 2024 18:33:07 -0300 > Thiago Jung Bauermann <thiago.bauermann@linaro.org> wrote: > >> diff --git a/gdb/testsuite/gdb.base/memops-watchpoint.exp >> b/gdb/testsuite/gdb.base/memops-watchpoint.exp >> new file mode 100644 >> index 000000000000..6fc84eb469c4 >> --- /dev/null >> +++ b/gdb/testsuite/gdb.base/memops-watchpoint.exp >> @@ -0,0 +1,83 @@ >> +# 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/>. >> + >> +# Test a binary that uses standard libc memory operation functions. They are >> +# frequently optimized with specialized instructions, so make sure GDB behaves >> +# correctly in their presence. >> + >> +# It's not possible to check in which libc function the watchpoint triggers >> +# without its debug info. >> +require libc_has_debug_info > > I'm wondering about the need for this requirement. When I comment it > out and run it on a machine without libc debuginfo, I do see 3 FAILs, > but it seems to me that those could be turned into PASSes by changing > the regular expressions for the "continue until..." tests. > > E.g. for the first one, with libc debuginfo, I see: > > continue > Continuing. > > Hardware watchpoint 2: -location a[31] > > Old value = 101 'e' > New value = 0 '\000' > __memset_avx2_unaligned () at ../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:146 > 146 VMOVU %VMM(0), (%rdi) > (gdb) PASS: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits > > But, without libc debuginfo, the watchpoint still works: > > continue > Continuing. > > Hardware watchpoint 2: -location a[31] > > Old value = 101 'e' > New value = 0 '\000' > 0x00007ffff7e3553a in __memset_avx2_unaligned () from /lib64/libc.so.6 > (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits > > As stated earlier, this could be turned into a PASS by tweaking the RE. > > In both cases, we know that it's in a "memset" function. (The presence > of minimal symbols provides GDB with this information.) I added the requirement because in my aarch64-linux system without libc6 debug info I get: continue Continuing. Hardware watchpoint 2: -location a[28] Old value = 104 'h' New value = 0 '\000' 0x0000fffff7e90664 in ?? () from /lib/aarch64-linux-gnu/libc.so.6 (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits And I just tested removing libc6-dbg from my x86_64-linux laptop: continue Continuing. Hardware watchpoint 2: -location a[28] Old value = 104 'h' New value = 0 '\000' 0x00007ffff7d8e05f in ?? () from /lib/x86_64-linux-gnu/libc.so.6 (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits So it depends on the system. One alternative would be to not use the require statement and run the test until the watchpoint hits, and have a case in gdb_test_multiple to mark as UNRESOLVED if the function name is '??'. -- Thiago
On Sun, 21 Apr 2024 21:24:42 -0300 Thiago Jung Bauermann <thiago.bauermann@linaro.org> wrote: > >> +require libc_has_debug_info > > > > I'm wondering about the need for this requirement. When I comment it > > out and run it on a machine without libc debuginfo, I do see 3 FAILs, > > but it seems to me that those could be turned into PASSes by changing > > the regular expressions for the "continue until..." tests. [...] > > I added the requirement because in my aarch64-linux system without libc6 > debug info I get: > > continue > Continuing. > > Hardware watchpoint 2: -location a[28] > > Old value = 104 'h' > New value = 0 '\000' > 0x0000fffff7e90664 in ?? () from /lib/aarch64-linux-gnu/libc.so.6 > (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits > > And I just tested removing libc6-dbg from my x86_64-linux laptop: > > continue > Continuing. > > Hardware watchpoint 2: -location a[28] > > Old value = 104 'h' > New value = 0 '\000' > 0x00007ffff7d8e05f in ?? () from /lib/x86_64-linux-gnu/libc.so.6 > (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits > > So it depends on the system. What distro are you using? > One alternative would be to not use the require statement and run the > test until the watchpoint hits, and have a case in gdb_test_multiple to > mark as UNRESOLVED if the function name is '??'. I'm in favor of this approach. If we stick with the require statement, I think that Fedora testing will frequently show this new test as unsupported since installing debuginfo is less common / important that it used to be. (This is due to debuginfod doing it for you. But I think that debuginfod is mostly disabled when running the GDB tests.)
Thiago Jung Bauermann <thiago.bauermann@linaro.org> writes: > Kevin Buettner <kevinb@redhat.com> writes: > >> On Sat, 20 Apr 2024 18:33:07 -0300 >> Thiago Jung Bauermann <thiago.bauermann@linaro.org> wrote: >> >>> +# It's not possible to check in which libc function the watchpoint triggers >>> +# without its debug info. >>> +require libc_has_debug_info >> >> I'm wondering about the need for this requirement. When I comment it >> out and run it on a machine without libc debuginfo, I do see 3 FAILs, >> but it seems to me that those could be turned into PASSes by changing >> the regular expressions for the "continue until..." tests. >> >> E.g. for the first one, with libc debuginfo, I see: >> >> continue >> Continuing. >> >> Hardware watchpoint 2: -location a[31] >> >> Old value = 101 'e' >> New value = 0 '\000' >> __memset_avx2_unaligned () at >> ../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:146 >> 146 VMOVU %VMM(0), (%rdi) >> (gdb) PASS: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits >> >> But, without libc debuginfo, the watchpoint still works: >> >> continue >> Continuing. >> >> Hardware watchpoint 2: -location a[31] >> >> Old value = 101 'e' >> New value = 0 '\000' >> 0x00007ffff7e3553a in __memset_avx2_unaligned () from /lib64/libc.so.6 >> (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits >> >> As stated earlier, this could be turned into a PASS by tweaking the RE. >> >> In both cases, we know that it's in a "memset" function. (The presence >> of minimal symbols provides GDB with this information.) > > I added the requirement because in my aarch64-linux system without libc6 > debug info I get: > > continue > Continuing. > > Hardware watchpoint 2: -location a[28] > > Old value = 104 'h' > New value = 0 '\000' > 0x0000fffff7e90664 in ?? () from /lib/aarch64-linux-gnu/libc.so.6 > (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits <snip> > So it depends on the system. > > One alternative would be to not use the require statement and run the > test until the watchpoint hits, and have a case in gdb_test_multiple to > mark as UNRESOLVED if the function name is '??'. I was able to do this in v3. But I mark the test as UNSUPPORTED rather than UNRESOLVED. The difference between them isn't always clear in my mind, but I think UNSUPPORTED is better in this case. -- Thiago
Kevin Buettner <kevinb@redhat.com> writes: > On Sun, 21 Apr 2024 21:24:42 -0300 > Thiago Jung Bauermann <thiago.bauermann@linaro.org> wrote: > >> >> +require libc_has_debug_info >> > >> > I'm wondering about the need for this requirement. When I comment it >> > out and run it on a machine without libc debuginfo, I do see 3 FAILs, >> > but it seems to me that those could be turned into PASSes by changing >> > the regular expressions for the "continue until..." tests. > [...] >> >> I added the requirement because in my aarch64-linux system without libc6 >> debug info I get: >> >> continue >> Continuing. >> >> Hardware watchpoint 2: -location a[28] >> >> Old value = 104 'h' >> New value = 0 '\000' >> 0x0000fffff7e90664 in ?? () from /lib/aarch64-linux-gnu/libc.so.6 >> (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits >> >> And I just tested removing libc6-dbg from my x86_64-linux laptop: >> >> continue >> Continuing. >> >> Hardware watchpoint 2: -location a[28] >> >> Old value = 104 'h' >> New value = 0 '\000' >> 0x00007ffff7d8e05f in ?? () from /lib/x86_64-linux-gnu/libc.so.6 >> (gdb) FAIL: gdb.base/memops-watchpoint.exp: continue until memset watchpoint hits >> >> So it depends on the system. > > What distro are you using? Ubuntu. Version 22.04 on some machines, and version 23.10 on others. >> One alternative would be to not use the require statement and run the >> test until the watchpoint hits, and have a case in gdb_test_multiple to >> mark as UNRESOLVED if the function name is '??'. > > I'm in favor of this approach. > > If we stick with the require statement, I think that Fedora testing > will frequently show this new test as unsupported since installing > debuginfo is less common / important that it used to be. (This is > due to debuginfod doing it for you. But I think that debuginfod is > mostly disabled when running the GDB tests.) Yes, that is a good point. I think v3 should work with the output you pasted in a previous email. Thank you for bring this up. -- Thiago
diff --git a/gdb/testsuite/gdb.base/memops-watchpoint.c b/gdb/testsuite/gdb.base/memops-watchpoint.c new file mode 100644 index 000000000000..13e923faa1e9 --- /dev/null +++ b/gdb/testsuite/gdb.base/memops-watchpoint.c @@ -0,0 +1,40 @@ +/* This test program is part of GDB, the GNU debugger. + + 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/>. */ + +#include <stdio.h> +#include <string.h> + +int +main (void) +{ + char s[40] = "This is a relatively long string..."; + char a[40] = "String to be overwritten with zeroes"; + char b[40] = "Another string to be memcopied..."; + char c[40] = "Another string to be memmoved..."; + + /* Break here. */ + memset (a, 0, sizeof (a)); + + memcpy (b, s, sizeof (b)); + + memmove (c, s, sizeof (c)); + + printf ("b = '%s'\n", b); + printf ("c = '%s'\n", c); + + return 0; +} diff --git a/gdb/testsuite/gdb.base/memops-watchpoint.exp b/gdb/testsuite/gdb.base/memops-watchpoint.exp new file mode 100644 index 000000000000..6fc84eb469c4 --- /dev/null +++ b/gdb/testsuite/gdb.base/memops-watchpoint.exp @@ -0,0 +1,83 @@ +# 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/>. + +# Test a binary that uses standard libc memory operation functions. They are +# frequently optimized with specialized instructions, so make sure GDB behaves +# correctly in their presence. + +# It's not possible to check in which libc function the watchpoint triggers +# without its debug info. +require libc_has_debug_info + +standard_testfile + +set options "-fno-builtin-memset -fno-builtin-memcpy -fno-builtin-memmove" +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile} \ + [list debug additional_flags=$options]] } { + return -1 +} + +set linespec ${srcfile}:[gdb_get_line_number "Break here"] + +if ![runto ${linespec}] { + return +} + +gdb_test "watch -location a\[31\]" \ + "(Hardware w|W)atchpoint ${decimal}: -location a\\\[31\\\]" \ + "set watch on a" +gdb_test "watch -location b\[31\]" \ + "(Hardware w|W)atchpoint ${decimal}: -location b\\\[31\\\]" \ + "set watchpoint on b" +gdb_test "watch -location c\[31\]" \ + "(Hardware w|W)atchpoint ${decimal}: -location c\\\[31\\\]" \ + "set watchpoint on c" + +gdb_test "continue" \ + [multi_line \ + "Continuing\\." \ + "" \ + "(Hardware w|W)atchpoint ${decimal}: -location a\\\[31\\\]" \ + "" \ + "Old value = 101 'e'" \ + "New value = 0 '\\\\000'" \ + ".*memset.* \\(\\) at .*:$decimal" \ + ".*"] \ + "continue until memset watchpoint hits" + +gdb_test "continue" \ + [multi_line \ + "Continuing\\." \ + "" \ + "(Hardware w|W)atchpoint ${decimal}: -location b\\\[31\\\]" \ + "" \ + "Old value = 46 '\\.'" \ + "New value = 103 'g'" \ + ".*memcpy.* \\(\\) at .*:$decimal" \ + ".*"] \ + "continue until memcpy watchpoint hits" + +# Note: Some architectures use memcpy for memmove. +gdb_test "continue" \ + [multi_line \ + "Continuing\\." \ + "" \ + "(Hardware w|W)atchpoint ${decimal}: -location c\\\[31\\\]" \ + "" \ + "Old value = 46 '\\.'" \ + "New value = 103 'g'" \ + ".*(memmove|memcpy).* \\(\\) at .*:$decimal" \ + ".*"] \ + "continue until memmove watchpoint hits"