Created attachment 1701683 [details]
Kernel patch for GDB regression

1. Please describe the problem:

I'm seeing a GDB regression when running gdb.base/corefile.exp.  This is NOT a GDB problem.  I have isolated the problem to a specific recent kernel patch and found a typo in that patch.  I have a fix for this problem.

2. What is the Version-Release number of the kernel:

Fedora 32: 5.6.16-300.fc32, 5.6.19-300.fc32, 5.7.8-200.fc32. (And, to the best of my knowledge, all versions in between the above listed versions too.)

Rawhide: 5.8.0-0.rc5.20200717git07a56bb875af.1.fc33

3. Did it work previously in Fedora? If so, what kernel version did the issue
   *first* appear?  Old kernels are available for download at
   https://koji.fedoraproject.org/koji/packageinfo?packageID=8 :

The GDB regression occurs in 5.6.16-300.fc32.

No GDB regression in 5.6.15-300.fc32.

4. Can you reproduce this issue? If so, please provide the steps to reproduce
   the issue below:

4a) Fetch upstream GDB sources from git:

git clone git://sourceware.org/git/binutils-gdb.git

4b) Build gdb:

cd binutils-gdb
mkdir inst
./configure --prefix=`pwd`/inst
make -j24 && make install

4c) Run gdb.base/corefile.exp test:

cd gdb
make check RUNTESTFLAGS="gdb.base/corefile.exp"

4d) Observe failure and summary output:

FAIL: gdb.base/corefile.exp: core-file warning-free

		=== gdb Summary ===

# of expected passes		27
# of unexpected failures	1

There should be no unexpected failures when running this test.

4e) Run gdb by hand to see problem when loading core file:

./gdb -q testsuite/outputs/gdb.base/corefile/corefile{,.core}

4f) Observe output from above command:

Reading symbols from testsuite/outputs/gdb.base/corefile/corefile...
[New LWP 64835]

warning: Unexpected size of section `.reg-xstate/64835' in core file.
Core was generated by `/mesquite2/scratch/binutils-gdb/gdb/testsuite/outputs/gdb.base/corefile/corefil'.
Program terminated with signal SIGABRT, Aborted.

warning: Unexpected size of section `.reg-xstate/64835' in core file.
#0  0x00007ff548f5ea25 in raise () from /lib64/libc.so.6
(gdb) 

4g) Discussion:

The problem in the above output is the warning:
  "warning: Unexpected size of section `.reg-xstate/64835' in core file."

The rest of the output is okay / expected.

5. Does this problem occur with the latest Rawhide kernel? To install the
   Rawhide kernel, run ``sudo dnf install fedora-repos-rawhide`` followed by
   ``sudo dnf update --enablerepo=rawhide kernel``:

Yes, I've tested on Rawhide using the latest kernel available at the time of this bug report: 5.8.0-0.rc5.20200717git07a56bb875af.1.fc33.  I see the same problem in that rawhide kernel.

6. Are you running any modules that not shipped with directly Fedora's kernel?:

No.

7. Please attach the kernel logs. You can get the complete kernel log
   for a boot with ``journalctl --no-hostname -k > dmesg.txt``. If the
   issue occurred on a previous boot, use the journalctl ``-b`` flag.

Kernel logs are not relevant for this problem.

I'm an upstream GDB developer.  I've investigated this regression to see where the problem lies.  The problem is *not* in GDB.

I found that the GDB regression goes away when I revert upstream commit 9e4636545933131de15e1ecd06733538ae939b2f :

https://github.com/torvalds/linux/commit/9e4636545933131de15e1ecd06733538ae939b2f

It's a patch by Al Viro titled "copy_xstate_to_kernel(): don't leave parts of destination uninitialized".  There is a bug/typo in this patch in accessing xstate_offsets[].  XFEATURE_MASK_SSE is used as an index, which is not correct; the index XFEATURE_SSE should have been used instead.

I've sent a patch fixing this problem to Al Viro <viro.org.uk> and the main linux kernel mailing list (linux-kernel.org).  (I don't do this very often; it's possible that some other list might have been more appropriate.)  Here's a link to my patch:

https://lkml.org/lkml/2020/7/18/96

I've added my patch as an attachment too.

I've tested my patch against 5.8.0-0.rc5.20200715gite9919e11e219.1 on rawhide and 5.7.8-200 on Fedora 32. In both cases, the GDB regression reported above has been fixed when using my patch.