Re: [PATCH] io_thread/x86: don't reset 'cs', 'ss', 'ds' and 'es' registers for io_threads

Am 04.05.21 um 17:55 schrieb Simon Marchi:

On 2021-05-04 11:35 a.m., Borislav Petkov wrote:

quoted

On Tue, May 04, 2021 at 10:39:23AM +0200, Peter Zijlstra wrote:

quoted

Anybody on toolchains that can help get GDB fixed?

In the meantime, Tom is looking at fixing this, in case people wanna try
gdb patches or give him a test case or so...

https://sourceware.org/bugzilla/show_bug.cgi?id=27822

Yes, please provide reproducing steps in that bug.  Unlike what was said
in this thread, some people do work on gdb and are willing to fix
things, but they can only do so if they know about the problem.

See https://lore.kernel.org/io-uring/0375b37f-2e1e-7999-53b8-c567422aa181@samba.org/ (local)
and https://lore.kernel.org/io-uring/20210411152705.2448053-1-metze@samba.org/T/#m461f280e8c3d32a49bc7da7bb5e214e90d97cf65 (local)

The question is why does inferior_ptid doesn't represent the thread
that was specified by 'gdb --pid PIDVAL'

https://www.samba.org/~metze/strace-uring-fail.txt
used "gdb --pid 1396" and does the following ptrace calls:

# grep ptrace strace-uring-fail.txt

18:46:35.319925 ptrace(PTRACE_ATTACH, 1396) = 0 <0.000048>
18:46:35.321622 ptrace(PTRACE_ATTACH, 1397) = 0 <0.000059>
18:46:35.322813 ptrace(PTRACE_ATTACH, 1398) = 0 <0.003052>
18:46:35.327287 ptrace(PTRACE_ATTACH, 1399) = 0 <0.000028>
18:46:35.334920 ptrace(PTRACE_GETREGS, 1396, NULL, 0x7ffed6173ea0) = 0 <0.000067>
18:46:35.341506 ptrace(PTRACE_SETOPTIONS, 1410, NULL, PTRACE_O_TRACESYSGOOD) = 0 <0.000056>
18:46:35.341681 ptrace(PTRACE_SETOPTIONS, 1410, NULL, PTRACE_O_TRACEFORK) = 0 <0.000051>
18:46:35.341816 ptrace(PTRACE_SETOPTIONS, 1410, NULL, PTRACE_O_TRACEFORK|PTRACE_O_TRACEVFORKDONE) = 0 <0.000054>
18:46:35.341957 ptrace(PTRACE_CONT, 1410, NULL, 0) = 0 <0.000056>
18:46:35.345568 ptrace(PTRACE_GETEVENTMSG, 1410, NULL, [1411]) = 0 <0.000081>
18:46:35.350541 ptrace(PTRACE_SETOPTIONS, 1410, NULL, PTRACE_O_EXITKILL) = 0 <0.000019>
18:46:35.354010 ptrace(PTRACE_SETOPTIONS, 1397, NULL, PTRACE_O_TRACESYSGOOD|PTRACE_O_TRACEFORK|PTRACE_O_TRACEVFORK|PTRACE_O_TRACECLONE|PTRACE_O_TRACEEXEC|PTRACE_O_TRACEVFORKDONE) = 0 <0.000019>
18:46:35.415730 ptrace(PTRACE_SETOPTIONS, 1396, NULL, PTRACE_O_TRACESYSGOOD|PTRACE_O_TRACEFORK|PTRACE_O_TRACEVFORK|PTRACE_O_TRACECLONE|PTRACE_O_TRACEEXEC|PTRACE_O_TRACEVFORKDONE) = 0 <0.000076>
18:46:35.421076 ptrace(PTRACE_GETREGS, 1412, NULL, 0x7ffed6174980) = 0 <0.000088>
18:46:35.429498 ptrace(PTRACE_PEEKUSER, 1397, 8*CS, [NULL]) = 0 <0.000022>
18:46:35.429632 ptrace(PTRACE_PEEKUSER, 1397, 8*SS + 24, [NULL]) = 0 <0.000019>
18:46:35.429732 ptrace(PTRACE_GETREGSET, 1397, NT_X86_XSTATE, [{iov_base=0x7ffed6174780, iov_len=576}]) = 0 <0.000030>
18:46:35.435507 ptrace(PTRACE_GETREGS, 1397, NULL, 0x7ffed6173cb0) = 0 <0.000019>
18:46:35.445877 ptrace(PTRACE_PEEKTEXT, 1397, 0x56357e99de00, [0x7f49d572b160]) = 0 <0.000057>
18:46:35.446043 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572b168, [0x7f49d572b190]) = 0 <0.000049>
18:46:35.447192 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572bbf0, [0x64762d78756e696c]) = 0 <0.000060>
18:46:35.447368 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572bbf0, [0x64762d78756e696c]) = 0 <0.000075>
18:46:35.447571 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572bbf8, [0x312e6f732e6f73]) = 0 <0.000070>
18:46:35.447762 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572bbf8, [0x312e6f732e6f73]) = 0 <0.000067>
18:46:35.448658 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be10, [0x3638782f62696c2f]) = 0 <0.000076>
18:46:35.448917 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be10, [0x3638782f62696c2f]) = 0 <0.000050>
18:46:35.449051 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be18, [0x756e696c2d34365f]) = 0 <0.000045>
18:46:35.449173 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be18, [0x756e696c2d34365f]) = 0 <0.000043>
18:46:35.449292 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be20, [0x696c2f756e672d78]) = 0 <0.000042>
18:46:35.449414 ptrace(PTRACE_PEEKTEXT, 1397, 0x7f49d572be20, [0x696c2f756e672d78]) = 0 <0.000048>

ptrace(PTRACE_GETREGS, 1396, ... looks expected to me, but
starting with ptrace(PTRACE_PEEKUSER, 1397, 8*CS, [NULL]) (which triggers the actual problem)
it's unexpected to me why 1397 is used instead of 1396.

1397 is the iou-mgr-1396 iothread.

I hope that helps!
metze