[Bug 58789] New: Changed behaviour when running a single wine process with valgrind since wine-10.9.

[WineHQ Bugzilla]

http://bugs.winehq.org/show_bug.cgi?id=58789

Bug ID: 58789
           Summary: Changed behaviour when running a single wine process
                    with valgrind since wine-10.9.
           Product: Wine
           Version: 10.9
          Hardware: x86-64
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ntdll
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

Before 5e70a4d81b I was able to start the background
wine processes by e.g. `wine taskmgr &`.
Then start just the target process with valgrind.

Since 5e70a4d81b this causes an segv_handler loop
starting with this message:

vex x86->IR: unhandled instruction bytes: 0xF 0xAE 0x6C 0x24
  ==833752== valgrind: Unrecognised instruction at address 0x470ed67.
  ==833752==    at 0x470ED67: __wine_syscall_dispatcher (in
dlls/ntdll/ntdll.so)

This instruction is the `xrstor 0x40(%esp)` from __wine_syscall_dispatcher,
which seems not yet supported by valgrind.

If I see it right, before 5e70a4d81b each process did its own lookup
of cpu flags, but since 5e70a4d81b this is just done by the first one
and then shared via user_shared_data.

As a workaround preventing the assignment of `features[PF_XSAVE_ENABLED]`
allows me start it with a recent wine git build.

I am uncertain if this qualifies as regression as I am uncertain,
if this is a valid mode of operation when splitting wine processes that way.