Re: [uml-devel] [PATCH] um: v3: Fix FP register size for XSTATE/XSAVE

[Thomas M. <th...@m3...>]

On Wed, Jul 12, 2017 at 09:18:26PM +0200, Richard Weinberger wrote:
> Thomas,
> 
> Am 12.07.2017 um 17:11 schrieb Thomas Meyer:
> > Am 10.07.2017 9:06 nachm. schrieb Richard Weinberger <ri...@no...>:
> > 
> >     Thomas,
> > 
> >     Am 10.07.2017 um 00:33 schrieb Thomas Meyer:
> >     > Hard code max size. Taken from
> >     > https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=gdb/common/x86-xstate.h
> >     >
> >     > v3: use static fp_regs for get_skas_faultinfo
> >     >
> >     > Signed-off-by: Thomas Meyer <th...@m3...>
> >     > ---
> >     >  arch/um/os-Linux/skas/process.c  |  7 ++++---
> >     >  arch/x86/um/os-Linux/registers.c | 18 ++++++++++++------
> >     >  arch/x86/um/user-offsets.c       |  2 +-
> >     >  3 files changed, 17 insertions(+), 10 deletions(-)
> >     >
> >     > diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
> >     > index 03b3c4cc7735..b5fdef795985 100644
> >     > --- a/arch/um/os-Linux/skas/process.c
> >     > +++ b/arch/um/os-Linux/skas/process.c
> >     > @@ -88,12 +88,13 @@ void wait_stub_done(int pid)
> >     > 
> >     >  extern unsigned long current_stub_stack(void);
> >     > 
> >     > +static unsigned long skas_faultinfo_fpregs[FP_SIZE];
> >     > +
> > 
> >     Uhmm, are you sure that this does not race with other userspace() instances?
> > 
> > Oh, is it really possible that a task gets schedule while we are in userspace signal handling from ptrace?
> 
> I hoped that you'd double check. ;)

I just did write a little test program, which seems to survive this
patch. Please have a look at this test program. As far as I understand
the problem, the test program should cover above test case.

Feel free to have a look at it:
https://github.com/thomasmey/xmm-test/

> From my knowledge it is not possible, but this code was not written by me and UML internal
> semantics are more than tricky.
> Let's postpone this patch for -rc2 and check again.
> 
> > Another question: when malloc a memory area in userspace: when does a userspace process end in userspace function!? Where would you free the malloced memory?
> 
> It ends via handle_trap() when userspac does an exit syscall.
> I agree this is tricky to catch and needs more effort.

Ah, okay, that makes sense!
> 
> Thanks,
> //richard