Re: [Qemu-devel] [PATCH v2 0/6] user-exec: cpu_resume_from_signal() cleanups

[Riku Voipio]

On Mon, Jun 06, 2016 at 05:57:35PM +0100, Peter Maydell wrote:
> On 6 June 2016 at 15:55, Peter Maydell <address@hidden> wrote:
> > Ping!
> 
> Thanks for the review, Sergey. Unless anybody else wants to review
> or wants to take it through their tree (Riku?), I propose to apply
> this to master sometime later this week.

Feel free to apply these yourself,
Acked-by: Riku Voipio <address@hidden>

> thanks
> -- PMM
> 
> > On 17 May 2016 at 15:18, Peter Maydell <address@hidden> wrote:
> >> I was trying to reason about user-mode's handling of signal masks,
> >> and I found our current code a bit confusing, so I cleaned it up.
> >>
> >> At the moment for user-only mode cpu_resume_from_signal() takes a
> >> usercontext pointer; if this is non-NULL then it has some awkward
> >> OS and CPU specific code to set the signal mask from something
> >> inside the usercontext before doing the same kind of siglongjmp()
> >> that the softmmu cpu_resume_from_signal() does.
> >>
> >> In fact the two use cases are completely separate:
> >>  * almost all calls to cpu_resume_from_signal() pass a NULL puc
> >>    argument (and most of those are softmmu-only anyway)
> >>  * only the code path handle_cpu_signal -> page_unprotect ->
> >>    tb_invalidate_phys_page -> cpu_resume_from_signal will pass
> >>    a non-NULL puc.
> >>
> >> The cleanups are:
> >>  * pull the call to cpu_resume_from_signal() up through the
> >>    callstack so we do the signal mask manipulation in
> >>    handle_cpu_signal()
> >>  * drop the OS/CPU spceific code to get a signal mask out of
> >>    a usercontext, because in the specific case of handle_cpu_signal()
> >>    we already have the signal mask value and can just use it
> >>  * rename cpu_resume_from_signal() to cpu_loop_exit_noexc(),
> >>    since all the remaining callsites are not in fact signal handlers
> >>    or even called from signal handlers
> >>  * get rid of an ugly TARGET_I386 ifdef in user-exec.c by moving
> >>    the i386-specific code into its handle_mmu_fault hook.
> >>
> >> Changes v1->v2:
> >>  * patches 1-4 are the same and already reviewed
> >>  * patch 5 is new, and just adds a clarifying comment to
> >>    do_interrupt_user()
> >>  * patch 6 is the old patch 5, and now sets env->exception_next_eip
> >>    to -1 as a clear indication that the value is not going to be used
> >>    (as noted in the comment in the new patch 5)
> >>
> >> thanks
> >> -- PMM
> >>
> >>
> >> Peter Maydell (6):
> >>   translate-all.c: Don't pass puc, locked to tb_invalidate_phys_page()
> >>   user-exec: Push resume-from-signal code out to handle_cpu_signal()
> >>   cpu-exec: Rename cpu_resume_from_signal() to cpu_loop_exit_noexc()
> >>   user-exec: Don't reextract sigmask from usercontext pointer
> >>   target-i386: Add comment about do_interrupt_user() next_eip argument
> >>   target-i386: Move user-mode exception actions out of user-exec.c
> >>
> >>  cpu-exec-common.c        |  8 ++---
> >>  exec.c                   |  2 +-
> >>  hw/i386/kvmvapic.c       |  2 +-
> >>  include/exec/exec-all.h  |  2 +-
> >>  target-i386/bpt_helper.c |  2 +-
> >>  target-i386/helper.c     |  2 ++
> >>  target-i386/seg_helper.c |  6 +++-
> >>  target-lm32/helper.c     |  2 +-
> >>  target-s390x/helper.c    |  2 +-
> >>  target-xtensa/helper.c   |  2 +-
> >>  translate-all.c          | 40 ++++++++++++---------
> >>  translate-all.h          |  2 +-
> >>  user-exec.c              | 93 
> >> +++++++++++++++++++++---------------------------
> >>  13 files changed, 82 insertions(+), 83 deletions(-)
> >>
> >> --
> >> 1.9.1