Daniel Jacobowitz wrote:
> On Wed, Aug 15, 2001 at 02:53:12PM +0200, Carsten Langgaard wrote:
> > There has been some reports regarding FP emulator failures, which the
> > attached patch should solve.
> > The patch include a fix for emulation of instructions in a COP1
> > delay-slot, a fix for FP context switching and some additional stuff ,
> > which was needed to pass our torture test.
> >
> > Ralf could you please apply this patch.
>
> Two comments, especially since parts of this seem to be the patch I
> posted here over a month ago.
You are absolutely right, it's your old patch.
>
> > Index: linux/arch/mips/kernel/signal.c
>
> > @@ -353,12 +355,11 @@
> > owned_fp = (current == last_task_used_math);
> > err |= __put_user(owned_fp, &sc->sc_ownedfp);
> >
> > - if (current->used_math) { /* fp is active. */
> > + if (owned_fp) { /* fp is active. */
> > set_cp0_status(ST0_CU1);
> > err |= save_fp_context(sc);
> > last_task_used_math = NULL;
> > regs->cp0_status &= ~ST0_CU1;
> > - current->used_math = 0;
> > }
> >
> > return err;
>
> This is absolutely not right. It's righter than the status quo. If we
> don't own the FP, you don't save the FP. Then we can use FP in the
> signal handler, corrupting the process's original floating point
> context.
You are probably right, this is not a proper fix, but at least it solves the problems people
has been seeing.
So until we come up with a better fix, this is still better than the current sources.
>
> > Index: linux/include/asm-mips/processor.h
>
> > @@ -235,8 +215,8 @@
> > * Do necessary setup to start up a newly executed thread.
> > */
> > #define start_thread(regs, new_pc, new_sp) do { \
> > - /* New thread looses kernel privileges. */ \
> > - regs->cp0_status = (regs->cp0_status & ~(ST0_CU0|ST0_KSU)) | KU_USER;\
> > + /* New thread loses kernel and FPU privileges. */ \
> > + regs->cp0_status = (regs->cp0_status & ~(ST0_CU0|ST0_KSU|ST0_CU1)) | KU_USER;\
> > regs->cp0_epc = new_pc; \
> > regs->regs[29] = new_sp; \
> > current->thread.current_ds = USER_DS; \
>
> I could be misremembering, but I believe that Ralf said this should be
> unnecessary and the problem was somewhere else. On the other hand, I
> still think it's a good idea.
>
> --
> Daniel Jacobowitz Carnegie Mellon University
> MontaVista Software Debian GNU/Linux Developer
--
_ _ ____ ___ Carsten Langgaard Mailto:carstenl@mips.com
|\ /|||___)(___ MIPS Denmark Direct: +45 4486 5527
| \/ ||| ____) Lautrupvang 4B Switch: +45 4486 5555
TECHNOLOGIES 2750 Ballerup Fax...: +45 4486 5556
Denmark http://www.mips.com
