On Thu, 8 Feb 2007 09:53:18 +0100, "Franck Bui-Huu" <vagabon.xyz@xxxxxxxxx> wrote: > I tried the following patch: > > diff --git a/arch/mips/kernel/signal.c b/arch/mips/kernel/signal.c > index 229276a..046fb1b 100644 > --- a/arch/mips/kernel/signal.c > +++ b/arch/mips/kernel/signal.c > @@ -68,7 +68,9 @@ int setup_sigcontext(struct pt_regs *regs, struct > sigcontext __user *sc) > err |= __put_user(regs->cp0_epc, &sc->sc_pc); > > err |= __put_user(0, &sc->sc_regs[0]); > - for (i = 1; i < 32; i++) > + for (i = 1; i < 16; i++) > + err |= __put_user(regs->regs[i], &sc->sc_regs[i]); > + for (i = 24; i < 32; i++) > err |= __put_user(regs->regs[i], &sc->sc_regs[i]); > > err |= __put_user(regs->hi, &sc->sc_mdhi); > @@ -126,7 +128,9 @@ int restore_sigcontext(struct pt_regs *regs, > struct sigcontext __user *sc) > err |= __get_user(treg, &sc->sc_dsp); wrdsp(treg, DSP_MASK); > } > > - for (i = 1; i < 32; i++) > + for (i = 1; i < 16; i++) > + err |= __get_user(regs->regs[i], &sc->sc_regs[i]); > + for (i = 24; i < 32; i++) > err |= __get_user(regs->regs[i], &sc->sc_regs[i]); > > err |= __get_user(used_math, &sc->sc_used_math); > > ...and it still passes LTP tests. > > Someone reported that not saving/restoring static registers may break > user tools but the gain is important I think. NO! This change might silently corrupt static registers! If you did not restore static registers in kernel stack on restore_sigcontext(), succeeding RESTORE_STATIC in restore_all will load garbages to static registers. Note that any hardware interrupts in middle of signal handler overwrite pt_regs area in kernel stack. I can still remember random static register corruption bug and how hard to debug ... --- Atsushi Nemoto
