On Tue, 10 Apr 2012 at 17:23, Benjamin Herrenschmidt wrote: > can you test the patch below ?
When applied to 3.4-rc2, Linux boots again, yay! :-) Tested-by: Christian Kujau <li...@nerdbynature.de> Thanks! Christian. > From 08f1ec8a594c60bf3856e3c45b6d15fd691d90bb Mon Sep 17 00:00:00 2001 > From: Benjamin Herrenschmidt <b...@kernel.crashing.org> > Date: Tue, 10 Apr 2012 17:21:35 +1000 > Subject: [PATCH] powerpc: Fix page fault with lockdep regression > > commit a546498f3bf9aac311c66f965186373aee2ca0b0 > introduced a regression on 32-bit when irq tracing > is enabled by exposing an old bug in our irq tracing > code for exception entry. > > The code would save and restore some GPRs around the > calls to the C lockdep code, however, it tries to be > too smart for its own good and restores some of the > GPRs from the exception frame (as saved there on > exception entry). > > However, for page faults, we do replace those GPRs with > arguments to do_page_fault before we call transfer_to_handler > and so restoring from the exception frame is plain wrong in > this case. > > This was fine as long as we didn't touch the interrupt state > when taking page fault, but when I started doing it, it would > trigger the lockdep calls and the bug. > > This fixes it by cleaning up that code a bit. It did create > a small stack frame for the sake of backtraces, so let's > make it a bit bigger and use it to save and restore the > stuff we care about. > > Signed-off-by: Benjamin Herrenschmidt <b...@kernel.crashing.org> > --- > arch/powerpc/kernel/entry_32.S | 39 +++++++++++++++++++++------------------ > 1 file changed, 21 insertions(+), 18 deletions(-) > > diff --git a/arch/powerpc/kernel/entry_32.S b/arch/powerpc/kernel/entry_32.S > index 3e57a00..ba3aeb4 100644 > --- a/arch/powerpc/kernel/entry_32.S > +++ b/arch/powerpc/kernel/entry_32.S > @@ -206,40 +206,43 @@ reenable_mmu: /* re-enable > mmu so we can */ > andi. r10,r10,MSR_EE /* Did EE change? */ > beq 1f > > - /* Save handler and return address into the 2 unused words > - * of the STACK_FRAME_OVERHEAD (sneak sneak sneak). Everything > - * else can be recovered from the pt_regs except r3 which for > - * normal interrupts has been set to pt_regs and for syscalls > - * is an argument, so we temporarily use ORIG_GPR3 to save it > - */ > - stw r9,8(r1) > - stw r11,12(r1) > - stw r3,ORIG_GPR3(r1) > /* > * The trace_hardirqs_off will use CALLER_ADDR0 and CALLER_ADDR1. > * If from user mode there is only one stack frame on the stack, and > * accessing CALLER_ADDR1 will cause oops. So we need create a dummy > * stack frame to make trace_hardirqs_off happy. > + * > + * This is handy because we also need to save a bunch of GPRs, > + * r3 can be different from GPR3(r1) at this point, r9 and r11 > + * contains the old MSR and handler address respectively, > + * r4 & r5 can contain page fault arguments that need to be passed > + * along as well. r12, CCR, CTR, XER etc... are left clobbered as > + * they aren't useful past this point (aren't syscall arguments), > + * the rest is restored from the exception frame. > */ > + stwu r1,-32(r1) > + stw r9,8(r1) > + stw r11,12(r1) > + stw r3,16(r1) > + stw r4,20(r1) > + stw r5,24(r1) > andi. r12,r12,MSR_PR > - beq 11f > - stwu r1,-16(r1) > + b 11f > bl trace_hardirqs_off > - addi r1,r1,16 > b 12f > - > 11: > bl trace_hardirqs_off > 12: > + lwz r5,24(r1) > + lwz r4,20(r1) > + lwz r3,16(r1) > + lwz r11,12(r1) > + lwz r9,8(r1) > + addi r1,r1,32 > lwz r0,GPR0(r1) > - lwz r3,ORIG_GPR3(r1) > - lwz r4,GPR4(r1) > - lwz r5,GPR5(r1) > lwz r6,GPR6(r1) > lwz r7,GPR7(r1) > lwz r8,GPR8(r1) > - lwz r9,8(r1) > - lwz r11,12(r1) > 1: mtctr r11 > mtlr r9 > bctr /* jump to handler */ > -- > 1.7.9.5 > -- BOFH excuse #292: We ran out of dial tone and we're and waiting for the phone company to deliver another bottle. _______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
