On Mon, Oct 09, 2017 at 08:55:04PM +0800, Fengguang Wu wrote: > On Mon, Oct 09, 2017 at 08:21:13PM +0800, Fengguang Wu wrote: > > On Mon, Oct 09, 2017 at 12:50:55PM +0200, Peter Zijlstra wrote: > > > > Fengguang, if you're still listening, could you please rerun the tests > > > > on top of ce07a9415f26, with the attached patches also applied? > > > > > > Ping!? it would be very good to get feedback on this asap. > > > > Sorry for the delay! > > > > > > From e7840ad76515f0b5061fcdd098b57b7c01b61482 Mon Sep 17 00:00:00 2001 > > > > Message-Id: > > > > <e7840ad76515f0b5061fcdd098b57b7c01b61482.1507215196.git.jpoim...@redhat.com> > > > > From: Josh Poimboeuf <jpoim...@redhat.com> > > > > Date: Thu, 5 Oct 2017 09:43:59 -0500 > > > > Subject: [PATCH 1/2] unwinder fixes > > > > > > > > --- > > > > arch/x86/kernel/unwind_frame.c | 33 ++++++++++++++++++++++++++++++--- > > > > I just test 316 boots and see 7 WARNINGs: > > > > [ 404.948035] WARNING: kernel stack frame pointer at c6ea3ecd in init:212 > > has bad value (null) > > [ 298.118383] WARNING: kernel stack frame pointer at cde07dad in init:1 > > has bad value bc000000 > > [ 112.848677] WARNING: kernel stack frame pointer at cde07dbd in > > swapper/0:1 has bad value c2000000 > > [ 127.942417] WARNING: kernel stack frame pointer at cf95de71 in > > rb_producer:50 has bad value 03cf95de > > [ 4.736938] WARNING: kernel stack frame pointer at bf643d59 in > > kworker/0:1:15 has bad value b5000000 > > [ 308.260066] WARNING: kernel stack frame pointer at bde07da5 in udevd:155 > > has bad value b5bfa17b > > > > [ 277.473596] WARNING: CPU: 0 PID: 520 at kernel/locking/lockdep.c:3841 > > check_flags+0x119/0x1b0
The unwinder patch I sent had a few bugs: it broke frame pointer encoding (causing the '?' entries on the lockdep stack trace) and it didn't disable the frame pointer warnings. Here's the fixed version. Fengguang, can you do a round of tests with this patch and the lockdep patch I sent before? Thanks! diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S index 8a13d468635a..50e0d2bc4528 100644 --- a/arch/x86/entry/entry_32.S +++ b/arch/x86/entry/entry_32.S @@ -176,7 +176,7 @@ /* * This is a sneaky trick to help the unwinder find pt_regs on the stack. The * frame pointer is replaced with an encoded pointer to pt_regs. The encoding - * is just setting the LSB, which makes it an invalid stack address and is also + * is just clearing the MSB, which makes it an invalid stack address and is also * a signal to the unwinder that it's a pt_regs pointer in disguise. * * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the @@ -185,7 +185,7 @@ .macro ENCODE_FRAME_POINTER #ifdef CONFIG_FRAME_POINTER mov %esp, %ebp - orl $0x1, %ebp + andl $0x7fffffff, %ebp #endif .endm diff --git a/arch/x86/kernel/unwind_frame.c b/arch/x86/kernel/unwind_frame.c index d145a0b1f529..f157238528a6 100644 --- a/arch/x86/kernel/unwind_frame.c +++ b/arch/x86/kernel/unwind_frame.c @@ -44,7 +44,8 @@ static void unwind_dump(struct unwind_state *state) state->stack_info.type, state->stack_info.next_sp, state->stack_mask, state->graph_idx); - for (sp = state->orig_sp; sp; sp = PTR_ALIGN(stack_info.next_sp, sizeof(long))) { + for (sp = PTR_ALIGN(state->orig_sp, sizeof(long)); sp; + sp = PTR_ALIGN(stack_info.next_sp, sizeof(long))) { if (get_stack_info(sp, state->task, &stack_info, &visit_mask)) break; @@ -77,6 +78,12 @@ static size_t regs_size(struct pt_regs *regs) return sizeof(*regs); } +#ifdef CONFIG_X86_32 +#define KERNEL_REGS_SIZE (sizeof(struct pt_regs) - 2*sizeof(long)) +#else +#define KERNEL_REGS_SIZE (sizeof(struct pt_regs)) +#endif + static bool in_entry_code(unsigned long ip) { char *addr = (char *)ip; @@ -174,6 +181,7 @@ static bool is_last_task_frame(struct unwind_state *state) * This determines if the frame pointer actually contains an encoded pointer to * pt_regs on the stack. See ENCODE_FRAME_POINTER. */ +#ifdef CONFIG_X86_64 static struct pt_regs *decode_frame_pointer(unsigned long *bp) { unsigned long regs = (unsigned long)bp; @@ -183,6 +191,17 @@ static struct pt_regs *decode_frame_pointer(unsigned long *bp) return (struct pt_regs *)(regs & ~0x1); } +#else +static struct pt_regs *decode_frame_pointer(unsigned long *bp) +{ + unsigned long regs = (unsigned long)bp; + + if (regs & 0x80000000) + return NULL; + + return (struct pt_regs *)(regs | 0x80000000); +} +#endif static bool update_stack_state(struct unwind_state *state, unsigned long *next_bp) @@ -202,7 +221,7 @@ static bool update_stack_state(struct unwind_state *state, regs = decode_frame_pointer(next_bp); if (regs) { frame = (unsigned long *)regs; - len = regs_size(regs); + len = KERNEL_REGS_SIZE; state->got_irq = true; } else { frame = next_bp; @@ -226,6 +245,14 @@ static bool update_stack_state(struct unwind_state *state, frame < prev_frame_end) return false; + /* + * On 32-bit with user mode regs, make sure the last two regs are safe + * to access: + */ + if (IS_ENABLED(CONFIG_X86_32) && regs && user_mode(regs) && + !on_stack(info, frame, len + 2*sizeof(long))) + return false; + /* Move state to the next frame: */ if (regs) { state->regs = regs; @@ -328,6 +355,13 @@ bool unwind_next_frame(struct unwind_state *state) state->regs->sp < (unsigned long)task_pt_regs(state->task)) goto the_end; + /* + * There are some known frame pointer issues on 32-bit. Disable + * unwinder warnings until it gets objtool support. + */ + if (IS_ENABLED(CONFIG_X86_32)) + goto the_end; + if (state->regs) { printk_deferred_once(KERN_WARNING "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n",