diff options
| author | Matt Fleming <matt@console-pimps.org> | 2010-01-30 18:36:20 +0100 |
|---|---|---|
| committer | Paul Mundt <lethal@linux-sh.org> | 2010-02-08 02:47:04 +0100 |
| commit | 944a3438615da65f11e2559840404a2cac5f65ea (patch) | |
| tree | 44b77dbb19ee1ac55d0a9d7c174e1ef04dcf6f71 /arch | |
| parent | sh: Setup frame pointer in handle_exception path (diff) | |
| download | linux-944a3438615da65f11e2559840404a2cac5f65ea.tar.xz linux-944a3438615da65f11e2559840404a2cac5f65ea.zip | |
sh: Don't continue unwinding across interrupts
Unfortunately, due to poor DWARF info in current toolchains, unwinding
through interrutps cannot be done reliably. The problem is that the
DWARF info for function epilogues is wrong.
Take this standard epilogue sequence,
80003cc4: e3 6f mov r14,r15
80003cc6: 26 4f lds.l @r15+,pr
80003cc8: f6 6e mov.l @r15+,r14
<---- interrupt here
80003cca: f6 6b mov.l @r15+,r11
80003ccc: f6 6a mov.l @r15+,r10
80003cce: f6 69 mov.l @r15+,r9
80003cd0: 0b 00 rts
If we take an interrupt at the highlighted point, the DWARF info will
bogusly claim that the return address can be found at some offset from
the frame pointer, even though the frame pointer was just restored. The
worst part is if the unwinder finds a text address at the bogus stack
address - unwinding will continue, for a bit, until it finally comes
across an unexpected address on the stack and blows up.
The only solution is to stop unwinding once we've calculated the
function that was executing when the interrupt occurred. This PC can be
easily calculated from pt_regs->pc.
Signed-off-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Diffstat (limited to 'arch')
| -rw-r--r-- | arch/sh/kernel/dwarf.c | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/arch/sh/kernel/dwarf.c b/arch/sh/kernel/dwarf.c index 88d28ec3780a..e51168064e56 100644 --- a/arch/sh/kernel/dwarf.c +++ b/arch/sh/kernel/dwarf.c @@ -540,6 +540,8 @@ void dwarf_free_frame(struct dwarf_frame *frame) mempool_free(frame, dwarf_frame_pool); } +extern void ret_from_irq(void); + /** * dwarf_unwind_stack - unwind the stack * @@ -678,6 +680,24 @@ struct dwarf_frame * dwarf_unwind_stack(unsigned long pc, addr = frame->cfa + reg->addr; frame->return_addr = __raw_readl(addr); + /* + * Ah, the joys of unwinding through interrupts. + * + * Interrupts are tricky - the DWARF info needs to be _really_ + * accurate and unfortunately I'm seeing a lot of bogus DWARF + * info. For example, I've seen interrupts occur in epilogues + * just after the frame pointer (r14) had been restored. The + * problem was that the DWARF info claimed that the CFA could be + * reached by using the value of the frame pointer before it was + * restored. + * + * So until the compiler can be trusted to produce reliable + * DWARF info when it really matters, let's stop unwinding once + * we've calculated the function that was interrupted. + */ + if (prev && prev->pc == (unsigned long)ret_from_irq) + frame->return_addr = 0; + return frame; bail: |