hujun260 opened a new pull request, #21689:
URL: https://github.com/apache/nuttx/pull/21689

   # NuttX PR Description
   
   ## Title
   
   arm: prevent unwinding to invalid addresses from signal handlers
   
   ## Description
   
   This PR clears the link register in the saved signal-delivery context before
   entering the signal trampoline. This gives the ARM stack unwinder an explicit
   termination point and prevents it from following a stale LR value into an
   invalid address after unwinding through `arm_sigdeliver()`.
   
   The change is applied consistently to the ARM, ARMv6-M, ARMv7-A, ARMv7-M,
   ARMv7-R, ARMv8-M, and ARMv8-R signal scheduling implementations.
   
   ## Problem
   
   Before this change, a backtrace captured from a signal callback could 
continue
   past `arm_sigdeliver()` using an unrelated LR value. The unwinder could then
   report invalid frames or fault while trying to unwind the corrupted chain.
   
   Example invalid unwind result before the change:
   
   ```text
   (gdb) bt
   #7  _assert (...) at misc/assert.c:952
   #8  arm_dataabort (...) at armv7-a/arm_dataabort.c:172
   #9  arm_vectordata () at armv7-a/arm_vectors.S:453
   #10 unwind_frame (...) at common/arm_backtrace_unwind.c:568
   #11 backtrace_unwind (...) at common/arm_backtrace_unwind.c:652
   #12 up_backtrace (...) at common/arm_backtrace_unwind.c:768
   #13 sched_backtrace (...) at sched/sched_backtrace.c:106
   ...
   #28 arm_sigdeliver () at armv7-a/arm_sigdeliver.c:77
   #29 nxsched_add_prioritized (...)
   #30 nxsem_wait_slow (...)
   #31 0x55555554 in ?? ()
   Backtrace stopped: previous frame identical to this frame (corrupt stack?)
   ```
   
   The captured unwind buffer also contained an invalid address after
   `arm_sigdeliver()`:
   
   ```text
   0x43158a5c: 0x68fa18 <file_readv+212> 0x68fade <read+78>
   0x43158a64: 0x72c494 <uv__signal_handler+84> 0x60ebc2 <nxsig_deliver+482>
   0x43158a6c: 0x6505d2 <arm_sigdeliver+78> 0x6cf02a <nxsem_wait_slow+1018>
   0x43158a74: 0x55555552 0x0
   ```
   
   ## Changes
   
   - Clear the saved LR value when preparing a signal-delivery context.
   - Stop the unwinder at the signal trampoline instead of following stale 
state.
   - Apply the fix to:
     - `arch/arm/src/arm/arm_schedulesigaction.c`
     - `arch/arm/src/armv6-m/arm_schedulesigaction.c`
     - `arch/arm/src/armv7-a/arm_schedulesigaction.c`
     - `arch/arm/src/armv7-m/arm_schedulesigaction.c`
     - `arch/arm/src/armv7-r/arm_schedulesigaction.c`
     - `arch/arm/src/armv8-m/arm_schedulesigaction.c`
     - `arch/arm/src/armv8-r/arm_schedulesigaction.c`
   
   ## Result
   
   After the change, sending signal 10 to task 3 produces a bounded and valid
   backtrace from the signal callback:
   
   ```text
   nsh> kill -10 3
   nsh> sched_dumpstack: backtrace:
   sched_dumpstack: [ 3] [<0x1b0f6>] backtrace_unwind+0x157/0x15a
   sched_dumpstack: [ 3] [<0xb262>] sched_backtrace+0x39/0x88
   sched_dumpstack: [ 3] [<0x6e3e>] sched_dumpstack+0x19/0x98
   sched_dumpstack: [ 3] [<0x37850>] siguser_action+0x13/0x20
   sched_dumpstack: [ 3] [<0x17a46>] nxsig_deliver+0x103/0x204
   sched_dumpstack: [ 3] [<0x1986>] arm_sigdeliver+0x3d/0xa4
   nsh>
   ```
   
   The unwind terminates at `arm_sigdeliver()` without continuing to invalid or
   unrelated addresses.
   
   ## Testing Status
   
   - [x] Triggered a signal callback with `kill -10 3`
   - [x] Captured the signal callback backtrace
   - [x] Confirmed that unwinding terminates at `arm_sigdeliver()`
   - [x] Confirmed that no invalid address is reported after the signal frame
   - [ ] Compile-tested all affected ARM architecture variants
   
   ## Commit
   
   - `3107efd51823f4d09cacdb8d9086f4e0e3a46309` arm: fix unwind to invalid 
addresses in signal
   
   ## Change Summary
   
   The commit modifies 7 files with 11 insertions.
   
   Signed-off-by: xuxingliang <[email protected]>
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to