ddb: get the unwinder to handle function prologue/epilogue exceptions
ClosedPublic
Actions

Authored by markj on Jun 21 2015, 11:08 PM.

Details

Reviewers

Commits

rS285776: Let the unwinder handle faults during function prologues or epilogues.

Summary

The trapframe argument to db_backtrace() was previously always NULL, so
this code was never executing. Modify db_trace_thread() to pass
kdb_frame if we're being asked to trace kdb_thread's stack.

Get rid of get_rsp(). It looks like this came over from i386, which
needs to take into account whether the exception triggered a CPL switch,
since SS:ESP is only pushed onto the stack if so. On amd64, SS:RSP is
pushed regardless of whether the CPL changed, so get_rsp() was doing the
wrong thing for exceptions in ring 0.

Test Plan

Tested on amd64 and i386 by placing breakpoints on a sysctl handler and making sure that we get the right backtrace when:

the bp is on the first instruction (push rbp)
the bp is on the second instruction (mov rsp, rbp)
the bp is on a ret instruction, after a "pop rbp"

Diff Detail

Lint

Lint Passed

Unit

No Test Coverage

Event Timeline

markj updated this revision to Diff 6370.Jun 21 2015, 11:08 PM

markj retitled this revision from to ddb: get the unwinder to handle function prologue/epilogue exceptions.

markj edited the test plan for this revision. (Show Details)

markj updated this object.

Herald added a subscriber: imp. · View Herald TranscriptJun 21 2015, 11:08 PM

markj edited the test plan for this revision. (Show Details)Jun 21 2015, 11:11 PM

markj added a reviewer: jhb.

jhb added inline comments.Jul 21 2015, 4:58 AM

sys/amd64/amd64/db_trace.c
146–147	Good catch on removing get_rsp(). However, I think you can remove db_rsp entirely and just use db_frame. Is the same true of db_ss() btw? Ah, it is. You can remove db_ss() and just use db_frame() to read/write tf_ss as well.