[0/2] riscv: ftrace: make stack walk more robust.

Message

Puranjay Mohan March 28, 2024, 6:40 p.m. UTC

The current stack walker in riscv implemented in walk_stackframe() provides
the PC to a callback function when it unwinds the stacks. This doesn't
allow implementing stack walkers that need access to more information like
the frame pointer, etc.

This series makes walk_stackframe() provide a unwinde_state structure to
callback functions. This structure has all the information that
walk_stackframe() can provide.

Currently, there are four users of walk_stackframe(): return_address(),
perf_callchain_kernel(), dump_backtrace(), and __get_wchan(). All of these
have been converted to use arch_stack_walk() rather than calling
walk_stackframe() directly.

We need this to implement arch_bpf_stack_walk() that provides a callback
that needs the FP, SP, and PC. This will be needed for implementing BFP
exceptions for RISCV.

There are no functional changes in this series.

I have tested this by crashing the kernel and looking at the stack trace
with and without CONFIG_FRAME_POINTER

Puranjay Mohan (2):
  riscv: stacktrace: use arch_stack_walk() in place of walk_stackframe
  riscv: stacktrace: make walk_stackframe() more robust

 arch/riscv/include/asm/stacktrace.h |  2 -
 arch/riscv/kernel/perf_callchain.c  |  2 +-
 arch/riscv/kernel/stacktrace.c      | 75 +++++++++++++++++++++++------
 3 files changed, 61 insertions(+), 18 deletions(-)

Comments

Björn Töpel April 2, 2024, 1:18 p.m. UTC | #1

Puranjay Mohan <[e-mail]> writes:

> The current stack walker in riscv implemented in walk_stackframe() provides
> the PC to a callback function when it unwinds the stacks. This doesn't
> allow implementing stack walkers that need access to more information like
> the frame pointer, etc.
>
> This series makes walk_stackframe() provide a unwinde_state structure to
> callback functions. This structure has all the information that
> walk_stackframe() can provide.
>
> Currently, there are four users of walk_stackframe(): return_address(),
> perf_callchain_kernel(), dump_backtrace(), and __get_wchan(). All of these
> have been converted to use arch_stack_walk() rather than calling
> walk_stackframe() directly.
>
> We need this to implement arch_bpf_stack_walk() that provides a callback
> that needs the FP, SP, and PC. This will be needed for implementing BFP
> exceptions for RISCV.

Hmm, I wonder if it's easier to have these two patches as part of the
BPF exception series, instead of having the dependencies be cross-tree?

> There are no functional changes in this series.
>
> I have tested this by crashing the kernel and looking at the stack trace
> with and without CONFIG_FRAME_POINTER

I have two really minor style nits, but regardless if they're fixed or
not:

Reviewed-by: Björn Töpel <[e-mail]>