Re: [RFC PATCH v8 1/4] arm64: Make all stack walking functions use arch_stack_walk()

From: Mark Rutland
Date: Tue Aug 24 2021 - 09:13:53 EST


On Thu, Aug 12, 2021 at 02:06:00PM -0500, madvenka@xxxxxxxxxxxxxxxxxxx wrote:
> From: "Madhavan T. Venkataraman" <madvenka@xxxxxxxxxxxxxxxxxxx>
>
> Currently, there are multiple functions in ARM64 code that walk the
> stack using start_backtrace() and unwind_frame(). Convert all of
> them to use arch_stack_walk(). This makes maintenance easier.

It would be good to split this into a series of patches as Mark Brown
suggested in v7.

> Here is the list of functions:
>
> perf_callchain_kernel()
> get_wchan()
> return_address()
> dump_backtrace()
> profile_pc()

Note that arch_stack_walk() depends on CONFIG_STACKTRACE (which is not in
defconfig), so we'll need to reorganise things such that it's always defined,
or factor out the core of that function and add a wrapper such that we
can always use it.

> Signed-off-by: Madhavan T. Venkataraman <madvenka@xxxxxxxxxxxxxxxxxxx>
> ---
> arch/arm64/include/asm/stacktrace.h | 3 ---
> arch/arm64/kernel/perf_callchain.c | 5 +---
> arch/arm64/kernel/process.c | 39 ++++++++++++++++++-----------
> arch/arm64/kernel/return_address.c | 6 +----
> arch/arm64/kernel/stacktrace.c | 38 +++-------------------------
> arch/arm64/kernel/time.c | 22 +++++++++-------
> 6 files changed, 43 insertions(+), 70 deletions(-)
>
> diff --git a/arch/arm64/include/asm/stacktrace.h b/arch/arm64/include/asm/stacktrace.h
> index 8aebc00c1718..e43dea1c6b41 100644
> --- a/arch/arm64/include/asm/stacktrace.h
> +++ b/arch/arm64/include/asm/stacktrace.h
> @@ -61,9 +61,6 @@ struct stackframe {
> #endif
> };
>
> -extern int unwind_frame(struct task_struct *tsk, struct stackframe *frame);
> -extern void walk_stackframe(struct task_struct *tsk, struct stackframe *frame,
> - bool (*fn)(void *, unsigned long), void *data);
> extern void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk,
> const char *loglvl);
>
> diff --git a/arch/arm64/kernel/perf_callchain.c b/arch/arm64/kernel/perf_callchain.c
> index 4a72c2727309..2f289013c9c9 100644
> --- a/arch/arm64/kernel/perf_callchain.c
> +++ b/arch/arm64/kernel/perf_callchain.c
> @@ -147,15 +147,12 @@ static bool callchain_trace(void *data, unsigned long pc)
> void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry,
> struct pt_regs *regs)
> {
> - struct stackframe frame;
> -
> if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
> /* We don't support guest os callchain now */
> return;
> }
>
> - start_backtrace(&frame, regs->regs[29], regs->pc);
> - walk_stackframe(current, &frame, callchain_trace, entry);
> + arch_stack_walk(callchain_trace, entry, current, regs);
> }

We can also update callchain_trace take the return value of
perf_callchain_store into acount, e.g.

| static bool callchain_trace(void *data, unsigned long pc)
| {
| struct perf_callchain_entry_ctx *entry = data;
| return perf_callchain_store(entry, pc) == 0;
| }

>
> unsigned long perf_instruction_pointer(struct pt_regs *regs)
> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> index c8989b999250..52c12fd26407 100644
> --- a/arch/arm64/kernel/process.c
> +++ b/arch/arm64/kernel/process.c
> @@ -544,11 +544,28 @@ __notrace_funcgraph struct task_struct *__switch_to(struct task_struct *prev,
> return last;
> }
>
> +struct wchan_info {
> + unsigned long pc;
> + int count;
> +};
> +
> +static bool get_wchan_cb(void *arg, unsigned long pc)
> +{
> + struct wchan_info *wchan_info = arg;
> +
> + if (!in_sched_functions(pc)) {
> + wchan_info->pc = pc;
> + return false;
> + }
> + wchan_info->count--;
> + return !!wchan_info->count;
> +}

This will terminate one entry earlier than the old logic since we used
to use a post-increment (testing the prior value), and now we're
effectively using a pre-decrement (testing the new value).

I don't think that matters all that much in practice, but it might be
best to keep the original logic, e.g. initialize `count` to 0 and here
do:

return wchan_info->count++ < 16;

> +
> unsigned long get_wchan(struct task_struct *p)
> {
> - struct stackframe frame;
> - unsigned long stack_page, ret = 0;
> - int count = 0;
> + unsigned long stack_page;
> + struct wchan_info wchan_info;
> +
> if (!p || p == current || task_is_running(p))
> return 0;
>
> @@ -556,20 +573,12 @@ unsigned long get_wchan(struct task_struct *p)
> if (!stack_page)
> return 0;
>
> - start_backtrace(&frame, thread_saved_fp(p), thread_saved_pc(p));
> + wchan_info.pc = 0;
> + wchan_info.count = 16;
> + arch_stack_walk(get_wchan_cb, &wchan_info, p, NULL);
>
> - do {
> - if (unwind_frame(p, &frame))
> - goto out;
> - if (!in_sched_functions(frame.pc)) {
> - ret = frame.pc;
> - goto out;
> - }
> - } while (count++ < 16);
> -
> -out:
> put_task_stack(p);
> - return ret;
> + return wchan_info.pc;
> }

Other than the comment above, this looks good to me.

> unsigned long arch_align_stack(unsigned long sp)
> diff --git a/arch/arm64/kernel/return_address.c b/arch/arm64/kernel/return_address.c
> index a6d18755652f..92a0f4d434e4 100644
> --- a/arch/arm64/kernel/return_address.c
> +++ b/arch/arm64/kernel/return_address.c
> @@ -35,15 +35,11 @@ NOKPROBE_SYMBOL(save_return_addr);
> void *return_address(unsigned int level)
> {
> struct return_address_data data;
> - struct stackframe frame;
>
> data.level = level + 2;
> data.addr = NULL;
>
> - start_backtrace(&frame,
> - (unsigned long)__builtin_frame_address(0),
> - (unsigned long)return_address);
> - walk_stackframe(current, &frame, save_return_addr, &data);
> + arch_stack_walk(save_return_addr, &data, current, NULL);
>
> if (!data.level)
> return data.addr;

Nor that arch_stack_walk() will start with it's caller, so
return_address() will be included in the trace where it wasn't
previously, which implies we need to skip an additional level.

That said, I'm not entirely sure why we need to skip 2 levels today; it
might be worth checking that's correct.

We should also mark return_address() as noinline to avoid surprises with
LTO.

> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index 8982a2b78acf..1800310f92be 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -151,23 +151,21 @@ void notrace walk_stackframe(struct task_struct *tsk, struct stackframe *frame,
> }
> NOKPROBE_SYMBOL(walk_stackframe);
>
> -static void dump_backtrace_entry(unsigned long where, const char *loglvl)
> +static bool dump_backtrace_entry(void *arg, unsigned long where)
> {
> + char *loglvl = arg;
> printk("%s %pSb\n", loglvl, (void *)where);
> + return true;
> }
>
> void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk,
> const char *loglvl)
> {
> - struct stackframe frame;
> - int skip = 0;
> -
> pr_debug("%s(regs = %p tsk = %p)\n", __func__, regs, tsk);
>
> if (regs) {
> if (user_mode(regs))
> return;
> - skip = 1;
> }

We can simplifiy this to:

if (regs && user_mode(regs))
return;

>
> if (!tsk)
> @@ -176,36 +174,8 @@ void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk,
> if (!try_get_task_stack(tsk))
> return;
>
> - if (tsk == current) {
> - start_backtrace(&frame,
> - (unsigned long)__builtin_frame_address(0),
> - (unsigned long)dump_backtrace);
> - } else {
> - /*
> - * task blocked in __switch_to
> - */
> - start_backtrace(&frame,
> - thread_saved_fp(tsk),
> - thread_saved_pc(tsk));
> - }
> -
> printk("%sCall trace:\n", loglvl);
> - do {
> - /* skip until specified stack frame */
> - if (!skip) {
> - dump_backtrace_entry(frame.pc, loglvl);
> - } else if (frame.fp == regs->regs[29]) {
> - skip = 0;
> - /*
> - * Mostly, this is the case where this function is
> - * called in panic/abort. As exception handler's
> - * stack frame does not contain the corresponding pc
> - * at which an exception has taken place, use regs->pc
> - * instead.
> - */
> - dump_backtrace_entry(regs->pc, loglvl);
> - }
> - } while (!unwind_frame(tsk, &frame));
> + arch_stack_walk(dump_backtrace_entry, (void *)loglvl, tsk, regs);

It turns out we currently need this skipping to get the balance the
ftrace call stack, and arch_stack_walk() doesn't currently do the right
thing when starting from regs. That balancing isn't quite right, and
will be wrong in some case when unwinding across exception boundaries;
we could implement HAVE_FUNCTION_GRAPH_RET_ADDR_PTR using the FP to
solve that.

>
> put_task_stack(tsk);
> }
> diff --git a/arch/arm64/kernel/time.c b/arch/arm64/kernel/time.c
> index eebbc8d7123e..671b3038a772 100644
> --- a/arch/arm64/kernel/time.c
> +++ b/arch/arm64/kernel/time.c
> @@ -32,22 +32,26 @@
> #include <asm/stacktrace.h>
> #include <asm/paravirt.h>
>
> +static bool profile_pc_cb(void *arg, unsigned long pc)
> +{
> + unsigned long *prof_pc = arg;
> +
> + if (in_lock_functions(pc))
> + return true;
> + *prof_pc = pc;
> + return false;
> +}
> +
> unsigned long profile_pc(struct pt_regs *regs)
> {
> - struct stackframe frame;
> + unsigned long prof_pc = 0;
>
> if (!in_lock_functions(regs->pc))
> return regs->pc;
>
> - start_backtrace(&frame, regs->regs[29], regs->pc);
> -
> - do {
> - int ret = unwind_frame(NULL, &frame);
> - if (ret < 0)
> - return 0;
> - } while (in_lock_functions(frame.pc));
> + arch_stack_walk(profile_pc_cb, &prof_pc, current, regs);
>
> - return frame.pc;
> + return prof_pc;
> }
> EXPORT_SYMBOL(profile_pc);

Mdoulo the problem above w.r.t. unwinding from regs, this looks good.

Thanks,
Mark.

>
> --
> 2.25.1
>