Re: [PATCH 4/4] perf tools: determine if LR is the return address

[Alexandre Truong]

On 2/8/21 3:39 PM, James Clark wrote:
> On 22/01/2021 18:18, Alexandre Truong wrote:
>
> > +}
> > +
> > +static int add_entry(struct unwind_entry *entry, void *arg)
> > +{
> > +    struct entries *entries = arg;
> > +
> > +    entries->stack[entries->i++] = entry->ip;
> > +    return 0;
> > +}
> > +
> > +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread)
> > +{
> > +    u64 leaf_frame;
> > +    struct entries entries = {{0, 0}, 0};
> > +
> > +    if (get_leaf_frame_caller_enabled(sample))
> > +            return 0;
> > +
> > +    unwind__get_entries(add_entry, &entries, thread, sample, 2);
> > +    leaf_frame = callchain_param.order == ORDER_CALLER ?
> > +            entries.stack[0] : entries.stack[1];
> > +
> > +    if (leaf_frame + 1 == sample->user_regs.regs[PERF_REG_ARM64_LR])
> > +            return sample->user_regs.regs[PERF_REG_ARM64_LR];
>
> Hi Alex,
>
>  From your other reply about your investigation it looks like the check against PERF_REG_ARM64_LR isn't
> required because libunwind won't return a value if it's not correct. Whether it's equal to the LR or not.
>
> And PERF_REG_ARM64_LR points to the instruction _after_ the call site. i.e. where to return to,
> not where the call was made from. So just leaf_frame rather than leaf_frame+1 would be more accurate.
>
> I was also looking at unwind_entry in machine.c which is similar to your add_entry function and saw that it
> does some extra bits like this:
>
>       if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
>               return 0;
>
>       if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
>               return 0;
>
>       /*
>        * Convert entry->ip from a virtual address to an offset in
>        * its corresponding binary.
>        */
>       if (entry->ms.map)
>               addr = map__map_ip(entry->ms.map, entry->ip);
>
> I have a feeling you will also need to do those on your values returned from libunwind to make it 100%
> equivalent.
>
> James

Hi James,

Thanks for your reply.

The check against PERF_REG_ARM64_LR is indeed not required and I can
check if libunwind goes successfully.

I am going to follow up with a v2 of the patch applying these changes.

I think the bits you mentioned don't need to be added because this check
is already done in add_callchain_ip() called afterwards in machine.c :

    if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
        return 0;

For the second one, I don't think it needs to be added either because
append_inlines() appends ip on the cursor which is also already done by
add_callchain_ip().

For the last one, the conversion from a virtual address to a binary one
isn't required.


Also for the expansion to all platforms, it doesn't work on x86 so I'll
leave it just for arm for now.

Regards,

Alexandre

> > +    return 0;
> > +}
> > diff --git a/tools/perf/util/arm-frame-pointer-unwind-support.h b/tools/perf/util/arm-frame-pointer-unwind-support.h
> > new file mode 100644
> > index 000000000000..16dc03fa9abe
> > --- /dev/null
> > +++ b/tools/perf/util/arm-frame-pointer-unwind-support.h
> > @@ -0,0 +1,7 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +#ifndef __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> > +#define __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> > +
> > +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread);
> > +
> > +#endif /* __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H */
> > diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
> > index 40082d70eec1..bc6147e46c89 100644
> > --- a/tools/perf/util/machine.c
> > +++ b/tools/perf/util/machine.c
> > @@ -34,6 +34,7 @@
> >   #include "bpf-event.h"
> >   #include <internal/lib.h> // page_size
> >   #include "cgroup.h"
> > +#include "arm-frame-pointer-unwind-support.h"
> >
> >   #include <linux/ctype.h>
> >   #include <symbol/kallsyms.h>
> > @@ -2671,10 +2672,12 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
> >      return err;
> >   }
> >
> > -static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
> > -            struct thread *thread __maybe_unused)
> > +static u64 get_leaf_frame_caller(struct perf_sample *sample, struct thread *thread)
> >   {
> > -    return 0;
> > +    if (strncmp(thread->maps->machine->env->arch, "aarch64", 7) == 0)
> > +            return get_leaf_frame_caller_aarch64(sample, thread);
> > +    else
> > +            return 0;
> >   }
> >
> >   static int thread__resolve_callchain_sample(struct thread *thread,
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.