Re: [PATCH RFC 04/10] perf: Introduce deferred user callchains

[Josh Poimboeuf]

On Sat, Nov 11, 2023 at 10:49:10AM -0800, Josh Poimboeuf wrote:
> On Fri, Nov 10, 2023 at 10:57:58PM -0800, Namhyung Kim wrote:
> > > +static void perf_pending_task_unwind(struct perf_event *event)
> > > +{
> > > +       struct pt_regs *regs = task_pt_regs(current);
> > > +       struct perf_output_handle handle;
> > > +       struct perf_event_header header;
> > > +       struct perf_sample_data data;
> > > +       struct perf_callchain_entry *callchain;
> > > +
> > > +       callchain = kmalloc(sizeof(struct perf_callchain_entry) +
> > > +                           (sizeof(__u64) * event->attr.sample_max_stack) +
> > > +                           (sizeof(__u64) * 1) /* one context */,
> > > +                           GFP_KERNEL);
> > 
> > Any chance it can reuse get_perf_callchain() instead of
> > allocating the callchains every time?
> 
> I don't think so, because if it gets preempted, the new task might also
> need to do an unwind.  But there's only one task-context callchain per
> CPU.

BTW it's not just preemption, this code can also block when the unwinder
tries to copy from user space.  So disabling preemption isn't an option.

-- 
Josh