Re: [PATCH V6] perf: Add PERF_SAMPLE_PHYS_ADDR

From: Peter Zijlstra
Date: Thu Aug 24 2017 - 04:56:41 EST


On Thu, Aug 24, 2017 at 11:26:17AM +1000, Michael Ellerman wrote:
> Peter Zijlstra <peterz@xxxxxxxxxxxxx> writes:
> >> diff --git a/kernel/events/core.c b/kernel/events/core.c
> >> index d704e23..b991af3 100644
> >> --- a/kernel/events/core.c
> >> +++ b/kernel/events/core.c
> >> @@ -6027,6 +6033,38 @@ void perf_output_sample(struct perf_output_handle *handle,
> >> }
> >> }
> >>
> >> +static u64 perf_virt_to_phys(u64 virt)
> >> +{
> >> + u64 phys_addr = 0;
> >> + struct page *p = NULL;
> >> +
> >> + if (!virt)
> >> + return 0;
> >> +
> >> + if (virt >= TASK_SIZE) {
> >> + /* If it's vmalloc()d memory, leave phys_addr as 0 */
> >> + if (virt_addr_valid(virt) &&
> >> + !(virt >= VMALLOC_START && virt < VMALLOC_END))
> >> + phys_addr = (u64)virt_to_phys((void *)(uintptr_t)virt);
> >> + } else {
> >> + /*
> >> + * Walking the pages tables for user address.
> >> + * Interrupts are disabled, so it prevents any tear down
> >> + * of the page tables.
> >> + * Try IRQ-safe __get_user_pages_fast first.
> >> + * If failed, leave phys_addr as 0.
> >> + */
> >> + if ((current->mm != NULL) &&
> >> + (__get_user_pages_fast(virt, 1, 0, &p) == 1))
> >> + phys_addr = page_to_phys(p) + virt % PAGE_SIZE;
> >> +
> >> + if (p)
> >> + put_page(p);
> >> + }
> >> +
> >> + return phys_addr;
> >> +}
> >
> > Michael, does this work for PPC as is?
>
> I think so.
>
> We have about 8 MMUs and 32-bit and 64-bit so it's a bit hard to say for
> sure.

You can limit it to those that include a PMU that supports data->addr,
which is book3s, whatever that gets you.

> I'm pretty sure virt_addr_valid() will exclude everything except the
> linear mapping for us, so the vmalloc check is redundant but that's
> fine. So that looks safe AFAICS.
>
> I'm not an expert on GUP fast, but AFAIK there's nothing special
> required for us.

Yeah, gup fast should work for you.