On Wed, Oct 24, 2018 at 5:23 PM Peter Zijlstra <pet...@infradead.org> wrote: > > On Wed, Oct 24, 2018 at 12:30:52PM -0700, Stephane Eranian wrote: > > Hi, > > > > On Wed, Oct 24, 2018 at 8:12 AM <kan.li...@linux.intel.com> wrote: > > > > > > From: Kan Liang <kan.li...@linux.intel.com> > > > > > > To calculate the physical address, perf needs to walk the pages tables. > > > The related mapping may has already been removed from pages table in > > > some cases (e.g. large PEBS). The virtual address recorded in the first > > > PEBS records may already be unmapped before draining PEBS buffers. > > > > > > Add a munmap callback to notify the PMU of any unmapping, which only be > > > invoked when the munmap is implemented. > > > > > The need for this new record type extends beyond physical address > > conversions > > and PEBS. A long while ago, someone reported issues with symbolization > > related > > to perf lacking munmap tracking. It had to do with vma merging. I think the > > sequence of mmaps was as follows in the problematic case: > > 1. addr1 = mmap(8192); > > 2. munmap(addr1 + 4096, 4096) > > 3. addr2 = mmap(addr1+4096, 4096) > > > > If successful, that yields addr2 = addr1 + 4096 (could also get the > > same without forcing the address). > > That is actually a different problem. And you're right, we never did fix > that. > it is a different problem but the solution is the same: PERF_RECORD_UNMAP! That's why I mentioned it here. To show that this is needed for more than one reason ;-)
> I agree with you that Kan's Changelog is somewhat cryptic; it took me at > least 3 times reading and looking at what the patches actually do to > understand :/
