Messages in this thread | | | Date | Fri, 14 Mar 2014 11:41:00 +0100 | From | Peter Zijlstra <> | Subject | Re: [PATCH v1 03/11] perf: Allow for multiple ring buffers per event |
| |
On Wed, Feb 19, 2014 at 02:02:36PM -0800, Dave Hansen wrote: > On 02/17/2014 06:33 AM, Peter Zijlstra wrote: > > Then write the PERF_RECORD_DATA structure into the normal ring-buffer > > location; set data_offset to point to the first page boundary, data_size > > to 1mb. > > > > Then frob things such that perf_mmap_to_page() for the next 1mb of pages > > points to your buffer pages and wipe the page-table entries. > > Wouldn't we have to teach a ton of code how to be IRQ safe for this to > work? Just step one: how do we go modifying page tables safely from an > interrupt? mm->page_table_lock is a plain non-irq spinlock.
One could modify existing page tables the same way we do the lockless lookup for GUP. But instead of doing the get_page() we do a pte modification.
But I suppose we can push all that to task context by having the polling task do it before it gets to userspace again.
| |