* Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote: > On Fri, Jul 12, 2013 at 10:13:48AM +0200, Peter Zijlstra wrote: > > On Fri, Jul 12, 2013 at 08:39:14AM +0300, Pekka Enberg wrote: > > > On 07/12/2013 05:34 AM, Colin Cross wrote: > > > >Userspace processes often have multiple allocators that each do > > > >anonymous mmaps to get memory. When examining memory usage of > > > >individual processes or systems as a whole, it is useful to be > > > >able to break down the various heaps that were allocated by > > > >each layer and examine their size, RSS, and physical memory > > > >usage. > > > > > > > >This patch adds a user pointer to the shared union in > > > >vm_area_struct that points to a null terminated string inside > > > >the user process containing a name for the vma. vmas that > > > >point to the same address will be merged, but vmas that > > > >point to equivalent strings at different addresses will > > > >not be merged. > > > > > > > >Userspace can set the name for a region of memory by calling > > > >prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); > > > >Setting the name to NULL clears it. > > > > > > > >The names of named anonymous vmas are shown in /proc/pid/maps > > > >as [anon:<name>] and in /proc/pid/smaps in a new "Name" field > > > >that is only present for named vmas. If the userspace pointer > > > >is no longer valid all or part of the name will be replaced > > > >with "<fault>". > > > > > > > >The idea to store a userspace pointer to reduce the complexity > > > >within mm (at the expense of the complexity of reading > > > >/proc/pid/mem) came from Dave Hansen. This results in no > > > >runtime overhead in the mm subsystem other than comparing > > > >the anon_name pointers when considering vma merging. The pointer > > > >is stored in a union with fieds that are only used on file-backed > > > >mappings, so it does not increase memory usage. > > > > > > > >Signed-off-by: Colin Cross <ccross@xxxxxxxxxxx> > > > > > > Ingo, PeterZ, is this something worthwhile for replacing our > > > current JIT symbol hack with perf? > > > > I really don't see the point of this stuff; in fact I intensely > > dislike it as I don't think this is something the kernel needs to do > > at all. > > > > Why can't these allocators Collin talks about use file maps and/or > > write their own meta-data to file? He is after all only interested in > > Android and they have complete control over the entire userspace > > stack. > > In fact, nowhere in his entire Changelog does he explain why this needs > be in the kernel; _why_ can't userspace do this? > > He needs to go change his allocators to use the new madv syscall anyway, > he might as well change them to write the stuff to a local file and be > done with it. > > what gives? It makes tons of sense. Just like we have a task's cmd-name it makes a lot of sense to name objects in a human readable fashion, to help debugging, instrumentation, performance analysis, etc. Yes, in theory user-space could do all that. That's not the point: the point is to make it fast, easy enough and to have a central version (the kernel). Doing it via temporary files has various disadvantages: - many tools really like to be filesystem invariant (not touch any files even in tmpfs, be able to run in a readonly environment, etc.) - the overhead of opening, writing to and closing a file is an order of magnitude larger than a single prctl() call. [I'd even argue for such user-space tags to be attached to do_mmap(), unfortunately the mmap system call argument space is already pretty full. ] - stray files hang around (even in tmpfs). Point of instrumentation is to be non-intrusive and as fool-proof as possible. When we are debugging problems the last thing we want are extra problems and unreliable instrumentation introduced by a fragile temporary file solution... - user space also tends to get the security model of temporary files wrong. static linking makes the user-space version iteration of such facilities harder. etc. etc. - there's other disadvantages as well. So using temporary files is an instrumentation and debugging nightmare really. A simple self-contained prctl() variant, with the info stored by the kernel is as convenient as it gets. I guess the real question is not whether it's useful, I think it clearly is. The question should be: are there real downsides? Does the addition to the anon mmap field blow up the size of vma_struct by a pointer, or is there still space? Thanks, Ingo -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>