Messages in this thread | | | Date | Fri, 9 May 2008 10:02:49 +1000 | From | David Gibson <> | Subject | Re: [PATCH 0/3] Guarantee faults for processes that call mmap(MAP_PRIVATE) on hugetlbfs v2 |
| |
On Thu, May 08, 2008 at 12:14:08PM +0100, Andy Whitcroft wrote: > On Thu, May 08, 2008 at 11:48:22AM +1000, David Gibson wrote: > > On Wed, May 07, 2008 at 08:38:26PM +0100, Mel Gorman wrote: > > > MAP_SHARED mappings on hugetlbfs reserve huge pages at mmap() time. > > > This guarantees all future faults against the mapping will succeed. > > > This allows local allocations at first use improving NUMA locality whilst > > > retaining reliability. > > > > > > MAP_PRIVATE mappings do not reserve pages. This can result in an application > > > being SIGKILLed later if a huge page is not available at fault time. This > > > makes huge pages usage very ill-advised in some cases as the unexpected > > > application failure cannot be detected and handled as it is immediately fatal. > > > Although an application may force instantiation of the pages using mlock(), > > > this may lead to poor memory placement and the process may still be killed > > > when performing COW. > > > > > > This patchset introduces a reliability guarantee for the process which creates > > > a private mapping, i.e. the process that calls mmap() on a hugetlbfs file > > > successfully. The first patch of the set is purely mechanical code move to > > > make later diffs easier to read. The second patch will guarantee faults up > > > until the process calls fork(). After patch two, as long as the child keeps > > > the mappings, the parent is no longer guaranteed to be reliable. Patch > > > 3 guarantees that the parent will always successfully COW by unmapping > > > the pages from the child in the event there are insufficient pages in the > > > hugepage pool in allocate a new page, be it via a static or dynamic pool. > > > > I don't think patch 3 is a good idea. It's a fair bit of code to > > implement a pretty bizarre semantic that I really don't think is all > > that useful. Patches 1-2 are already sufficient to cover the > > fork()/exec() case and a fair proportion of fork()/minor > > frobbing/exit() cases. If the child also needs to write the hugepage > > area, chances are it's doing real work and we care about its > > reliability too. > > Without patch 3 the parent is still vunerable during the period the > child exists. Even if that child does nothing with the pages not even > referencing them, and then execs immediatly. As soon as we fork any > reference from the parent will trigger a COW, at which point there may > be no pages available and the parent will have to be killed. That is > regardless of the fact the child is not going to reference the page and > leave the address space shortly. With patch 3 on COW if we find no memory > available the page may be stolen for the parent saving it, and the _risk_ > of reference death moves to the child; the child is killed only should it > then re-reference the page.
Yes, thinko, sorry. Forgot that a COW would be triggered even if the child never wrote the pages. I see the point of patch 3 now. Damn, but it's still a weird semantic to be implementing though.
-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
| |