Messages in this thread | | | Date | Wed, 27 Jun 2007 08:59:52 -0700 (PDT) | From | Davide Libenzi <> | Subject | Re: [patch 2/3] MAP_NOZERO - implement sys_brk2() |
| |
On Wed, 27 Jun 2007, Hugh Dickins wrote:
> On Tue, 26 Jun 2007, Ulrich Drepper wrote: > > On 6/26/07, Davide Libenzi <davidel@xmailserver.org> wrote: > > > > > OTOH glibc could implement __morecore using mmap(MAP_NOZERO), and hence > > > brk2() would not be needed, no? > > > > No. mmap calls create individual VMAs which gets expensive. There > > are also some hardware drivers which get more expensive the more VMAs > > there are. I want to go away as much as possible from mmap for > > malloc. > > Not so: if an mmap can be done by extending either adjacent vma (prot > and flags and file and offset all match up), that's what's done and no > separate vma is created. (And adjacent vmas get merged when mprotect > removes the difference in protection.) > > I don't think there's any such reason to prefer brk to mmap. But you > may have encountered something which we in the kernel are thinking of > as an insignificant corner case, which is actually breaking things up > badly in practice (I recall the kernel's internal VM_ACCOUNT flag, > relating to non-overcommit accounting, which might get turned on at > any time, sometimes preventing a vma merge you'd otherwise expect). > Please let me know if you've a test case which shows more vmas than > expected.
The only way I can see vma fragmentation happen in that case, is if userspace uses a mixture of mmaps and mallocs, and flags+prots of the two does not match. The glibc allocator seems to support it just fine. There's a macro where you specify if the heaps are contiguous or not.
- Davide
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |