From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Mon, 11 Oct 1999 01:28:36 +0200 (CEST) From: Andrea Arcangeli Subject: Re: locking question: do_mmap(), do_munmap() In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-linux-mm@kvack.org Return-Path: To: Alexander Viro Cc: Manfred Spraul , linux-kernel@vger.rutgers.edu, Ingo Molnar , linux-mm@kvack.org List-ID: On Sun, 10 Oct 1999, Alexander Viro wrote: >mm - no list modifications, no vma removal, etc. We could introduce a new >semaphore (spinlocks are not going to work - ->swapout gets vma as We could add a reference count to each vma protected by a per-mm spinlock. Then we could drop the spinlock in swap_out_mm as soon as we incremented the refcount of the vma. I am talking by memory, I am not sure if it can be really done and if it's the right thing to do this time. (I'll check the code ASAP). >argument and it can sleep. The question being: where can we trigger >__get_free_pages() with __GFP_WAIT if the mmap_sem is held? And another All userspace allocations do exactly that. >one - where do we modify ->mmap? If they can be easily separated - We modify mmap in the mmap.c and infact we hold the semaphore there too. The reason they are not separated is that during all the page fault path you can't have _your_ vmas to change under you if another thread is running munmap in parallel. >eat the fs on the testbox ;-), but I'ld be really grateful if some of VM >people would check the results. I can check them of course ;). Andrea -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://humbolt.geo.uu.nl/Linux-MM/