From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail202.messagelabs.com (mail202.messagelabs.com [216.82.254.227]) by kanga.kvack.org (Postfix) with SMTP id D2DEE6B006A for ; Thu, 21 Jan 2010 00:05:26 -0500 (EST) Date: Thu, 21 Jan 2010 13:05:21 +0800 From: Wu Fengguang Subject: Re: [PATCH 5/8] vmalloc: simplify vread()/vwrite() Message-ID: <20100121050521.GB24236@localhost> References: <20100113135305.013124116@intel.com> <20100113135957.833222772@intel.com> <20100114124526.GB7518@laptop> <20100118133512.GC721@localhost> <20100118142359.GA14472@laptop> <20100119013303.GA12513@localhost> <20100119112343.04f4eff5.kamezawa.hiroyu@jp.fujitsu.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20100119112343.04f4eff5.kamezawa.hiroyu@jp.fujitsu.com> Sender: owner-linux-mm@kvack.org To: KAMEZAWA Hiroyuki Cc: Nick Piggin , Andrew Morton , LKML , Tejun Heo , Ingo Molnar , Andi Kleen , Hugh Dickins , Christoph Lameter , Linux Memory Management List List-ID: On Mon, Jan 18, 2010 at 07:23:43PM -0700, KAMEZAWA Hiroyuki wrote: > On Tue, 19 Jan 2010 09:33:03 +0800 > Wu Fengguang wrote: > > > The whole thing looks stupid though, apparently kmap is used to avoid "the > > > lock". But the lock is already held. We should just use the vmap > > > address. > > > > Yes. I wonder why Kame introduced kmap_atomic() in d0107eb07 -- given > > that he at the same time fixed the order of removing vm_struct and > > vmap in dd32c279983b. > > > Hmm...I must check my thinking again before answering.. > > vmalloc/vmap is constructed by 2 layer. > - vmalloc layer....guarded by vmlist_lock. > - vmap layer ....gurderd by purge_lock. etc. > > Now, let's see how vmalloc() works. It does job in 2 steps. > vmalloc(): > - allocate vmalloc area to the list under vmlist_lock. > - map pages. > vfree() > - free vmalloc area from the list under vmlist_lock. > - unmap pages under purge_lock. > > Now. vread(), vwrite() just take vmlist_lock, doesn't take purge_lock(). > It walks page table and find pte entry, page, kmap and access it. > > Oh, yes. It seems it's safe without kmap. But My concern is percpu allocator. > > It uses get_vm_area() and controls mapped pages by themselves and > map/unmap pages by with their own logic. vmalloc.c is just used for > alloc/free virtual address. > > Now, vread()/vwrite() just holds vmlist_lock() and walk page table > without no guarantee that the found page is stably mapped. So, I used kmap. > > If I miss something, I'm very sorry to add such kmap. Ah Thanks for explanation! I did some audit and find that - set_memory_uc(), set_memory_array_uc(), set_pages_uc(), set_pages_array_uc() are called EFI code and various video drivers, all of them don't touch HIGHMEM RAM - Kame: ioremap() won't allow remap of physical RAM So kmap_atomic() is safe. Let's just settle on this patch? Thanks, Fengguang -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org