From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.0 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_HELO_NONE,SPF_PASS, UNPARSEABLE_RELAY autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 1B815C433E0 for ; Fri, 3 Jul 2020 01:34:42 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id DB313207E8 for ; Fri, 3 Jul 2020 01:34:41 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org DB313207E8 Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=linux.alibaba.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=owner-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix) id 56E538D0037; Thu, 2 Jul 2020 21:34:41 -0400 (EDT) Received: by kanga.kvack.org (Postfix, from userid 40) id 51F0A8D0021; Thu, 2 Jul 2020 21:34:41 -0400 (EDT) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 40D0B8D0037; Thu, 2 Jul 2020 21:34:41 -0400 (EDT) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0100.hostedemail.com [216.40.44.100]) by kanga.kvack.org (Postfix) with ESMTP id 2BE948D0021 for ; Thu, 2 Jul 2020 21:34:41 -0400 (EDT) Received: from smtpin05.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay03.hostedemail.com (Postfix) with ESMTP id C6CD1824556B for ; Fri, 3 Jul 2020 01:34:40 +0000 (UTC) X-FDA: 76995045120.05.heart92_540549726e8e Received: from filter.hostedemail.com (10.5.16.251.rfc1918.com [10.5.16.251]) by smtpin05.hostedemail.com (Postfix) with ESMTP id A1D9F1802820A for ; Fri, 3 Jul 2020 01:34:40 +0000 (UTC) X-HE-Tag: heart92_540549726e8e X-Filterd-Recvd-Size: 7024 Received: from out30-133.freemail.mail.aliyun.com (out30-133.freemail.mail.aliyun.com [115.124.30.133]) by imf48.hostedemail.com (Postfix) with ESMTP for ; Fri, 3 Jul 2020 01:34:39 +0000 (UTC) X-Alimail-AntiSpam:AC=PASS;BC=-1|-1;BR=01201311R131e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e04357;MF=richard.weiyang@linux.alibaba.com;NM=1;PH=DS;RN=13;SR=0;TI=SMTPD_---0U1XFdqr_1593740075; Received: from localhost(mailfrom:richard.weiyang@linux.alibaba.com fp:SMTPD_---0U1XFdqr_1593740075) by smtp.aliyun-inc.com(127.0.0.1); Fri, 03 Jul 2020 09:34:36 +0800 Date: Fri, 3 Jul 2020 09:34:35 +0800 From: Wei Yang To: David Hildenbrand Cc: Wei Yang , dave.hansen@linux.intel.com, luto@kernel.org, peterz@infradead.org, tglx@linutronix.de, mingo@redhat.com, bp@alien8.de, akpm@linux-foundation.org, x86@kernel.org, linux-kernel@vger.kernel.org, kasan-dev@googlegroups.com, linux-mm@kvack.org Subject: Re: [PATCH] mm: define pte_add_end for consistency Message-ID: <20200703013435.GA11340@L-31X9LVDL-1304.local> Reply-To: Wei Yang References: <20200630031852.45383-1-richard.weiyang@linux.alibaba.com> <40362e99-a354-c44f-8645-e2326a6df680@redhat.com> <20200701021113.GA51306@L-31X9LVDL-1304.local> <20200701115441.GA4979@L-31X9LVDL-1304.local> <7562991b-c1e7-4037-a3f0-124acd0669b7@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <7562991b-c1e7-4037-a3f0-124acd0669b7@redhat.com> X-Rspamd-Queue-Id: A1D9F1802820A X-Spamd-Result: default: False [0.00 / 100.00] X-Rspamd-Server: rspam04 X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: On Thu, Jul 02, 2020 at 06:28:19PM +0200, David Hildenbrand wrote: >On 01.07.20 13:54, Wei Yang wrote: >> On Wed, Jul 01, 2020 at 10:29:08AM +0200, David Hildenbrand wrote: >>> On 01.07.20 04:11, Wei Yang wrote: >>>> On Tue, Jun 30, 2020 at 02:44:00PM +0200, David Hildenbrand wrote: >>>>> On 30.06.20 05:18, Wei Yang wrote: >>>>>> When walking page tables, we define several helpers to get the address of >>>>>> the next boundary. But we don't have one for pte level. >>>>>> >>>>>> Let's define it and consolidate the code in several places. >>>>>> >>>>>> Signed-off-by: Wei Yang >>>>>> --- >>>>>> arch/x86/mm/init_64.c | 6 ++---- >>>>>> include/linux/pgtable.h | 7 +++++++ >>>>>> mm/kasan/init.c | 4 +--- >>>>>> 3 files changed, 10 insertions(+), 7 deletions(-) >>>>>> >>>>>> diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c >>>>>> index dbae185511cd..f902fbd17f27 100644 >>>>>> --- a/arch/x86/mm/init_64.c >>>>>> +++ b/arch/x86/mm/init_64.c >>>>>> @@ -973,9 +973,7 @@ remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end, >>>>>> >>>>>> pte = pte_start + pte_index(addr); >>>>>> for (; addr < end; addr = next, pte++) { >>>>>> - next = (addr + PAGE_SIZE) & PAGE_MASK; >>>>>> - if (next > end) >>>>>> - next = end; >>>>>> + next = pte_addr_end(addr, end); >>>>>> >>>>>> if (!pte_present(*pte)) >>>>>> continue; >>>>>> @@ -1558,7 +1556,7 @@ void register_page_bootmem_memmap(unsigned long section_nr, >>>>>> get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO); >>>>>> >>>>>> if (!boot_cpu_has(X86_FEATURE_PSE)) { >>>>>> - next = (addr + PAGE_SIZE) & PAGE_MASK; >>>>>> + next = pte_addr_end(addr, end); >>>>>> pmd = pmd_offset(pud, addr); >>>>>> if (pmd_none(*pmd)) >>>>>> continue; >>>>>> diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h >>>>>> index 32b6c52d41b9..0de09c6c89d2 100644 >>>>>> --- a/include/linux/pgtable.h >>>>>> +++ b/include/linux/pgtable.h >>>>>> @@ -706,6 +706,13 @@ static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot) >>>>>> }) >>>>>> #endif >>>>>> >>>>>> +#ifndef pte_addr_end >>>>>> +#define pte_addr_end(addr, end) \ >>>>>> +({ unsigned long __boundary = ((addr) + PAGE_SIZE) & PAGE_MASK; \ >>>>>> + (__boundary - 1 < (end) - 1) ? __boundary : (end); \ >>>>>> +}) >>>>>> +#endif >>>>>> + >>>>>> /* >>>>>> * When walking page tables, we usually want to skip any p?d_none entries; >>>>>> * and any p?d_bad entries - reporting the error before resetting to none. >>>>>> diff --git a/mm/kasan/init.c b/mm/kasan/init.c >>>>>> index fe6be0be1f76..89f748601f74 100644 >>>>>> --- a/mm/kasan/init.c >>>>>> +++ b/mm/kasan/init.c >>>>>> @@ -349,9 +349,7 @@ static void kasan_remove_pte_table(pte_t *pte, unsigned long addr, >>>>>> unsigned long next; >>>>>> >>>>>> for (; addr < end; addr = next, pte++) { >>>>>> - next = (addr + PAGE_SIZE) & PAGE_MASK; >>>>>> - if (next > end) >>>>>> - next = end; >>>>>> + next = pte_addr_end(addr, end); >>>>>> >>>>>> if (!pte_present(*pte)) >>>>>> continue; >>>>>> >>>>> >>>>> I'm not really a friend of this I have to say. We're simply iterating >>>>> over single pages, not much magic .... >>>> >>>> Hmm... yes, we are iterating on Page boundary, while we many have the case >>>> when addr or end is not PAGE_ALIGN. >>> >>> I really do wonder if not having page aligned addresses actually happens >>> in real life. Page tables operate on page granularity, and >>> adding/removing unaligned parts feels wrong ... and that's also why I >>> dislike such a helper. >>> >>> 1. kasan_add_zero_shadow()/kasan_remove_zero_shadow(). If I understand >>> the logic (WARN_ON()) correctly, we bail out in case we would ever end >>> up in such a scenario, where we would want to add/remove things not >>> aligned to PAGE_SIZE. >>> >>> 2. remove_pagetable()...->remove_pte_table() >>> >>> vmemmap_free() should never try to de-populate sub-pages. Even with >>> sub-section hot-add/remove (2MB / 512 pages), with valid struct page >>> sizes (56, 64, 72, 80), we always end up with full pages. >>> >>> kernel_physical_mapping_remove() is only called via >>> arch_remove_memory(). That will never remove unaligned parts. >>> >> >> I don't have a very clear mind now, while when you look into >> remove_pte_table(), it has two cases based on alignment of addr and next. >> >> If we always remove a page, the second case won't happen? > >So, the code talks about that the second case can only happen for >vmemmap, never for direct mappings. > >I don't see a way how this could ever happen with current page sizes, >even with sub-section hotadd (2MB). Maybe that is a legacy leftover or >was never relevant? Or I am missing something important, where we could >have sub-4k-page vmemmap data. > I took a calculation on the sub-section page struct size, it is page size (4K) aligned. This means you are right, which we won't depopulate a sub-page. And yes, I am not sure all those variants would fit this case. So I would like to leave as it now. How about your opinion? >-- >Thanks, > >David / dhildenb -- Wei Yang Help you, Help me