linux-kernel - Re: [PATCH 2/2] mm: zero remaining unavailable struct pages

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 31 Aug 2018 02:55:36 +0000
From:   Naoya Horiguchi <n-horiguchi@...jp.nec.com>
To:     Masayoshi Mizuma <msys.mizuma@...il.com>
CC:     "Pavel.Tatashin@...rosoft.com" <Pavel.Tatashin@...rosoft.com>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        "mhocko@...nel.org" <mhocko@...nel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "x86@...nel.org" <x86@...nel.org>
Subject: Re: [PATCH 2/2] mm: zero remaining unavailable struct pages

On Wed, Aug 29, 2018 at 11:16:30AM -0400, Masayoshi Mizuma wrote:
> Hi Horiguchi-san and Pavel
> 
> Thank you for your comments!
> The Pavel's additional patch looks good to me, so I will add it to this series.
> 
> However, unfortunately, the movable_node option has something wrong yet...
> When I offline the memory which belongs to movable zone, I got the following
> warning. I'm trying to debug it.
> 
> I try to describe the issue as following. 
> If you have any comments, please let me know.
> 
> WARNING: CPU: 156 PID: 25611 at mm/page_alloc.c:7730 has_unmovable_pages+0x1bf/0x200
> RIP: 0010:has_unmovable_pages+0x1bf/0x200
> ...
> Call Trace:
>  is_mem_section_removable+0xd3/0x160
>  show_mem_removable+0x8e/0xb0
>  dev_attr_show+0x1c/0x50
>  sysfs_kf_seq_show+0xb3/0x110
>  seq_read+0xee/0x480
>  __vfs_read+0x36/0x190
>  vfs_read+0x89/0x130
>  ksys_read+0x52/0xc0
>  do_syscall_64+0x5b/0x180
>  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> RIP: 0033:0x7fe7b7823f70
> ...
> 
> I added a printk to catch the unmovable page.
> ---
> @@ -7713,8 +7719,12 @@ bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
>                  * is set to both of a memory hole page and a _used_ kernel
>                  * page at boot.
>                  */
> -               if (found > count)
> +               if (found > count) {
> +                       pr_info("DEBUG: %s zone: %lx page: %lx pfn: %lx flags: %lx found: %ld count: %ld \n",
> +                               __func__, zone, page, page_to_pfn(page), page->flags, found, count);
>                         goto unmovable;
> +               }
> ---
> 
> Then I got the following. The page (PFN: 0x1c0ff130d) flag is 
> 0xdfffffc0040048 (uptodate|active|swapbacked)
> 
> ---
> DEBUG: has_unmovable_pages zone: 0xffff8c0ffff80380 page: 0xffffea703fc4c340 pfn: 0x1c0ff130d flags: 0xdfffffc0040048 found: 1 count: 0 
> ---
> 
> And I got the owner from /sys/kernel/debug/page_owner.
> 
> Page allocated via order 0, mask 0x6280ca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO)
> PFN 7532909325 type Movable Block 14712713 type Movable Flags 0xdfffffc0040048(uptodate|active|swapbacked)
>  __alloc_pages_nodemask+0xfc/0x270
>  alloc_pages_vma+0x7c/0x1e0
>  handle_pte_fault+0x399/0xe50
>  __handle_mm_fault+0x38e/0x520
>  handle_mm_fault+0xdc/0x210
>  __do_page_fault+0x243/0x4c0
>  do_page_fault+0x31/0x130
>  page_fault+0x1e/0x30
> 
> The page is allocated as anonymous page via page fault.
> I'm not sure, but lru flag should be added to the page...?

There is a small window of no PageLRU flag just after page allocation
until the page is linked to some LRU list.
This kind of unmovability is transient, so retrying can work.

I guess that this warning seems to be visible since commit 15c30bc09085
("mm, memory_hotplug: make has_unmovable_pages more robust")
which turned off the optimization based on the assumption that pages
under ZONE_MOVABLE are always movable.
I think that it helps developers find the issue that permanently
unmovable pages are accidentally located in ZONE_MOVABLE zone.
But even ZONE_MOVABLE zone could have transiently unmovable pages,
so the reported warning seems to me a false charge and should be avoided.
Doing lru_add_drain_all()/drain_all_pages() before has_unmovable_pages()
might be helpful?

Thanks,
Naoya Horiguchi