linux-kernel - Re: [PATCH v2] mm/hugetlb: Don't put_page in lock of hugetlb

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:   Mon, 6 May 2019 16:20:01 +0200
From:   Michal Hocko <mhocko@...nel.org>
To:     Zhiqiang Liu <liuzhiqiang26@...wei.com>
Cc:     mike.kravetz@...cle.com, shenkai8@...wei.com,
        linfeilong@...wei.com, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, wangwang2@...wei.com,
        "Zhoukang (A)" <zhoukang7@...wei.com>,
        Mingfangsen <mingfangsen@...wei.com>, agl@...ibm.com,
        nacc@...ibm.com, Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH v2] mm/hugetlb: Don't put_page in lock of hugetlb_lock

On Mon 06-05-19 22:06:38, Zhiqiang Liu wrote:
> From: Kai Shen <shenkai8@...wei.com>
> 
> spinlock recursion happened when do LTP test:
> #!/bin/bash
> ./runltp -p -f hugetlb &
> ./runltp -p -f hugetlb &
> ./runltp -p -f hugetlb &
> ./runltp -p -f hugetlb &
> ./runltp -p -f hugetlb &
> 
> The dtor returned by get_compound_page_dtor in __put_compound_page
> may be the function of free_huge_page which will lock the hugetlb_lock,
> so don't put_page in lock of hugetlb_lock.
> 
>  BUG: spinlock recursion on CPU#0, hugemmap05/1079
>   lock: hugetlb_lock+0x0/0x18, .magic: dead4ead, .owner: hugemmap05/1079, .owner_cpu: 0
>  Call trace:
>   dump_backtrace+0x0/0x198
>   show_stack+0x24/0x30
>   dump_stack+0xa4/0xcc
>   spin_dump+0x84/0xa8
>   do_raw_spin_lock+0xd0/0x108
>   _raw_spin_lock+0x20/0x30
>   free_huge_page+0x9c/0x260
>   __put_compound_page+0x44/0x50
>   __put_page+0x2c/0x60
>   alloc_surplus_huge_page.constprop.19+0xf0/0x140
>   hugetlb_acct_memory+0x104/0x378
>   hugetlb_reserve_pages+0xe0/0x250
>   hugetlbfs_file_mmap+0xc0/0x140
>   mmap_region+0x3e8/0x5b0
>   do_mmap+0x280/0x460
>   vm_mmap_pgoff+0xf4/0x128
>   ksys_mmap_pgoff+0xb4/0x258
>   __arm64_sys_mmap+0x34/0x48
>   el0_svc_common+0x78/0x130
>   el0_svc_handler+0x38/0x78
>   el0_svc+0x8/0xc
> 
> Fixes: 9980d744a0 ("mm, hugetlb: get rid of surplus page accounting tricks")
> Signed-off-by: Kai Shen <shenkai8@...wei.com>
> Signed-off-by: Feilong Lin <linfeilong@...wei.com>
> Reported-by: Wang Wang <wangwang2@...wei.com>
> Acked-by: Michal Hocko <mhocko@...e.com>
> ---
> v1->v2: add Acked-by: Michal Hocko <mhocko@...e.com>

A new version for single ack is usually an overkill and only makes the
situation more confusing. You have also didn't add Cc: stable as
suggested during the review. That part is arguably more important.

You also haven't CCed Andrew (now done) and your patch will not get
merged without him applying it. Anyway, let's wait for Andrew to pick
this patch up.

> 
>  mm/hugetlb.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 6cdc7b2..c1e7b81 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1574,8 +1574,9 @@ static struct page *alloc_surplus_huge_page(struct hstate *h, gfp_t gfp_mask,
>  	 */
>  	if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) {
>  		SetPageHugeTemporary(page);
> +		spin_unlock(&hugetlb_lock);
>  		put_page(page);
> -		page = NULL;
> +		return NULL;
>  	} else {
>  		h->surplus_huge_pages++;
>  		h->surplus_huge_pages_node[page_to_nid(page)]++;
> -- 
> 1.8.3.1
> 

-- 
Michal Hocko
SUSE Labs