Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Sender: Minchan Kim <minchan.kim@gmail.com>
Date:   Wed, 11 May 2022 16:08:56 -0700
From:   Minchan Kim <minchan@kernel.org>
To:     John Hubbard <jhubbard@nvidia.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        linux-mm <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>,
        "Paul E . McKenney" <paulmck@kernel.org>,
        John Dias <joaodias@google.com>,
        David Hildenbrand <david@redhat.com>
Subject: Re: [PATCH v4] mm: fix is_pinnable_page against on cma page
Message-ID: <YnxCCPZUfgQNXSg6@google.com>
References: <20220510211743.95831-1-minchan@kernel.org>
 <857d21da-5de2-fa3e-b1ce-41cc1cfb0191@nvidia.com>
 <Ynr10y6pkH6WwbQo@google.com>
 <bb6e76f1-2cbc-893d-c8ab-3ecd3fcae2a5@nvidia.com>
 <Ynr+wTCQpyh8+vOD@google.com>
 <2ffa7670-04ea-bb28-28f8-93a9b9eea7e8@nvidia.com>
 <YnwupNzDNv7IbjRQ@google.com>
 <54b5d177-f2f4-cef2-3a68-cd3b0b276f86@nvidia.com>
 <Ynw6mauQuNhrOAHy@google.com>
 <f7bcda22-8ffe-67be-36bc-fcde58d8884a@nvidia.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <f7bcda22-8ffe-67be-36bc-fcde58d8884a@nvidia.com>
Precedence: bulk

On Wed, May 11, 2022 at 03:49:06PM -0700, John Hubbard wrote:
> On 5/11/22 15:37, Minchan Kim wrote:
> > > Yes. But one thing that is still unanswered, that I think you can
> > > answer, is: even if the compiler *did* re-read the mt variable, what
> > > problems could that cause? I claim "no problems", because there is
> > > no combination of 0, _CMA, _ISOLATE, _CMA|ISOLATE that will cause
> > > problems here.
> > 
> > What scenario I am concerning with __READ_ONCE so compiler
> > inlining get_pageblock_migratetype two times are
> > 
> >          CPU 0                                                       CPU 1
> >                                                                  alloc_contig_range
> > is_pinnable_page                                                start_isolate_page_range
> >                                                                    set_pageblock_migratetype(MIGRATE_ISOLATE)
> >     if (get_pageeblock_migratetype(page) == MIGRATE_CMA)
> >         so it's false
> >                                                                  undo:
> >                                                                    set_pageblock_migratetype(MIGRATE_CMA)
> >     if (get_pageeblock_migratetype(page) == MIGRATE_ISOLATE)
> >         so it's false
> > 
> > In the end, CMA memory would be pinned by CPU 0 process
> > so CMA allocation keep failed until the process release the
> > refcount.
> > 
> 
> OK, so the code checks the wrong item each time. But the code really
> only needs to know "is either _CMA or _ISOLATE set?". And so you

Yes.

> can just sidestep the entire question by writing it like this:
> 
> int mt = get_pageblock_migratetype(page);
> 
> if (mt & (MIGRATE_ISOLATE | MIGRATE_CMA))
> 	return false;

I am confused. Isn't it same question?

                                                    set_pageblock_migratetype(MIGRATE_ISOLATE)
if (get_pageblock_migrate(page) & MIGRATE_CMA)

                                                    set_pageblock_migratetype(MIGRATE_CMA)

if (get_pageblock_migrate(page) & MIGRATE_ISOLATE)