Re: [PATCH v2 2/3] mm, page_owner: decouple freeing stack trace from debug_pagealloc

From: "Kirill A. Shutemov" <kirill@shutemov.name>
To: Qian Cai <cai@lca.pw>
Cc: Vlastimil Babka <vbabka@suse.cz>,
	Andrew Morton <akpm@linux-foundation.org>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	kasan-dev@googlegroups.com,
	"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
	Matthew Wilcox <willy@infradead.org>,
	Mel Gorman <mgorman@techsingularity.net>,
	Michal Hocko <mhocko@kernel.org>,
	Dmitry Vyukov <dvyukov@google.com>,
	Walter Wu <walter-zh.wu@mediatek.com>,
	Andrey Ryabinin <aryabinin@virtuozzo.com>
Subject: Re: [PATCH v2 2/3] mm, page_owner: decouple freeing stack trace from debug_pagealloc
Date: Tue, 1 Oct 2019 15:31:51 +0300	[thread overview]
Message-ID: <20191001123151.qmv4ist3enq65uha@box> (raw)
In-Reply-To: <1569932788.5576.247.camel@lca.pw>

On Tue, Oct 01, 2019 at 08:26:28AM -0400, Qian Cai wrote:
> On Tue, 2019-10-01 at 14:51 +0300, Kirill A. Shutemov wrote:
> > On Tue, Oct 01, 2019 at 10:07:44AM +0200, Vlastimil Babka wrote:
> > > On 10/1/19 1:49 AM, Qian Cai wrote:
> > > > 
> > > > 
> > > > > On Sep 30, 2019, at 5:43 PM, Vlastimil Babka <vbabka@suse.cz> wrote:
> > > > > 
> > > > > Well, my use case is shipping production kernels with CONFIG_PAGE_OWNER
> > > > > and CONFIG_DEBUG_PAGEALLOC enabled, and instructing users to boot-time
> > > > > enable only for troubleshooting a crash or memory leak, without a need
> > > > > to install a debug kernel. Things like static keys and page_ext
> > > > > allocations makes this possible without CPU and memory overhead when not
> > > > > boot-time enabled. I don't know too much about KASAN internals, but I
> > > > > assume it's not possible to use it that way on production kernels yet?
> > > > 
> > > > In that case, why can’t users just simply enable page_owner=on and
> > > > debug_pagealloc=on for troubleshooting? The later makes the kernel
> > > > slower, but I am not sure if it is worth optimization by adding a new
> > > > parameter. There have already been quite a few MM-related kernel
> > > > parameters that could tidy up a bit in the future.
> > > 
> > > They can do that and it was intention, yes. The extra parameter was
> > > requested by Kirill, so I'll defer the answer to him :)
> > 
> > DEBUG_PAGEALLOC is much more intrusive debug option. Not all architectures
> > support it in an efficient way. Some require hibernation.
> > 
> > I don't see a reason to tie these two option together.
> 
> Make sense. How about page_owner=on will have page_owner_free=on by default?
> That way we don't need the extra parameter.

It's 0.1% of system memory. Does it matter for a debug option? I don't know.

-- 
 Kirill A. Shutemov