linux-kernel - Re: [PATCH 1/4] mm/mm_init: use deferred_init_memmap_chunk() in deferred_grow

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]

Message-ID: <aKRX9iIe8h9fFi9v@kernel.org>
Date: Tue, 19 Aug 2025 13:54:46 +0300
From: Mike Rapoport <rppt@...nel.org>
To: Wei Yang <richard.weiyang@...il.com>
Cc: linux-mm@...ck.org, Andrew Morton <akpm@...ux-foundation.org>,
	Bill Wendling <morbo@...gle.com>,
	Daniel Jordan <daniel.m.jordan@...cle.com>,
	Justin Stitt <justinstitt@...gle.com>,
	Michael Ellerman <mpe@...erman.id.au>,
	Miguel Ojeda <ojeda@...nel.org>,
	Nathan Chancellor <nathan@...nel.org>,
	Nick Desaulniers <nick.desaulniers+lkml@...il.com>,
	linux-kernel@...r.kernel.org, llvm@...ts.linux.dev
Subject: Re: [PATCH 1/4] mm/mm_init: use deferred_init_memmap_chunk() in
 deferred_grow_zone()

On Tue, Aug 19, 2025 at 09:52:23AM +0000, Wei Yang wrote:
> Hi, Mike
> 
> After going through the code again, I have some trivial thoughts to discuss
> with you. If not right, please let me know.
> 
> On Mon, Aug 18, 2025 at 09:46:12AM +0300, Mike Rapoport wrote:
> [...]
> > bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
> > {
> >-	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
> >+	unsigned long nr_pages_needed = SECTION_ALIGN_UP(1 << order);
> > 	pg_data_t *pgdat = zone->zone_pgdat;
> > 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
> > 	unsigned long spfn, epfn, flags;
> > 	unsigned long nr_pages = 0;
> >-	u64 i = 0;
> > 
> > 	/* Only the last zone may have deferred pages */
> > 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
> >@@ -2262,37 +2272,26 @@ bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
> > 		return true;
> > 	}
> 
> In the file above this line, there is a compare between first_deferred_pfn and
> its original value after grab pgdat_resize_lock.

Do you mean this one:

	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
		pgdat_resize_unlock(pgdat, &flags);
		return true;
	}
 
> I am thinking to compare first_deferred_pfn with ULONG_MAX, as it compared in
> deferred_init_memmap(). This indicate this zone has already been initialized
> totally.

It may be another CPU ran deferred_grow_zone() and won the race for resize
lock. Then pgdat->first_deferred_pfn will be larger than
first_deferred_pfn, but still not entire zone would be initialized.
 
> Current code guard this by spfn < zone_end_pfn(zone). Maybe a check ahead
> would be more clear?

Not sure I follow you here. The check that we don't pass zone_end_pfn is
inside the loop for every section we initialize.
 
> > 
> >-	/* If the zone is empty somebody else may have cleared out the zone */
> >-	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
> >-						 first_deferred_pfn)) {
> >-		pgdat->first_deferred_pfn = ULONG_MAX;
> >-		pgdat_resize_unlock(pgdat, &flags);
> >-		/* Retry only once. */
> >-		return first_deferred_pfn != ULONG_MAX;
> >+	/*
> >+	 * Initialize at least nr_pages_needed in section chunks.
> >+	 * If a section has less free memory than nr_pages_needed, the next
> >+	 * section will be also initalized.
> >+	 * Note, that it still does not guarantee that allocation of order can
> >+	 * be satisfied if the sections are fragmented because of memblock
> >+	 * allocations.
> >+	 */
> >+	for (spfn = first_deferred_pfn, epfn = SECTION_ALIGN_UP(spfn + 1);
> 
> I am expecting first_deferred_pfn is section aligned. So epfn += PAGES_PER_SECTION
> is fine?

It should be, but I'd prefer to be on the safe side and keep it this way.
 
> Maybe I missed something.
> 
> >+	     nr_pages < nr_pages_needed && spfn < zone_end_pfn(zone);
> >+	     spfn = epfn, epfn += PAGES_PER_SECTION) {
> >+		nr_pages += deferred_init_memmap_chunk(spfn, epfn, zone);
> > 	}
> > 
> > 	/*
> >-	 * Initialize and free pages in MAX_PAGE_ORDER sized increments so
> >-	 * that we can avoid introducing any issues with the buddy
> >-	 * allocator.
> >+	 * There were no pages to initialize and free which means the zone's
> >+	 * memory map is completely initialized.
> > 	 */
> >-	while (spfn < epfn) {
> >-		/* update our first deferred PFN for this section */
> >-		first_deferred_pfn = spfn;
> >-
> >-		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
> >-		touch_nmi_watchdog();
> >-
> >-		/* We should only stop along section boundaries */
> >-		if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
> >-			continue;
> >-
> >-		/* If our quota has been met we can stop here */
> >-		if (nr_pages >= nr_pages_needed)
> >-			break;
> >-	}
> >+	pgdat->first_deferred_pfn = nr_pages ? spfn : ULONG_MAX;
> 
> If we come here because spfn >= zone_end_pfn(zone), first_deferred_pfn is left
> a "valid" value and deferred_init_memmap() will try to do its job. But
> actually nothing left to initialize.

We anyway run a thread for each node with memory. In the very unlikely case
we've completely initialized a deferred zone that thread will finish much
faster :)
 
> For this case, I suggest to set it ULONG_MAX too. But this is really corner
> case.

-- 
Sincerely yours,
Mike.