linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Mike Rapoport <rppt@linux.ibm.com>
To: David Hildenbrand <david@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>,
	Vlastimil Babka <vbabka@suse.cz>, Mel Gorman <mgorman@suse.de>,
	Andrew Morton <akpm@linux-foundation.org>,
	linux-mm@kvack.org, Qian Cai <cai@lca.pw>,
	Michal Hocko <mhocko@kernel.org>,
	linux-kernel@vger.kernel.org, Baoquan He <bhe@redhat.com>
Subject: Re: [PATCH 1/1] mm: compaction: avoid fast_isolate_around() to set pageblock_skip on reserved pages
Date: Thu, 26 Nov 2020 19:46:01 +0200	[thread overview]
Message-ID: <20201126174601.GT123287@linux.ibm.com> (raw)
In-Reply-To: <3bb709a7-6100-aa5c-4125-7ed80c6d9643@redhat.com>

On Thu, Nov 26, 2020 at 11:05:14AM +0100, David Hildenbrand wrote:
> On 26.11.20 10:36, Mike Rapoport wrote:
> > On Wed, Nov 25, 2020 at 04:38:16PM -0500, Andrea Arcangeli wrote:
> > 
> >> At best memory.reserved should be calculated automatically by the
> >> page_alloc.c based on the zone_start_pfn/zone_end_pfn and not passed
> >> by the e820 caller, instead of adding the memory_reserve call for type
> >> 20 we should delete the memory_reserve function.
> >  
> > memory.reserved cannot be calculated automatically. It represents all
> > the memory allocations made before page allocator is up. And as
> > memblock_reserve() is the most basic to allocate memory early at boot we
> > cannot really delete it ;-)
> > 
> > As for e820 and type 20, unless it is in memblock, page_alloc.c has no
> > way to properly initialize memmap for it. It can continue to guess, like
> > it does with init_unavailable_memory().
> 
> AFAIS, init_unavailable_mem() has to logic to setup *any* such struct
> page properly up - except the node/zid, because that's unknown at that
> time. It is documented that (big comment above init_unavailable_mem())

// The comment is a bit inaccurate, but that's another story :)

> which pages won't get initialized again. They *are* initialized, just
> the node/zone is "fake" - that's the real problem we seem to have.

Let's try to merge init_unavailable_memory() into memmap_init().
Than it'll be able to set zone/nid for those nasty pfns that BIOS
decided to keep to itself, like in Andrea's case and will also take care
of struct pages that do not really have a frame in DRAM, but are there
because of arbitrary section size.

Something like this:

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index eaa227a479e4..072e94042a11 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6185,24 +6185,84 @@ static void __meminit zone_init_free_lists(struct zone *zone)
 	}
 }
 
-void __meminit __weak memmap_init(unsigned long size, int nid,
-				  unsigned long zone,
-				  unsigned long range_start_pfn)
+#if !defined(CONFIG_FLAT_NODE_MEM_MAP)
+/*
+ * Only struct pages that are backed by physical memory available to the
+ * kernel are zeroed and initialized by memmap_init_zone().
+ * But, there are some struct pages that are either reserved by firmware or
+ * do not correspond to physical page frames becuase actual memory bank is
+ * not a multiple of SECTION_SIZE. Fields of those struct pages may be
+ * accessed (for example page_to_pfn() on some configuration accesses
+ * flags) so we must explicitly initialize those struct pages.
+ */
+static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn,
+					 int zone, int node)
 {
-	unsigned long start_pfn, end_pfn;
+	unsigned long pfn;
+	u64 pgcnt = 0;
+
+	for (pfn = spfn; pfn < epfn; pfn++) {
+		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
+			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
+				+ pageblock_nr_pages - 1;
+			continue;
+		}
+		__init_single_page(pfn_to_page(pfn), pfn, zone, node);
+		__SetPageReserved(pfn_to_page(pfn));
+		pgcnt++;
+	}
+
+	return pgcnt;
+}
+#else
+static inline u64 init_unavailable_range(unsigned long spfn, unsigned long epfn,
+					 int zone, int node)
+{
+	return 0;
+}
+#endif
+
+void __init __weak memmap_init(unsigned long size, int nid,
+			       unsigned long zone,
+			       unsigned long range_start_pfn)
+{
+	unsigned long start_pfn, end_pfn, next_pfn = 0;
 	unsigned long range_end_pfn = range_start_pfn + size;
+	u64 pgcnt = 0;
 	int i;
 
 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
+		next_pfn = clamp(next_pfn, range_start_pfn, range_end_pfn);
 
 		if (end_pfn > start_pfn) {
 			size = end_pfn - start_pfn;
 			memmap_init_zone(size, nid, zone, start_pfn,
 					 MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
 		}
+
+		if (next_pfn < start_pfn)
+			pgcnt += init_unavailable_range(next_pfn, start_pfn,
+							zone, nid);
+		next_pfn = end_pfn;
 	}
+
+	/*
+	 * Early sections always have a fully populated memmap for the whole
+	 * section - see pfn_valid(). If the last section has holes at the
+	 * end and that section is marked "online", the memmap will be
+	 * considered initialized. Make sure that memmap has a well defined
+	 * state.
+	 */
+	if (next_pfn < range_end_pfn)
+		pgcnt += init_unavailable_range(next_pfn, range_end_pfn,
+						zone, nid);
+
+	if (pgcnt)
+		pr_info("%s: Zeroed struct page in unavailable ranges: %lld\n",
+			zone_names[zone], pgcnt);
+
 }
 
 static int zone_batchsize(struct zone *zone)
@@ -6995,88 +7055,6 @@ void __init free_area_init_memoryless_node(int nid)
 	free_area_init_node(nid);
 }
 
-#if !defined(CONFIG_FLAT_NODE_MEM_MAP)
-/*
- * Initialize all valid struct pages in the range [spfn, epfn) and mark them
- * PageReserved(). Return the number of struct pages that were initialized.
- */
-static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn)
-{
-	unsigned long pfn;
-	u64 pgcnt = 0;
-
-	for (pfn = spfn; pfn < epfn; pfn++) {
-		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
-			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
-				+ pageblock_nr_pages - 1;
-			continue;
-		}
-		/*
-		 * Use a fake node/zone (0) for now. Some of these pages
-		 * (in memblock.reserved but not in memblock.memory) will
-		 * get re-initialized via reserve_bootmem_region() later.
-		 */
-		__init_single_page(pfn_to_page(pfn), pfn, 0, 0);
-		__SetPageReserved(pfn_to_page(pfn));
-		pgcnt++;
-	}
-
-	return pgcnt;
-}
-
-/*
- * Only struct pages that are backed by physical memory are zeroed and
- * initialized by going through __init_single_page(). But, there are some
- * struct pages which are reserved in memblock allocator and their fields
- * may be accessed (for example page_to_pfn() on some configuration accesses
- * flags). We must explicitly initialize those struct pages.
- *
- * This function also addresses a similar issue where struct pages are left
- * uninitialized because the physical address range is not covered by
- * memblock.memory or memblock.reserved. That could happen when memblock
- * layout is manually configured via memmap=, or when the highest physical
- * address (max_pfn) does not end on a section boundary.
- */
-static void __init init_unavailable_mem(void)
-{
-	phys_addr_t start, end;
-	u64 i, pgcnt;
-	phys_addr_t next = 0;
-
-	/*
-	 * Loop through unavailable ranges not covered by memblock.memory.
-	 */
-	pgcnt = 0;
-	for_each_mem_range(i, &start, &end) {
-		if (next < start)
-			pgcnt += init_unavailable_range(PFN_DOWN(next),
-							PFN_UP(start));
-		next = end;
-	}
-
-	/*
-	 * Early sections always have a fully populated memmap for the whole
-	 * section - see pfn_valid(). If the last section has holes at the
-	 * end and that section is marked "online", the memmap will be
-	 * considered initialized. Make sure that memmap has a well defined
-	 * state.
-	 */
-	pgcnt += init_unavailable_range(PFN_DOWN(next),
-					round_up(max_pfn, PAGES_PER_SECTION));
-
-	/*
-	 * Struct pages that do not have backing memory. This could be because
-	 * firmware is using some of this memory, or for some other reasons.
-	 */
-	if (pgcnt)
-		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
-}
-#else
-static inline void __init init_unavailable_mem(void)
-{
-}
-#endif /* !CONFIG_FLAT_NODE_MEM_MAP */
-
 #if MAX_NUMNODES > 1
 /*
  * Figure out the number of possible node ids.
@@ -7500,7 +7478,6 @@ void __init free_area_init(unsigned long *max_zone_pfn)
 	/* Initialise every node */
 	mminit_verify_pageflags_layout();
 	setup_nr_node_ids();
-	init_unavailable_mem();
 	for_each_online_node(nid) {
 		pg_data_t *pgdat = NODE_DATA(nid);
 		free_area_init_node(nid);


> I agree that this is sub-optimal, as such pages are impossible to detect
> (PageReserved is just not clear as discussed with Andrea). The basic
> question is how we want to proceed:
> 
> a) Make sure any online struct page has a valid nid/zid, and is spanned
> by the nid/zid.
> b) Use a fake nid that will bail out when used for page_zone() and
> page_pgdat(), and make pfn walkers detect that.
> 
> AFAIU, Andrea seems to prefer a). I thing b) might be easier in corner
> cases. Thoughts?

I'd also prefer (a). 

The hardware defines what physical addresses correspond to which node,
so for any populated DIMM (or soldered DRAM for that matter) we can
detect page <-> node relationship.

As for the struct pages that just "hang" in the end of a section (your
example with 4000M), the addresses of these pages still obey the same
rules, so again, we have page <-> node correspondence.

The zones are software construct but they also correspond to some
hardware defined restrictions - each zone has a maximal PFN that HW
allows. Here again we can always know which zone spans that or another
page.

We'd have to fix a couple of things to get there, though :)

> ---
> 
> The tricky thing for b) is to fix all call places that do a
> page_zone()/page_pgdat() without checking if they are allowed to do so.
> We would have to audit all callers of page_zone() / page_zone_id() /
> page_pgdat() ...
> 
> 
> E.g., using a pattern like
> mm/memory_hotplug.c:find_biggest_section_pfn() is fine
> 
> 
> if (unlikely(!pfn_to_online_page(pfn)))
> 	continue;
> if (unlikely(pfn_to_nid(pfn) != nid))
> 	continue;
> 
> 
> e.g., mm/page_alloc.c:__pageblock_pfn_to_page() would have to be adjusted
> 
> start_page = pfn_to_online_page(start_pfn);
> if (!start_page)
> 	return NULL;
> 
> // necessary to avoid ...
> if (page_to_nid(page) == NODE_INVALID)
> 	return NULL;
> 
> // page_zone() from bailing out
> if (page_zone(start_page) != zone)
> 	return NULL;
> 
> -- 
> Thanks,
> 
> David / dhildenb
> 

-- 
Sincerely yours,
Mike.

  reply	other threads:[~2020-11-26 17:46 UTC|newest]

Thread overview: 61+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-04-23 21:25 compaction: VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn)) Qian Cai
2020-04-24  3:43 ` Baoquan He
2020-04-24 13:45   ` Qian Cai
2020-05-05 12:43     ` Baoquan He
2020-05-05 13:20       ` Qian Cai
2020-05-11  1:21         ` Baoquan He
2020-04-26 14:41 ` Mike Rapoport
2020-04-27 13:45   ` Qian Cai
2020-11-21 19:45 ` [PATCH 0/1] VM_BUG_ON_PAGE(!zone_spans_pfn) in set_pfnblock_flags_mask Andrea Arcangeli
2020-11-21 19:45   ` [PATCH 1/1] mm: compaction: avoid fast_isolate_around() to set pageblock_skip on reserved pages Andrea Arcangeli
2020-11-21 19:53     ` Andrea Arcangeli
2020-11-23 11:26       ` David Hildenbrand
2020-11-23 13:01     ` Vlastimil Babka
2020-11-24 13:32       ` Mel Gorman
2020-11-24 20:56         ` Andrea Arcangeli
2020-11-25 10:30           ` Mel Gorman
2020-11-25 17:59             ` Andrea Arcangeli
2020-11-26 10:47               ` Mel Gorman
2020-12-06  2:26                 ` Andrea Arcangeli
2020-12-06 23:47                   ` Mel Gorman
2020-11-25  5:34       ` Andrea Arcangeli
2020-11-25  6:45         ` David Hildenbrand
2020-11-25  8:51           ` Mike Rapoport
2020-11-25 10:39           ` Mel Gorman
2020-11-25 11:04             ` David Hildenbrand
2020-11-25 11:41               ` David Hildenbrand
2020-11-25 18:47                 ` Andrea Arcangeli
2020-11-25 13:33               ` Mel Gorman
2020-11-25 13:41                 ` David Hildenbrand
2020-11-25 18:28           ` Andrea Arcangeli
2020-11-25 19:27             ` David Hildenbrand
2020-11-25 20:41               ` Andrea Arcangeli
2020-11-25 21:13                 ` David Hildenbrand
2020-11-25 21:04               ` Mike Rapoport
2020-11-25 21:38                 ` Andrea Arcangeli
2020-11-26  9:36                   ` Mike Rapoport
2020-11-26 10:05                     ` David Hildenbrand
2020-11-26 17:46                       ` Mike Rapoport [this message]
2020-11-29 12:32                         ` Mike Rapoport
2020-12-02  0:44                           ` Andrea Arcangeli
2020-12-02 17:39                             ` Mike Rapoport
2020-12-03  6:23                               ` Andrea Arcangeli
2020-12-03 10:51                                 ` Mike Rapoport
2020-12-03 17:31                                   ` Andrea Arcangeli
2020-12-06  8:09                                     ` Mike Rapoport
2020-11-26 18:15                       ` Andrea Arcangeli
2020-11-26 18:29                     ` Andrea Arcangeli
2020-11-26 19:44                       ` Mike Rapoport
2020-11-26 20:30                         ` Andrea Arcangeli
2020-11-26 21:03                           ` Mike Rapoport
2020-11-26 19:21                     ` Andrea Arcangeli
2020-11-25 12:08         ` Vlastimil Babka
2020-11-25 13:32           ` David Hildenbrand
2020-11-25 14:13             ` Mike Rapoport
2020-11-25 14:42               ` David Hildenbrand
2020-11-26 10:51                 ` Mel Gorman
2020-11-25 19:14               ` Andrea Arcangeli
2020-11-25 19:01           ` Andrea Arcangeli
2020-11-25 19:33             ` David Hildenbrand
2020-11-26  3:40         ` Andrea Arcangeli
2020-11-26 10:43           ` Mike Rapoport

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20201126174601.GT123287@linux.ibm.com \
    --to=rppt@linux.ibm.com \
    --cc=aarcange@redhat.com \
    --cc=akpm@linux-foundation.org \
    --cc=bhe@redhat.com \
    --cc=cai@lca.pw \
    --cc=david@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=mgorman@suse.de \
    --cc=mhocko@kernel.org \
    --cc=vbabka@suse.cz \
    --subject='Re: [PATCH 1/1] mm: compaction: avoid fast_isolate_around() to set pageblock_skip on reserved pages' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).