[PATCH v2 0/2] Fixes for sub-section hotplug

[PATCH v2 0/2] Fixes for sub-section hotplug

[Oscar Salvador]

v2 -> v1: Go the easy way and just adapt the check (Dan/Aneesh)

Hi all,

these two patches address a couple of issues I found while working on my
vmemmap-patchset.
The issues are:

        1) section_deactivate mistakenly zeroes ms->section_mem_map and then
           tries to check whether the section is an early section, but since
           section_mem_map might have been zeroed, we will return false
           when it is really an early section.
           In order to fix this, let us check whether the section is early
           at function entry, so we do not neet check it again later.

        2) shrink_{node,zone}_span work on sub-section granularity now.
           The problem is that since deactivation of the section occurs later
           on in sparse_remove_section, so the pfn_valid()->pfn_section_valid()
           check will always return true for every sub-section chunk.
           In order to avoid that, let us adapt the check and skip the whole
           range to be removed.
           The user visible effect of this is that we are always left with,
           at least, PAGES_PER_SECTION spanned, even if we got to remove all
           memory linked to a zone/node

Oscar Salvador (2):
  mm,sparse: Fix deactivate_section for early sections
  mm,memory_hotplug: Fix shrink_{zone,node}_span

 mm/memory_hotplug.c | 8 ++++----
 mm/sparse.c         | 5 +++--
 2 files changed, 7 insertions(+), 6 deletions(-)

-- 
2.12.3

[PATCH v2 1/2] mm,sparse: Fix deactivate_section for early sections

[Oscar Salvador]

deactivate_section checks whether a section is early or not
in order to either call free_map_bootmem() or depopulate_section_memmap().
Being the former for sections added at boot time, and the latter for
sections hotplugged.

The problem is that we zero section_mem_map, so the last early_section()
will always report false and the section will not be removed.

Fix this checking whether a section is early or not at function
entry.

Fixes: mmotm ("mm/sparsemem: Support sub-section hotplug")
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Dan Williams <dan.j.wiliams@intel.com>
---
 mm/sparse.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/mm/sparse.c b/mm/sparse.c
index 3267c4001c6d..1e224149aab6 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -738,6 +738,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
 	DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
 	DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
 	struct mem_section *ms = __pfn_to_section(pfn);
+	bool section_is_early = early_section(ms);
 	struct page *memmap = NULL;
 	unsigned long *subsection_map = ms->usage
 		? &ms->usage->subsection_map[0] : NULL;
@@ -772,7 +773,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
 	if (bitmap_empty(subsection_map, SUBSECTIONS_PER_SECTION)) {
 		unsigned long section_nr = pfn_to_section_nr(pfn);
 
-		if (!early_section(ms)) {
+		if (!section_is_early) {
 			kfree(ms->usage);
 			ms->usage = NULL;
 		}
@@ -780,7 +781,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
 		ms->section_mem_map = sparse_encode_mem_map(NULL, section_nr);
 	}
 
-	if (early_section(ms) && memmap)
+	if (section_is_early && memmap)
 		free_map_bootmem(memmap);
 	else
 		depopulate_section_memmap(pfn, nr_pages, altmap);
-- 
2.12.3

[PATCH v2 2/2] mm,memory_hotplug: Fix shrink_{zone,node}_span

[Oscar Salvador]

Since [1], shrink_{zone,node}_span work on PAGES_PER_SUBSECTION granularity.
We need to adapt the loop that checks whether a zone/node contains only holes,
and skip the whole range to be removed.

Otherwise, since sub-sections belonging to the range to be removed have not yet
been deactivated, pfn_valid() will return true on those and we will be left
with a wrong accounting of spanned_pages, both for the zone and the node.

Fixes: mmotm ("mm/hotplug: prepare shrink_{zone, pgdat}_span for sub-section removal")
Signed-off-by: Oscar Salvador <osalvador@suse.de>
---
 mm/memory_hotplug.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index b9ba5b85f9f7..2a9bbddb0e55 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -422,8 +422,8 @@ static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 		if (page_zone(pfn_to_page(pfn)) != zone)
 			continue;
 
-		 /* If the section is current section, it continues the loop */
-		if (start_pfn == pfn)
+		/* Skip range to be removed */
+		if (pfn >= start_pfn && pfn < end_pfn)
 			continue;
 
 		/* If we find valid section, we have nothing to do */
@@ -487,8 +487,8 @@ static void shrink_pgdat_span(struct pglist_data *pgdat,
 		if (pfn_to_nid(pfn) != nid)
 			continue;
 
-		 /* If the section is current section, it continues the loop */
-		if (start_pfn == pfn)
+		/* Skip range to be removed */
+		if (pfn >= start_pfn && pfn < end_pfn)
 			continue;
 
 		/* If we find valid section, we have nothing to do */
-- 
2.12.3

Re: [PATCH v2 2/2] mm,memory_hotplug: Fix shrink_{zone,node}_span

[Dan Williams]

On Wed, Jul 17, 2019 at 2:07 AM Oscar Salvador <osalvador@suse.de> wrote:
>
> Since [1], shrink_{zone,node}_span work on PAGES_PER_SUBSECTION granularity.
> We need to adapt the loop that checks whether a zone/node contains only holes,
> and skip the whole range to be removed.
>
> Otherwise, since sub-sections belonging to the range to be removed have not yet
> been deactivated, pfn_valid() will return true on those and we will be left
> with a wrong accounting of spanned_pages, both for the zone and the node.
>
> Fixes: mmotm ("mm/hotplug: prepare shrink_{zone, pgdat}_span for sub-section removal")
> Signed-off-by: Oscar Salvador <osalvador@suse.de>

Looks good,

Reviewed-by: Dan Williams <dan.j.williams@intel.com>