[PATCH v1 10/12] mm: convert MAX_ORDER sized static arrays to dynamic ones.

From: Zi Yan
Date: Wed Sep 21 2022 - 21:13:47 EST


From: Zi Yan <ziy@xxxxxxxxxx>

This prepares for the upcoming changes to make MAX_ORDER a boot time
parameter instead of compilation time constant. All static arrays with
MAX_ORDER size are converted to pointers and their memory is allocated
at runtime.

free_area array in struct zone is allocated using memblock_alloc_node()
at boot time and using kzalloc() when memory is hot-added.

Signed-off-by: Zi Yan <ziy@xxxxxxxxxx>
Cc: Dave Young <dyoung@xxxxxxxxxx>
Cc: Jonathan Corbet <corbet@xxxxxxx>
Cc: Christian Koenig <christian.koenig@xxxxxxx>
Cc: David Airlie <airlied@xxxxxxxx>
Cc: kexec@xxxxxxxxxxxxxxxxxxx
Cc: linux-doc@xxxxxxxxxxxxxxx
Cc: dri-devel@xxxxxxxxxxxxxxxxxxxxx
Cc: linux-mm@xxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx
---
.../admin-guide/kdump/vmcoreinfo.rst | 2 +-
drivers/gpu/drm/ttm/ttm_device.c | 7 ++-
drivers/gpu/drm/ttm/ttm_pool.c | 58 +++++++++++++++++--
include/drm/ttm/ttm_pool.h | 4 +-
include/linux/mmzone.h | 2 +-
mm/kmsan/init.c | 12 +++-
mm/page_alloc.c | 32 ++++++++--
7 files changed, 98 insertions(+), 19 deletions(-)

diff --git a/Documentation/admin-guide/kdump/vmcoreinfo.rst b/Documentation/admin-guide/kdump/vmcoreinfo.rst
index c572b5230fe0..a775462aa7c7 100644
--- a/Documentation/admin-guide/kdump/vmcoreinfo.rst
+++ b/Documentation/admin-guide/kdump/vmcoreinfo.rst
@@ -172,7 +172,7 @@ variables.
Offset of the free_list's member. This value is used to compute the number
of free pages.

-Each zone has a free_area structure array called free_area[MAX_ORDER + 1].
+Each zone has a free_area structure array called free_area with length of MAX_ORDER + 1.
The free_list represents a linked list of free page blocks.

(list_head, next|prev)
diff --git a/drivers/gpu/drm/ttm/ttm_device.c b/drivers/gpu/drm/ttm/ttm_device.c
index e7147e304637..442a77bb5b4f 100644
--- a/drivers/gpu/drm/ttm/ttm_device.c
+++ b/drivers/gpu/drm/ttm/ttm_device.c
@@ -92,7 +92,9 @@ static int ttm_global_init(void)
>> PAGE_SHIFT;
num_dma32 = min(num_dma32, 2UL << (30 - PAGE_SHIFT));

- ttm_pool_mgr_init(num_pages);
+ ret = ttm_pool_mgr_init(num_pages);
+ if (ret)
+ goto out;
ttm_tt_mgr_init(num_pages, num_dma32);

glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);
@@ -218,7 +220,8 @@ int ttm_device_init(struct ttm_device *bdev, struct ttm_device_funcs *funcs,
bdev->funcs = funcs;

ttm_sys_man_init(bdev);
- ttm_pool_init(&bdev->pool, dev, use_dma_alloc, use_dma32);
+ if (ttm_pool_init(&bdev->pool, dev, use_dma_alloc, use_dma32))
+ return -ENOMEM;

bdev->vma_manager = vma_manager;
INIT_DELAYED_WORK(&bdev->wq, ttm_device_delayed_workqueue);
diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
index 85d19f425af6..d76f7d476421 100644
--- a/drivers/gpu/drm/ttm/ttm_pool.c
+++ b/drivers/gpu/drm/ttm/ttm_pool.c
@@ -64,11 +64,11 @@ module_param(page_pool_size, ulong, 0644);

static atomic_long_t allocated_pages;

-static struct ttm_pool_type global_write_combined[MAX_ORDER + 1];
-static struct ttm_pool_type global_uncached[MAX_ORDER + 1];
+static struct ttm_pool_type *global_write_combined;
+static struct ttm_pool_type *global_uncached;

-static struct ttm_pool_type global_dma32_write_combined[MAX_ORDER + 1];
-static struct ttm_pool_type global_dma32_uncached[MAX_ORDER + 1];
+static struct ttm_pool_type *global_dma32_write_combined;
+static struct ttm_pool_type *global_dma32_uncached;

static spinlock_t shrinker_lock;
static struct list_head shrinker_list;
@@ -493,8 +493,10 @@ EXPORT_SYMBOL(ttm_pool_free);
* @use_dma32: true if GFP_DMA32 should be used
*
* Initialize the pool and its pool types.
+ *
+ * Returns: 0 on successe, negative error code otherwise
*/
-void ttm_pool_init(struct ttm_pool *pool, struct device *dev,
+int ttm_pool_init(struct ttm_pool *pool, struct device *dev,
bool use_dma_alloc, bool use_dma32)
{
unsigned int i, j;
@@ -506,11 +508,30 @@ void ttm_pool_init(struct ttm_pool *pool, struct device *dev,
pool->use_dma32 = use_dma32;

if (use_dma_alloc) {
- for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i)
+ for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i) {
+ pool->caching[i].orders =
+ kvcalloc(MAX_ORDER + 1, sizeof(struct ttm_pool_type),
+ GFP_KERNEL);
+ if (!pool->caching[i].orders) {
+ i--;
+ goto failed;
+ }
for (j = 0; j <= MAX_ORDER; ++j)
ttm_pool_type_init(&pool->caching[i].orders[j],
pool, i, j);
+
+ }
+ return 0;
+
+failed:
+ for (; i >= 0; i--) {
+ for (j = 0; j <= MAX_ORDER; ++j)
+ ttm_pool_type_fini(&pool->caching[i].orders[j]);
+ kfree(pool->caching[i].orders);
+ }
+ return -ENOMEM;
}
+ return 0;
}

/**
@@ -701,6 +722,31 @@ int ttm_pool_mgr_init(unsigned long num_pages)
spin_lock_init(&shrinker_lock);
INIT_LIST_HEAD(&shrinker_list);

+ if (!global_write_combined) {
+ global_write_combined = kvcalloc(MAX_ORDER + 1, sizeof(struct ttm_pool_type),
+ GFP_KERNEL);
+ if (!global_write_combined)
+ return -ENOMEM;
+ }
+ if (!global_uncached) {
+ global_uncached = kvcalloc(MAX_ORDER + 1, sizeof(struct ttm_pool_type),
+ GFP_KERNEL);
+ if (!global_uncached)
+ return -ENOMEM;
+ }
+ if (!global_dma32_write_combined) {
+ global_dma32_write_combined = kvcalloc(MAX_ORDER + 1, sizeof(struct ttm_pool_type),
+ GFP_KERNEL);
+ if (!global_dma32_write_combined)
+ return -ENOMEM;
+ }
+ if (!global_dma32_uncached) {
+ global_dma32_uncached = kvcalloc(MAX_ORDER + 1, sizeof(struct ttm_pool_type),
+ GFP_KERNEL);
+ if (!global_dma32_uncached)
+ return -ENOMEM;
+ }
+
for (i = 0; i <= MAX_ORDER; ++i) {
ttm_pool_type_init(&global_write_combined[i], NULL,
ttm_write_combined, i);
diff --git a/include/drm/ttm/ttm_pool.h b/include/drm/ttm/ttm_pool.h
index 8ce14f9d202a..f5ce60f629ae 100644
--- a/include/drm/ttm/ttm_pool.h
+++ b/include/drm/ttm/ttm_pool.h
@@ -72,7 +72,7 @@ struct ttm_pool {
bool use_dma32;

struct {
- struct ttm_pool_type orders[MAX_ORDER + 1];
+ struct ttm_pool_type *orders;
} caching[TTM_NUM_CACHING_TYPES];
};

@@ -80,7 +80,7 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
struct ttm_operation_ctx *ctx);
void ttm_pool_free(struct ttm_pool *pool, struct ttm_tt *tt);

-void ttm_pool_init(struct ttm_pool *pool, struct device *dev,
+int ttm_pool_init(struct ttm_pool *pool, struct device *dev,
bool use_dma_alloc, bool use_dma32);
void ttm_pool_fini(struct ttm_pool *pool);

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index da5745fa15c3..032d347f36dd 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -822,7 +822,7 @@ struct zone {
CACHELINE_PADDING(_pad1_);

/* free areas of different sizes */
- struct free_area free_area[MAX_ORDER + 1];
+ struct free_area *free_area;

/* zone flags, see below */
unsigned long flags;
diff --git a/mm/kmsan/init.c b/mm/kmsan/init.c
index 6c0834274316..d3777146e7c4 100644
--- a/mm/kmsan/init.c
+++ b/mm/kmsan/init.c
@@ -96,7 +96,7 @@ void __init kmsan_init_shadow(void)
struct metadata_page_pair {
struct page *shadow, *origin;
};
-static struct metadata_page_pair held_back[MAX_ORDER + 1] __initdata;
+static struct metadata_page_pair *held_back __initdata;

/*
* Eager metadata allocation. When the memblock allocator is freeing pages to
@@ -115,6 +115,16 @@ bool kmsan_memblock_free_pages(struct page *page, unsigned int order)
{
struct page *shadow, *origin;

+ if (!held_back) {
+ held_back = memblock_alloc((MAX_ORDER + 1) * sizeof(struct metadata_page_pair),
+ sizeof(struct metadata_page_pair));
+ /* held_back cannot be allocated, kmsan will not take the page */
+ if (!held_back) {
+ WARN_ONCE(1, "held_back array cannot be allocated, kmsan will not work");
+ return true;
+ }
+ }
+
if (!held_back[order].shadow) {
held_back[order].shadow = page;
return false;
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index e3af87d89ebf..ba7c284ba3d3 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6249,13 +6249,23 @@ void __show_free_areas(unsigned int filter, nodemask_t *nodemask, int max_zone_i

for_each_populated_zone(zone) {
unsigned int order;
- unsigned long nr[MAX_ORDER + 1], flags, total = 0;
- unsigned char types[MAX_ORDER + 1];
+ unsigned long *nr, flags, total = 0;
+ unsigned char *types;

if (zone_idx(zone) > max_zone_idx)
continue;
if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
continue;
+
+ nr = kmalloc_array(MAX_ORDER + 1, sizeof(unsigned long), GFP_KERNEL);
+ if (!nr)
+ break;
+ types = kmalloc_array(MAX_ORDER + 1, sizeof(unsigned char), GFP_KERNEL);
+ if (!types) {
+ kfree(nr);
+ break;
+ }
+
show_node(zone);
printk(KERN_CONT "%s: ", zone->name);

@@ -7710,8 +7720,8 @@ static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
lruvec_init(&pgdat->__lruvec);
}

-static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
- unsigned long remaining_pages)
+static void __init zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
+ unsigned long remaining_pages, bool hotplug)
{
atomic_long_set(&zone->managed_pages, remaining_pages);
zone_set_nid(zone, nid);
@@ -7720,6 +7730,16 @@ static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx,
spin_lock_init(&zone->lock);
zone_seqlock_init(zone);
zone_pcp_init(zone);
+ if (hotplug)
+ zone->free_area =
+ kcalloc_node(MAX_ORDER + 1, sizeof(struct free_area),
+ GFP_KERNEL, nid);
+ else
+ zone->free_area =
+ memblock_alloc_node(sizeof(struct free_area) * (MAX_ORDER + 1),
+ sizeof(struct free_area), nid);
+ BUG_ON(!zone->free_area);
+
}

/*
@@ -7758,7 +7778,7 @@ void __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
}

for (z = 0; z < MAX_NR_ZONES; z++)
- zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
+ zone_init_internals(&pgdat->node_zones[z], z, nid, 0, true);
}
#endif

@@ -7821,7 +7841,7 @@ static void __init free_area_init_core(struct pglist_data *pgdat)
* when the bootmem allocator frees pages into the buddy system.
* And all highmem pages will be managed by the buddy system.
*/
- zone_init_internals(zone, j, nid, freesize);
+ zone_init_internals(zone, j, nid, freesize, false);

if (!size)
continue;
--
2.35.1