Skip to contentSkip to conversation

HMAT node online fixes

PATCH In review Keith Busch 4 participants 8 messages Aug 5, 2019 – Feb 12, 2020 ACPI

lore
Share & tools
lore
#0 lore raw

cover HMAT node online fixes

From: Keith Busch <[email protected]>

Hi Rafael,

These are just some fixes from a while ago to work correctly with memory node onlining, but haven't been merged in yet. I've included a fix from Dan, but had to modify it slightly for conflicts. I think it makes most sense for this to go through the acpi tree, but please let me know if you think this should go through a different route.

Thanks!

Dan Williams (1):
acpi/hmat: Skip publishing target info for nodes with no online memory

Keith Busch (2):
hmat: Register memory-side cache after parsing
hmat: Register attributes for memory hot add

 drivers/acpi/hmat/hmat.c | 143 +++++++++++++++++++++++++++++++++++++----------
 1 file changed, 114 insertions(+), 29 deletions(-)
Signature · 1 line
2.14.4
#1 lore raw

PATCH 1/3 hmat: Register memory-side cache after parsing

Instead of registering the hmat cache attributes in line with parsing the table, save the attributes in the memory target and register them after parsing completes. This will make it easier to register the attributes later when hot add is supported.

---
 drivers/acpi/hmat/hmat.c | 70 +++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 55 insertions(+), 15 deletions(-)
drivers/acpi/hmat/hmat.c +55 -15
index 96b7d39a97c6..bf23c9a27958 100644
@@ -36,11 +36,17 @@ enum locality_types {
 
 static struct memory_locality *localities_types[4];
 
+struct target_cache {
+	struct list_head node;
+	struct node_cache_attrs cache_attrs;
+};
+
 struct memory_target {
 	struct list_head node;
 	unsigned int memory_pxm;
 	unsigned int processor_pxm;
 	struct node_hmem_attrs hmem_attrs;
+	struct list_head caches;
 };
 
 struct memory_initiator {
@@ -110,6 +116,7 @@ static __init void alloc_memory_target(unsigned int mem_pxm)
 	target->memory_pxm = mem_pxm;
 	target->processor_pxm = PXM_INVAL;
 	list_add_tail(&target->node, &targets);
+	INIT_LIST_HEAD(&target->caches);
 }
 
 static __init const char *hmat_data_type(u8 type)
@@ -314,7 +321,8 @@ static __init int hmat_parse_cache(union acpi_subtable_headers *header,
 				   const unsigned long end)
 {
 	struct acpi_hmat_cache *cache = (void *)header;
-	struct node_cache_attrs cache_attrs;
+	struct memory_target *target;
+	struct target_cache *tcache;
 	u32 attrs;
 
 	if (cache->header.length < sizeof(*cache)) {
@@ -328,37 +336,47 @@ static __init int hmat_parse_cache(union acpi_subtable_headers *header,
 		cache->memory_PD, cache->cache_size, attrs,
 		cache->number_of_SMBIOShandles);
 
-	cache_attrs.size = cache->cache_size;
-	cache_attrs.level = (attrs & ACPI_HMAT_CACHE_LEVEL) >> 4;
-	cache_attrs.line_size = (attrs & ACPI_HMAT_CACHE_LINE_SIZE) >> 16;
+	target = find_mem_target(cache->memory_PD);
+	if (!target)
+		return 0;
+
+	tcache = kzalloc(sizeof(*tcache), GFP_KERNEL);
+	if (!tcache) {
+		pr_notice_once("Failed to allocate HMAT cache info\n");
+		return 0;
+	}
+
+	tcache->cache_attrs.size = cache->cache_size;
+	tcache->cache_attrs.level = (attrs & ACPI_HMAT_CACHE_LEVEL) >> 4;
+	tcache->cache_attrs.line_size = (attrs & ACPI_HMAT_CACHE_LINE_SIZE) >> 16;
 
 	switch ((attrs & ACPI_HMAT_CACHE_ASSOCIATIVITY) >> 8) {
 	case ACPI_HMAT_CA_DIRECT_MAPPED:
-		cache_attrs.indexing = NODE_CACHE_DIRECT_MAP;
+		tcache->cache_attrs.indexing = NODE_CACHE_DIRECT_MAP;
 		break;
 	case ACPI_HMAT_CA_COMPLEX_CACHE_INDEXING:
-		cache_attrs.indexing = NODE_CACHE_INDEXED;
+		tcache->cache_attrs.indexing = NODE_CACHE_INDEXED;
 		break;
 	case ACPI_HMAT_CA_NONE:
 	default:
-		cache_attrs.indexing = NODE_CACHE_OTHER;
+		tcache->cache_attrs.indexing = NODE_CACHE_OTHER;
 		break;
 	}
 
 	switch ((attrs & ACPI_HMAT_WRITE_POLICY) >> 12) {
 	case ACPI_HMAT_CP_WB:
-		cache_attrs.write_policy = NODE_CACHE_WRITE_BACK;
+		tcache->cache_attrs.write_policy = NODE_CACHE_WRITE_BACK;
 		break;
 	case ACPI_HMAT_CP_WT:
-		cache_attrs.write_policy = NODE_CACHE_WRITE_THROUGH;
+		tcache->cache_attrs.write_policy = NODE_CACHE_WRITE_THROUGH;
 		break;
 	case ACPI_HMAT_CP_NONE:
 	default:
-		cache_attrs.write_policy = NODE_CACHE_WRITE_OTHER;
+		tcache->cache_attrs.write_policy = NODE_CACHE_WRITE_OTHER;
 		break;
 	}
+	list_add_tail(&tcache->node, &target->caches);
 
-	node_add_cache(pxm_to_node(cache->memory_PD), &cache_attrs);
 	return 0;
 }
 
@@ -577,20 +595,37 @@ static __init void hmat_register_target_initiators(struct memory_target *target)
 	}
 }
 
+static __init void hmat_register_target_cache(struct memory_target *target)
+{
+	unsigned mem_nid = pxm_to_node(target->memory_pxm);
+	struct target_cache *tcache;
+
+	list_for_each_entry(tcache, &target->caches, node)
+		node_add_cache(mem_nid, &tcache->cache_attrs);
+}
+
 static __init void hmat_register_target_perf(struct memory_target *target)
 {
 	unsigned mem_nid = pxm_to_node(target->memory_pxm);
 	node_set_perf_attrs(mem_nid, &target->hmem_attrs, 0);
 }
 
+static __init void hmat_register_target(struct memory_target *target)
+{
+	if (!node_online(pxm_to_node(target->memory_pxm)))
+		return;
+
+	hmat_register_target_initiators(target);
+	hmat_register_target_cache(target);
+	hmat_register_target_perf(target);
+}
+
 static __init void hmat_register_targets(void)
 {
 	struct memory_target *target;
 
-	list_for_each_entry(target, &targets, node) {
-		hmat_register_target_initiators(target);
-		hmat_register_target_perf(target);
-	}
+	list_for_each_entry(target, &targets, node)
+		hmat_register_target(target);
 }
 
 static __init void hmat_free_structures(void)
@@ -598,8 +633,13 @@ static __init void hmat_free_structures(void)
 	struct memory_target *target, *tnext;
 	struct memory_locality *loc, *lnext;
 	struct memory_initiator *initiator, *inext;
+	struct target_cache *tcache, *cnext;
 
 	list_for_each_entry_safe(target, tnext, &targets, node) {
+		list_for_each_entry_safe(tcache, cnext, &target->caches, node) {
+			list_del(&tcache->node);
+			kfree(tcache);
+		}
 		list_del(&target->node);
 		kfree(target);
 	}
Signature · 1 line
2.14.4
#2 lore raw

PATCH 2/3 hmat: Register attributes for memory hot add

In reply to Keith Busch

Some of the memory nodes described in HMAT may not be online at the time the hmat subsystem parses their nodes' attributes. Should the node be set to online later, as can happen when using PMEM as RAM after boot, the nodes will be missing their initiator links and performance attributes.

Regsiter a memory notifier callback and register the memory attributes the first time its node is brought online if it wasn't registered.

---
 drivers/acpi/hmat/hmat.c | 75 ++++++++++++++++++++++++++++++++++++------------
 1 file changed, 57 insertions(+), 18 deletions(-)
drivers/acpi/hmat/hmat.c +57 -18
index bf23c9a27958..f86fe7130736 100644
@@ -14,14 +14,18 @@
 #include <linux/init.h>
 #include <linux/list.h>
 #include <linux/list_sort.h>
+#include <linux/memory.h>
+#include <linux/mutex.h>
 #include <linux/node.h>
 #include <linux/sysfs.h>
 
-static __initdata u8 hmat_revision;
+static u8 hmat_revision;
 
-static __initdata LIST_HEAD(targets);
-static __initdata LIST_HEAD(initiators);
-static __initdata LIST_HEAD(localities);
+static LIST_HEAD(targets);
+static LIST_HEAD(initiators);
+static LIST_HEAD(localities);
+
+static DEFINE_MUTEX(target_lock);
 
 /*
  * The defined enum order is used to prioritize attributes to break ties when
@@ -47,6 +51,8 @@ struct memory_target {
 	unsigned int processor_pxm;
 	struct node_hmem_attrs hmem_attrs;
 	struct list_head caches;
+	struct node_cache_attrs cache_attrs;
+	bool registered;
 };
 
 struct memory_initiator {
@@ -59,7 +65,7 @@ struct memory_locality {
 	struct acpi_hmat_locality *hmat_loc;
 };
 
-static __init struct memory_initiator *find_mem_initiator(unsigned int cpu_pxm)
+static struct memory_initiator *find_mem_initiator(unsigned int cpu_pxm)
 {
 	struct memory_initiator *initiator;
 
@@ -69,7 +75,7 @@ static __init struct memory_initiator *find_mem_initiator(unsigned int cpu_pxm)
 	return NULL;
 }
 
-static __init struct memory_target *find_mem_target(unsigned int mem_pxm)
+static struct memory_target *find_mem_target(unsigned int mem_pxm)
 {
 	struct memory_target *target;
 
@@ -155,7 +161,7 @@ static __init const char *hmat_data_type_suffix(u8 type)
 	}
 }
 
-static __init u32 hmat_normalize(u16 entry, u64 base, u8 type)
+static u32 hmat_normalize(u16 entry, u64 base, u8 type)
 {
 	u32 value;
 
@@ -190,7 +196,7 @@ static __init u32 hmat_normalize(u16 entry, u64 base, u8 type)
 	return value;
 }
 
-static __init void hmat_update_target_access(struct memory_target *target,
+static void hmat_update_target_access(struct memory_target *target,
 					     u8 type, u32 value)
 {
 	switch (type) {
@@ -453,7 +459,7 @@ static __init int srat_parse_mem_affinity(union acpi_subtable_headers *header,
 	return 0;
 }
 
-static __init u32 hmat_initiator_perf(struct memory_target *target,
+static u32 hmat_initiator_perf(struct memory_target *target,
 			       struct memory_initiator *initiator,
 			       struct acpi_hmat_locality *hmat_loc)
 {
@@ -491,7 +497,7 @@ static __init u32 hmat_initiator_perf(struct memory_target *target,
 			      hmat_loc->data_type);
 }
 
-static __init bool hmat_update_best(u8 type, u32 value, u32 *best)
+static bool hmat_update_best(u8 type, u32 value, u32 *best)
 {
 	bool updated = false;
 
@@ -535,7 +541,7 @@ static int initiator_cmp(void *priv, struct list_head *a, struct list_head *b)
 	return ia->processor_pxm - ib->processor_pxm;
 }
 
-static __init void hmat_register_target_initiators(struct memory_target *target)
+static void hmat_register_target_initiators(struct memory_target *target)
 {
 	static DECLARE_BITMAP(p_nodes, MAX_NUMNODES);
 	struct memory_initiator *initiator;
@@ -595,7 +601,7 @@ static __init void hmat_register_target_initiators(struct memory_target *target)
 	}
 }
 
-static __init void hmat_register_target_cache(struct memory_target *target)
+static void hmat_register_target_cache(struct memory_target *target)
 {
 	unsigned mem_nid = pxm_to_node(target->memory_pxm);
 	struct target_cache *tcache;
@@ -604,23 +610,28 @@ static __init void hmat_register_target_cache(struct memory_target *target)
 		node_add_cache(mem_nid, &tcache->cache_attrs);
 }
 
-static __init void hmat_register_target_perf(struct memory_target *target)
+static void hmat_register_target_perf(struct memory_target *target)
 {
 	unsigned mem_nid = pxm_to_node(target->memory_pxm);
 	node_set_perf_attrs(mem_nid, &target->hmem_attrs, 0);
 }
 
-static __init void hmat_register_target(struct memory_target *target)
+static void hmat_register_target(struct memory_target *target)
 {
 	if (!node_online(pxm_to_node(target->memory_pxm)))
 		return;
 
-	hmat_register_target_initiators(target);
-	hmat_register_target_cache(target);
-	hmat_register_target_perf(target);
+	mutex_lock(&target_lock);
+	if (!target->registered) {
+		hmat_register_target_initiators(target);
+		hmat_register_target_cache(target);
+		hmat_register_target_perf(target);
+		target->registered = true;
+	}
+	mutex_unlock(&target_lock);
 }
 
-static __init void hmat_register_targets(void)
+static void hmat_register_targets(void)
 {
 	struct memory_target *target;
 
@@ -628,6 +639,30 @@ static __init void hmat_register_targets(void)
 		hmat_register_target(target);
 }
 
+static int hmat_callback(struct notifier_block *self,
+			 unsigned long action, void *arg)
+{
+	struct memory_target *target;
+	struct memory_notify *mnb = arg;
+	int pxm, nid = mnb->status_change_nid;
+
+	if (nid == NUMA_NO_NODE || action != MEM_ONLINE)
+		return NOTIFY_OK;
+
+	pxm = node_to_pxm(nid);
+	target = find_mem_target(pxm);
+	if (!target)
+		return NOTIFY_OK;
+
+	hmat_register_target(target);
+	return NOTIFY_OK;
+}
+
+static struct notifier_block hmat_callback_nb = {
+	.notifier_call = hmat_callback,
+	.priority = 2,
+};
+
 static __init void hmat_free_structures(void)
 {
 	struct memory_target *target, *tnext;
@@ -698,6 +733,10 @@ static __init int hmat_init(void)
 		}
 	}
 	hmat_register_targets();
+
+	/* Keep the table and structures if the notifier may use them */
+	if (!register_hotmemory_notifier(&hmat_callback_nb))
+		return 0;
 out_put:
 	hmat_free_structures();
 	acpi_put_table(tbl);
Signature · 1 line
2.14.4
#3 lore raw

PATCH 3/3 acpi/hmat: Skip publishing target info for nodes with no online memory

In reply to Keith Busch

From: Dan Williams <[email protected]>

There are multiple scenarios where the HMAT may contain information about proximity domains that are not currently online. Rather than fail to report any HMAT data just elide those offline domains.

If and when those domains are later onlined they can be added to the HMEM reporting at that point.

This was found while testing EFI_MEMORY_SP support which reserves "specific purpose" memory from the general allocation pool. If that reservation results in an empty numa-node then the node is not marked online leading a spurious:

"acpi/hmat: Ignoring HMAT: Invalid table"

...result for HMAT parsing.

---
 drivers/acpi/hmat/hmat.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
drivers/acpi/hmat/hmat.c +10 -4
index f86fe7130736..8f9a28a870b0 100644
@@ -108,9 +108,6 @@ static __init void alloc_memory_target(unsigned int mem_pxm)
 {
 	struct memory_target *target;
 
-	if (pxm_to_node(mem_pxm) == NUMA_NO_NODE)
-		return;
-
 	target = find_mem_target(mem_pxm);
 	if (target)
 		return;
@@ -618,7 +615,16 @@ static void hmat_register_target_perf(struct memory_target *target)
 
 static void hmat_register_target(struct memory_target *target)
 {
-	if (!node_online(pxm_to_node(target->memory_pxm)))
+	int nid = pxm_to_node(target->memory_pxm);
+
+	/*
+	 * Skip offline nodes. This can happen when memory
+	 * marked EFI_MEMORY_SP, "specific purpose", is applied
+	 * to all the memory in a promixity domain leading to
+	 * the node being marked offline / unplugged, or if
+	 * memory-only "hotplug" node is offline.
+	 */
+	if (nid == NUMA_NO_NODE || !node_online(nid))
 		return;
 
 	mutex_lock(&target_lock);
Signature · 1 line
2.14.4
#4 lore raw

Re: [PATCH 3/3] acpi/hmat: Skip publishing target info for nodes with no online memory

Quoted 15 lines · Keith Busch

From: Dan Williams <[email protected]>

There are multiple scenarios where the HMAT may contain information about proximity domains that are not currently online. Rather than fail to report any HMAT data just elide those offline domains.

If and when those domains are later onlined they can be added to the HMEM reporting at that point.

This was found while testing EFI_MEMORY_SP support which reserves "specific purpose" memory from the general allocation pool. If that reservation results in an empty numa-node then the node is not marked online leading a spurious:

Quoted 8 lines

"acpi/hmat: Ignoring HMAT: Invalid table"

...result for HMAT parsing.

When you send somebody else's patches, you should sign them off as a rule, but since you sent this one with your own R-by, I converted that to a S-o-b.

#5 lore raw

Re: [PATCH 3/3] acpi/hmat: Skip publishing target info for nodes with no online memory

Quoted 20 lines · CEST Rafael J. Wysocki
Quoted 19 lines · Keith Busch

From: Dan Williams <[email protected]>

There are multiple scenarios where the HMAT may contain information about proximity domains that are not currently online. Rather than fail to report any HMAT data just elide those offline domains.

If and when those domains are later onlined they can be added to the HMEM reporting at that point.

This was found while testing EFI_MEMORY_SP support which reserves "specific purpose" memory from the general allocation pool. If that reservation results in an empty numa-node then the node is not marked online leading a spurious:

"acpi/hmat: Ignoring HMAT: Invalid table"

...result for HMAT parsing.

Quoted 9 lines
Quoted 4 lines

When you send somebody else's patches, you should sign them off as a rule, but since you sent this one with your own R-by, I converted that to a S-o-b.

And all patches in the series have been applied.

Thanks!

#6 lore raw

Re: [PATCH 3/3] acpi/hmat: Skip publishing target info for nodes with no online memory

Quoted 27 lines · Rafael J. Wysocki
Quoted 25 lines · CEST Rafael J. Wysocki
Quoted 23 lines · Keith Busch

From: Dan Williams <[email protected]>

There are multiple scenarios where the HMAT may contain information about proximity domains that are not currently online. Rather than fail to report any HMAT data just elide those offline domains.

If and when those domains are later onlined they can be added to the HMEM reporting at that point.

This was found while testing EFI_MEMORY_SP support which reserves "specific purpose" memory from the general allocation pool. If that reservation results in an empty numa-node then the node is not marked online leading a spurious:

"acpi/hmat: Ignoring HMAT: Invalid table"

...result for HMAT parsing.

Quoted 6 lines
Quoted 4 lines

When you send somebody else's patches, you should sign them off as a rule, but since you sent this one with your own R-by, I converted that to a S-o-b.

And all patches in the series have been applied.

I want to flag this patch (commit 5c7ed4385424 "HMAT: Skip publishing target info for nodes with no online memory")
for -stable to cleanup a spurious WARN_ON:

WARNING: CPU: 7 PID: 1 at drivers/base/node.c:191 node_set_perf_attrs+0x90/0xa0
CPU: 7 PID: 1 Comm: swapper/0 Not tainted 5.3.6-100.fc29.x86_64 #1
RIP: 0010:node_set_perf_attrs+0x90/0xa0
Call Trace:
 ? do_early_param+0x8e/0x8e
 hmat_init+0x2ff/0x443
 ? hmat_parse_subtable+0x55a/0x55a
 ? do_early_param+0x8e/0x8e
 do_one_initcall+0x46/0x1f4

Do you mind if I forward to stable@, or do you collect ACPI patches to send to stable@?

#7 lore raw

Re: [PATCH 3/3] acpi/hmat: Skip publishing target info for nodes with no online memory

Quoted 50 lines · Dan Williams
Quoted 27 lines · Rafael J. Wysocki
Quoted 25 lines · CEST Rafael J. Wysocki
Quoted 23 lines · Keith Busch

From: Dan Williams <[email protected]>

There are multiple scenarios where the HMAT may contain information about proximity domains that are not currently online. Rather than fail to report any HMAT data just elide those offline domains.

If and when those domains are later onlined they can be added to the HMEM reporting at that point.

This was found while testing EFI_MEMORY_SP support which reserves "specific purpose" memory from the general allocation pool. If that reservation results in an empty numa-node then the node is not marked online leading a spurious:

"acpi/hmat: Ignoring HMAT: Invalid table"

...result for HMAT parsing.

Quoted 6 lines
Quoted 4 lines

When you send somebody else's patches, you should sign them off as a rule, but since you sent this one with your own R-by, I converted that to a S-o-b.

And all patches in the series have been applied.

I want to flag this patch (commit 5c7ed4385424 "HMAT: Skip publishing target info for nodes with no online memory")
for -stable to cleanup a spurious WARN_ON:

WARNING: CPU: 7 PID: 1 at drivers/base/node.c:191 node_set_perf_attrs+0x90/0xa0
CPU: 7 PID: 1 Comm: swapper/0 Not tainted 5.3.6-100.fc29.x86_64 #1
RIP: 0010:node_set_perf_attrs+0x90/0xa0
Call Trace:
 ? do_early_param+0x8e/0x8e
 hmat_init+0x2ff/0x443
 ? hmat_parse_subtable+0x55a/0x55a
 ? do_early_param+0x8e/0x8e
 do_one_initcall+0x46/0x1f4
Quoted 2 lines

Do you mind if I forward to stable@, or do you collect ACPI patches to send to stable@?

Please forward it, thanks!