This is much the same as the original version, accept there's an initial
patch to add hotplug support for memory caching. The first version only
provided the attributes for locality and performance.
The second patch adds the hotplug support via memory notifier. The
difference from v1 is the added lock, ensuring onlining multiple regions
is single threaded to prevent duplicate hmat registration races.
Keith Busch (2):
hmat: Register memory-side cache after parsing
hmat: Register attributes for memory hot add
drivers/acpi/hmat/hmat.c | 108 +++++++++++++++++++++++++++++++++++------------
1 file changed, 82 insertions(+), 26 deletions(-)
--
2.14.4
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.
Signed-off-by: Keith Busch <[email protected]>
---
drivers/acpi/hmat/hmat.c | 48 +++++++++++++++++++++++++++++++++---------------
1 file changed, 33 insertions(+), 15 deletions(-)
diff --git a/drivers/acpi/hmat/hmat.c b/drivers/acpi/hmat/hmat.c
index b7824a0309f7..bdb167c026ff 100644
--- a/drivers/acpi/hmat/hmat.c
+++ b/drivers/acpi/hmat/hmat.c
@@ -41,6 +41,7 @@ struct memory_target {
unsigned int memory_pxm;
unsigned int processor_pxm;
struct node_hmem_attrs hmem_attrs;
+ struct node_cache_attrs cache_attrs;
};
struct memory_initiator {
@@ -314,7 +315,7 @@ 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;
u32 attrs;
if (cache->header.length < sizeof(*cache)) {
@@ -328,37 +329,40 @@ 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;
+
+ target->cache_attrs.size = cache->cache_size;
+ target->cache_attrs.level = (attrs & ACPI_HMAT_CACHE_LEVEL) >> 4;
+ target->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;
+ target->cache_attrs.indexing = NODE_CACHE_DIRECT_MAP;
break;
case ACPI_HMAT_CA_COMPLEX_CACHE_INDEXING:
- cache_attrs.indexing = NODE_CACHE_INDEXED;
+ target->cache_attrs.indexing = NODE_CACHE_INDEXED;
break;
case ACPI_HMAT_CA_NONE:
default:
- cache_attrs.indexing = NODE_CACHE_OTHER;
+ target->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;
+ target->cache_attrs.write_policy = NODE_CACHE_WRITE_BACK;
break;
case ACPI_HMAT_CP_WT:
- cache_attrs.write_policy = NODE_CACHE_WRITE_THROUGH;
+ target->cache_attrs.write_policy = NODE_CACHE_WRITE_THROUGH;
break;
case ACPI_HMAT_CP_NONE:
default:
- cache_attrs.write_policy = NODE_CACHE_WRITE_OTHER;
+ target->cache_attrs.write_policy = NODE_CACHE_WRITE_OTHER;
break;
}
- node_add_cache(pxm_to_node(cache->memory_PD), &cache_attrs);
return 0;
}
@@ -577,20 +581,34 @@ 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);
+ node_add_cache(mem_nid, &target->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)
--
2.14.4
Some memory nodes described in HMAT may not be online at the time the
we parse the subtables. 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.
Register a memory notifier callback and register the memory attributes
the first time its node is brought online if it wasn't registered,
ensuring a node's attributes may be registered only once.
Reported-by: Brice Goglin <[email protected]>
Signed-off-by: Keith Busch <[email protected]>
---
drivers/acpi/hmat/hmat.c | 72 ++++++++++++++++++++++++++++++++++++------------
1 file changed, 55 insertions(+), 17 deletions(-)
diff --git a/drivers/acpi/hmat/hmat.c b/drivers/acpi/hmat/hmat.c
index bdb167c026ff..4fcfad6c2181 100644
--- a/drivers/acpi/hmat/hmat.c
+++ b/drivers/acpi/hmat/hmat.c
@@ -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
@@ -42,6 +46,7 @@ struct memory_target {
unsigned int processor_pxm;
struct node_hmem_attrs hmem_attrs;
struct node_cache_attrs cache_attrs;
+ bool registered;
};
struct memory_initiator {
@@ -54,7 +59,7 @@ 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;
@@ -64,7 +69,7 @@ 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;
@@ -149,7 +154,7 @@ 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;
@@ -184,7 +189,7 @@ 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) {
@@ -439,7 +444,7 @@ 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)
{
@@ -477,7 +482,7 @@ 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;
@@ -521,7 +526,7 @@ 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;
@@ -581,13 +586,13 @@ 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);
node_add_cache(mem_nid, &target->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);
@@ -598,12 +603,17 @@ 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);
+ 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;
@@ -611,6 +621,30 @@ 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;
@@ -676,6 +710,10 @@ 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);
--
2.14.4
On Tue, Apr 16, 2019 at 04:55:21PM +0200, Brice Goglin wrote:
> Hello Keith
>
> Several issues:
>
> * We always get a memory_side_cache, even if nothing was found in ACPI.
> You should at least ignore the cache if size==0?
>
> * Your code seems to only work with a single level of cache, since
> there's a single cache_attrs entry in each target structure.
>
> * I was getting a section mismatch warning and a crash on PMEM node
> hotplug until I applied the patch below.
>
> WARNING: vmlinux.o(.text+0x47d3f7): Section mismatch in reference from the function hmat_callback() to the function .init.text:hmat_register_target()
> The function hmat_callback() references
> the function __init hmat_register_target().
> This is often because hmat_callback lacks a __init
> annotation or the annotation of hmat_register_target is wrong.
>
> Thanks
>
> Brice
Oh, thanks for the notice. I'll add multi-level and no-caches into my
test, as I had it fixed to one. Will need to respin this series.