* [patch 1/5] x86: reduce minimum fake node size to 32M
@ 2010-11-12 21:53 David Rientjes
2010-11-12 21:54 ` [patch 2/5] x86: avoid compiling NUMA emulation functions without CONFIG_NUMA_EMU David Rientjes
` (3 more replies)
0 siblings, 4 replies; 6+ messages in thread
From: David Rientjes @ 2010-11-12 21:53 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
This patch changes the minimum fake node size from 64MB to 32MB so it is
possible to test NUMA code at a greater scale on smaller machines
(64 nodes on a 2G machine, 1024 nodes on 32G machine with
CONFIG_NODES_SHIFT=10).
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/include/asm/numa_64.h | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/arch/x86/include/asm/numa_64.h b/arch/x86/include/asm/numa_64.h
--- a/arch/x86/include/asm/numa_64.h
+++ b/arch/x86/include/asm/numa_64.h
@@ -38,7 +38,7 @@ extern void __cpuinit numa_add_cpu(int cpu);
extern void __cpuinit numa_remove_cpu(int cpu);
#ifdef CONFIG_NUMA_EMU
-#define FAKE_NODE_MIN_SIZE ((u64)64 << 20)
+#define FAKE_NODE_MIN_SIZE ((u64)32 << 20)
#define FAKE_NODE_MIN_HASH_MASK (~(FAKE_NODE_MIN_SIZE - 1UL))
#endif /* CONFIG_NUMA_EMU */
#else
^ permalink raw reply [flat|nested] 6+ messages in thread
* [patch 2/5] x86: avoid compiling NUMA emulation functions without CONFIG_NUMA_EMU
2010-11-12 21:53 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
@ 2010-11-12 21:54 ` David Rientjes
2010-11-12 21:54 ` [patch 3/5] x86: fake apicid and pxm mappings for NUMA emulation David Rientjes
` (2 subsequent siblings)
3 siblings, 0 replies; 6+ messages in thread
From: David Rientjes @ 2010-11-12 21:54 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
Both acpi_get_nodes() and k8_get_nodes() are only necessary when
CONFIG_NUMA_EMU is enabled, so avoid compiling them when the option is
disabled.
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/include/asm/acpi.h | 5 ++++-
arch/x86/include/asm/amd_nb.h | 5 ++++-
arch/x86/mm/k8topology_64.c | 2 ++
arch/x86/mm/srat_64.c | 2 ++
4 files changed, 12 insertions(+), 2 deletions(-)
diff --git a/arch/x86/include/asm/acpi.h b/arch/x86/include/asm/acpi.h
--- a/arch/x86/include/asm/acpi.h
+++ b/arch/x86/include/asm/acpi.h
@@ -188,14 +188,17 @@ extern int acpi_numa;
extern int acpi_get_nodes(struct bootnode *physnodes);
extern int acpi_scan_nodes(unsigned long start, unsigned long end);
#define NR_NODE_MEMBLKS (MAX_NUMNODES*2)
+
+#ifdef CONFIG_NUMA_EMU
extern void acpi_fake_nodes(const struct bootnode *fake_nodes,
int num_nodes);
+#endif
#else
static inline void acpi_fake_nodes(const struct bootnode *fake_nodes,
int num_nodes)
{
}
-#endif
+#endif /* CONFIG_ACPI_NUMA */
#define acpi_unlazy_tlb(x) leave_mm(x)
diff --git a/arch/x86/include/asm/amd_nb.h b/arch/x86/include/asm/amd_nb.h
--- a/arch/x86/include/asm/amd_nb.h
+++ b/arch/x86/include/asm/amd_nb.h
@@ -9,10 +9,13 @@ struct bootnode;
extern int early_is_k8_nb(u32 value);
extern int cache_k8_northbridges(void);
extern void k8_flush_garts(void);
-extern int k8_get_nodes(struct bootnode *nodes);
extern int k8_numa_init(unsigned long start_pfn, unsigned long end_pfn);
extern int k8_scan_nodes(void);
+#ifdef CONFIG_NUMA_EMU
+extern int k8_get_nodes(struct bootnode *nodes);
+#endif
+
struct k8_northbridge_info {
u16 num;
u8 gart_supported;
diff --git a/arch/x86/mm/k8topology_64.c b/arch/x86/mm/k8topology_64.c
--- a/arch/x86/mm/k8topology_64.c
+++ b/arch/x86/mm/k8topology_64.c
@@ -69,6 +69,7 @@ static __init void early_get_boot_cpu_id(void)
early_init_lapic_mapping();
}
+#ifdef CONFIG_NUMA_EMU
int __init k8_get_nodes(struct bootnode *physnodes)
{
int i;
@@ -81,6 +82,7 @@ int __init k8_get_nodes(struct bootnode *physnodes)
}
return ret;
}
+#endif /* CONFIG_NUMA_EMU */
int __init k8_numa_init(unsigned long start_pfn, unsigned long end_pfn)
{
diff --git a/arch/x86/mm/srat_64.c b/arch/x86/mm/srat_64.c
--- a/arch/x86/mm/srat_64.c
+++ b/arch/x86/mm/srat_64.c
@@ -339,6 +339,7 @@ static int __init nodes_cover_memory(const struct bootnode *nodes)
void __init acpi_numa_arch_fixup(void) {}
+#ifdef CONFIG_NUMA_EMU
int __init acpi_get_nodes(struct bootnode *physnodes)
{
int i;
@@ -351,6 +352,7 @@ int __init acpi_get_nodes(struct bootnode *physnodes)
}
return ret;
}
+#endif /* CONFIG_NUMA_EMU */
/* Use the information discovered above to actually set up the nodes. */
int __init acpi_scan_nodes(unsigned long start, unsigned long end)
^ permalink raw reply [flat|nested] 6+ messages in thread
* [patch 3/5] x86: fake apicid and pxm mappings for NUMA emulation
2010-11-12 21:53 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
2010-11-12 21:54 ` [patch 2/5] x86: avoid compiling NUMA emulation functions without CONFIG_NUMA_EMU David Rientjes
@ 2010-11-12 21:54 ` David Rientjes
2010-11-12 21:54 ` [patch 4/5] x86: fake node-to-cpumask " David Rientjes
2010-11-12 21:54 ` [patch 5/5] x86: fix cpu to node mapping for sparse node ids David Rientjes
3 siblings, 0 replies; 6+ messages in thread
From: David Rientjes @ 2010-11-12 21:54 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
This patch adds the equivalent of acpi_fake_nodes() for AMD Northbridge
platforms. The goal is to fake the apicid-to-node mappings for NUMA
emulation so the physical topology of the machine is correctly maintained
within the kernel.
This change also fakes proximity domains for both ACPI and k8 code so the
physical distance between emulated nodes is maintained via
node_distance(). This exports the correct distances via
/sys/devices/system/node/.../distance based on the underlying topology.
A new helper function, fake_physnodes(), is introduced to correctly
invoke the correct NUMA code to fake these two mappings based on the
system type. If there is no underlying NUMA configuration, all cpus are
mapped to node 0 for local distance.
Since acpi_fake_nodes() is no longer called with CONFIG_ACPI_NUMA, it's
prototype can be removed from the header file for such a configuration.
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/include/asm/acpi.h | 5 --
arch/x86/include/asm/amd_nb.h | 1 +
arch/x86/mm/k8topology_64.c | 91 +++++++++++++++++++++++++++++++++-------
arch/x86/mm/numa_64.c | 20 +++++++++-
arch/x86/mm/srat_64.c | 2 -
5 files changed, 95 insertions(+), 24 deletions(-)
diff --git a/arch/x86/include/asm/acpi.h b/arch/x86/include/asm/acpi.h
--- a/arch/x86/include/asm/acpi.h
+++ b/arch/x86/include/asm/acpi.h
@@ -193,11 +193,6 @@ extern int acpi_scan_nodes(unsigned long start, unsigned long end);
extern void acpi_fake_nodes(const struct bootnode *fake_nodes,
int num_nodes);
#endif
-#else
-static inline void acpi_fake_nodes(const struct bootnode *fake_nodes,
- int num_nodes)
-{
-}
#endif /* CONFIG_ACPI_NUMA */
#define acpi_unlazy_tlb(x) leave_mm(x)
diff --git a/arch/x86/include/asm/amd_nb.h b/arch/x86/include/asm/amd_nb.h
--- a/arch/x86/include/asm/amd_nb.h
+++ b/arch/x86/include/asm/amd_nb.h
@@ -13,6 +13,7 @@ extern int k8_numa_init(unsigned long start_pfn, unsigned long end_pfn);
extern int k8_scan_nodes(void);
#ifdef CONFIG_NUMA_EMU
+extern void k8_fake_nodes(const struct bootnode *nodes, int nr_nodes);
extern int k8_get_nodes(struct bootnode *nodes);
#endif
diff --git a/arch/x86/mm/k8topology_64.c b/arch/x86/mm/k8topology_64.c
--- a/arch/x86/mm/k8topology_64.c
+++ b/arch/x86/mm/k8topology_64.c
@@ -27,6 +27,7 @@
#include <asm/amd_nb.h>
static struct bootnode __initdata nodes[8];
+static unsigned char __initdata nodeids[8];
static nodemask_t __initdata nodes_parsed = NODE_MASK_NONE;
static __init int find_northbridge(void)
@@ -69,21 +70,6 @@ static __init void early_get_boot_cpu_id(void)
early_init_lapic_mapping();
}
-#ifdef CONFIG_NUMA_EMU
-int __init k8_get_nodes(struct bootnode *physnodes)
-{
- int i;
- int ret = 0;
-
- for_each_node_mask(i, nodes_parsed) {
- physnodes[ret].start = nodes[i].start;
- physnodes[ret].end = nodes[i].end;
- ret++;
- }
- return ret;
-}
-#endif /* CONFIG_NUMA_EMU */
-
int __init k8_numa_init(unsigned long start_pfn, unsigned long end_pfn)
{
unsigned long start = PFN_PHYS(start_pfn);
@@ -116,7 +102,7 @@ int __init k8_numa_init(unsigned long start_pfn, unsigned long end_pfn)
base = read_pci_config(0, nb, 1, 0x40 + i*8);
limit = read_pci_config(0, nb, 1, 0x44 + i*8);
- nodeid = limit & 7;
+ nodeids[i] = nodeid = limit & 7;
if ((base & 3) == 0) {
if (i < numnodes)
pr_info("Skipping disabled node %d\n", i);
@@ -196,6 +182,79 @@ int __init k8_numa_init(unsigned long start_pfn, unsigned long end_pfn)
return 0;
}
+#ifdef CONFIG_NUMA_EMU
+static s16 fake_apicid_to_node[MAX_LOCAL_APIC] __initdata = {
+ [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
+};
+
+int __init k8_get_nodes(struct bootnode *physnodes)
+{
+ int i;
+ int ret = 0;
+
+ for_each_node_mask(i, nodes_parsed) {
+ physnodes[ret].start = nodes[i].start;
+ physnodes[ret].end = nodes[i].end;
+ ret++;
+ }
+ return ret;
+}
+
+static int __init find_node_by_addr(unsigned long addr)
+{
+ int ret = NUMA_NO_NODE;
+ int i;
+
+ for (i = 0; i < 8; i++)
+ if (addr >= nodes[i].start && addr < nodes[i].end) {
+ ret = i;
+ break;
+ }
+ return ret;
+}
+
+/*
+ * For NUMA emulation, fake proximity domain (_PXM) to node id mappings must be
+ * setup to represent the physical topology but reflect the emulated
+ * environment. For each emulated node, the real node which it appears on is
+ * found and a fake pxm to nid mapping is created which mirrors the actual
+ * locality. node_distance() then represents the correct distances between
+ * emulated nodes by using the fake acpi mappings to pxms.
+ */
+void __init k8_fake_nodes(const struct bootnode *nodes, int nr_nodes)
+{
+ unsigned int bits;
+ unsigned int cores;
+ unsigned int apicid_base = 0;
+ int i;
+
+ bits = boot_cpu_data.x86_coreid_bits;
+ cores = 1 << bits;
+ early_get_boot_cpu_id();
+ if (boot_cpu_physical_apicid > 0)
+ apicid_base = boot_cpu_physical_apicid;
+
+ for (i = 0; i < nr_nodes; i++) {
+ int index;
+ int nid;
+ int j;
+
+ nid = find_node_by_addr(nodes[i].start);
+ if (nid == NUMA_NO_NODE)
+ continue;
+
+ index = nodeids[nid] << bits;
+ if (fake_apicid_to_node[index + apicid_base] == NUMA_NO_NODE)
+ for (j = apicid_base; j < cores + apicid_base; j++)
+ fake_apicid_to_node[index + j] = i;
+#ifdef CONFIG_ACPI_NUMA
+ __acpi_map_pxm_to_node(nid, i);
+#endif
+ }
+ memcpy(apicid_to_node, fake_apicid_to_node, sizeof(apicid_to_node));
+}
+#endif /* CONFIG_NUMA_EMU */
+
int __init k8_scan_nodes(void)
{
unsigned int bits;
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -324,6 +324,24 @@ static int __init setup_physnodes(unsigned long start, unsigned long end,
return ret;
}
+static void __init fake_physnodes(int acpi, int k8, int nr_nodes)
+{
+ int i;
+
+ BUG_ON(acpi && k8);
+#ifdef CONFIG_ACPI_NUMA
+ if (acpi)
+ acpi_fake_nodes(nodes, nr_nodes);
+#endif
+#ifdef CONFIG_K8_NUMA
+ if (k8)
+ k8_fake_nodes(nodes, nr_nodes);
+#endif
+ if (!acpi && !k8)
+ for (i = 0; i < nr_cpu_ids; i++)
+ numa_set_node(i, 0);
+}
+
/*
* Setups up nid to range from addr to addr + size. If the end
* boundary is greater than max_addr, then max_addr is used instead.
@@ -595,7 +613,7 @@ static int __init numa_emulation(unsigned long start_pfn,
nodes[i].end >> PAGE_SHIFT);
setup_node_bootmem(i, nodes[i].start, nodes[i].end);
}
- acpi_fake_nodes(nodes, num_nodes);
+ fake_physnodes(acpi, k8, num_nodes);
numa_init_array();
return 0;
}
diff --git a/arch/x86/mm/srat_64.c b/arch/x86/mm/srat_64.c
--- a/arch/x86/mm/srat_64.c
+++ b/arch/x86/mm/srat_64.c
@@ -497,8 +497,6 @@ void __init acpi_fake_nodes(const struct bootnode *fake_nodes, int num_nodes)
{
int i, j;
- printk(KERN_INFO "Faking PXM affinity for fake nodes on real "
- "topology.\n");
for (i = 0; i < num_nodes; i++) {
int nid, pxm;
^ permalink raw reply [flat|nested] 6+ messages in thread
* [patch 4/5] x86: fake node-to-cpumask for NUMA emulation
2010-11-12 21:53 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
2010-11-12 21:54 ` [patch 2/5] x86: avoid compiling NUMA emulation functions without CONFIG_NUMA_EMU David Rientjes
2010-11-12 21:54 ` [patch 3/5] x86: fake apicid and pxm mappings for NUMA emulation David Rientjes
@ 2010-11-12 21:54 ` David Rientjes
2010-11-12 21:54 ` [patch 5/5] x86: fix cpu to node mapping for sparse node ids David Rientjes
3 siblings, 0 replies; 6+ messages in thread
From: David Rientjes @ 2010-11-12 21:54 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
It's necessary to fake the node-to-cpumask mapping so that an emulated
node ID returns a cpumask that includes all cpus that have affinity to
the memory it represents.
This is a little intrusive because it requires knowledge of the physical
topology of the system. setup_physnodes() gives us that information, but
since NUMA emulation ends up altering the physnodes array, it's necessary
to reset it before cpus are brought online.
Accordingly, the physnodes array is moved out of init.data and into
cpuinit.data since it will be needed on cpuup callbacks.
This works regardless of whether numa=fake is used on the command line,
or the setup of the fake node succeeds or fails. The physnodes array
always contains the physical topology of the machine if CONFIG_NUMA_EMU
is enabled and can be used to setup the correct node-to-cpumask mappings
in all cases since setup_physnodes() is called whenever the array needs
to be repopulated with the correct data.
To fake the actual mappings, numa_add_cpu() and numa_remove_cpu() are
rewritten for CONFIG_NUMA_EMU so that we first find the physical node to
which each cpu has local affinity, then iterate through all online nodes
to find the emulated nodes that have local affinity to that physical
node, and then finally map the cpu to each of those emulated nodes.
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/mm/numa_64.c | 99 +++++++++++++++++++++++++++++++++++++++----------
1 files changed, 79 insertions(+), 20 deletions(-)
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -260,7 +260,7 @@ void __init numa_init_array(void)
#ifdef CONFIG_NUMA_EMU
/* Numa emulation */
static struct bootnode nodes[MAX_NUMNODES] __initdata;
-static struct bootnode physnodes[MAX_NUMNODES] __initdata;
+static struct bootnode physnodes[MAX_NUMNODES] __cpuinitdata;
static char *cmdline __initdata;
static int __init setup_physnodes(unsigned long start, unsigned long end,
@@ -270,6 +270,7 @@ static int __init setup_physnodes(unsigned long start, unsigned long end,
int ret = 0;
int i;
+ memset(physnodes, 0, sizeof(physnodes));
#ifdef CONFIG_ACPI_NUMA
if (acpi)
nr_nodes = acpi_get_nodes(physnodes);
@@ -370,8 +371,7 @@ static int __init setup_node_range(int nid, u64 *addr, u64 size, u64 max_addr)
* Sets up nr_nodes fake nodes interleaved over physical nodes ranging from addr
* to max_addr. The return value is the number of nodes allocated.
*/
-static int __init split_nodes_interleave(u64 addr, u64 max_addr,
- int nr_phys_nodes, int nr_nodes)
+static int __init split_nodes_interleave(u64 addr, u64 max_addr, int nr_nodes)
{
nodemask_t physnode_mask = NODE_MASK_NONE;
u64 size;
@@ -402,7 +402,7 @@ static int __init split_nodes_interleave(u64 addr, u64 max_addr,
return -1;
}
- for (i = 0; i < nr_phys_nodes; i++)
+ for (i = 0; i < MAX_NUMNODES; i++)
if (physnodes[i].start != physnodes[i].end)
node_set(i, physnode_mask);
@@ -571,11 +571,9 @@ static int __init numa_emulation(unsigned long start_pfn,
{
u64 addr = start_pfn << PAGE_SHIFT;
u64 max_addr = last_pfn << PAGE_SHIFT;
- int num_phys_nodes;
int num_nodes;
int i;
- num_phys_nodes = setup_physnodes(addr, max_addr, acpi, k8);
/*
* If the numa=fake command-line contains a 'M' or 'G', it represents
* the fixed node size. Otherwise, if it is just a single number N,
@@ -590,7 +588,7 @@ static int __init numa_emulation(unsigned long start_pfn,
unsigned long n;
n = simple_strtoul(cmdline, NULL, 0);
- num_nodes = split_nodes_interleave(addr, max_addr, num_phys_nodes, n);
+ num_nodes = split_nodes_interleave(addr, max_addr, n);
}
if (num_nodes < 0)
@@ -613,6 +611,7 @@ static int __init numa_emulation(unsigned long start_pfn,
nodes[i].end >> PAGE_SHIFT);
setup_node_bootmem(i, nodes[i].start, nodes[i].end);
}
+ setup_physnodes(addr, max_addr, acpi, k8);
fake_physnodes(acpi, k8, num_nodes);
numa_init_array();
return 0;
@@ -628,8 +627,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
nodes_clear(node_online_map);
#ifdef CONFIG_NUMA_EMU
+ setup_physnodes(start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT,
+ acpi, k8);
if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8))
return;
+ setup_physnodes(start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT,
+ acpi, k8);
nodes_clear(node_possible_map);
nodes_clear(node_online_map);
#endif
@@ -785,6 +788,7 @@ void __cpuinit numa_clear_node(int cpu)
#ifndef CONFIG_DEBUG_PER_CPU_MAPS
+#ifndef CONFIG_NUMA_EMU
void __cpuinit numa_add_cpu(int cpu)
{
cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
@@ -794,6 +798,51 @@ void __cpuinit numa_remove_cpu(int cpu)
{
cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
}
+#else
+void __cpuinit numa_add_cpu(int cpu)
+{
+ unsigned long addr;
+ u16 apicid;
+ int physnid;
+ int nid = NUMA_NO_NODE;
+
+ apicid = early_per_cpu(x86_cpu_to_apicid, cpu);
+ if (apicid != BAD_APICID)
+ nid = apicid_to_node[apicid];
+ if (nid == NUMA_NO_NODE)
+ nid = early_cpu_to_node(cpu);
+ BUG_ON(nid == NUMA_NO_NODE || !node_online(nid));
+
+ /*
+ * Use the starting address of the emulated node to find which physical
+ * node it is allocated on.
+ */
+ addr = node_start_pfn(nid) << PAGE_SHIFT;
+ for (physnid = 0; physnid < MAX_NUMNODES; physnid++)
+ if (addr >= physnodes[physnid].start &&
+ addr < physnodes[physnid].end)
+ break;
+
+ /*
+ * Map the cpu to each emulated node that is allocated on the physical
+ * node of the cpu's apic id.
+ */
+ for_each_online_node(nid) {
+ addr = node_start_pfn(nid) << PAGE_SHIFT;
+ if (addr >= physnodes[physnid].start &&
+ addr < physnodes[physnid].end)
+ cpumask_set_cpu(cpu, node_to_cpumask_map[nid]);
+ }
+}
+
+void __cpuinit numa_remove_cpu(int cpu)
+{
+ int i;
+
+ for_each_online_node(i)
+ cpumask_clear_cpu(cpu, node_to_cpumask_map[i]);
+}
+#endif /* !CONFIG_NUMA_EMU */
#else /* CONFIG_DEBUG_PER_CPU_MAPS */
@@ -805,22 +854,32 @@ static void __cpuinit numa_set_cpumask(int cpu, int enable)
int node = early_cpu_to_node(cpu);
struct cpumask *mask;
char buf[64];
+ int i;
- mask = node_to_cpumask_map[node];
- if (mask == NULL) {
- printk(KERN_ERR "node_to_cpumask_map[%i] NULL\n", node);
- dump_stack();
- return;
- }
+ for_each_online_node(i) {
+ unsigned long addr;
- if (enable)
- cpumask_set_cpu(cpu, mask);
- else
- cpumask_clear_cpu(cpu, mask);
+ addr = node_start_pfn(i) << PAGE_SHIFT;
+ if (addr < physnodes[node].start ||
+ addr >= physnodes[node].end)
+ continue;
+ mask = node_to_cpumask_map[node];
+ if (mask == NULL) {
+ pr_err("node_to_cpumask_map[%i] NULL\n", i);
+ dump_stack();
+ return;
+ }
+
+ if (enable)
+ cpumask_set_cpu(cpu, mask);
+ else
+ cpumask_clear_cpu(cpu, mask);
- cpulist_scnprintf(buf, sizeof(buf), mask);
- printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
- enable ? "numa_add_cpu" : "numa_remove_cpu", cpu, node, buf);
+ cpulist_scnprintf(buf, sizeof(buf), mask);
+ printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
+ enable ? "numa_add_cpu" : "numa_remove_cpu",
+ cpu, node, buf);
+ }
}
void __cpuinit numa_add_cpu(int cpu)
^ permalink raw reply [flat|nested] 6+ messages in thread
* [patch 5/5] x86: fix cpu to node mapping for sparse node ids
2010-11-12 21:53 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
` (2 preceding siblings ...)
2010-11-12 21:54 ` [patch 4/5] x86: fake node-to-cpumask " David Rientjes
@ 2010-11-12 21:54 ` David Rientjes
3 siblings, 0 replies; 6+ messages in thread
From: David Rientjes @ 2010-11-12 21:54 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
NUMA boot code assumes that physical node ids start at 0, but the DIMMs
that the apic id represents may not be reachable. If this is the case,
node 0 is never online and cpus never end up getting appropriately
assigned to a node. This causes the cpumask of all online nodes to be
empty and machines crash with kernel code assuming online nodes have
valid cpus.
The fix is to appropriately map all the address ranges for physical nodes
and ensure the cpu to node mapping function checks all possible nodes (up
to MAX_NUMNODES) instead of simply checking nodes 0-N, where N is the
number of physical nodes, for valid address ranges.
This requires no longer "compressing" the address ranges of nodes in the
physical node map from 0-N, but rather leave indices in physnodes[] to
represent the actual node id of the physical node. Accordingly, the
topology exported by both k8_get_nodes() and acpi_get_nodes() no longer
must return the number of nodes to iterate through; all such iterations
will now be to MAX_NUMNODES.
This change also passes the end address of system RAM (which may be
different from normal operation if mem= is specified on the command line)
before the physnodes[] array is populated. ACPI parsed nodes are
truncated to fit within the address range that respect the mem=
boundaries and even some physical nodes may become unreachable in such
cases.
When NUMA emulation does succeed, any apicid to node mapping that exists
for unreachable nodes are given default values so that proximity domains
can still be assigned. This is important for node_distance() to
function as desired.
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/include/asm/acpi.h | 3 ++-
arch/x86/include/asm/amd_nb.h | 2 +-
arch/x86/mm/k8topology_64.c | 9 +++------
arch/x86/mm/numa_64.c | 18 +++---------------
arch/x86/mm/srat_64.c | 22 ++++++++++++++++------
5 files changed, 25 insertions(+), 29 deletions(-)
diff --git a/arch/x86/include/asm/acpi.h b/arch/x86/include/asm/acpi.h
--- a/arch/x86/include/asm/acpi.h
+++ b/arch/x86/include/asm/acpi.h
@@ -185,7 +185,8 @@ struct bootnode;
#ifdef CONFIG_ACPI_NUMA
extern int acpi_numa;
-extern int acpi_get_nodes(struct bootnode *physnodes);
+extern void acpi_get_nodes(struct bootnode *physnodes, unsigned long start,
+ unsigned long end);
extern int acpi_scan_nodes(unsigned long start, unsigned long end);
#define NR_NODE_MEMBLKS (MAX_NUMNODES*2)
diff --git a/arch/x86/include/asm/amd_nb.h b/arch/x86/include/asm/amd_nb.h
--- a/arch/x86/include/asm/amd_nb.h
+++ b/arch/x86/include/asm/amd_nb.h
@@ -14,7 +14,7 @@ extern int k8_scan_nodes(void);
#ifdef CONFIG_NUMA_EMU
extern void k8_fake_nodes(const struct bootnode *nodes, int nr_nodes);
-extern int k8_get_nodes(struct bootnode *nodes);
+extern void k8_get_nodes(struct bootnode *nodes);
#endif
struct k8_northbridge_info {
diff --git a/arch/x86/mm/k8topology_64.c b/arch/x86/mm/k8topology_64.c
--- a/arch/x86/mm/k8topology_64.c
+++ b/arch/x86/mm/k8topology_64.c
@@ -187,17 +187,14 @@ static s16 fake_apicid_to_node[MAX_LOCAL_APIC] __initdata = {
[0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
};
-int __init k8_get_nodes(struct bootnode *physnodes)
+void __init k8_get_nodes(struct bootnode *physnodes)
{
int i;
- int ret = 0;
for_each_node_mask(i, nodes_parsed) {
- physnodes[ret].start = nodes[i].start;
- physnodes[ret].end = nodes[i].end;
- ret++;
+ physnodes[i].start = nodes[i].start;
+ physnodes[i].end = nodes[i].end;
}
- return ret;
}
static int __init find_node_by_addr(unsigned long addr)
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -266,25 +266,24 @@ static char *cmdline __initdata;
static int __init setup_physnodes(unsigned long start, unsigned long end,
int acpi, int k8)
{
- int nr_nodes = 0;
int ret = 0;
int i;
memset(physnodes, 0, sizeof(physnodes));
#ifdef CONFIG_ACPI_NUMA
if (acpi)
- nr_nodes = acpi_get_nodes(physnodes);
+ acpi_get_nodes(physnodes, start, end);
#endif
#ifdef CONFIG_K8_NUMA
if (k8)
- nr_nodes = k8_get_nodes(physnodes);
+ k8_get_nodes(physnodes);
#endif
/*
* Basic sanity checking on the physical node map: there may be errors
* if the SRAT or K8 incorrectly reported the topology or the mem=
* kernel parameter is used.
*/
- for (i = 0; i < nr_nodes; i++) {
+ for (i = 0; i < MAX_NUMNODES; i++) {
if (physnodes[i].start == physnodes[i].end)
continue;
if (physnodes[i].start > end) {
@@ -299,17 +298,6 @@ static int __init setup_physnodes(unsigned long start, unsigned long end,
physnodes[i].start = start;
if (physnodes[i].end > end)
physnodes[i].end = end;
- }
-
- /*
- * Remove all nodes that have no memory or were truncated because of the
- * limited address range.
- */
- for (i = 0; i < nr_nodes; i++) {
- if (physnodes[i].start == physnodes[i].end)
- continue;
- physnodes[ret].start = physnodes[i].start;
- physnodes[ret].end = physnodes[i].end;
ret++;
}
diff --git a/arch/x86/mm/srat_64.c b/arch/x86/mm/srat_64.c
--- a/arch/x86/mm/srat_64.c
+++ b/arch/x86/mm/srat_64.c
@@ -340,17 +340,16 @@ static int __init nodes_cover_memory(const struct bootnode *nodes)
void __init acpi_numa_arch_fixup(void) {}
#ifdef CONFIG_NUMA_EMU
-int __init acpi_get_nodes(struct bootnode *physnodes)
+void __init acpi_get_nodes(struct bootnode *physnodes, unsigned long start,
+ unsigned long end)
{
int i;
- int ret = 0;
for_each_node_mask(i, nodes_parsed) {
- physnodes[ret].start = nodes[i].start;
- physnodes[ret].end = nodes[i].end;
- ret++;
+ cutoff_node(i, start, end);
+ physnodes[i].start = nodes[i].start;
+ physnodes[i].end = nodes[i].end;
}
- return ret;
}
#endif /* CONFIG_NUMA_EMU */
@@ -516,6 +515,17 @@ void __init acpi_fake_nodes(const struct bootnode *fake_nodes, int num_nodes)
fake_apicid_to_node[j] == NUMA_NO_NODE)
fake_apicid_to_node[j] = i;
}
+
+ /*
+ * If there are apicid-to-node mappings for physical nodes that do not
+ * have a corresponding emulated node, it should default to a guaranteed
+ * value.
+ */
+ for (i = 0; i < MAX_LOCAL_APIC; i++)
+ if (apicid_to_node[i] != NUMA_NO_NODE &&
+ fake_apicid_to_node[i] == NUMA_NO_NODE)
+ fake_apicid_to_node[i] = 0;
+
for (i = 0; i < num_nodes; i++)
__acpi_map_pxm_to_node(fake_node_to_pxm_map[i], i);
memcpy(apicid_to_node, fake_apicid_to_node, sizeof(apicid_to_node));
^ permalink raw reply [flat|nested] 6+ messages in thread
* [patch 4/5] x86: fake node-to-cpumask for NUMA emulation
2010-12-02 1:36 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
@ 2010-12-02 1:36 ` David Rientjes
0 siblings, 0 replies; 6+ messages in thread
From: David Rientjes @ 2010-12-02 1:36 UTC (permalink / raw)
To: Ingo Molnar, H. Peter Anvin, Thomas Gleixner; +Cc: x86, linux-kernel
It's necessary to fake the node-to-cpumask mapping so that an emulated
node ID returns a cpumask that includes all cpus that have affinity to
the memory it represents.
This is a little intrusive because it requires knowledge of the physical
topology of the system. setup_physnodes() gives us that information, but
since NUMA emulation ends up altering the physnodes array, it's necessary
to reset it before cpus are brought online.
Accordingly, the physnodes array is moved out of init.data and into
cpuinit.data since it will be needed on cpuup callbacks.
This works regardless of whether numa=fake is used on the command line,
or the setup of the fake node succeeds or fails. The physnodes array
always contains the physical topology of the machine if CONFIG_NUMA_EMU
is enabled and can be used to setup the correct node-to-cpumask mappings
in all cases since setup_physnodes() is called whenever the array needs
to be repopulated with the correct data.
To fake the actual mappings, numa_add_cpu() and numa_remove_cpu() are
rewritten for CONFIG_NUMA_EMU so that we first find the physical node to
which each cpu has local affinity, then iterate through all online nodes
to find the emulated nodes that have local affinity to that physical
node, and then finally map the cpu to each of those emulated nodes.
Signed-off-by: David Rientjes <rientjes@google.com>
---
arch/x86/mm/numa_64.c | 99 +++++++++++++++++++++++++++++++++++++++----------
1 files changed, 79 insertions(+), 20 deletions(-)
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -260,7 +260,7 @@ void __init numa_init_array(void)
#ifdef CONFIG_NUMA_EMU
/* Numa emulation */
static struct bootnode nodes[MAX_NUMNODES] __initdata;
-static struct bootnode physnodes[MAX_NUMNODES] __initdata;
+static struct bootnode physnodes[MAX_NUMNODES] __cpuinitdata;
static char *cmdline __initdata;
static int __init setup_physnodes(unsigned long start, unsigned long end,
@@ -270,6 +270,7 @@ static int __init setup_physnodes(unsigned long start, unsigned long end,
int ret = 0;
int i;
+ memset(physnodes, 0, sizeof(physnodes));
#ifdef CONFIG_ACPI_NUMA
if (acpi)
nr_nodes = acpi_get_nodes(physnodes);
@@ -370,8 +371,7 @@ static int __init setup_node_range(int nid, u64 *addr, u64 size, u64 max_addr)
* Sets up nr_nodes fake nodes interleaved over physical nodes ranging from addr
* to max_addr. The return value is the number of nodes allocated.
*/
-static int __init split_nodes_interleave(u64 addr, u64 max_addr,
- int nr_phys_nodes, int nr_nodes)
+static int __init split_nodes_interleave(u64 addr, u64 max_addr, int nr_nodes)
{
nodemask_t physnode_mask = NODE_MASK_NONE;
u64 size;
@@ -402,7 +402,7 @@ static int __init split_nodes_interleave(u64 addr, u64 max_addr,
return -1;
}
- for (i = 0; i < nr_phys_nodes; i++)
+ for (i = 0; i < MAX_NUMNODES; i++)
if (physnodes[i].start != physnodes[i].end)
node_set(i, physnode_mask);
@@ -571,11 +571,9 @@ static int __init numa_emulation(unsigned long start_pfn,
{
u64 addr = start_pfn << PAGE_SHIFT;
u64 max_addr = last_pfn << PAGE_SHIFT;
- int num_phys_nodes;
int num_nodes;
int i;
- num_phys_nodes = setup_physnodes(addr, max_addr, acpi, k8);
/*
* If the numa=fake command-line contains a 'M' or 'G', it represents
* the fixed node size. Otherwise, if it is just a single number N,
@@ -590,7 +588,7 @@ static int __init numa_emulation(unsigned long start_pfn,
unsigned long n;
n = simple_strtoul(cmdline, NULL, 0);
- num_nodes = split_nodes_interleave(addr, max_addr, num_phys_nodes, n);
+ num_nodes = split_nodes_interleave(addr, max_addr, n);
}
if (num_nodes < 0)
@@ -613,6 +611,7 @@ static int __init numa_emulation(unsigned long start_pfn,
nodes[i].end >> PAGE_SHIFT);
setup_node_bootmem(i, nodes[i].start, nodes[i].end);
}
+ setup_physnodes(addr, max_addr, acpi, k8);
fake_physnodes(acpi, k8, num_nodes);
numa_init_array();
return 0;
@@ -628,8 +627,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
nodes_clear(node_online_map);
#ifdef CONFIG_NUMA_EMU
+ setup_physnodes(start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT,
+ acpi, k8);
if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8))
return;
+ setup_physnodes(start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT,
+ acpi, k8);
nodes_clear(node_possible_map);
nodes_clear(node_online_map);
#endif
@@ -785,6 +788,7 @@ void __cpuinit numa_clear_node(int cpu)
#ifndef CONFIG_DEBUG_PER_CPU_MAPS
+#ifndef CONFIG_NUMA_EMU
void __cpuinit numa_add_cpu(int cpu)
{
cpumask_set_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
@@ -794,6 +798,51 @@ void __cpuinit numa_remove_cpu(int cpu)
{
cpumask_clear_cpu(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
}
+#else
+void __cpuinit numa_add_cpu(int cpu)
+{
+ unsigned long addr;
+ u16 apicid;
+ int physnid;
+ int nid = NUMA_NO_NODE;
+
+ apicid = early_per_cpu(x86_cpu_to_apicid, cpu);
+ if (apicid != BAD_APICID)
+ nid = apicid_to_node[apicid];
+ if (nid == NUMA_NO_NODE)
+ nid = early_cpu_to_node(cpu);
+ BUG_ON(nid == NUMA_NO_NODE || !node_online(nid));
+
+ /*
+ * Use the starting address of the emulated node to find which physical
+ * node it is allocated on.
+ */
+ addr = node_start_pfn(nid) << PAGE_SHIFT;
+ for (physnid = 0; physnid < MAX_NUMNODES; physnid++)
+ if (addr >= physnodes[physnid].start &&
+ addr < physnodes[physnid].end)
+ break;
+
+ /*
+ * Map the cpu to each emulated node that is allocated on the physical
+ * node of the cpu's apic id.
+ */
+ for_each_online_node(nid) {
+ addr = node_start_pfn(nid) << PAGE_SHIFT;
+ if (addr >= physnodes[physnid].start &&
+ addr < physnodes[physnid].end)
+ cpumask_set_cpu(cpu, node_to_cpumask_map[nid]);
+ }
+}
+
+void __cpuinit numa_remove_cpu(int cpu)
+{
+ int i;
+
+ for_each_online_node(i)
+ cpumask_clear_cpu(cpu, node_to_cpumask_map[i]);
+}
+#endif /* !CONFIG_NUMA_EMU */
#else /* CONFIG_DEBUG_PER_CPU_MAPS */
@@ -805,22 +854,32 @@ static void __cpuinit numa_set_cpumask(int cpu, int enable)
int node = early_cpu_to_node(cpu);
struct cpumask *mask;
char buf[64];
+ int i;
- mask = node_to_cpumask_map[node];
- if (mask == NULL) {
- printk(KERN_ERR "node_to_cpumask_map[%i] NULL\n", node);
- dump_stack();
- return;
- }
+ for_each_online_node(i) {
+ unsigned long addr;
- if (enable)
- cpumask_set_cpu(cpu, mask);
- else
- cpumask_clear_cpu(cpu, mask);
+ addr = node_start_pfn(i) << PAGE_SHIFT;
+ if (addr < physnodes[node].start ||
+ addr >= physnodes[node].end)
+ continue;
+ mask = node_to_cpumask_map[node];
+ if (mask == NULL) {
+ pr_err("node_to_cpumask_map[%i] NULL\n", i);
+ dump_stack();
+ return;
+ }
+
+ if (enable)
+ cpumask_set_cpu(cpu, mask);
+ else
+ cpumask_clear_cpu(cpu, mask);
- cpulist_scnprintf(buf, sizeof(buf), mask);
- printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
- enable ? "numa_add_cpu" : "numa_remove_cpu", cpu, node, buf);
+ cpulist_scnprintf(buf, sizeof(buf), mask);
+ printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
+ enable ? "numa_add_cpu" : "numa_remove_cpu",
+ cpu, node, buf);
+ }
}
void __cpuinit numa_add_cpu(int cpu)
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2010-12-02 1:36 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-11-12 21:53 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
2010-11-12 21:54 ` [patch 2/5] x86: avoid compiling NUMA emulation functions without CONFIG_NUMA_EMU David Rientjes
2010-11-12 21:54 ` [patch 3/5] x86: fake apicid and pxm mappings for NUMA emulation David Rientjes
2010-11-12 21:54 ` [patch 4/5] x86: fake node-to-cpumask " David Rientjes
2010-11-12 21:54 ` [patch 5/5] x86: fix cpu to node mapping for sparse node ids David Rientjes
2010-12-02 1:36 [patch 1/5] x86: reduce minimum fake node size to 32M David Rientjes
2010-12-02 1:36 ` [patch 4/5] x86: fake node-to-cpumask for NUMA emulation David Rientjes
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).