Re: [PATCH v7 0/5] Make cpuid <-> nodeid mapping persistent
Hi, On 05/19/2016 10:46 PM, Peter Zijlstra wrote: On Thu, May 19, 2016 at 06:39:41PM +0800, Zhu Guihua wrote: [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So why are you not fixing up wq_numa_possible_cpumask instead? That seems the far easier solution. We tried to do that. You can see our patch at http://www.gossamer-threads.com/lists/linux/kernel/2116748 But maintainer thought, we should establish persistent cpuid<->nodeid relationship, there is no need to change the map. Cc TJ, Could we return to workqueue to fix this? Thanks, Zhu Do all the other archs that support NUMA and HOTPLUG have the mapping stable, or will you now go fix each and every one of them? .
[PATCH v7 4/5] x86, acpi, cpu-hotplug: Enable MADT APIs to return disabled apicid.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 3. There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm(persistent) 2. apicid (physical cpu id) <-> nodeid (persistent) 3. cpuid (logical cpu id) <-> apicid (not persistent, now persistent by step 2) 4. cpuid (logical cpu id) <-> nodeid (not persistent) So, in order to setup persistent cpuid <-> nodeid mapping for all possible CPUs, we should: 1. Setup cpuid <-> apicid mapping for all possible CPUs, which has been done in step 1, 2. 2. Setup cpuid <-> nodeid mapping for all possible CPUs. But before that, we should obtain all apicids from MADT. All processors' apicids can be obtained by _MAT method or from MADT in ACPI. The current code ignores disabled processors and returns -ENODEV. After this patch, a new parameter will be added to MADT APIs so that caller is able to control if disabled processors are ignored. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- drivers/acpi/acpi_processor.c | 5 +++- drivers/acpi/processor_core.c | 57 +++ 2 files changed, 40 insertions(+), 22 deletions(-) diff --git a/drivers/acpi/acpi_processor.c b/drivers/acpi/acpi_processor.c index 0d92d0f..9aa2dd1 100644 --- a/drivers/acpi/acpi_processor.c +++ b/drivers/acpi/acpi_processor.c @@ -300,8 +300,11 @@ static int acpi_processor_get_info(struct acpi_device *device) * Extra Processor objects may be enumerated on MP systems with * less than the max # of CPUs. They should be ignored _iff * they are physically not present. +* +* NOTE: Even if the processor has a cpuid, it may not present because +* cpuid <-> apicid mapping is persistent now. */ - if (invalid_logical_cpuid(pr->id)) { + if (invalid_logical_cpuid(pr->id) || !cpu_present(pr->id)) { int ret = acpi_processor_hotadd_init(pr); if (ret) return ret; diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 33a38d6..824b98b 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -32,12 +32,12 @@ static struct acpi_table_madt *get_madt_table(void) } static int map_lapic_id(struct acpi_subtable_header *entry, -u32 acpi_id, phys_cpuid_t *apic_id) +u32 acpi_id, phys_cpuid_t *apic_id, bool ignore_disabled) { struct acpi_madt_local_apic *lapic = container_of(entry, struct acpi_madt_local_apic, header); - if (!(lapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (lapic->processor_id != acpi_id) @@ -48,12 +48,13 @@ static int map_lapic_id(struct acpi_subtable_header *entry, } static int map_x2apic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_x2apic *apic = container_of(entry, struct acpi_madt_local_x2apic, header); - if (!(apic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(apic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration && (apic->uid == acpi_id)) { @@ -65,12 +66,13 @@ static int map_x2apic_id(struct acpi_subtable_header *entry, } static int map_lsapic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_sapic *lsapic = container_of(entry, struct acpi_madt_local_sapic, header); - if (!(lsapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lsapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration) { @@ -87,12 +89,13 @@ static int map_lsapic_id(struct acpi_subtable_header *entry, * Retrieve the ARM CPU physical identifie
[PATCH v7 5/5] x86, acpi, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 4. This patch set the persistent cpuid <-> nodeid mapping for all enabled/disabled processors at boot time via an additional acpi namespace walk for processors. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/ia64/kernel/acpi.c | 3 +- arch/x86/kernel/acpi/boot.c | 4 ++- drivers/acpi/acpi_processor.c | 5 drivers/acpi/bus.c| 3 ++ drivers/acpi/processor_core.c | 65 +++ include/linux/acpi.h | 2 ++ 6 files changed, 80 insertions(+), 2 deletions(-) diff --git a/arch/ia64/kernel/acpi.c b/arch/ia64/kernel/acpi.c index b1698bc..bb36515 100644 --- a/arch/ia64/kernel/acpi.c +++ b/arch/ia64/kernel/acpi.c @@ -796,7 +796,7 @@ int acpi_isa_irq_to_gsi(unsigned isa_irq, u32 *gsi) * ACPI based hotplug CPU support */ #ifdef CONFIG_ACPI_HOTPLUG_CPU -static int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA /* @@ -811,6 +811,7 @@ static int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) #endif return 0; } +EXPORT_SYMBOL(acpi_map_cpu2node); int additional_cpus __initdata = -1; diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index 37248c3..0900264f 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -695,7 +695,7 @@ static void __init acpi_set_irq_model_ioapic(void) #ifdef CONFIG_ACPI_HOTPLUG_CPU #include -static void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA int nid; @@ -706,7 +706,9 @@ static void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) numa_set_node(cpu, nid); } #endif + return 0; } +EXPORT_SYMBOL(acpi_map_cpu2node); int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, int *pcpu) { diff --git a/drivers/acpi/acpi_processor.c b/drivers/acpi/acpi_processor.c index 9aa2dd1..b20f417 100644 --- a/drivers/acpi/acpi_processor.c +++ b/drivers/acpi/acpi_processor.c @@ -182,6 +182,11 @@ int __weak arch_register_cpu(int cpu) void __weak arch_unregister_cpu(int cpu) {} +int __weak acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +{ + return -ENODEV; +} + static int acpi_processor_hotadd_init(struct acpi_processor *pr) { unsigned long long sta; diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c index 31e8da6..02f5721 100644 --- a/drivers/acpi/bus.c +++ b/drivers/acpi/bus.c @@ -1124,6 +1124,9 @@ static int __init acpi_init(void) acpi_sleep_proc_init(); acpi_wakeup_device_init(); acpi_debugger_init(); +#ifdef CONFIG_ACPI_HOTPLUG_CPU + acpi_set_processor_mapping(); +#endif return 0; } diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 824b98b..69fb027 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -261,6 +261,71 @@ int acpi_get_cpuid(acpi_handle handle, int type, u32 acpi_id) } EXPORT_SYMBOL_GPL(acpi_get_cpuid); +#ifdef CONFIG_ACPI_HOTPLUG_CPU +static bool map_processor(acpi_handle handle, phys_cpuid_t *phys_id, int *cpuid) +{ + int type; + u32 acpi_id; + acpi_status status; + acpi_object_type acpi_type; + unsigned long long tmp; + union acpi_object object = { 0 }; + struct acpi_buffer buffer = { sizeof(union acpi_object), &object }; + + status = acpi_get_type(handle, &acpi_type); + if (ACPI_FAILURE(status)) + return false; + + switch (acpi_type) { + case ACPI_TYPE_PROCESSOR: + status = acpi_evaluate_object(handle, NULL, NULL, &buffer); + if (ACPI_FAILURE(status)) + return false; + acpi_id = object.processor.proc_id; + break; + case ACPI_TYPE_DEVICE: + status = acpi_evaluate_integer(handle, "_UID", NULL, &tmp); + if (ACPI_FAILURE(status)) + return false; + acpi_id = tmp; + break; + default: + return false; + } + + type = (acpi_type == ACPI_TYPE_DEVICE) ? 1 : 0; + + *phys_id = __acpi_get_phys_id(handle, ty
[PATCH v7 3/5] x86, acpi, cpu-hotplug: Introduce cpuid_to_apicid[] array to store persistent cpuid <-> apicid mapping.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 2. In this patch, we introduce a new static array named cpuid_to_apicid[], which is large enough to store info for all possible cpus. And then, we modify the cpuid calculation. In generic_processor_info(), it simply finds the next unused cpuid. And it is also why the cpuid <-> nodeid mapping changes with node hotplug. After this patch, we find the next unused cpuid, map it to an apicid, and store the mapping in cpuid_to_apicid[], so that cpuid <-> apicid mapping will be persistent. And finally we will use this array to make cpuid <-> nodeid persistent. cpuid <-> apicid mapping is established at local apic registeration time. But non-present or disabled cpus are ignored. In this patch, we establish all possible cpuid <-> apicid mapping when registering local apic. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/include/asm/mpspec.h | 1 + arch/x86/kernel/acpi/boot.c | 6 ++--- arch/x86/kernel/apic/apic.c | 61 --- 3 files changed, 61 insertions(+), 7 deletions(-) diff --git a/arch/x86/include/asm/mpspec.h b/arch/x86/include/asm/mpspec.h index b07233b..db902d8 100644 --- a/arch/x86/include/asm/mpspec.h +++ b/arch/x86/include/asm/mpspec.h @@ -86,6 +86,7 @@ static inline void early_reserve_e820_mpc_new(void) { } #endif int generic_processor_info(int apicid, int version); +int __generic_processor_info(int apicid, int version, bool enabled); #define PHYSID_ARRAY_SIZE BITS_TO_LONGS(MAX_LOCAL_APIC) diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index 9414f84..37248c3 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -174,15 +174,13 @@ static int acpi_register_lapic(int id, u8 enabled) return -EINVAL; } - if (!enabled) { + if (!enabled) ++disabled_cpus; - return -EINVAL; - } if (boot_cpu_physical_apicid != -1U) ver = apic_version[boot_cpu_physical_apicid]; - return generic_processor_info(id, ver); + return __generic_processor_info(id, ver, enabled); } static int __init diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index 8e3c377..366fbbc 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -1998,7 +1998,53 @@ void disconnect_bsp_APIC(int virt_wire_setup) apic_write(APIC_LVT1, value); } -static int __generic_processor_info(int apicid, int version, bool enabled) +/* + * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated + * contiguously, it equals to current allocated max logical CPU ID plus 1. + * All allocated CPU ID should be in [0, nr_logical_cpuidi), so the maximum of + * nr_logical_cpuids is nr_cpu_ids. + * + * NOTE: Reserve 0 for BSP. + */ +static int nr_logical_cpuids = 1; + +/* + * Used to store mapping between logical CPU IDs and APIC IDs. + */ +static int cpuid_to_apicid[] = { + [0 ... NR_CPUS - 1] = -1, +}; + +/* + * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids + * and cpuid_to_apicid[] synchronized. + */ +static int allocate_logical_cpuid(int apicid) +{ + int i; + + /* +* cpuid <-> apicid mapping is persistent, so when a cpu is up, +* check if the kernel has allocated a cpuid for it. +*/ + for (i = 0; i < nr_logical_cpuids; i++) { + if (cpuid_to_apicid[i] == apicid) + return i; + } + + /* Allocate a new cpuid. */ + if (nr_logical_cpuids >= nr_cpu_ids) { + WARN_ONCE(1, "Only %d processors supported." +"Processor %d/0x%x and the rest are ignored.\n", +nr_cpu_ids - 1, nr_logical_cpuids, apicid); + return -1; + } + + cpuid_to_apicid[nr_logical_cpuids] = apicid; + return nr_logical_cpuids++; +} + +int __generic_processor_info(int apicid, int version, bool enabled) { int cpu, max = nr_cpu_ids; bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid, @@ -2079,8 +2125,17 @@ static int __generic_processor_info(int apicid, int version, bool enabled) * for BSP. */ cpu = 0; - } else -
Re: [PATCH v7 0/5] Make cpuid <-> nodeid mapping persistent
Hi Rafael, This patch set was reported a kernel panic from Intel LKP team. I do some investigation for this. I found that this panic was caused because of Intel test machine. On their machine, the acpi tables has something wrong. The proc_id of processors which are not present cannot be assigned correctly, they are assigned the same value. The wrong value will be used by our patch, and lead to panic. Thanks, Zhu On 05/19/2016 06:39 PM, Zhu Guihua wrote: [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid.
[PATCH v7 1/5] x86, memhp, numa: Online memory-less nodes at boot time.
From: Tang Chen For now, x86 does not support memory-less node. A node without memory will not be onlined, and the cpus on it will be mapped to the other online nodes with memory in init_cpu_to_node(). The reason of doing this is to ensure each cpu has mapped to a node with memory, so that it will be able to allocate local memory for that cpu. But we don't have to do it in this way. In this series of patches, we are going to construct cpu <-> node mapping for all possible cpus at boot time, which is a 1-1 mapping. It means the cpu will be mapped to the node it belongs to, and will never be changed. If a node has only cpus but no memory, the cpus on it will be mapped to a memory-less node. And the memory-less node should be onlined. This patch allocate pgdats for all memory-less nodes and online them at boot time. Then build zonelists for these nodes. As a result, when cpus on these memory-less nodes try to allocate memory from local node, it will automatically fall back to the proper zones in the zonelists. Signed-off-by: Zhu Guihua --- arch/x86/mm/numa.c | 27 +-- 1 file changed, 13 insertions(+), 14 deletions(-) diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c index f70c1ff..257dd56 100644 --- a/arch/x86/mm/numa.c +++ b/arch/x86/mm/numa.c @@ -725,22 +725,19 @@ void __init x86_numa_init(void) numa_init(dummy_numa_init); } -static __init int find_near_online_node(int node) +static void __init init_memory_less_node(int nid) { - int n, val; - int min_val = INT_MAX; - int best_node = -1; + unsigned long zones_size[MAX_NR_ZONES] = {0}; + unsigned long zholes_size[MAX_NR_ZONES] = {0}; - for_each_online_node(n) { - val = node_distance(node, n); + /* Allocate and initialize node data. Memory-less node is now online.*/ + alloc_node_data(nid); + free_area_init_node(nid, zones_size, 0, zholes_size); - if (val < min_val) { - min_val = val; - best_node = n; - } - } - - return best_node; + /* +* All zonelists will be built later in start_kernel() after per cpu +* areas are initialized. +*/ } /* @@ -769,8 +766,10 @@ void __init init_cpu_to_node(void) if (node == NUMA_NO_NODE) continue; + if (!node_online(node)) - node = find_near_online_node(node); + init_memory_less_node(node); + numa_set_node(cpu, node); } } -- 1.9.3
[PATCH v7 0/5] Make cpuid <-> nodeid mapping persistent
[Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. For previous discussion, please refer to: https://lkml.org/lkml/2015/2/27/145 https://lkml.org/lkml/2015/3/25/989 https://lkml.org/lkml/2015/5/14/244 https://lkml.org/lkml/2015/7/7/200 https://lkml.org/lkml/2015/9/27/209 Change log v6 -> v7: 1. Fix arm64 build failure. Change log v5 -> v6: 1. Define func acpi_map_cpu2node() for x86 and ia64 respectively. Change log v4 -> v5: 1. Remove usele
[PATCH v7 2/5] x86, acpi, cpu-hotplug: Enable acpi to register all possible cpus at boot time.
From: Gu Zheng [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. This patch finished step 1. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/kernel/apic/apic.c | 26 +++--- 1 file changed, 19 insertions(+), 7 deletions(-) diff -
Re: [PATCH v6 5/5] x86, acpi, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting.
Hi Lorenzo, Thanks for your test and detailed suggestions. I will update those later. Thanks, Zhu On 04/06/2016 10:29 PM, Lorenzo Pieralisi wrote: [+Dennis since he reported ARM64 build breakage] On Thu, Mar 17, 2016 at 09:32:40AM +0800, Zhu Guihua wrote: From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 4. And it breaks the build on ARM64. drivers/acpi/processor_core.c: In function 'set_processor_node_mapping': drivers/acpi/processor_core.c:316:2: error: implicit declaration of function 'acpi_map_cpu2node' [-Werror=implicit-function-declaration] [...] diff --git a/arch/ia64/kernel/acpi.c b/arch/ia64/kernel/acpi.c index b1698bc..7db5563 100644 --- a/arch/ia64/kernel/acpi.c +++ b/arch/ia64/kernel/acpi.c @@ -796,7 +796,7 @@ int acpi_isa_irq_to_gsi(unsigned isa_irq, u32 *gsi) * ACPI based hotplug CPU support */ #ifdef CONFIG_ACPI_HOTPLUG_CPU -static int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) Return value seems to be ignored on IA64 so you can get rid of it, see below. { #ifdef CONFIG_ACPI_NUMA /* diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index 0ce06ee..7d45261 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -696,7 +696,7 @@ static void __init acpi_set_irq_model_ioapic(void) #ifdef CONFIG_ACPI_HOTPLUG_CPU #include -static void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA int nid; diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c index 0e85678..215177a 100644 --- a/drivers/acpi/bus.c +++ b/drivers/acpi/bus.c @@ -1110,6 +1110,9 @@ static int __init acpi_init(void) acpi_sleep_proc_init(); acpi_wakeup_device_init(); acpi_debugger_init(); +#ifdef CONFIG_ACPI_HOTPLUG_CPU + acpi_set_processor_mapping(); +#endif return 0; } diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 824b98b..45580ff 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -261,6 +261,71 @@ int acpi_get_cpuid(acpi_handle handle, int type, u32 acpi_id) } EXPORT_SYMBOL_GPL(acpi_get_cpuid); +#ifdef CONFIG_ACPI_HOTPLUG_CPU +static bool map_processor(acpi_handle handle, int *phys_id, int *cpuid) phys_id size is 64 bits (phys_cpuid_t) on ARM64, (phys_cpuid_t *phys_id) is what you have to have here. +{ + int type; + u32 acpi_id; + acpi_status status; + acpi_object_type acpi_type; + unsigned long long tmp; + union acpi_object object = { 0 }; + struct acpi_buffer buffer = { sizeof(union acpi_object), &object }; + + status = acpi_get_type(handle, &acpi_type); + if (ACPI_FAILURE(status)) + return false; + + switch (acpi_type) { + case ACPI_TYPE_PROCESSOR: + status = acpi_evaluate_object(handle, NULL, NULL, &buffer); + if (ACPI_FAILURE(status)) + return false; + acpi_id = object.processor.proc_id; + break; + case ACPI_TYPE_DEVICE: + status = acpi_evaluate_integer(handle, "_UID", NULL, &tmp); + if (ACPI_FAILURE(status)) + return false; + acpi_id = tmp; + break; + default: + return false; + } + + type = (acpi_type == ACPI_TYPE_DEVICE) ? 1 : 0; + + *phys_id = __acpi_get_phys_id(handle, type, acpi_id, false); Wrong on ARM64, see above. + *cpuid = acpi_map_cpuid(*phys_id, acpi_id); + if (*cpuid == -1) + return false; + + return true; +} + +static acpi_status __init +set_processor_node_mapping(acpi_handle handle, u32 lvl, void *context, + void **rv) +{ + u32 apic_id; - You can't use u32 here see above - This is generic code and on ARM64 I have no idea what apic_id means, choose another variable name please (phys_id ?) + int cpu_id; + + if (!map_processor(handle, &apic_id, &cpu_id)) + return AE_ERROR; + + acpi_map_cpu2node(handle, cpu_id, apic_id); + return AE_OK; +} + +void __init acpi_set_processor_mapping(void) +{ + /* Set persistent cpu <-> node ma
Re: [lkp] [x86, ACPI, cpu] f962c29c2f: BUG: unable to handle kernel paging request at 0000000000001b00
Hi Rafael, On 04/06/2016 11:21 PM, Rafael J. Wysocki wrote: On 4/6/2016 2:43 AM, kernel test robot wrote: FYI, we noticed the below changes on https://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm.git linux-next commit f962c29c2f5d80be28bd76e2c3fadf1ce97ccd76 ("x86, ACPI, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting") Thanks for the report. I've dropped the patch series that commit is part of from the linux-next branch of linux-pm.git because of this problem. I tried to reproduce this failure, but I failed. I compiled linux-next branch source code of linux-pm.git with this config, then I used qemu to boot a guest with my compiled kernel, it worked well. QEMU command line is: ./x86_64-softmmu/qemu-system-x86_64 -hda /home/fedora21.img \ -kernel /home/linux-pm/arch/x86_64/boot/bzImage \ -m 512,slots=100,maxmem=100G -smp 2,maxcpus=100 -serial stdio -enable-kvm \ -initrd /home/initramfs-4.6.0-rc2-00250-g8767898.img -append "root=/dev/sda3 console=ttyS0 Anyone succeed? Thanks, Zhu +--+++ | | 40d4d2e6c5 | f962c29c2f | +--+++ | boot_successes | 12 | 4 | | boot_failures| 0 | 8 | | BUG:unable_to_handle_kernel | 0 | 8 | | Oops | 0 | 8 | | RIP:__alloc_pages_nodemask | 0 | 8 | | Kernel_panic-not_syncing:Fatal_exception | 0 | 8 | | backtrace:pcpu_balance_workfn| 0 | 8 | +--+++ [ 10.413471] RAPL PMU: hw unit of domain pp0-core 2^-16 Joules [ 10.419886] RAPL PMU: hw unit of domain package 2^-16 Joules [ 10.426206] RAPL PMU: hw unit of domain dram 2^-16 Joules [ 10.433905] BUG: unable to handle kernel paging request at 1b00 [ 10.441699] IP: [] __alloc_pages_nodemask+0x210/0xc10 [ 10.449097] PGD 0 [ 10.451350] Oops: [#1] SMP [ 10.454972] Modules linked in: [ 10.458390] CPU: 12 PID: 407 Comm: kworker/12:1 Not tainted 4.6.0-rc1-7-gf962c29 #1 [ 10.467326] Hardware name: Intel Corporation S2600WP/S2600WP, BIOS SE5C600.86B.02.02.0002.122320131210 12/23/2013 [ 10.478776] Workqueue: events pcpu_balance_workfn [ 10.484035] task: 88100d858000 ti: 88100d86 task.ti: 88100d86 [ 10.492381] RIP: 0010:[] [] __alloc_pages_nodemask+0x210/0xc10 [ 10.502491] RSP: :88100d863c28 EFLAGS: 00010286 [ 10.508419] RAX: RBX: RCX: 0004 [ 10.516383] RDX: 8000 RSI: 0d06 RDI: 81c9d430 [ 10.524348] RBP: 88100d863d40 R08: R09: [ 10.532312] R10: 81ca1e87 R11: c9076000 R12: 0003 [ 10.540278] R13: 1b00 R14: R15: 88100d43e480 [ 10.548243] FS: () GS:88101340() knlGS: [ 10.557275] CS: 0010 DS: ES: CR0: 80050033 [ 10.563687] CR2: 1b00 CR3: 00103ee06000 CR4: 001406e0 [ 10.571652] Stack: [ 10.573894] c9090fff c9091000 881013002008 0001 [ 10.582196] 8163 024082c2 0040 [ 10.590492] 001e 880ffe57c0c0 88100d863c88 811b7766 [ 10.598794] Call Trace: [ 10.601527] [] ? map_vm_area+0x36/0x50 [ 10.607553] [] pcpu_populate_chunk+0xae/0x340 [ 10.614257] [] pcpu_balance_workfn+0x578/0x5b0 [ 10.621060] [] process_one_work+0x155/0x440 [ 10.627562] [] worker_thread+0x4e/0x4c0 [ 10.633687] [] ? __schedule+0x34b/0x8b0 [ 10.639809] [] ? rescuer_thread+0x350/0x350 [ 10.646319] [] ? rescuer_thread+0x350/0x350 [ 10.652831] [] kthread+0xd4/0xf0 [ 10.658276] [] ret_from_fork+0x22/0x40 [ 10.664302] [] ? kthread_park+0x60/0x60 [ 10.670423] Code: d0 49 8b 04 24 48 85 c0 75 e0 65 ff 0d 22 f0 e8 7e eb 8b 31 d2 be 06 0d 00 00 48 c7 c7 30 d4 c9 81 e8 35 2c f2 ff e8 50 40 77 00 <49> 83 7d 00 00 0f 85 7c fe ff ff 31 c0 e9 6d ff ff ff 0f 1f 44 [ 10.692137] RIP [] __alloc_pages_nodemask+0x210/0xc10 [ 10.699627] RSP [ 10.703518] CR2: 1b00 [ 10.707220] ---[ end trace bc83ef9ded88f808 ]--- [ 10.712373] Kernel panic - not syncing: Fatal exception Thanks, Kernel Test Robot .
[PATCH v6 1/5] x86, memhp, numa: Online memory-less nodes at boot time.
From: Tang Chen For now, x86 does not support memory-less node. A node without memory will not be onlined, and the cpus on it will be mapped to the other online nodes with memory in init_cpu_to_node(). The reason of doing this is to ensure each cpu has mapped to a node with memory, so that it will be able to allocate local memory for that cpu. But we don't have to do it in this way. In this series of patches, we are going to construct cpu <-> node mapping for all possible cpus at boot time, which is a 1-1 mapping. It means the cpu will be mapped to the node it belongs to, and will never be changed. If a node has only cpus but no memory, the cpus on it will be mapped to a memory-less node. And the memory-less node should be onlined. This patch allocate pgdats for all memory-less nodes and online them at boot time. Then build zonelists for these nodes. As a result, when cpus on these memory-less nodes try to allocate memory from local node, it will automatically fall back to the proper zones in the zonelists. Signed-off-by: Zhu Guihua --- arch/x86/mm/numa.c | 27 +-- 1 file changed, 13 insertions(+), 14 deletions(-) diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c index f70c1ff..257dd56 100644 --- a/arch/x86/mm/numa.c +++ b/arch/x86/mm/numa.c @@ -725,22 +725,19 @@ void __init x86_numa_init(void) numa_init(dummy_numa_init); } -static __init int find_near_online_node(int node) +static void __init init_memory_less_node(int nid) { - int n, val; - int min_val = INT_MAX; - int best_node = -1; + unsigned long zones_size[MAX_NR_ZONES] = {0}; + unsigned long zholes_size[MAX_NR_ZONES] = {0}; - for_each_online_node(n) { - val = node_distance(node, n); + /* Allocate and initialize node data. Memory-less node is now online.*/ + alloc_node_data(nid); + free_area_init_node(nid, zones_size, 0, zholes_size); - if (val < min_val) { - min_val = val; - best_node = n; - } - } - - return best_node; + /* +* All zonelists will be built later in start_kernel() after per cpu +* areas are initialized. +*/ } /* @@ -769,8 +766,10 @@ void __init init_cpu_to_node(void) if (node == NUMA_NO_NODE) continue; + if (!node_online(node)) - node = find_near_online_node(node); + init_memory_less_node(node); + numa_set_node(cpu, node); } } -- 1.9.3
[PATCH v6 0/5] Make cpuid <-> nodeid mapping persistent
[Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. For previous discussion, please refer to: https://lkml.org/lkml/2015/2/27/145 https://lkml.org/lkml/2015/3/25/989 https://lkml.org/lkml/2015/5/14/244 https://lkml.org/lkml/2015/7/7/200 https://lkml.org/lkml/2015/9/27/209 Change log v5 -> v6: 1. Define func acpi_map_cpu2node() for x86 and ia64 respectively. Change log v4 -> v5: 1. Remove useless code in patch 1. 2. Small improvement of commit
[PATCH v6 3/5] x86, acpi, cpu-hotplug: Introduce cpuid_to_apicid[] array to store persistent cpuid <-> apicid mapping.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 2. In this patch, we introduce a new static array named cpuid_to_apicid[], which is large enough to store info for all possible cpus. And then, we modify the cpuid calculation. In generic_processor_info(), it simply finds the next unused cpuid. And it is also why the cpuid <-> nodeid mapping changes with node hotplug. After this patch, we find the next unused cpuid, map it to an apicid, and store the mapping in cpuid_to_apicid[], so that cpuid <-> apicid mapping will be persistent. And finally we will use this array to make cpuid <-> nodeid persistent. cpuid <-> apicid mapping is established at local apic registeration time. But non-present or disabled cpus are ignored. In this patch, we establish all possible cpuid <-> apicid mapping when registering local apic. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/include/asm/mpspec.h | 1 + arch/x86/kernel/acpi/boot.c | 6 ++--- arch/x86/kernel/apic/apic.c | 61 --- 3 files changed, 61 insertions(+), 7 deletions(-) diff --git a/arch/x86/include/asm/mpspec.h b/arch/x86/include/asm/mpspec.h index b07233b..db902d8 100644 --- a/arch/x86/include/asm/mpspec.h +++ b/arch/x86/include/asm/mpspec.h @@ -86,6 +86,7 @@ static inline void early_reserve_e820_mpc_new(void) { } #endif int generic_processor_info(int apicid, int version); +int __generic_processor_info(int apicid, int version, bool enabled); #define PHYSID_ARRAY_SIZE BITS_TO_LONGS(MAX_LOCAL_APIC) diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index e759076..0ce06ee 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -174,15 +174,13 @@ static int acpi_register_lapic(int id, u8 enabled) return -EINVAL; } - if (!enabled) { + if (!enabled) ++disabled_cpus; - return -EINVAL; - } if (boot_cpu_physical_apicid != -1U) ver = apic_version[boot_cpu_physical_apicid]; - return generic_processor_info(id, ver); + return __generic_processor_info(id, ver, enabled); } static int __init diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index 55cb80e..bedb674 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -1998,7 +1998,53 @@ void disconnect_bsp_APIC(int virt_wire_setup) apic_write(APIC_LVT1, value); } -static int __generic_processor_info(int apicid, int version, bool enabled) +/* + * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated + * contiguously, it equals to current allocated max logical CPU ID plus 1. + * All allocated CPU ID should be in [0, nr_logical_cpuidi), so the maximum of + * nr_logical_cpuids is nr_cpu_ids. + * + * NOTE: Reserve 0 for BSP. + */ +static int nr_logical_cpuids = 1; + +/* + * Used to store mapping between logical CPU IDs and APIC IDs. + */ +static int cpuid_to_apicid[] = { + [0 ... NR_CPUS - 1] = -1, +}; + +/* + * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids + * and cpuid_to_apicid[] synchronized. + */ +static int allocate_logical_cpuid(int apicid) +{ + int i; + + /* +* cpuid <-> apicid mapping is persistent, so when a cpu is up, +* check if the kernel has allocated a cpuid for it. +*/ + for (i = 0; i < nr_logical_cpuids; i++) { + if (cpuid_to_apicid[i] == apicid) + return i; + } + + /* Allocate a new cpuid. */ + if (nr_logical_cpuids >= nr_cpu_ids) { + WARN_ONCE(1, "Only %d processors supported." +"Processor %d/0x%x and the rest are ignored.\n", +nr_cpu_ids - 1, nr_logical_cpuids, apicid); + return -1; + } + + cpuid_to_apicid[nr_logical_cpuids] = apicid; + return nr_logical_cpuids++; +} + +int __generic_processor_info(int apicid, int version, bool enabled) { int cpu, max = nr_cpu_ids; bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid, @@ -2079,8 +2125,17 @@ static int __generic_processor_info(int apicid, int version, bool enabled) * for BSP. */ cpu = 0; - } else -
[PATCH v6 2/5] x86, acpi, cpu-hotplug: Enable acpi to register all possible cpus at boot time.
From: Gu Zheng [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. This patch finished step 1. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/kernel/apic/apic.c | 26 +++--- 1 file changed, 19 insertions(+), 7 deletions(-) diff -
[PATCH v6 4/5] x86, acpi, cpu-hotplug: Enable MADT APIs to return disabled apicid.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 3. There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm(persistent) 2. apicid (physical cpu id) <-> nodeid (persistent) 3. cpuid (logical cpu id) <-> apicid (not persistent, now persistent by step 2) 4. cpuid (logical cpu id) <-> nodeid (not persistent) So, in order to setup persistent cpuid <-> nodeid mapping for all possible CPUs, we should: 1. Setup cpuid <-> apicid mapping for all possible CPUs, which has been done in step 1, 2. 2. Setup cpuid <-> nodeid mapping for all possible CPUs. But before that, we should obtain all apicids from MADT. All processors' apicids can be obtained by _MAT method or from MADT in ACPI. The current code ignores disabled processors and returns -ENODEV. After this patch, a new parameter will be added to MADT APIs so that caller is able to control if disabled processors are ignored. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- drivers/acpi/acpi_processor.c | 5 +++- drivers/acpi/processor_core.c | 57 +++ 2 files changed, 40 insertions(+), 22 deletions(-) diff --git a/drivers/acpi/acpi_processor.c b/drivers/acpi/acpi_processor.c index b5e54f2..3ccdd8f 100644 --- a/drivers/acpi/acpi_processor.c +++ b/drivers/acpi/acpi_processor.c @@ -300,8 +300,11 @@ static int acpi_processor_get_info(struct acpi_device *device) * Extra Processor objects may be enumerated on MP systems with * less than the max # of CPUs. They should be ignored _iff * they are physically not present. +* +* NOTE: Even if the processor has a cpuid, it may not present because +* cpuid <-> apicid mapping is persistent now. */ - if (invalid_logical_cpuid(pr->id)) { + if (invalid_logical_cpuid(pr->id) || !cpu_present(pr->id)) { int ret = acpi_processor_hotadd_init(pr); if (ret) return ret; diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 33a38d6..824b98b 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -32,12 +32,12 @@ static struct acpi_table_madt *get_madt_table(void) } static int map_lapic_id(struct acpi_subtable_header *entry, -u32 acpi_id, phys_cpuid_t *apic_id) +u32 acpi_id, phys_cpuid_t *apic_id, bool ignore_disabled) { struct acpi_madt_local_apic *lapic = container_of(entry, struct acpi_madt_local_apic, header); - if (!(lapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (lapic->processor_id != acpi_id) @@ -48,12 +48,13 @@ static int map_lapic_id(struct acpi_subtable_header *entry, } static int map_x2apic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_x2apic *apic = container_of(entry, struct acpi_madt_local_x2apic, header); - if (!(apic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(apic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration && (apic->uid == acpi_id)) { @@ -65,12 +66,13 @@ static int map_x2apic_id(struct acpi_subtable_header *entry, } static int map_lsapic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_sapic *lsapic = container_of(entry, struct acpi_madt_local_sapic, header); - if (!(lsapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lsapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration) { @@ -87,12 +89,13 @@ static int map_lsapic_id(struct acpi_subtable_header *entry, * Retrieve the ARM CPU physical identifie
[PATCH v6 5/5] x86, acpi, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 4. This patch set the persistent cpuid <-> nodeid mapping for all enabled/disabled processors at boot time via an additional acpi namespace walk for processors. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/ia64/kernel/acpi.c | 2 +- arch/x86/kernel/acpi/boot.c | 2 +- drivers/acpi/bus.c| 3 ++ drivers/acpi/processor_core.c | 65 +++ include/linux/acpi.h | 6 5 files changed, 76 insertions(+), 2 deletions(-) diff --git a/arch/ia64/kernel/acpi.c b/arch/ia64/kernel/acpi.c index b1698bc..7db5563 100644 --- a/arch/ia64/kernel/acpi.c +++ b/arch/ia64/kernel/acpi.c @@ -796,7 +796,7 @@ int acpi_isa_irq_to_gsi(unsigned isa_irq, u32 *gsi) * ACPI based hotplug CPU support */ #ifdef CONFIG_ACPI_HOTPLUG_CPU -static int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA /* diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index 0ce06ee..7d45261 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -696,7 +696,7 @@ static void __init acpi_set_irq_model_ioapic(void) #ifdef CONFIG_ACPI_HOTPLUG_CPU #include -static void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA int nid; diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c index 0e85678..215177a 100644 --- a/drivers/acpi/bus.c +++ b/drivers/acpi/bus.c @@ -1110,6 +1110,9 @@ static int __init acpi_init(void) acpi_sleep_proc_init(); acpi_wakeup_device_init(); acpi_debugger_init(); +#ifdef CONFIG_ACPI_HOTPLUG_CPU + acpi_set_processor_mapping(); +#endif return 0; } diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 824b98b..45580ff 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -261,6 +261,71 @@ int acpi_get_cpuid(acpi_handle handle, int type, u32 acpi_id) } EXPORT_SYMBOL_GPL(acpi_get_cpuid); +#ifdef CONFIG_ACPI_HOTPLUG_CPU +static bool map_processor(acpi_handle handle, int *phys_id, int *cpuid) +{ + int type; + u32 acpi_id; + acpi_status status; + acpi_object_type acpi_type; + unsigned long long tmp; + union acpi_object object = { 0 }; + struct acpi_buffer buffer = { sizeof(union acpi_object), &object }; + + status = acpi_get_type(handle, &acpi_type); + if (ACPI_FAILURE(status)) + return false; + + switch (acpi_type) { + case ACPI_TYPE_PROCESSOR: + status = acpi_evaluate_object(handle, NULL, NULL, &buffer); + if (ACPI_FAILURE(status)) + return false; + acpi_id = object.processor.proc_id; + break; + case ACPI_TYPE_DEVICE: + status = acpi_evaluate_integer(handle, "_UID", NULL, &tmp); + if (ACPI_FAILURE(status)) + return false; + acpi_id = tmp; + break; + default: + return false; + } + + type = (acpi_type == ACPI_TYPE_DEVICE) ? 1 : 0; + + *phys_id = __acpi_get_phys_id(handle, type, acpi_id, false); + *cpuid = acpi_map_cpuid(*phys_id, acpi_id); + if (*cpuid == -1) + return false; + + return true; +} + +static acpi_status __init +set_processor_node_mapping(acpi_handle handle, u32 lvl, void *context, + void **rv) +{ + u32 apic_id; + int cpu_id; + + if (!map_processor(handle, &apic_id, &cpu_id)) + return AE_ERROR; + + acpi_map_cpu2node(handle, cpu_id, apic_id); + return AE_OK; +} + +void __init acpi_set_processor_mapping(void) +{ + /* Set persistent cpu <-> node mapping for all processors. */ + acpi_walk_namespace(ACPI_TYPE_PROCESSOR, ACPI_ROOT_OBJECT, + ACPI_UINT32_MAX, set_processor_node_mapping, + NULL, NULL, NULL); +} +#endif + #ifdef CONFIG_ACPI_HOTPLUG_IOAPIC static int get_ioapic_id(struct acpi_subtable_header *entry, u32 gsi_base, u64 *phys_addr, int *ioapic_id) diff --git
Re: [RESEND PATCH v5 5/5] x86, acpi, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting.
On 03/03/2016 10:11 AM, kbuild test robot wrote: Hi Gu, [auto build test ERROR on tip/x86/core] [also build test ERROR on v4.5-rc6 next-20160302] [if your patch is applied to the wrong git tree, please drop us a note to help improving the system] url: https://github.com/0day-ci/linux/commits/Zhu-Guihua/Make-cpuid-nodeid-mapping-persistent/20160303-094713 config: ia64-allyesconfig (attached as .config) reproduce: wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross # save the attached .config to linux build tree make.cross ARCH=ia64 All errors (new ones prefixed by >>): arch/ia64/kernel/acpi.c:799:5: error: conflicting types for 'acpi_map_cpu2node' int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) ^ In file included from arch/ia64/kernel/acpi.c:43:0: include/linux/acpi.h:268:6: note: previous declaration of 'acpi_map_cpu2node' was here void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid); ^ vim +/acpi_map_cpu2node +799 arch/ia64/kernel/acpi.c 793 } 794 795 /* 796 * ACPI based hotplug CPU support 797 */ 798 #ifdef CONFIG_ACPI_HOTPLUG_CPU > 799 int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) 800 { 801 #ifdef CONFIG_ACPI_NUMA 802 /* --- 0-DAY kernel test infrastructureOpen Source Technology Center https://lists.01.org/pipermail/kbuild-all Intel Corporation Thanks for your test. I will investigate this. Thanks, Zhu
Re: [RESEND PATCH v5 0/5] Make cpuid <-> nodeid mapping persistent
Hi, On 03/03/2016 10:08 AM, Rafael J. Wysocki wrote: Hi, On Thu, Mar 3, 2016 at 2:42 AM, Zhu Guihua wrote: [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. Are there any changes in this version relative to the previous one? No, there are no changes in this version. Thanks, Zhu Thanks, Rafael .
Re: [RESEND PATCH v5 0/5] Make cpuid <-> nodeid mapping persistent
Hi Thomas, Ingo, hpa, Would you please help to review the X86 part of this patch-set ? Thanks, Zhu On 03/03/2016 09:42 AM, Zhu Guihua wrote: [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. For previous discussion, please refer to: https://lkml.org/lkml
[RESEND PATCH v5 1/5] x86, memhp, numa: Online memory-less nodes at boot time.
From: Tang Chen For now, x86 does not support memory-less node. A node without memory will not be onlined, and the cpus on it will be mapped to the other online nodes with memory in init_cpu_to_node(). The reason of doing this is to ensure each cpu has mapped to a node with memory, so that it will be able to allocate local memory for that cpu. But we don't have to do it in this way. In this series of patches, we are going to construct cpu <-> node mapping for all possible cpus at boot time, which is a 1-1 mapping. It means the cpu will be mapped to the node it belongs to, and will never be changed. If a node has only cpus but no memory, the cpus on it will be mapped to a memory-less node. And the memory-less node should be onlined. This patch allocate pgdats for all memory-less nodes and online them at boot time. Then build zonelists for these nodes. As a result, when cpus on these memory-less nodes try to allocate memory from local node, it will automatically fall back to the proper zones in the zonelists. Signed-off-by: Zhu Guihua --- arch/x86/mm/numa.c | 27 +-- 1 file changed, 13 insertions(+), 14 deletions(-) diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c index d04f809..baf3d72 100644 --- a/arch/x86/mm/numa.c +++ b/arch/x86/mm/numa.c @@ -704,22 +704,19 @@ void __init x86_numa_init(void) numa_init(dummy_numa_init); } -static __init int find_near_online_node(int node) +static void __init init_memory_less_node(int nid) { - int n, val; - int min_val = INT_MAX; - int best_node = -1; + unsigned long zones_size[MAX_NR_ZONES] = {0}; + unsigned long zholes_size[MAX_NR_ZONES] = {0}; - for_each_online_node(n) { - val = node_distance(node, n); + /* Allocate and initialize node data. Memory-less node is now online.*/ + alloc_node_data(nid); + free_area_init_node(nid, zones_size, 0, zholes_size); - if (val < min_val) { - min_val = val; - best_node = n; - } - } - - return best_node; + /* +* All zonelists will be built later in start_kernel() after per cpu +* areas are initialized. +*/ } /* @@ -748,8 +745,10 @@ void __init init_cpu_to_node(void) if (node == NUMA_NO_NODE) continue; + if (!node_online(node)) - node = find_near_online_node(node); + init_memory_less_node(node); + numa_set_node(cpu, node); } } -- 1.9.3
[RESEND PATCH v5 0/5] Make cpuid <-> nodeid mapping persistent
[Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. For previous discussion, please refer to: https://lkml.org/lkml/2015/2/27/145 https://lkml.org/lkml/2015/3/25/989 https://lkml.org/lkml/2015/5/14/244 https://lkml.org/lkml/2015/7/7/200 https://lkml.org/lkml/2015/9/27/209 Change log v4 -> v5: 1. Remove useless code in patch 1. 2. Small improvement of commit message. Change log v3 -> v4: 1. Fix the kernel panic at boot time. The cause is that
[RESEND PATCH v5 2/5] x86, acpi, cpu-hotplug: Enable acpi to register all possible cpus at boot time.
From: Gu Zheng [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. This patch finished step 1. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/kernel/apic/apic.c | 26 +++--- 1 file changed, 19 insertions(+), 7 deletions(-) diff -
[RESEND PATCH v5 4/5] x86, acpi, cpu-hotplug: Enable MADT APIs to return disabled apicid.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 3. There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm(persistent) 2. apicid (physical cpu id) <-> nodeid (persistent) 3. cpuid (logical cpu id) <-> apicid (not persistent, now persistent by step 2) 4. cpuid (logical cpu id) <-> nodeid (not persistent) So, in order to setup persistent cpuid <-> nodeid mapping for all possible CPUs, we should: 1. Setup cpuid <-> apicid mapping for all possible CPUs, which has been done in step 1, 2. 2. Setup cpuid <-> nodeid mapping for all possible CPUs. But before that, we should obtain all apicids from MADT. All processors' apicids can be obtained by _MAT method or from MADT in ACPI. The current code ignores disabled processors and returns -ENODEV. After this patch, a new parameter will be added to MADT APIs so that caller is able to control if disabled processors are ignored. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- drivers/acpi/acpi_processor.c | 5 +++- drivers/acpi/processor_core.c | 57 +++ 2 files changed, 40 insertions(+), 22 deletions(-) diff --git a/drivers/acpi/acpi_processor.c b/drivers/acpi/acpi_processor.c index 6979186..d30111a 100644 --- a/drivers/acpi/acpi_processor.c +++ b/drivers/acpi/acpi_processor.c @@ -300,8 +300,11 @@ static int acpi_processor_get_info(struct acpi_device *device) * Extra Processor objects may be enumerated on MP systems with * less than the max # of CPUs. They should be ignored _iff * they are physically not present. +* +* NOTE: Even if the processor has a cpuid, it may not present because +* cpuid <-> apicid mapping is persistent now. */ - if (invalid_logical_cpuid(pr->id)) { + if (invalid_logical_cpuid(pr->id) || !cpu_present(pr->id)) { int ret = acpi_processor_hotadd_init(pr); if (ret) return ret; diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 33a38d6..824b98b 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -32,12 +32,12 @@ static struct acpi_table_madt *get_madt_table(void) } static int map_lapic_id(struct acpi_subtable_header *entry, -u32 acpi_id, phys_cpuid_t *apic_id) +u32 acpi_id, phys_cpuid_t *apic_id, bool ignore_disabled) { struct acpi_madt_local_apic *lapic = container_of(entry, struct acpi_madt_local_apic, header); - if (!(lapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (lapic->processor_id != acpi_id) @@ -48,12 +48,13 @@ static int map_lapic_id(struct acpi_subtable_header *entry, } static int map_x2apic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_x2apic *apic = container_of(entry, struct acpi_madt_local_x2apic, header); - if (!(apic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(apic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration && (apic->uid == acpi_id)) { @@ -65,12 +66,13 @@ static int map_x2apic_id(struct acpi_subtable_header *entry, } static int map_lsapic_id(struct acpi_subtable_header *entry, - int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id) + int device_declaration, u32 acpi_id, phys_cpuid_t *apic_id, + bool ignore_disabled) { struct acpi_madt_local_sapic *lsapic = container_of(entry, struct acpi_madt_local_sapic, header); - if (!(lsapic->lapic_flags & ACPI_MADT_ENABLED)) + if (ignore_disabled && !(lsapic->lapic_flags & ACPI_MADT_ENABLED)) return -ENODEV; if (device_declaration) { @@ -87,12 +89,13 @@ static int map_lsapic_id(struct acpi_subtable_header *entry, * Retrieve the ARM CPU physical identifie
[RESEND PATCH v5 3/5] x86, acpi, cpu-hotplug: Introduce cpuid_to_apicid[] array to store persistent cpuid <-> apicid mapping.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 2. In this patch, we introduce a new static array named cpuid_to_apicid[], which is large enough to store info for all possible cpus. And then, we modify the cpuid calculation. In generic_processor_info(), it simply finds the next unused cpuid. And it is also why the cpuid <-> nodeid mapping changes with node hotplug. After this patch, we find the next unused cpuid, map it to an apicid, and store the mapping in cpuid_to_apicid[], so that cpuid <-> apicid mapping will be persistent. And finally we will use this array to make cpuid <-> nodeid persistent. cpuid <-> apicid mapping is established at local apic registeration time. But non-present or disabled cpus are ignored. In this patch, we establish all possible cpuid <-> apicid mapping when registering local apic. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/x86/include/asm/mpspec.h | 1 + arch/x86/kernel/acpi/boot.c | 6 ++--- arch/x86/kernel/apic/apic.c | 61 --- 3 files changed, 61 insertions(+), 7 deletions(-) diff --git a/arch/x86/include/asm/mpspec.h b/arch/x86/include/asm/mpspec.h index b07233b..db902d8 100644 --- a/arch/x86/include/asm/mpspec.h +++ b/arch/x86/include/asm/mpspec.h @@ -86,6 +86,7 @@ static inline void early_reserve_e820_mpc_new(void) { } #endif int generic_processor_info(int apicid, int version); +int __generic_processor_info(int apicid, int version, bool enabled); #define PHYSID_ARRAY_SIZE BITS_TO_LONGS(MAX_LOCAL_APIC) diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index e759076..0ce06ee 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -174,15 +174,13 @@ static int acpi_register_lapic(int id, u8 enabled) return -EINVAL; } - if (!enabled) { + if (!enabled) ++disabled_cpus; - return -EINVAL; - } if (boot_cpu_physical_apicid != -1U) ver = apic_version[boot_cpu_physical_apicid]; - return generic_processor_info(id, ver); + return __generic_processor_info(id, ver, enabled); } static int __init diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index 1625778..4822cda 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -1998,7 +1998,53 @@ void disconnect_bsp_APIC(int virt_wire_setup) apic_write(APIC_LVT1, value); } -static int __generic_processor_info(int apicid, int version, bool enabled) +/* + * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated + * contiguously, it equals to current allocated max logical CPU ID plus 1. + * All allocated CPU ID should be in [0, nr_logical_cpuidi), so the maximum of + * nr_logical_cpuids is nr_cpu_ids. + * + * NOTE: Reserve 0 for BSP. + */ +static int nr_logical_cpuids = 1; + +/* + * Used to store mapping between logical CPU IDs and APIC IDs. + */ +static int cpuid_to_apicid[] = { + [0 ... NR_CPUS - 1] = -1, +}; + +/* + * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids + * and cpuid_to_apicid[] synchronized. + */ +static int allocate_logical_cpuid(int apicid) +{ + int i; + + /* +* cpuid <-> apicid mapping is persistent, so when a cpu is up, +* check if the kernel has allocated a cpuid for it. +*/ + for (i = 0; i < nr_logical_cpuids; i++) { + if (cpuid_to_apicid[i] == apicid) + return i; + } + + /* Allocate a new cpuid. */ + if (nr_logical_cpuids >= nr_cpu_ids) { + WARN_ONCE(1, "Only %d processors supported." +"Processor %d/0x%x and the rest are ignored.\n", +nr_cpu_ids - 1, nr_logical_cpuids, apicid); + return -1; + } + + cpuid_to_apicid[nr_logical_cpuids] = apicid; + return nr_logical_cpuids++; +} + +int __generic_processor_info(int apicid, int version, bool enabled) { int cpu, max = nr_cpu_ids; bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid, @@ -2079,8 +2125,17 @@ static int __generic_processor_info(int apicid, int version, bool enabled) * for BSP. */ cpu = 0; - } else -
[RESEND PATCH v5 5/5] x86, acpi, cpu-hotplug: Set persistent cpuid <-> nodeid mapping when booting.
From: Gu Zheng The whole patch-set aims at making cpuid <-> nodeid mapping persistent. So that, when node online/offline happens, cache based on cpuid <-> nodeid mapping such as wq_numa_possible_cpumask will not cause any problem. It contains 4 steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. 4. Establish all possible cpuid <-> nodeid mapping. This patch finishes step 4. This patch set the persistent cpuid <-> nodeid mapping for all enabled/disabled processors at boot time via an additional acpi namespace walk for processors. Signed-off-by: Gu Zheng Signed-off-by: Tang Chen Signed-off-by: Zhu Guihua --- arch/ia64/kernel/acpi.c | 2 +- arch/x86/kernel/acpi/boot.c | 2 +- drivers/acpi/bus.c| 3 ++ drivers/acpi/processor_core.c | 65 +++ include/linux/acpi.h | 2 ++ 5 files changed, 72 insertions(+), 2 deletions(-) diff --git a/arch/ia64/kernel/acpi.c b/arch/ia64/kernel/acpi.c index b1698bc..7db5563 100644 --- a/arch/ia64/kernel/acpi.c +++ b/arch/ia64/kernel/acpi.c @@ -796,7 +796,7 @@ int acpi_isa_irq_to_gsi(unsigned isa_irq, u32 *gsi) * ACPI based hotplug CPU support */ #ifdef CONFIG_ACPI_HOTPLUG_CPU -static int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +int acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA /* diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c index 0ce06ee..7d45261 100644 --- a/arch/x86/kernel/acpi/boot.c +++ b/arch/x86/kernel/acpi/boot.c @@ -696,7 +696,7 @@ static void __init acpi_set_irq_model_ioapic(void) #ifdef CONFIG_ACPI_HOTPLUG_CPU #include -static void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) +void acpi_map_cpu2node(acpi_handle handle, int cpu, int physid) { #ifdef CONFIG_ACPI_NUMA int nid; diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c index 891c42d..d92f45f 100644 --- a/drivers/acpi/bus.c +++ b/drivers/acpi/bus.c @@ -1096,6 +1096,9 @@ static int __init acpi_init(void) acpi_sleep_proc_init(); acpi_wakeup_device_init(); acpi_debugger_init(); +#ifdef CONFIG_ACPI_HOTPLUG_CPU + acpi_set_processor_mapping(); +#endif return 0; } diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c index 824b98b..45580ff 100644 --- a/drivers/acpi/processor_core.c +++ b/drivers/acpi/processor_core.c @@ -261,6 +261,71 @@ int acpi_get_cpuid(acpi_handle handle, int type, u32 acpi_id) } EXPORT_SYMBOL_GPL(acpi_get_cpuid); +#ifdef CONFIG_ACPI_HOTPLUG_CPU +static bool map_processor(acpi_handle handle, int *phys_id, int *cpuid) +{ + int type; + u32 acpi_id; + acpi_status status; + acpi_object_type acpi_type; + unsigned long long tmp; + union acpi_object object = { 0 }; + struct acpi_buffer buffer = { sizeof(union acpi_object), &object }; + + status = acpi_get_type(handle, &acpi_type); + if (ACPI_FAILURE(status)) + return false; + + switch (acpi_type) { + case ACPI_TYPE_PROCESSOR: + status = acpi_evaluate_object(handle, NULL, NULL, &buffer); + if (ACPI_FAILURE(status)) + return false; + acpi_id = object.processor.proc_id; + break; + case ACPI_TYPE_DEVICE: + status = acpi_evaluate_integer(handle, "_UID", NULL, &tmp); + if (ACPI_FAILURE(status)) + return false; + acpi_id = tmp; + break; + default: + return false; + } + + type = (acpi_type == ACPI_TYPE_DEVICE) ? 1 : 0; + + *phys_id = __acpi_get_phys_id(handle, type, acpi_id, false); + *cpuid = acpi_map_cpuid(*phys_id, acpi_id); + if (*cpuid == -1) + return false; + + return true; +} + +static acpi_status __init +set_processor_node_mapping(acpi_handle handle, u32 lvl, void *context, + void **rv) +{ + u32 apic_id; + int cpu_id; + + if (!map_processor(handle, &apic_id, &cpu_id)) + return AE_ERROR; + + acpi_map_cpu2node(handle, cpu_id, apic_id); + return AE_OK; +} + +void __init acpi_set_processor_mapping(void) +{ + /* Set persistent cpu <-> node mapping for all processors. */ + acpi_walk_namespace(ACPI_TYPE_PROCESSOR, ACPI_ROOT_OBJECT, + ACPI_UINT32_MAX, set_processor_node_mapping, + NULL, NULL, NULL); +} +#endif + #ifdef CONFIG_ACPI_HOTPLUG_IOAPIC static int get_ioapic_id(struct acpi_subtable_header *entry, u32 gsi_base, u64 *phys_addr, int *ioapic_id) diff --git a/incl
Re: [PATCH v5 RESEND 0/5] Make cpuid <-> nodeid mapping persistent
Hi Rafael, On 02/03/2016 08:02 PM, Rafael J. Wysocki wrote: Hi, On Wed, Feb 3, 2016 at 10:14 AM, Zhu Guihua wrote: On 01/25/2016 02:12 PM, Tang Chen wrote: Hi Rafael, Len, Would you please help to review the ACPI part of this patch-set ? Can anyone help to review this? I'm planning to look into this more thoroughly in the next few days. Were you reviewing this ? Thanks. Thanks, Rafael .
Re: [PATCH v5 RESEND 0/5] Make cpuid <-> nodeid mapping persistent
On 01/25/2016 02:12 PM, Tang Chen wrote: Hi Rafael, Len, Would you please help to review the ACPI part of this patch-set ? Can anyone help to review this? Cc tj: Can you add acked-by into this patchset? Thanks. Thanks. On 01/25/2016 02:08 PM, Tang Chen wrote: [Problem] cpuid <-> nodeid mapping is firstly established at boot time. And workqueue caches the mapping in wq_numa_possible_cpumask in wq_numa_init() at boot time. When doing node online/offline, cpuid <-> nodeid mapping is established/destroyed, which means, cpuid <-> nodeid mapping will change if node hotplug happens. But workqueue does not update wq_numa_possible_cpumask. So here is the problem: Assume we have the following cpuid <-> nodeid in the beginning: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 2 | 30-44, 90-104 node 3 | 45-59, 105-119 and we hot-remove node2 and node3, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 and we hot-add node4 and node5, it becomes: Node | CPU node 0 | 0-14, 60-74 node 1 | 15-29, 75-89 node 4 | 30-59 node 5 | 90-119 But in wq_numa_possible_cpumask, cpu30 is still mapped to node2, and the like. When a pool workqueue is initialized, if its cpumask belongs to a node, its pool->node will be mapped to that node. And memory used by this workqueue will also be allocated on that node. static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs){ ... /* if cpumask is contained inside a NUMA node, we belong to that node */ if (wq_numa_enabled) { for_each_node(node) { if (cpumask_subset(pool->attrs->cpumask, wq_numa_possible_cpumask[node])) { pool->node = node; break; } } } Since wq_numa_possible_cpumask is not updated, it could be mapped to an offline node, which will lead to memory allocation failure: SLUB: Unable to allocate memory on node 2 (gfp=0x80d0) cache: kmalloc-192, object size: 192, buffer size: 192, default order: 1, min order: 0 node 0: slabs: 6172, objs: 259224, free: 245741 node 1: slabs: 3261, objs: 136962, free: 127656 It happens here: create_worker(struct worker_pool *pool) |--> worker = alloc_worker(pool->node); static struct worker *alloc_worker(int node) { struct worker *worker; worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); --> Here, useing the wrong node. .. return worker; } [Solution] There are four mappings in the kernel: 1. nodeid (logical node id) <-> pxm 2. apicid (physical cpu id) <-> nodeid 3. cpuid (logical cpu id) <-> apicid 4. cpuid (logical cpu id) <-> nodeid 1. pxm (proximity domain) is provided by ACPI firmware in SRAT, and nodeid <-> pxm mapping is setup at boot time. This mapping is persistent, won't change. 2. apicid <-> nodeid mapping is setup using info in 1. The mapping is setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. This mapping is also persistent. 3. cpuid <-> apicid mapping is setup at boot time and CPU hotadd time. cpuid is allocated, lower ids first, and released at CPU hotremove time, reused for other hotadded CPUs. So this mapping is not persistent. 4. cpuid <-> nodeid mapping is also setup at boot time and CPU hotadd time, and cleared at CPU hotremove time. As a result of 3, this mapping is not persistent. To fix this problem, we establish cpuid <-> nodeid mapping for all the possible cpus at boot time, and make it persistent. And according to init_cpu_to_node(), cpuid <-> nodeid mapping is based on apicid <-> nodeid mapping and cpuid <-> apicid mapping. So the key point is obtaining all cpus' apicid. apicid can be obtained by _MAT (Multiple APIC Table Entry) method or found in MADT (Multiple APIC Description Table). So we finish the job in the following steps: 1. Enable apic registeration flow to handle both enabled and disabled cpus. This is done by introducing an extra parameter to generic_processor_info to let the caller control if disabled cpus are ignored. 2. Introduce a new array storing all possible cpuid <-> apicid mapping. And also modify the way cpuid is calculated. Establish all possible cpuid <-> apicid mapping when registering local apic. Store the mapping in this array. 3. Enable _MAT and MADT relative apis to return non-presnet or disabled cpus' apicid. This is also done by introducing an extra parameter to these apis to let the caller control if disabled cpus are ignored. 4. Establish all possible cpuid <-> nodeid mapping. This is done via an additional acpi namespace walk for processors. For previous discussion, please refer to: https://lkml.org/lkml/2015/2/27/145 https://lkml.org/lkml/201
Re: [RFC] theoretical race between memory hotplug and pfn iterator
On 12/21/2015 03:17 PM, Joonsoo Kim wrote: On Mon, Dec 21, 2015 at 03:00:08PM +0800, Zhu Guihua wrote: On 12/21/2015 11:15 AM, Joonsoo Kim wrote: Hello, memory-hotplug folks. I found theoretical problems between memory hotplug and pfn iterator. For example, pfn iterator works something like below. for (pfn = zone_start_pfn; pfn < zone_end_pfn; pfn++) { if (!pfn_valid(pfn)) continue; page = pfn_to_page(pfn); /* Do whatever we want */ } Sequence of hotplug is something like below. 1) add memmap (after then, pfn_valid will return valid) 2) memmap_init_zone() So, if pfn iterator runs between 1) and 2), it could access uninitialized page information. This problem could be solved by re-ordering initialization steps. Hot-remove also has a problem. If memory is hot-removed after pfn_valid() succeed in pfn iterator, access to page would cause NULL deference because hot-remove frees corresponding memmap. There is no guard against free in any pfn iterators. This problem can be solved by inserting get_online_mems() in all pfn iterators but this looks error-prone for future usage. Another idea is that delaying free corresponding memmap until synchronization point such as system suspend. It will guarantee that there is no running pfn iterator. Do any have a better idea? Btw, I tried to memory-hotremove with QEMU 2.5.5 but it didn't work. I followed sequences in doc/memory-hotplug. Do you have any comment on this? I tried memory hot remove with qemu 2.5.5 and RHEL 7, it works well. Maybe you can provide more details, such as guest version, err log. I'm testing with qemu 2.5.5 and linux-next-20151209 with reverting following two patches. "mm/memblock.c: use memblock_insert_region() for the empty array" "mm-memblock-use-memblock_insert_region-for-the-empty-array-checkpatch-fixes" When I type "device_del dimm1" in qemu monitor, there is no err log in kernel and it looks like command has no effect. I inserted log to acpi_memory_device_remove() but there is no message, too. Is there another way to check that device_del event is actually transmitted to kernel? You can use udev to monitor memory device remove event. (udevadm monitor) I launch the qemu with following command. ./qemu-system-x86_64-recent -enable-kvm -smp 8 -m 4096,slots=16,maxmem=8G ... Thanks. . -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [RFC] theoretical race between memory hotplug and pfn iterator
On 12/21/2015 11:15 AM, Joonsoo Kim wrote: Hello, memory-hotplug folks. I found theoretical problems between memory hotplug and pfn iterator. For example, pfn iterator works something like below. for (pfn = zone_start_pfn; pfn < zone_end_pfn; pfn++) { if (!pfn_valid(pfn)) continue; page = pfn_to_page(pfn); /* Do whatever we want */ } Sequence of hotplug is something like below. 1) add memmap (after then, pfn_valid will return valid) 2) memmap_init_zone() So, if pfn iterator runs between 1) and 2), it could access uninitialized page information. This problem could be solved by re-ordering initialization steps. Hot-remove also has a problem. If memory is hot-removed after pfn_valid() succeed in pfn iterator, access to page would cause NULL deference because hot-remove frees corresponding memmap. There is no guard against free in any pfn iterators. This problem can be solved by inserting get_online_mems() in all pfn iterators but this looks error-prone for future usage. Another idea is that delaying free corresponding memmap until synchronization point such as system suspend. It will guarantee that there is no running pfn iterator. Do any have a better idea? Btw, I tried to memory-hotremove with QEMU 2.5.5 but it didn't work. I followed sequences in doc/memory-hotplug. Do you have any comment on this? I tried memory hot remove with qemu 2.5.5 and RHEL 7, it works well. Maybe you can provide more details, such as guest version, err log. Thanks, Zhu Thanks. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ . -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[tip:x86/urgent] x86/espfix: Init espfix on the boot CPU side
Commit-ID: 20d5e4a9cd453991e2590a4c25230a99b42dee0c Gitweb: http://git.kernel.org/tip/20d5e4a9cd453991e2590a4c25230a99b42dee0c Author: Zhu Guihua AuthorDate: Fri, 3 Jul 2015 17:37:19 +0800 Committer: Ingo Molnar CommitDate: Mon, 6 Jul 2015 15:00:34 +0200 x86/espfix: Init espfix on the boot CPU side As we alloc pages with GFP_KERNEL in init_espfix_ap() which is called before we enable local irqs, so the lockdep sub-system would (correctly) trigger a warning about the potentially blocking API. So we allocate them on the boot CPU side when the secondary CPU is brought up by the boot CPU, and hand them over to the secondary CPU. And we use alloc_pages_node() with the secondary CPU's node, to make sure the espfix stack is NUMA-local to the CPU that is going to use it. Signed-off-by: Zhu Guihua Cc: Cc: Cc: Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Link: http://lkml.kernel.org/r/c97add2670e9abebb90095369f0cfc172373ac94.1435824469.git.zhugh.f...@cn.fujitsu.com Signed-off-by: Ingo Molnar --- arch/x86/kernel/espfix_64.c | 21 + arch/x86/kernel/smpboot.c | 14 +++--- 2 files changed, 20 insertions(+), 15 deletions(-) diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index 1fb0e00..ce95676 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -141,12 +141,12 @@ void init_espfix_ap(int cpu) pud_t pud, *pud_p; pmd_t pmd, *pmd_p; pte_t pte, *pte_p; - int n; + int n, node; void *stack_page; pteval_t ptemask; /* We only have to do this once... */ - if (likely(this_cpu_read(espfix_stack))) + if (likely(per_cpu(espfix_stack, cpu))) return; /* Already initialized */ addr = espfix_base_addr(cpu); @@ -164,12 +164,15 @@ void init_espfix_ap(int cpu) if (stack_page) goto unlock_done; + node = cpu_to_node(cpu); ptemask = __supported_pte_mask; pud_p = &espfix_pud_page[pud_index(addr)]; pud = *pud_p; if (!pud_present(pud)) { - pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pmd_p = (pmd_t *)page_address(page); pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pmd(&init_mm, __pa(pmd_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PUD_CLONES; n++) @@ -179,7 +182,9 @@ void init_espfix_ap(int cpu) pmd_p = pmd_offset(&pud, addr); pmd = *pmd_p; if (!pmd_present(pmd)) { - pte_p = (pte_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pte_p = (pte_t *)page_address(page); pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pte(&init_mm, __pa(pte_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PMD_CLONES; n++) @@ -187,7 +192,7 @@ void init_espfix_ap(int cpu) } pte_p = pte_offset_kernel(&pmd, addr); - stack_page = (void *)__get_free_page(GFP_KERNEL); + stack_page = page_address(alloc_pages_node(node, GFP_KERNEL, 0)); pte = __pte(__pa(stack_page) | (__PAGE_KERNEL_RO & ptemask)); for (n = 0; n < ESPFIX_PTE_CLONES; n++) set_pte(&pte_p[n*PTE_STRIDE], pte); @@ -198,7 +203,7 @@ void init_espfix_ap(int cpu) unlock_done: mutex_unlock(&espfix_init_mutex); done: - this_cpu_write(espfix_stack, addr); - this_cpu_write(espfix_waddr, (unsigned long)stack_page - + (addr & ~PAGE_MASK)); + per_cpu(espfix_stack, cpu) = addr; + per_cpu(espfix_waddr, cpu) = (unsigned long)stack_page + + (addr & ~PAGE_MASK); } diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index 6f5abac..0bd8c1d 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -239,13 +239,6 @@ static void notrace start_secondary(void *unused) check_tsc_sync_target(); /* -* Enable the espfix hack for this CPU -*/ -#ifdef CONFIG_X86_ESPFIX64 - init_espfix_ap(smp_processor_id()); -#endif - - /* * We need to hold vector_lock so there the set of online cpus * does not change while we are assigning vectors to cpus. Holding * this lock ensures we don't half assign or remove an irq from a cpu. @@ -854,6 +847,13 @@ static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle) initial_code = (unsigned long)start_secondary; stack_start = idle->thread.sp; + /* +* Enable the espfix hack for this CPU + */ +#ifdef CONFIG_X86_ESPFIX64 + init_espfix_ap(cpu); +#endif + /* So we see what's up
[tip:x86/urgent] x86/espfix: Add 'cpu' parameter to init_espfix_ap()
Commit-ID: 1db875631f8d5bbf497f67e47f254eece888d51d Gitweb: http://git.kernel.org/tip/1db875631f8d5bbf497f67e47f254eece888d51d Author: Zhu Guihua AuthorDate: Fri, 3 Jul 2015 17:37:18 +0800 Committer: Ingo Molnar CommitDate: Mon, 6 Jul 2015 15:00:33 +0200 x86/espfix: Add 'cpu' parameter to init_espfix_ap() Add a CPU index parameter to init_espfix_ap(), so that the parameter could be propagated to the function for espfix page allocation. Signed-off-by: Zhu Guihua Cc: Cc: Cc: Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Link: http://lkml.kernel.org/r/cde3fcf1b3211f3f03feb1a995bce3fee850f0fc.1435824469.git.zhugh.f...@cn.fujitsu.com Signed-off-by: Ingo Molnar --- arch/x86/include/asm/espfix.h | 2 +- arch/x86/kernel/espfix_64.c | 7 +++ arch/x86/kernel/smpboot.c | 2 +- 3 files changed, 5 insertions(+), 6 deletions(-) diff --git a/arch/x86/include/asm/espfix.h b/arch/x86/include/asm/espfix.h index 99efebb..ca3ce9a 100644 --- a/arch/x86/include/asm/espfix.h +++ b/arch/x86/include/asm/espfix.h @@ -9,7 +9,7 @@ DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack); DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr); extern void init_espfix_bsp(void); -extern void init_espfix_ap(void); +extern void init_espfix_ap(int cpu); #endif /* CONFIG_X86_64 */ diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index f5d0730..1fb0e00 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -131,12 +131,12 @@ void __init init_espfix_bsp(void) init_espfix_random(); /* The rest is the same as for any other processor */ - init_espfix_ap(); + init_espfix_ap(0); } -void init_espfix_ap(void) +void init_espfix_ap(int cpu) { - unsigned int cpu, page; + unsigned int page; unsigned long addr; pud_t pud, *pud_p; pmd_t pmd, *pmd_p; @@ -149,7 +149,6 @@ void init_espfix_ap(void) if (likely(this_cpu_read(espfix_stack))) return; /* Already initialized */ - cpu = smp_processor_id(); addr = espfix_base_addr(cpu); page = cpu/ESPFIX_STACKS_PER_PAGE; diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index 8add66b..6f5abac 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -242,7 +242,7 @@ static void notrace start_secondary(void *unused) * Enable the espfix hack for this CPU */ #ifdef CONFIG_X86_ESPFIX64 - init_espfix_ap(); + init_espfix_ap(smp_processor_id()); #endif /* -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH v3 0/2] x86, espfix: silence warning on espfix initialization for ap
The following lockdep warning occurrs when running with latest kernel: [3.178000] [ cut here ] [3.183000] WARNING: CPU: 128 PID: 0 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0xdd/0xe0() [3.193000] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) [3.199000] Modules linked in: [3.203000] CPU: 128 PID: 0 Comm: swapper/128 Not tainted 4.1.0-rc3 #70 [3.221000] 2d6601fb3e6d4e4c 88086fd5fc38 81773f0a [3.23] 88086fd5fc90 88086fd5fc78 8108c85a [3.238000] 88086fd6 0092 88086fd6 00d0 [3.246000] Call Trace: [3.249000] [] dump_stack+0x4c/0x65 [3.255000] [] warn_slowpath_common+0x8a/0xc0 [3.261000] [] warn_slowpath_fmt+0x55/0x70 [3.268000] [] lockdep_trace_alloc+0xdd/0xe0 [3.274000] [] __alloc_pages_nodemask+0xad/0xca0 [3.281000] [] ? __lock_acquire+0xf6d/0x1560 [3.288000] [] alloc_page_interleave+0x3a/0x90 [3.295000] [] alloc_pages_current+0x17d/0x1a0 [3.301000] [] ? __get_free_pages+0xe/0x50 [3.308000] [] __get_free_pages+0xe/0x50 [3.314000] [] init_espfix_ap+0x17b/0x320 [3.32] [] start_secondary+0xf1/0x1f0 [3.327000] ---[ end trace 1b3327d9d6a1d62c ]--- As we alloc pages with GFP_KERNEL in init_espfix_ap() which is called before enabled local irq, and the lockdep sub-system considers this behaviour as allocating memory with GFP_FS with local irq disabled, then trigger the warning as mentioned about. So we allocate them on the boot CPU side when the target CPU is bringing up by the primary CPU, and hand them over to the secondary CPU. v3: -split the original path into two parts v2: -allocate espfix stack pages when the targert CPU is bringing up by the primary CPU -commit messages changed v1: -Alloc the page on the node the target CPU is on. Zhu Guihua (2): espfix: add cpu parameter to init_espfix_ap() x86, espfix: init espfix on the boot cpu side arch/x86/include/asm/espfix.h | 2 +- arch/x86/kernel/espfix_64.c | 28 arch/x86/kernel/smpboot.c | 14 +++--- 3 files changed, 24 insertions(+), 20 deletions(-) -- 1.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH v3 1/2] espfix: add cpu parameter to init_espfix_ap()
Add cpu index parameter to init_espfix_ap(), so that the parameter could be propagated to the function for pages allocation. Signed-off-by: Zhu Guihua --- arch/x86/include/asm/espfix.h | 2 +- arch/x86/kernel/espfix_64.c | 7 +++ arch/x86/kernel/smpboot.c | 2 +- 3 files changed, 5 insertions(+), 6 deletions(-) diff --git a/arch/x86/include/asm/espfix.h b/arch/x86/include/asm/espfix.h index 99efebb..ca3ce9a 100644 --- a/arch/x86/include/asm/espfix.h +++ b/arch/x86/include/asm/espfix.h @@ -9,7 +9,7 @@ DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack); DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr); extern void init_espfix_bsp(void); -extern void init_espfix_ap(void); +extern void init_espfix_ap(int cpu); #endif /* CONFIG_X86_64 */ diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index f5d0730..1fb0e00 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -131,12 +131,12 @@ void __init init_espfix_bsp(void) init_espfix_random(); /* The rest is the same as for any other processor */ - init_espfix_ap(); + init_espfix_ap(0); } -void init_espfix_ap(void) +void init_espfix_ap(int cpu) { - unsigned int cpu, page; + unsigned int page; unsigned long addr; pud_t pud, *pud_p; pmd_t pmd, *pmd_p; @@ -149,7 +149,6 @@ void init_espfix_ap(void) if (likely(this_cpu_read(espfix_stack))) return; /* Already initialized */ - cpu = smp_processor_id(); addr = espfix_base_addr(cpu); page = cpu/ESPFIX_STACKS_PER_PAGE; diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index 8add66b..6f5abac 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -242,7 +242,7 @@ static void notrace start_secondary(void *unused) * Enable the espfix hack for this CPU */ #ifdef CONFIG_X86_ESPFIX64 - init_espfix_ap(); + init_espfix_ap(smp_processor_id()); #endif /* -- 1.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH v3 2/2] x86, espfix: init espfix on the boot cpu side
As we alloc pages with GFP_KERNEL in init_espfix_ap() which is called before enabled local irq, the lockdep sub-system would trigger a warning. So we allocate them on the boot CPU side when the target CPU is bringing up by the primary CPU, and hand them over to the secondary CPU. And we use alloc_pages_node() with the secondary CPU's node, to make sure the espfix stack is NUMA-local to the CPU that is going to use it. Signed-off-by: Zhu Guihua --- arch/x86/kernel/espfix_64.c | 21 + arch/x86/kernel/smpboot.c | 14 +++--- 2 files changed, 20 insertions(+), 15 deletions(-) diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index 1fb0e00..ce95676 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -141,12 +141,12 @@ void init_espfix_ap(int cpu) pud_t pud, *pud_p; pmd_t pmd, *pmd_p; pte_t pte, *pte_p; - int n; + int n, node; void *stack_page; pteval_t ptemask; /* We only have to do this once... */ - if (likely(this_cpu_read(espfix_stack))) + if (likely(per_cpu(espfix_stack, cpu))) return; /* Already initialized */ addr = espfix_base_addr(cpu); @@ -164,12 +164,15 @@ void init_espfix_ap(int cpu) if (stack_page) goto unlock_done; + node = cpu_to_node(cpu); ptemask = __supported_pte_mask; pud_p = &espfix_pud_page[pud_index(addr)]; pud = *pud_p; if (!pud_present(pud)) { - pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pmd_p = (pmd_t *)page_address(page); pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pmd(&init_mm, __pa(pmd_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PUD_CLONES; n++) @@ -179,7 +182,9 @@ void init_espfix_ap(int cpu) pmd_p = pmd_offset(&pud, addr); pmd = *pmd_p; if (!pmd_present(pmd)) { - pte_p = (pte_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pte_p = (pte_t *)page_address(page); pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pte(&init_mm, __pa(pte_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PMD_CLONES; n++) @@ -187,7 +192,7 @@ void init_espfix_ap(int cpu) } pte_p = pte_offset_kernel(&pmd, addr); - stack_page = (void *)__get_free_page(GFP_KERNEL); + stack_page = page_address(alloc_pages_node(node, GFP_KERNEL, 0)); pte = __pte(__pa(stack_page) | (__PAGE_KERNEL_RO & ptemask)); for (n = 0; n < ESPFIX_PTE_CLONES; n++) set_pte(&pte_p[n*PTE_STRIDE], pte); @@ -198,7 +203,7 @@ void init_espfix_ap(int cpu) unlock_done: mutex_unlock(&espfix_init_mutex); done: - this_cpu_write(espfix_stack, addr); - this_cpu_write(espfix_waddr, (unsigned long)stack_page - + (addr & ~PAGE_MASK)); + per_cpu(espfix_stack, cpu) = addr; + per_cpu(espfix_waddr, cpu) = (unsigned long)stack_page + + (addr & ~PAGE_MASK); } diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c index 6f5abac..0bd8c1d 100644 --- a/arch/x86/kernel/smpboot.c +++ b/arch/x86/kernel/smpboot.c @@ -239,13 +239,6 @@ static void notrace start_secondary(void *unused) check_tsc_sync_target(); /* -* Enable the espfix hack for this CPU -*/ -#ifdef CONFIG_X86_ESPFIX64 - init_espfix_ap(smp_processor_id()); -#endif - - /* * We need to hold vector_lock so there the set of online cpus * does not change while we are assigning vectors to cpus. Holding * this lock ensures we don't half assign or remove an irq from a cpu. @@ -854,6 +847,13 @@ static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle) initial_code = (unsigned long)start_secondary; stack_start = idle->thread.sp; + /* +* Enable the espfix hack for this CPU + */ +#ifdef CONFIG_X86_ESPFIX64 + init_espfix_ap(cpu); +#endif + /* So we see what's up */ announce_cpu(cpu, apicid); -- 1.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH v2] x86, espfix: init espfix on the boot cpu side
The following lockdep warning occurrs when running with latest kernel: [3.178000] [ cut here ] [3.183000] WARNING: CPU: 128 PID: 0 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0xdd/0xe0() [3.193000] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) [3.199000] Modules linked in: [3.203000] CPU: 128 PID: 0 Comm: swapper/128 Not tainted 4.1.0-rc3 #70 [3.221000] 2d6601fb3e6d4e4c 88086fd5fc38 81773f0a [3.23] 88086fd5fc90 88086fd5fc78 8108c85a [3.238000] 88086fd6 0092 88086fd6 00d0 [3.246000] Call Trace: [3.249000] [] dump_stack+0x4c/0x65 [3.255000] [] warn_slowpath_common+0x8a/0xc0 [3.261000] [] warn_slowpath_fmt+0x55/0x70 [3.268000] [] lockdep_trace_alloc+0xdd/0xe0 [3.274000] [] __alloc_pages_nodemask+0xad/0xca0 [3.281000] [] ? __lock_acquire+0xf6d/0x1560 [3.288000] [] alloc_page_interleave+0x3a/0x90 [3.295000] [] alloc_pages_current+0x17d/0x1a0 [3.301000] [] ? __get_free_pages+0xe/0x50 [3.308000] [] __get_free_pages+0xe/0x50 [3.314000] [] init_espfix_ap+0x17b/0x320 [3.32] [] start_secondary+0xf1/0x1f0 [3.327000] ---[ end trace 1b3327d9d6a1d62c ]--- As we alloc pages with GFP_KERNEL in init_espfix_ap() which is called before enabled local irq, and the lockdep sub-system considers this behaviour as allocating memory with GFP_FS with local irq disabled, then trigger the warning as mentioned about. So we allocate them on the boot CPU side when the target CPU is bringing up by the primary CPU, and hand them over to the secondary CPU. Signed-off-by: Zhu Guihua --- v2: -allocate espfix stack pages when the targert CPU is bringing up by the primary CPU -commit messages changed v1: -Alloc the page on the node the target CPU is on. RFC v2: -Let the boot up routine init the espfix stack for the target cpu after it booted. --- arch/x86/include/asm/espfix.h | 2 +- arch/x86/kernel/espfix_64.c | 28 arch/x86/kernel/smpboot.c | 14 +++--- 3 files changed, 24 insertions(+), 20 deletions(-) diff --git a/arch/x86/include/asm/espfix.h b/arch/x86/include/asm/espfix.h index 99efebb..ca3ce9a 100644 --- a/arch/x86/include/asm/espfix.h +++ b/arch/x86/include/asm/espfix.h @@ -9,7 +9,7 @@ DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack); DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr); extern void init_espfix_bsp(void); -extern void init_espfix_ap(void); +extern void init_espfix_ap(int cpu); #endif /* CONFIG_X86_64 */ diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index f5d0730..f6dd9df 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -131,25 +131,24 @@ void __init init_espfix_bsp(void) init_espfix_random(); /* The rest is the same as for any other processor */ - init_espfix_ap(); + init_espfix_ap(0); } -void init_espfix_ap(void) +void init_espfix_ap(int cpu) { - unsigned int cpu, page; + unsigned int page; unsigned long addr; pud_t pud, *pud_p; pmd_t pmd, *pmd_p; pte_t pte, *pte_p; - int n; + int n, node; void *stack_page; pteval_t ptemask; /* We only have to do this once... */ - if (likely(this_cpu_read(espfix_stack))) + if (likely(per_cpu(espfix_stack, cpu))) return; /* Already initialized */ - cpu = smp_processor_id(); addr = espfix_base_addr(cpu); page = cpu/ESPFIX_STACKS_PER_PAGE; @@ -165,12 +164,15 @@ void init_espfix_ap(void) if (stack_page) goto unlock_done; + node = cpu_to_node(cpu); ptemask = __supported_pte_mask; pud_p = &espfix_pud_page[pud_index(addr)]; pud = *pud_p; if (!pud_present(pud)) { - pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pmd_p = (pmd_t *)page_address(page); pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pmd(&init_mm, __pa(pmd_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PUD_CLONES; n++) @@ -180,7 +182,9 @@ void init_espfix_ap(void) pmd_p = pmd_offset(&pud, addr); pmd = *pmd_p; if (!pmd_present(pmd)) { - pte_p = (pte_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pte_p = (pte_t *)page_address(page); pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pte(&init_mm, __pa(pte_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PMD_CLONES; n++) @@ -188,7 +192,7 @@ void init_espfix_ap(void) }
Re: [PATCH V1] x86, espfix: postpone the initialization of espfix stack for AP
Any feedback about this? On 06/04/2015 05:45 PM, Gu Zheng wrote: The following lockdep warning occurrs when running with latest kernel: [3.178000] [ cut here ] [3.183000] WARNING: CPU: 128 PID: 0 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0xdd/0xe0() [3.193000] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) [3.199000] Modules linked in: [3.203000] CPU: 128 PID: 0 Comm: swapper/128 Not tainted 4.1.0-rc3 #70 [3.221000] 2d6601fb3e6d4e4c 88086fd5fc38 81773f0a [3.23] 88086fd5fc90 88086fd5fc78 8108c85a [3.238000] 88086fd6 0092 88086fd6 00d0 [3.246000] Call Trace: [3.249000] [] dump_stack+0x4c/0x65 [3.255000] [] warn_slowpath_common+0x8a/0xc0 [3.261000] [] warn_slowpath_fmt+0x55/0x70 [3.268000] [] lockdep_trace_alloc+0xdd/0xe0 [3.274000] [] __alloc_pages_nodemask+0xad/0xca0 [3.281000] [] ? __lock_acquire+0xf6d/0x1560 [3.288000] [] alloc_page_interleave+0x3a/0x90 [3.295000] [] alloc_pages_current+0x17d/0x1a0 [3.301000] [] ? __get_free_pages+0xe/0x50 [3.308000] [] __get_free_pages+0xe/0x50 [3.314000] [] init_espfix_ap+0x17b/0x320 [3.32] [] start_secondary+0xf1/0x1f0 [3.327000] ---[ end trace 1b3327d9d6a1d62c ]--- As we alloc pages with GFP_KERNEL in init_espfix_ap() which is called before enabled local irq, and the lockdep sub-system considers this behaviour as allocating memory with GFP_FS with local irq disabled, then trigger the warning as mentioned about. Though we could allocate them on the boot CPU side and hand them over to the secondary CPU, but it seemes a bit waste if some of cpus are offline. As thers is no need to these pages(espfix stack) until we try to run user code, so we postpone the initialization of espfix stack, and let the boot up routine init the espfix stack for the target cpu after it booted to avoid the noise. Signed-off-by: Gu Zheng --- v1: Alloc the page on the node the target CPU is on. RFC: Let the boot up routine init the espfix stack for the target cpu after it booted. --- --- arch/x86/include/asm/espfix.h |2 +- arch/x86/kernel/espfix_64.c | 28 arch/x86/kernel/smpboot.c | 14 +++--- 3 files changed, 24 insertions(+), 20 deletions(-) diff --git a/arch/x86/include/asm/espfix.h b/arch/x86/include/asm/espfix.h index 99efebb..ca3ce9a 100644 --- a/arch/x86/include/asm/espfix.h +++ b/arch/x86/include/asm/espfix.h @@ -9,7 +9,7 @@ DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack); DECLARE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr); extern void init_espfix_bsp(void); -extern void init_espfix_ap(void); +extern void init_espfix_ap(int cpu); #endif /* CONFIG_X86_64 */ diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c index f5d0730..e397583 100644 --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -131,25 +131,24 @@ void __init init_espfix_bsp(void) init_espfix_random(); /* The rest is the same as for any other processor */ - init_espfix_ap(); + init_espfix_ap(0); } -void init_espfix_ap(void) +void init_espfix_ap(int cpu) { - unsigned int cpu, page; + unsigned int page; unsigned long addr; pud_t pud, *pud_p; pmd_t pmd, *pmd_p; pte_t pte, *pte_p; - int n; + int n, node; void *stack_page; pteval_t ptemask; /* We only have to do this once... */ - if (likely(this_cpu_read(espfix_stack))) + if (likely(per_cpu(espfix_stack, cpu))) return; /* Already initialized */ - cpu = smp_processor_id(); addr = espfix_base_addr(cpu); page = cpu/ESPFIX_STACKS_PER_PAGE; @@ -165,12 +164,15 @@ void init_espfix_ap(void) if (stack_page) goto unlock_done; + node = cpu_to_node(cpu); ptemask = __supported_pte_mask; pud_p = &espfix_pud_page[pud_index(addr)]; pud = *pud_p; if (!pud_present(pud)) { - pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pmd_p = (pmd_t *)page_address(page); pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc_pmd(&init_mm, __pa(pmd_p) >> PAGE_SHIFT); for (n = 0; n < ESPFIX_PUD_CLONES; n++) @@ -180,7 +182,9 @@ void init_espfix_ap(void) pmd_p = pmd_offset(&pud, addr); pmd = *pmd_p; if (!pmd_present(pmd)) { - pte_p = (pte_t *)__get_free_page(PGALLOC_GFP); + struct page *page = alloc_pages_node(node, PGALLOC_GFP, 0); + + pte_p = (pte_t *)page_address(page); pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask)); paravirt_alloc
Re: [PATCH] mm/memory hotplug: print the last vmemmap region at the end of hot add memory
On 06/10/2015 04:29 AM, Andrew Morton wrote: On Tue, 9 Jun 2015 11:41:28 +0800 Zhu Guihua wrote: --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -513,6 +513,7 @@ int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, break; err = 0; } + vmemmap_populate_print_last(); return err; } vmemmap_populate_print_last() is only available on x86_64, when CONFIG_SPARSEMEM_VMEMMAP=y. Are you sure this won't break builds? I tried this on i386 and on x86_64 when CONFIG_SPARSEMEM_VMEMMAP=n , it builds ok. With powerpc: akpm3:/usr/src/25> make allmodconfig akpm3:/usr/src/25> make mm/memory_hotplug.o akpm3:/usr/src/25> nm mm/memory_hotplug.o | grep vmemmap_populate_print_last U .vmemmap_populate_print_last akpm3:/usr/src/25> grep -r vmemmap_populate_print_last arch/powerpc akpm3:/usr/src/25> So I think that's going to break. I expect ia64 will break also, but I didn't investigate. . There is void __weak __meminit vmemmap_populate_print last(void) in /mm/sparse.c, so I think this won't break builds. And I found the function was invoked in void __init sparse_init(void) without CONFIG_SPARSEMEM_VMEMMAP=y. I also tried this on arm, it builds ok too. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] mm/memory hotplug: print the last vmemmap region at the end of hot add memory
On 06/09/2015 07:30 AM, Andrew Morton wrote: On Mon, 8 Jun 2015 14:44:41 +0800 Zhu Guihua wrote: When hot add two nodes continuously, we found the vmemmap region info is a bit messed. The last region of node 2 is printed when node 3 hot added, like the following: Initmem setup node 2 [mem 0x-0x] On node 2 totalpages: 0 Built 2 zonelists in Node order, mobility grouping on. Total pages: 16090539 Policy zone: Normal init_memory_mapping: [mem 0x400-0x407] [mem 0x400-0x407] page 1G [ea10-ea10001f] PMD -> [8a077d80-8a077d9f] on node 2 [ea100020-ea10003f] PMD -> [8a077de0-8a077dff] on node 2 ... [ea101f60-ea101f9f] PMD -> [8a074ac0-8a074aff] on node 2 [ea101fa0-ea101fdf] PMD -> [8a074a80-8a074abf] on node 2 Initmem setup node 3 [mem 0x-0x] On node 3 totalpages: 0 Built 3 zonelists in Node order, mobility grouping on. Total pages: 16090539 Policy zone: Normal init_memory_mapping: [mem 0x600-0x607] [mem 0x600-0x607] page 1G [ea101fe0-ea101fff] PMD -> [8a074a40-8a074a5f] on node 2 <=== node 2 ??? [ea18-ea18001f] PMD -> [8a074a60-8a074a7f] on node 3 [ea180020-ea18005f] PMD -> [8a074a00-8a074a3f] on node 3 [ea180060-ea18009f] PMD -> [8a0749c0-8a0749ff] on node 3 ... The cause is the last region was missed at the and of hot add memory, and p_start, p_end, node_start were not reset, so when hot add memory to a new node, it will consider they are not contiguous blocks and print the previous one. So we print the last vmemmap region at the end of hot add memory to avoid the confusion. ... --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -513,6 +513,7 @@ int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, break; err = 0; } + vmemmap_populate_print_last(); return err; } vmemmap_populate_print_last() is only available on x86_64, when CONFIG_SPARSEMEM_VMEMMAP=y. Are you sure this won't break builds? I tried this on i386 and on x86_64 when CONFIG_SPARSEMEM_VMEMMAP=n , it builds ok. Thanks, Zhu . -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[PATCH] mm/memory hotplug: print the last vmemmap region at the end of hot add memory
When hot add two nodes continuously, we found the vmemmap region info is a bit messed. The last region of node 2 is printed when node 3 hot added, like the following: Initmem setup node 2 [mem 0x-0x] On node 2 totalpages: 0 Built 2 zonelists in Node order, mobility grouping on. Total pages: 16090539 Policy zone: Normal init_memory_mapping: [mem 0x400-0x407] [mem 0x400-0x407] page 1G [ea10-ea10001f] PMD -> [8a077d80-8a077d9f] on node 2 [ea100020-ea10003f] PMD -> [8a077de0-8a077dff] on node 2 ... [ea101f60-ea101f9f] PMD -> [8a074ac0-8a074aff] on node 2 [ea101fa0-ea101fdf] PMD -> [8a074a80-8a074abf] on node 2 Initmem setup node 3 [mem 0x-0x] On node 3 totalpages: 0 Built 3 zonelists in Node order, mobility grouping on. Total pages: 16090539 Policy zone: Normal init_memory_mapping: [mem 0x600-0x607] [mem 0x600-0x607] page 1G [ea101fe0-ea101fff] PMD -> [8a074a40-8a074a5f] on node 2 <=== node 2 ??? [ea18-ea18001f] PMD -> [8a074a60-8a074a7f] on node 3 [ea180020-ea18005f] PMD -> [8a074a00-8a074a3f] on node 3 [ea180060-ea18009f] PMD -> [8a0749c0-8a0749ff] on node 3 ... The cause is the last region was missed at the and of hot add memory, and p_start, p_end, node_start were not reset, so when hot add memory to a new node, it will consider they are not contiguous blocks and print the previous one. So we print the last vmemmap region at the end of hot add memory to avoid the confusion. Signed-off-by: Zhu Guihua --- mm/memory_hotplug.c | 1 + 1 file changed, 1 insertion(+) diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index 457bde5..58fb223 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -513,6 +513,7 @@ int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, break; err = 0; } + vmemmap_populate_print_last(); return err; } -- 1.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/