From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754749AbZIVHlW (ORCPT ); Tue, 22 Sep 2009 03:41:22 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753997AbZIVHkv (ORCPT ); Tue, 22 Sep 2009 03:40:51 -0400 Received: from hera.kernel.org ([140.211.167.34]:44582 "EHLO hera.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752250AbZIVHkr (ORCPT ); Tue, 22 Sep 2009 03:40:47 -0400 From: Tejun Heo To: Nick Piggin , Tony Luck , Fenghua Yu , linux-ia64 , Ingo Molnar , Rusty Russell , Christoph Lameter , linux-kernel@vger.kernel.org Cc: Tejun Heo , Tony Luck , Fenghua Yu , linux-ia64 Subject: [PATCH 3/4] ia64: convert to dynamic percpu allocator Date: Tue, 22 Sep 2009 16:40:13 +0900 Message-Id: <1253605214-23210-4-git-send-email-tj@kernel.org> X-Mailer: git-send-email 1.6.4.2 In-Reply-To: <1253605214-23210-1-git-send-email-tj@kernel.org> References: <1253605214-23210-1-git-send-email-tj@kernel.org> X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.0 (hera.kernel.org [127.0.0.1]); Tue, 22 Sep 2009 07:40:30 +0000 (UTC) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Unlike other archs, ia64 reserves space for percpu areas during early memory initialization. These areas occupy a contiguous region indexed by cpu number on contiguous memory model or are grouped by node on discontiguous memory model. As allocation and initialization are done by the arch code, all that setup_per_cpu_areas() needs to do is communicating the determined layout to the percpu allocator. This patch implements setup_per_cpu_areas() for both contig and discontig memory models and drops HAVE_LEGACY_PER_CPU_AREA. Please note that for contig model, the allocation itself is modified only to allocate for possible cpus instead of NR_CPUS. As dynamic percpu allocator can handle non-direct mapping, there's no reason to allocate memory for cpus which aren't possible. Signed-off-by: Tejun Heo Cc: Tony Luck Cc: Fenghua Yu Cc: linux-ia64 --- arch/ia64/Kconfig | 3 -- arch/ia64/kernel/setup.c | 12 ------ arch/ia64/mm/contig.c | 50 ++++++++++++++++++++++++--- arch/ia64/mm/discontig.c | 85 ++++++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 130 insertions(+), 20 deletions(-) diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig index 011a1cd..e624611 100644 --- a/arch/ia64/Kconfig +++ b/arch/ia64/Kconfig @@ -89,9 +89,6 @@ config GENERIC_TIME_VSYSCALL bool default y -config HAVE_LEGACY_PER_CPU_AREA - def_bool y - config HAVE_SETUP_PER_CPU_AREA def_bool y diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c index 1de86c9..42f8a18 100644 --- a/arch/ia64/kernel/setup.c +++ b/arch/ia64/kernel/setup.c @@ -856,18 +856,6 @@ identify_cpu (struct cpuinfo_ia64 *c) } /* - * In UP configuration, setup_per_cpu_areas() is defined in - * include/linux/percpu.h - */ -#ifdef CONFIG_SMP -void __init -setup_per_cpu_areas (void) -{ - /* start_kernel() requires this... */ -} -#endif - -/* * Do the following calculations: * * 1. the max. cache line size. diff --git a/arch/ia64/mm/contig.c b/arch/ia64/mm/contig.c index 9493bbf..c86ce4f 100644 --- a/arch/ia64/mm/contig.c +++ b/arch/ia64/mm/contig.c @@ -163,11 +163,11 @@ per_cpu_init (void) first_time = false; /* - * get_free_pages() cannot be used before cpu_init() done. BSP - * allocates "NR_CPUS" pages for all CPUs to avoid that AP calls - * get_zeroed_page(). + * get_free_pages() cannot be used before cpu_init() done. + * BSP allocates PERCPU_PAGE_SIZE bytes for all possible CPUs + * to avoid that AP calls get_zeroed_page(). */ - for (cpu = 0; cpu < NR_CPUS; cpu++) { + for_each_possible_cpu(cpu) { void *src = cpu == 0 ? cpu0_data : __phys_per_cpu_start; memcpy(cpu_data, src, __per_cpu_end - __per_cpu_start); @@ -196,9 +196,49 @@ skip: static inline void alloc_per_cpu_data(void) { - cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, + cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * num_possible_cpus(), PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); } + +/** + * setup_per_cpu_areas - setup percpu areas + * + * Arch code has already allocated and initialized percpu areas. All + * this function has to do is to teach the determined layout to the + * dynamic percpu allocator, which happens to be more complex than + * creating whole new ones using helpers. + */ +void __init +setup_per_cpu_areas(void) +{ + struct pcpu_alloc_info *ai; + struct pcpu_group_info *gi; + unsigned int cpu; + int rc; + + ai = pcpu_alloc_alloc_info(1, num_possible_cpus()); + if (!ai) + panic("failed to allocate pcpu_alloc_info"); + gi = &ai->groups[0]; + + /* units are assigned consecutively to possible cpus */ + for_each_possible_cpu(cpu) + gi->cpu_map[gi->nr_units++] = cpu; + + /* set parameters */ + ai->static_size = __per_cpu_end - __per_cpu_start; + ai->reserved_size = PERCPU_MODULE_RESERVE; + ai->dyn_size = PERCPU_DYNAMIC_RESERVE; + ai->unit_size = PERCPU_PAGE_SIZE; + ai->atom_size = PAGE_SIZE; + ai->alloc_size = PERCPU_PAGE_SIZE; + + rc = pcpu_setup_first_chunk(ai, __per_cpu_start + __per_cpu_offset[0]); + if (rc) + panic("failed to setup percpu area (err=%d)", rc); + + pcpu_free_alloc_info(ai); +} #else #define alloc_per_cpu_data() do { } while (0) #endif /* CONFIG_SMP */ diff --git a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c index 35a61ec..69e9e91 100644 --- a/arch/ia64/mm/discontig.c +++ b/arch/ia64/mm/discontig.c @@ -172,6 +172,91 @@ static void *per_cpu_node_setup(void *cpu_data, int node) return cpu_data; } +#ifdef CONFIG_SMP +/** + * setup_per_cpu_areas - setup percpu areas + * + * Arch code has already allocated and initialized percpu areas. All + * this function has to do is to teach the determined layout to the + * dynamic percpu allocator, which happens to be more complex than + * creating whole new ones using helpers. + */ +void __init setup_per_cpu_areas(void) +{ + struct pcpu_alloc_info *ai; + struct pcpu_group_info *uninitialized_var(gi); + unsigned int *cpu_map; + void *base; + unsigned long base_offset; + unsigned int cpu; + ssize_t static_size, reserved_size, dyn_size; + int node, prev_node, unit, nr_units, rc; + + ai = pcpu_alloc_alloc_info(MAX_NUMNODES, nr_cpu_ids); + if (!ai) + panic("failed to allocate pcpu_alloc_info"); + cpu_map = ai->groups[0].cpu_map; + + /* determine base */ + base = (void *)ULONG_MAX; + for_each_possible_cpu(cpu) + base = min(base, + (void *)(__per_cpu_offset[cpu] + __per_cpu_start)); + base_offset = (void *)__per_cpu_start - base; + + /* build cpu_map, units are grouped by node */ + unit = 0; + for_each_node(node) + for_each_possible_cpu(cpu) + if (node == node_cpuid[cpu].nid) + cpu_map[unit++] = cpu; + nr_units = unit; + + /* set basic parameters */ + static_size = __per_cpu_end - __per_cpu_start; + reserved_size = PERCPU_MODULE_RESERVE; + dyn_size = PERCPU_PAGE_SIZE - static_size - reserved_size; + if (dyn_size < 0) + panic("percpu area overflow static=%zd reserved=%zd\n", + static_size, reserved_size); + + ai->static_size = static_size; + ai->reserved_size = reserved_size; + ai->dyn_size = dyn_size; + ai->unit_size = PERCPU_PAGE_SIZE; + ai->atom_size = PAGE_SIZE; + ai->alloc_size = PERCPU_PAGE_SIZE; + + /* + * CPUs are put into groups according to node. Walk cpu_map + * and create new groups at node boundaries. + */ + prev_node = -1; + ai->nr_groups = 0; + for (unit = 0; unit < nr_units; unit++) { + cpu = cpu_map[unit]; + node = node_cpuid[cpu].nid; + + if (node == prev_node) { + gi->nr_units++; + continue; + } + prev_node = node; + + gi = &ai->groups[ai->nr_groups++]; + gi->nr_units = 1; + gi->base_offset = __per_cpu_offset[cpu] + base_offset; + gi->cpu_map = &cpu_map[unit]; + } + + rc = pcpu_setup_first_chunk(ai, base); + if (rc) + panic("failed to setup percpu area (err=%d)", rc); + + pcpu_free_alloc_info(ai); +} +#endif + /** * fill_pernode - initialize pernode data. * @node: the node id. -- 1.6.4.2 From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tejun Heo Date: Tue, 22 Sep 2009 07:40:13 +0000 Subject: [PATCH 3/4] ia64: convert to dynamic percpu allocator Message-Id: <1253605214-23210-4-git-send-email-tj@kernel.org> List-Id: References: <1253605214-23210-1-git-send-email-tj@kernel.org> In-Reply-To: <1253605214-23210-1-git-send-email-tj@kernel.org> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Nick Piggin , Tony Luck , Fenghua Yu , linux-ia64 , Ingo Molnar , Rusty Russell , Christoph Lameter , linux-kernel@vger.kernel.org Cc: Tejun Heo , Tony Luck , Fenghua Yu , linux-ia64 Unlike other archs, ia64 reserves space for percpu areas during early memory initialization. These areas occupy a contiguous region indexed by cpu number on contiguous memory model or are grouped by node on discontiguous memory model. As allocation and initialization are done by the arch code, all that setup_per_cpu_areas() needs to do is communicating the determined layout to the percpu allocator. This patch implements setup_per_cpu_areas() for both contig and discontig memory models and drops HAVE_LEGACY_PER_CPU_AREA. Please note that for contig model, the allocation itself is modified only to allocate for possible cpus instead of NR_CPUS. As dynamic percpu allocator can handle non-direct mapping, there's no reason to allocate memory for cpus which aren't possible. Signed-off-by: Tejun Heo Cc: Tony Luck Cc: Fenghua Yu Cc: linux-ia64 --- arch/ia64/Kconfig | 3 -- arch/ia64/kernel/setup.c | 12 ------ arch/ia64/mm/contig.c | 50 ++++++++++++++++++++++++--- arch/ia64/mm/discontig.c | 85 ++++++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 130 insertions(+), 20 deletions(-) diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig index 011a1cd..e624611 100644 --- a/arch/ia64/Kconfig +++ b/arch/ia64/Kconfig @@ -89,9 +89,6 @@ config GENERIC_TIME_VSYSCALL bool default y -config HAVE_LEGACY_PER_CPU_AREA - def_bool y - config HAVE_SETUP_PER_CPU_AREA def_bool y diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c index 1de86c9..42f8a18 100644 --- a/arch/ia64/kernel/setup.c +++ b/arch/ia64/kernel/setup.c @@ -856,18 +856,6 @@ identify_cpu (struct cpuinfo_ia64 *c) } /* - * In UP configuration, setup_per_cpu_areas() is defined in - * include/linux/percpu.h - */ -#ifdef CONFIG_SMP -void __init -setup_per_cpu_areas (void) -{ - /* start_kernel() requires this... */ -} -#endif - -/* * Do the following calculations: * * 1. the max. cache line size. diff --git a/arch/ia64/mm/contig.c b/arch/ia64/mm/contig.c index 9493bbf..c86ce4f 100644 --- a/arch/ia64/mm/contig.c +++ b/arch/ia64/mm/contig.c @@ -163,11 +163,11 @@ per_cpu_init (void) first_time = false; /* - * get_free_pages() cannot be used before cpu_init() done. BSP - * allocates "NR_CPUS" pages for all CPUs to avoid that AP calls - * get_zeroed_page(). + * get_free_pages() cannot be used before cpu_init() done. + * BSP allocates PERCPU_PAGE_SIZE bytes for all possible CPUs + * to avoid that AP calls get_zeroed_page(). */ - for (cpu = 0; cpu < NR_CPUS; cpu++) { + for_each_possible_cpu(cpu) { void *src = cpu = 0 ? cpu0_data : __phys_per_cpu_start; memcpy(cpu_data, src, __per_cpu_end - __per_cpu_start); @@ -196,9 +196,49 @@ skip: static inline void alloc_per_cpu_data(void) { - cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, + cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * num_possible_cpus(), PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); } + +/** + * setup_per_cpu_areas - setup percpu areas + * + * Arch code has already allocated and initialized percpu areas. All + * this function has to do is to teach the determined layout to the + * dynamic percpu allocator, which happens to be more complex than + * creating whole new ones using helpers. + */ +void __init +setup_per_cpu_areas(void) +{ + struct pcpu_alloc_info *ai; + struct pcpu_group_info *gi; + unsigned int cpu; + int rc; + + ai = pcpu_alloc_alloc_info(1, num_possible_cpus()); + if (!ai) + panic("failed to allocate pcpu_alloc_info"); + gi = &ai->groups[0]; + + /* units are assigned consecutively to possible cpus */ + for_each_possible_cpu(cpu) + gi->cpu_map[gi->nr_units++] = cpu; + + /* set parameters */ + ai->static_size = __per_cpu_end - __per_cpu_start; + ai->reserved_size = PERCPU_MODULE_RESERVE; + ai->dyn_size = PERCPU_DYNAMIC_RESERVE; + ai->unit_size = PERCPU_PAGE_SIZE; + ai->atom_size = PAGE_SIZE; + ai->alloc_size = PERCPU_PAGE_SIZE; + + rc = pcpu_setup_first_chunk(ai, __per_cpu_start + __per_cpu_offset[0]); + if (rc) + panic("failed to setup percpu area (err=%d)", rc); + + pcpu_free_alloc_info(ai); +} #else #define alloc_per_cpu_data() do { } while (0) #endif /* CONFIG_SMP */ diff --git a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c index 35a61ec..69e9e91 100644 --- a/arch/ia64/mm/discontig.c +++ b/arch/ia64/mm/discontig.c @@ -172,6 +172,91 @@ static void *per_cpu_node_setup(void *cpu_data, int node) return cpu_data; } +#ifdef CONFIG_SMP +/** + * setup_per_cpu_areas - setup percpu areas + * + * Arch code has already allocated and initialized percpu areas. All + * this function has to do is to teach the determined layout to the + * dynamic percpu allocator, which happens to be more complex than + * creating whole new ones using helpers. + */ +void __init setup_per_cpu_areas(void) +{ + struct pcpu_alloc_info *ai; + struct pcpu_group_info *uninitialized_var(gi); + unsigned int *cpu_map; + void *base; + unsigned long base_offset; + unsigned int cpu; + ssize_t static_size, reserved_size, dyn_size; + int node, prev_node, unit, nr_units, rc; + + ai = pcpu_alloc_alloc_info(MAX_NUMNODES, nr_cpu_ids); + if (!ai) + panic("failed to allocate pcpu_alloc_info"); + cpu_map = ai->groups[0].cpu_map; + + /* determine base */ + base = (void *)ULONG_MAX; + for_each_possible_cpu(cpu) + base = min(base, + (void *)(__per_cpu_offset[cpu] + __per_cpu_start)); + base_offset = (void *)__per_cpu_start - base; + + /* build cpu_map, units are grouped by node */ + unit = 0; + for_each_node(node) + for_each_possible_cpu(cpu) + if (node = node_cpuid[cpu].nid) + cpu_map[unit++] = cpu; + nr_units = unit; + + /* set basic parameters */ + static_size = __per_cpu_end - __per_cpu_start; + reserved_size = PERCPU_MODULE_RESERVE; + dyn_size = PERCPU_PAGE_SIZE - static_size - reserved_size; + if (dyn_size < 0) + panic("percpu area overflow static=%zd reserved=%zd\n", + static_size, reserved_size); + + ai->static_size = static_size; + ai->reserved_size = reserved_size; + ai->dyn_size = dyn_size; + ai->unit_size = PERCPU_PAGE_SIZE; + ai->atom_size = PAGE_SIZE; + ai->alloc_size = PERCPU_PAGE_SIZE; + + /* + * CPUs are put into groups according to node. Walk cpu_map + * and create new groups at node boundaries. + */ + prev_node = -1; + ai->nr_groups = 0; + for (unit = 0; unit < nr_units; unit++) { + cpu = cpu_map[unit]; + node = node_cpuid[cpu].nid; + + if (node = prev_node) { + gi->nr_units++; + continue; + } + prev_node = node; + + gi = &ai->groups[ai->nr_groups++]; + gi->nr_units = 1; + gi->base_offset = __per_cpu_offset[cpu] + base_offset; + gi->cpu_map = &cpu_map[unit]; + } + + rc = pcpu_setup_first_chunk(ai, base); + if (rc) + panic("failed to setup percpu area (err=%d)", rc); + + pcpu_free_alloc_info(ai); +} +#endif + /** * fill_pernode - initialize pernode data. * @node: the node id. -- 1.6.4.2