From a2f1b424900715ed9d1699c3bb88a434a2b42bc0 Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Sat, 5 Nov 2005 17:25:53 +0100 Subject: [PATCH] x86_64: Add 4GB DMA32 zone Add a new 4GB GFP_DMA32 zone between the GFP_DMA and GFP_NORMAL zones. As a bit of historical background: when the x86-64 port was originally designed we had some discussion if we should use a 16MB DMA zone like i386 or a 4GB DMA zone like IA64 or both. Both was ruled out at this point because it was in early 2.4 when VM is still quite shakey and had bad troubles even dealing with one DMA zone. We settled on the 16MB DMA zone mainly because we worried about older soundcards and the floppy. But this has always caused problems since then because device drivers had trouble getting enough DMA able memory. These days the VM works much better and the wide use of NUMA has proven it can deal with many zones successfully. So this patch adds both zones. This helps drivers who need a lot of memory below 4GB because their hardware is not accessing more (graphic drivers - proprietary and free ones, video frame buffer drivers, sound drivers etc.). Previously they could only use IOMMU+16MB GFP_DMA, which was not enough memory. Another common problem is that hardware who has full memory addressing for >4GB misses it for some control structures in memory (like transmit rings or other metadata). They tended to allocate memory in the 16MB GFP_DMA or the IOMMU/swiotlb then using pci_alloc_consistent, but that can tie up a lot of precious 16MB GFPDMA/IOMMU/swiotlb memory (even on AMD systems the IOMMU tends to be quite small) especially if you have many devices. With the new zone pci_alloc_consistent can just put this stuff into memory below 4GB which works better. One argument was still if the zone should be 4GB or 2GB. The main motivation for 2GB would be an unnamed not so unpopular hardware raid controller (mostly found in older machines from a particular four letter company) who has a strange 2GB restriction in firmware. But that one works ok with swiotlb/IOMMU anyways, so it doesn't really need GFP_DMA32. I chose 4GB to be compatible with IA64 and because it seems to be the most common restriction. The new zone is so far added only for x86-64. For other architectures who don't set up this new zone nothing changes. Architectures can set a compatibility define in Kconfig CONFIG_DMA_IS_DMA32 that will define GFP_DMA32 as GFP_DMA. Otherwise it's a nop because on 32bit architectures it's normally not needed because GFP_NORMAL (=0) is DMA able enough. One problem is still that GFP_DMA means different things on different architectures. e.g. some drivers used to have #ifdef ia64 use GFP_DMA (trusting it to be 4GB) #elif __x86_64__ (use other hacks like the swiotlb because 16MB is not enough) ... . This was quite ugly and is now obsolete. These should be now converted to use GFP_DMA32 unconditionally. I haven't done this yet. Or best only use pci_alloc_consistent/dma_alloc_coherent which will use GFP_DMA32 transparently. Signed-off-by: Andi Kleen Signed-off-by: Linus Torvalds --- include/linux/mmzone.h | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) (limited to 'include/linux/mmzone.h') diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index f5fa3082fd6..da7a829f856 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -71,10 +71,11 @@ struct per_cpu_pageset { #endif #define ZONE_DMA 0 -#define ZONE_NORMAL 1 -#define ZONE_HIGHMEM 2 +#define ZONE_DMA32 1 +#define ZONE_NORMAL 2 +#define ZONE_HIGHMEM 3 -#define MAX_NR_ZONES 3 /* Sync this with ZONES_SHIFT */ +#define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ @@ -108,9 +109,10 @@ struct per_cpu_pageset { /* * On machines where it is needed (eg PCs) we divide physical memory - * into multiple physical zones. On a PC we have 3 zones: + * into multiple physical zones. On a PC we have 4 zones: * * ZONE_DMA < 16 MB ISA DMA capable memory + * ZONE_DMA32 0 MB Empty * ZONE_NORMAL 16-896 MB direct mapped by the kernel * ZONE_HIGHMEM > 896 MB only page cache and user processes */ @@ -455,10 +457,10 @@ extern struct pglist_data contig_page_data; #if BITS_PER_LONG == 32 || defined(ARCH_HAS_ATOMIC_UNSIGNED) /* - * with 32 bit page->flags field, we reserve 8 bits for node/zone info. - * there are 3 zones (2 bits) and this leaves 8-2=6 bits for nodes. + * with 32 bit page->flags field, we reserve 9 bits for node/zone info. + * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. */ -#define FLAGS_RESERVED 8 +#define FLAGS_RESERVED 9 #elif BITS_PER_LONG == 64 /* -- cgit v1.2.3-70-g09d2 From 07808b74e7dab1aa385e698795875337d72daf7d Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Sat, 5 Nov 2005 17:25:53 +0100 Subject: [PATCH] x86_64: Remove obsolete ARCH_HAS_ATOMIC_UNSIGNED and page_flags_t Has been introduced for x86-64 at some point to save memory in struct page, but has been obsolete for some time. Just remove it. Signed-off-by: Andi Kleen Signed-off-by: Linus Torvalds --- include/linux/mm.h | 10 ++-------- include/linux/mmzone.h | 2 +- mm/filemap.c | 2 +- mm/page_alloc.c | 2 +- 4 files changed, 5 insertions(+), 11 deletions(-) (limited to 'include/linux/mmzone.h') diff --git a/include/linux/mm.h b/include/linux/mm.h index 5c1fb0a2e80..23fad4dae23 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -206,12 +206,6 @@ struct vm_operations_struct { struct mmu_gather; struct inode; -#ifdef ARCH_HAS_ATOMIC_UNSIGNED -typedef unsigned page_flags_t; -#else -typedef unsigned long page_flags_t; -#endif - /* * Each physical page in the system has a struct page associated with * it to keep track of whatever it is we are using the page for at the @@ -219,7 +213,7 @@ typedef unsigned long page_flags_t; * a page. */ struct page { - page_flags_t flags; /* Atomic flags, some possibly + unsigned long flags; /* Atomic flags, some possibly * updated asynchronously */ atomic_t _count; /* Usage count, see below. */ atomic_t _mapcount; /* Count of ptes mapped in mms, @@ -435,7 +429,7 @@ static inline void put_page(struct page *page) #endif /* Page flags: | [SECTION] | [NODE] | ZONE | ... | FLAGS | */ -#define SECTIONS_PGOFF ((sizeof(page_flags_t)*8) - SECTIONS_WIDTH) +#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH) #define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH) #define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH) diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index da7a829f856..57fc99c67c3 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -455,7 +455,7 @@ extern struct pglist_data contig_page_data; #include #endif -#if BITS_PER_LONG == 32 || defined(ARCH_HAS_ATOMIC_UNSIGNED) +#if BITS_PER_LONG == 32 /* * with 32 bit page->flags field, we reserve 9 bits for node/zone info. * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. diff --git a/mm/filemap.c b/mm/filemap.c index 5d6e4c2000d..33a28bfde15 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -134,7 +134,7 @@ static int sync_page(void *word) struct address_space *mapping; struct page *page; - page = container_of((page_flags_t *)word, struct page, flags); + page = container_of((unsigned long *)word, struct page, flags); /* * page_mapping() is being called without PG_locked held. diff --git a/mm/page_alloc.c b/mm/page_alloc.c index e8810a7e026..259a71bacca 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -128,7 +128,7 @@ static void bad_page(const char *function, struct page *page) printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n", function, current->comm, page); printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n", - (int)(2*sizeof(page_flags_t)), (unsigned long)page->flags, + (int)(2*sizeof(unsigned long)), (unsigned long)page->flags, page->mapping, page_mapcount(page), page_count(page)); printk(KERN_EMERG "Backtrace:\n"); dump_stack(); -- cgit v1.2.3-70-g09d2 From 69d81fcde7797342417591ba7affb372b9c86eae Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Sat, 5 Nov 2005 17:25:53 +0100 Subject: [PATCH] x86_64: Speed up numa_node_id by putting it directly into the PDA Not go from the CPU number to an mapping array. Mode number is often used now in fast paths. This also adds a generic numa_node_id to all the topology includes Suggested by Eric Dumazet Signed-off-by: Andi Kleen Signed-off-by: Linus Torvalds --- arch/x86_64/kernel/setup.c | 4 ++-- arch/x86_64/mm/numa.c | 10 ++++++++-- arch/x86_64/mm/srat.c | 2 +- include/asm-x86_64/numa.h | 2 ++ include/asm-x86_64/pda.h | 1 + include/asm-x86_64/topology.h | 2 ++ include/linux/mmzone.h | 2 ++ 7 files changed, 18 insertions(+), 5 deletions(-) (limited to 'include/linux/mmzone.h') diff --git a/arch/x86_64/kernel/setup.c b/arch/x86_64/kernel/setup.c index f27731ac95c..99cfa751949 100644 --- a/arch/x86_64/kernel/setup.c +++ b/arch/x86_64/kernel/setup.c @@ -823,7 +823,7 @@ static void __init amd_detect_cmp(struct cpuinfo_x86 *c) if (!node_online(node)) node = nearby_node(apicid); } - cpu_to_node[cpu] = node; + numa_set_node(cpu, node); printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n", cpu, c->x86_num_cores, node, cpu_core_id[cpu]); @@ -975,7 +975,7 @@ static void srat_detect_node(void) node = apicid_to_node[hard_smp_processor_id()]; if (node == NUMA_NO_NODE) node = 0; - cpu_to_node[cpu] = node; + numa_set_node(cpu, node); if (acpi_numa > 0) printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node); diff --git a/arch/x86_64/mm/numa.c b/arch/x86_64/mm/numa.c index 18e86e2eac2..4bf64583ba3 100644 --- a/arch/x86_64/mm/numa.c +++ b/arch/x86_64/mm/numa.c @@ -156,7 +156,7 @@ void __init numa_init_array(void) for (i = 0; i < NR_CPUS; i++) { if (cpu_to_node[i] != NUMA_NO_NODE) continue; - cpu_to_node[i] = rr; + numa_set_node(i, rr); rr = next_node(rr, node_online_map); if (rr == MAX_NUMNODES) rr = first_node(node_online_map); @@ -242,7 +242,7 @@ void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn) nodes_clear(node_online_map); node_set_online(0); for (i = 0; i < NR_CPUS; i++) - cpu_to_node[i] = 0; + numa_set_node(i, 0); node_to_cpumask[0] = cpumask_of_cpu(0); setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); } @@ -252,6 +252,12 @@ __cpuinit void numa_add_cpu(int cpu) set_bit(cpu, &node_to_cpumask[cpu_to_node(cpu)]); } +void __cpuinit numa_set_node(int cpu, int node) +{ + cpu_pda[cpu].nodenumber = node; + cpu_to_node[cpu] = node; +} + unsigned long __init numa_free_all_bootmem(void) { int i; diff --git a/arch/x86_64/mm/srat.c b/arch/x86_64/mm/srat.c index 4b2e844c15a..c7aa08a5804 100644 --- a/arch/x86_64/mm/srat.c +++ b/arch/x86_64/mm/srat.c @@ -203,7 +203,7 @@ int __init acpi_scan_nodes(unsigned long start, unsigned long end) if (cpu_to_node[i] == NUMA_NO_NODE) continue; if (!node_isset(cpu_to_node[i], nodes_parsed)) - cpu_to_node[i] = NUMA_NO_NODE; + numa_set_node(i, NUMA_NO_NODE); } numa_init_array(); return 0; diff --git a/include/asm-x86_64/numa.h b/include/asm-x86_64/numa.h index bcf55c3f7f7..d51e56fdc3d 100644 --- a/include/asm-x86_64/numa.h +++ b/include/asm-x86_64/numa.h @@ -17,6 +17,8 @@ extern void numa_add_cpu(int cpu); extern void numa_init_array(void); extern int numa_off; +extern void numa_set_node(int cpu, int node); + extern unsigned char apicid_to_node[256]; #define NUMA_NO_NODE 0xff diff --git a/include/asm-x86_64/pda.h b/include/asm-x86_64/pda.h index bbf89aa8a1a..8733ccfa442 100644 --- a/include/asm-x86_64/pda.h +++ b/include/asm-x86_64/pda.h @@ -15,6 +15,7 @@ struct x8664_pda { int irqcount; /* Irq nesting counter. Starts with -1 */ int cpunumber; /* Logical CPU number */ char *irqstackptr; /* top of irqstack */ + int nodenumber; /* number of current node */ unsigned int __softirq_pending; unsigned int __nmi_count; /* number of NMI on this CPUs */ struct mm_struct *active_mm; diff --git a/include/asm-x86_64/topology.h b/include/asm-x86_64/topology.h index 1c603cd7e4d..d39ebd5263e 100644 --- a/include/asm-x86_64/topology.h +++ b/include/asm-x86_64/topology.h @@ -28,6 +28,8 @@ extern int __node_distance(int, int); #define pcibus_to_node(bus) ((long)(bus->sysdata)) #define pcibus_to_cpumask(bus) node_to_cpumask(pcibus_to_node(bus)); +#define numa_node_id() read_pda(nodenumber) + /* sched_domains SD_NODE_INIT for x86_64 machines */ #define SD_NODE_INIT (struct sched_domain) { \ .span = CPU_MASK_NONE, \ diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index 57fc99c67c3..f3cffc354de 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -435,7 +435,9 @@ int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, #include /* Returns the number of the current Node. */ +#ifndef numa_node_id #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) +#endif #ifndef CONFIG_NEED_MULTIPLE_NODES -- cgit v1.2.3-70-g09d2