diff options
Diffstat (limited to 'arch/ia64/mm')
-rw-r--r-- | arch/ia64/mm/contig.c | 4 | ||||
-rw-r--r-- | arch/ia64/mm/discontig.c | 17 | ||||
-rw-r--r-- | arch/ia64/mm/init.c | 14 | ||||
-rw-r--r-- | arch/ia64/mm/numa.c | 4 | ||||
-rw-r--r-- | arch/ia64/mm/tlb.c | 357 |
5 files changed, 355 insertions, 41 deletions
diff --git a/arch/ia64/mm/contig.c b/arch/ia64/mm/contig.c index 344f64eca7a9..798bf9835a51 100644 --- a/arch/ia64/mm/contig.c +++ b/arch/ia64/mm/contig.c @@ -45,8 +45,6 @@ void show_mem(void) printk(KERN_INFO "Mem-info:\n"); show_free_areas(); - printk(KERN_INFO "Free swap: %6ldkB\n", - nr_swap_pages<<(PAGE_SHIFT-10)); printk(KERN_INFO "Node memory in pages:\n"); for_each_online_pgdat(pgdat) { unsigned long present; @@ -255,7 +253,7 @@ paging_init (void) max_zone_pfns[ZONE_NORMAL] = max_low_pfn; #ifdef CONFIG_VIRTUAL_MEM_MAP - efi_memmap_walk(register_active_ranges, NULL); + efi_memmap_walk(filter_memory, register_active_ranges); efi_memmap_walk(find_largest_hole, (u64 *)&max_gap); if (max_gap < LARGE_GAP) { vmem_map = (struct page *) 0; diff --git a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c index ee5e68b2af94..544dc420c65e 100644 --- a/arch/ia64/mm/discontig.c +++ b/arch/ia64/mm/discontig.c @@ -104,7 +104,7 @@ static int __meminit early_nr_cpus_node(int node) { int cpu, n = 0; - for (cpu = 0; cpu < NR_CPUS; cpu++) + for_each_possible_early_cpu(cpu) if (node == node_cpuid[cpu].nid) n++; @@ -124,6 +124,7 @@ static unsigned long __meminit compute_pernodesize(int node) pernodesize += node * L1_CACHE_BYTES; pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t)); pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); + pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t)); pernodesize = PAGE_ALIGN(pernodesize); return pernodesize; } @@ -142,7 +143,7 @@ static void *per_cpu_node_setup(void *cpu_data, int node) #ifdef CONFIG_SMP int cpu; - for (cpu = 0; cpu < NR_CPUS; cpu++) { + for_each_possible_early_cpu(cpu) { if (node == node_cpuid[cpu].nid) { memcpy(__va(cpu_data), __phys_per_cpu_start, __per_cpu_end - __per_cpu_start); @@ -345,7 +346,7 @@ static void __init initialize_pernode_data(void) #ifdef CONFIG_SMP /* Set the node_data pointer for each per-cpu struct */ - for (cpu = 0; cpu < NR_CPUS; cpu++) { + for_each_possible_early_cpu(cpu) { node = node_cpuid[cpu].nid; per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data; } @@ -444,7 +445,7 @@ void __init find_memory(void) mem_data[node].min_pfn = ~0UL; } - efi_memmap_walk(register_active_ranges, NULL); + efi_memmap_walk(filter_memory, register_active_ranges); /* * Initialize the boot memory maps in reverse order since that's @@ -493,13 +494,9 @@ void __cpuinit *per_cpu_init(void) int cpu; static int first_time = 1; - - if (smp_processor_id() != 0) - return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; - if (first_time) { first_time = 0; - for (cpu = 0; cpu < NR_CPUS; cpu++) + for_each_possible_early_cpu(cpu) per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu]; } @@ -522,8 +519,6 @@ void show_mem(void) printk(KERN_INFO "Mem-info:\n"); show_free_areas(); - printk(KERN_INFO "Free swap: %6ldkB\n", - nr_swap_pages<<(PAGE_SHIFT-10)); printk(KERN_INFO "Node memory in pages:\n"); for_each_online_pgdat(pgdat) { unsigned long present; diff --git a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c index a4ca657c72c6..5c1de53c8c1c 100644 --- a/arch/ia64/mm/init.c +++ b/arch/ia64/mm/init.c @@ -58,7 +58,6 @@ __ia64_sync_icache_dcache (pte_t pte) { unsigned long addr; struct page *page; - unsigned long order; page = pte_page(pte); addr = (unsigned long) page_address(page); @@ -66,12 +65,7 @@ __ia64_sync_icache_dcache (pte_t pte) if (test_bit(PG_arch_1, &page->flags)) return; /* i-cache is already coherent with d-cache */ - if (PageCompound(page)) { - order = compound_order(page); - flush_icache_range(addr, addr + (1UL << order << PAGE_SHIFT)); - } - else - flush_icache_range(addr, addr + PAGE_SIZE); + flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page))); set_bit(PG_arch_1, &page->flags); /* mark page as clean */ } @@ -553,12 +547,10 @@ find_largest_hole (u64 start, u64 end, void *arg) #endif /* CONFIG_VIRTUAL_MEM_MAP */ int __init -register_active_ranges(u64 start, u64 end, void *arg) +register_active_ranges(u64 start, u64 len, int nid) { - int nid = paddr_to_nid(__pa(start)); + u64 end = start + len; - if (nid < 0) - nid = 0; #ifdef CONFIG_KEXEC if (start > crashk_res.start && start < crashk_res.end) start = crashk_res.end; diff --git a/arch/ia64/mm/numa.c b/arch/ia64/mm/numa.c index 7807fc5c0422..b73bf1838e57 100644 --- a/arch/ia64/mm/numa.c +++ b/arch/ia64/mm/numa.c @@ -27,7 +27,9 @@ */ int num_node_memblks; struct node_memblk_s node_memblk[NR_NODE_MEMBLKS]; -struct node_cpuid_s node_cpuid[NR_CPUS]; +struct node_cpuid_s node_cpuid[NR_CPUS] = + { [0 ... NR_CPUS-1] = { .phys_id = 0, .nid = NUMA_NO_NODE } }; + /* * This is a matrix with "distances" between nodes, they should be * proportional to the memory access latency ratios. diff --git a/arch/ia64/mm/tlb.c b/arch/ia64/mm/tlb.c index 655da240d13c..d52ec4e83409 100644 --- a/arch/ia64/mm/tlb.c +++ b/arch/ia64/mm/tlb.c @@ -11,6 +11,9 @@ * Rohit Seth <rohit.seth@intel.com> * Ken Chen <kenneth.w.chen@intel.com> * Christophe de Dinechin <ddd@hp.com>: Avoid ptc.e on memory allocation + * Copyright (C) 2007 Intel Corp + * Fenghua Yu <fenghua.yu@intel.com> + * Add multiple ptc.g/ptc.ga instruction support in global tlb purge. */ #include <linux/module.h> #include <linux/init.h> @@ -26,6 +29,9 @@ #include <asm/pal.h> #include <asm/tlbflush.h> #include <asm/dma.h> +#include <asm/processor.h> +#include <asm/sal.h> +#include <asm/tlb.h> static struct { unsigned long mask; /* mask of supported purge page-sizes */ @@ -39,6 +45,10 @@ struct ia64_ctx ia64_ctx = { }; DEFINE_PER_CPU(u8, ia64_need_tlb_flush); +DEFINE_PER_CPU(u8, ia64_tr_num); /*Number of TR slots in current processor*/ +DEFINE_PER_CPU(u8, ia64_tr_used); /*Max Slot number used by kernel*/ + +struct ia64_tr_entry __per_cpu_idtrs[NR_CPUS][2][IA64_TR_ALLOC_MAX]; /* * Initializes the ia64_ctx.bitmap array based on max_ctx+1. @@ -84,14 +94,140 @@ wrap_mmu_context (struct mm_struct *mm) local_flush_tlb_all(); } +/* + * Implement "spinaphores" ... like counting semaphores, but they + * spin instead of sleeping. If there are ever any other users for + * this primitive it can be moved up to a spinaphore.h header. + */ +struct spinaphore { + atomic_t cur; +}; + +static inline void spinaphore_init(struct spinaphore *ss, int val) +{ + atomic_set(&ss->cur, val); +} + +static inline void down_spin(struct spinaphore *ss) +{ + while (unlikely(!atomic_add_unless(&ss->cur, -1, 0))) + while (atomic_read(&ss->cur) == 0) + cpu_relax(); +} + +static inline void up_spin(struct spinaphore *ss) +{ + atomic_add(1, &ss->cur); +} + +static struct spinaphore ptcg_sem; +static u16 nptcg = 1; +static int need_ptcg_sem = 1; +static int toolatetochangeptcgsem = 0; + +/* + * Kernel parameter "nptcg=" overrides max number of concurrent global TLB + * purges which is reported from either PAL or SAL PALO. + * + * We don't have sanity checking for nptcg value. It's the user's responsibility + * for valid nptcg value on the platform. Otherwise, kernel may hang in some + * cases. + */ +static int __init +set_nptcg(char *str) +{ + int value = 0; + + get_option(&str, &value); + setup_ptcg_sem(value, NPTCG_FROM_KERNEL_PARAMETER); + + return 1; +} + +__setup("nptcg=", set_nptcg); + +/* + * Maximum number of simultaneous ptc.g purges in the system can + * be defined by PAL_VM_SUMMARY (in which case we should take + * the smallest value for any cpu in the system) or by the PAL + * override table (in which case we should ignore the value from + * PAL_VM_SUMMARY). + * + * Kernel parameter "nptcg=" overrides maximum number of simultanesous ptc.g + * purges defined in either PAL_VM_SUMMARY or PAL override table. In this case, + * we should ignore the value from either PAL_VM_SUMMARY or PAL override table. + * + * Complicating the logic here is the fact that num_possible_cpus() + * isn't fully setup until we start bringing cpus online. + */ +void +setup_ptcg_sem(int max_purges, int nptcg_from) +{ + static int kp_override; + static int palo_override; + static int firstcpu = 1; + + if (toolatetochangeptcgsem) { + BUG_ON(max_purges < nptcg); + return; + } + + if (nptcg_from == NPTCG_FROM_KERNEL_PARAMETER) { + kp_override = 1; + nptcg = max_purges; + goto resetsema; + } + if (kp_override) { + need_ptcg_sem = num_possible_cpus() > nptcg; + return; + } + + if (nptcg_from == NPTCG_FROM_PALO) { + palo_override = 1; + + /* In PALO max_purges == 0 really means it! */ + if (max_purges == 0) + panic("Whoa! Platform does not support global TLB purges.\n"); + nptcg = max_purges; + if (nptcg == PALO_MAX_TLB_PURGES) { + need_ptcg_sem = 0; + return; + } + goto resetsema; + } + if (palo_override) { + if (nptcg != PALO_MAX_TLB_PURGES) + need_ptcg_sem = (num_possible_cpus() > nptcg); + return; + } + + /* In PAL_VM_SUMMARY max_purges == 0 actually means 1 */ + if (max_purges == 0) max_purges = 1; + + if (firstcpu) { + nptcg = max_purges; + firstcpu = 0; + } + if (max_purges < nptcg) + nptcg = max_purges; + if (nptcg == PAL_MAX_PURGES) { + need_ptcg_sem = 0; + return; + } else + need_ptcg_sem = (num_possible_cpus() > nptcg); + +resetsema: + spinaphore_init(&ptcg_sem, max_purges); +} + void ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start, unsigned long end, unsigned long nbits) { - static DEFINE_SPINLOCK(ptcg_lock); - struct mm_struct *active_mm = current->active_mm; + toolatetochangeptcgsem = 1; + if (mm != active_mm) { /* Restore region IDs for mm */ if (mm && active_mm) { @@ -102,19 +238,20 @@ ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start, } } - /* HW requires global serialization of ptc.ga. */ - spin_lock(&ptcg_lock); - { - do { - /* - * Flush ALAT entries also. - */ - ia64_ptcga(start, (nbits<<2)); - ia64_srlz_i(); - start += (1UL << nbits); - } while (start < end); - } - spin_unlock(&ptcg_lock); + if (need_ptcg_sem) + down_spin(&ptcg_sem); + + do { + /* + * Flush ALAT entries also. + */ + ia64_ptcga(start, (nbits << 2)); + ia64_srlz_i(); + start += (1UL << nbits); + } while (start < end); + + if (need_ptcg_sem) + up_spin(&ptcg_sem); if (mm != active_mm) { activate_context(active_mm); @@ -190,6 +327,9 @@ ia64_tlb_init (void) ia64_ptce_info_t uninitialized_var(ptce_info); /* GCC be quiet */ unsigned long tr_pgbits; long status; + pal_vm_info_1_u_t vm_info_1; + pal_vm_info_2_u_t vm_info_2; + int cpu = smp_processor_id(); if ((status = ia64_pal_vm_page_size(&tr_pgbits, &purge.mask)) != 0) { printk(KERN_ERR "PAL_VM_PAGE_SIZE failed with status=%ld; " @@ -206,4 +346,191 @@ ia64_tlb_init (void) local_cpu_data->ptce_stride[1] = ptce_info.stride[1]; local_flush_tlb_all(); /* nuke left overs from bootstrapping... */ + status = ia64_pal_vm_summary(&vm_info_1, &vm_info_2); + + if (status) { + printk(KERN_ERR "ia64_pal_vm_summary=%ld\n", status); + per_cpu(ia64_tr_num, cpu) = 8; + return; + } + per_cpu(ia64_tr_num, cpu) = vm_info_1.pal_vm_info_1_s.max_itr_entry+1; + if (per_cpu(ia64_tr_num, cpu) > + (vm_info_1.pal_vm_info_1_s.max_dtr_entry+1)) + per_cpu(ia64_tr_num, cpu) = + vm_info_1.pal_vm_info_1_s.max_dtr_entry+1; + if (per_cpu(ia64_tr_num, cpu) > IA64_TR_ALLOC_MAX) { + per_cpu(ia64_tr_num, cpu) = IA64_TR_ALLOC_MAX; + printk(KERN_DEBUG "TR register number exceeds IA64_TR_ALLOC_MAX!" + "IA64_TR_ALLOC_MAX should be extended\n"); + } +} + +/* + * is_tr_overlap + * + * Check overlap with inserted TRs. + */ +static int is_tr_overlap(struct ia64_tr_entry *p, u64 va, u64 log_size) +{ + u64 tr_log_size; + u64 tr_end; + u64 va_rr = ia64_get_rr(va); + u64 va_rid = RR_TO_RID(va_rr); + u64 va_end = va + (1<<log_size) - 1; + + if (va_rid != RR_TO_RID(p->rr)) + return 0; + tr_log_size = (p->itir & 0xff) >> 2; + tr_end = p->ifa + (1<<tr_log_size) - 1; + + if (va > tr_end || p->ifa > va_end) + return 0; + return 1; + +} + +/* + * ia64_insert_tr in virtual mode. Allocate a TR slot + * + * target_mask : 0x1 : itr, 0x2 : dtr, 0x3 : idtr + * + * va : virtual address. + * pte : pte entries inserted. + * log_size: range to be covered. + * + * Return value: <0 : error No. + * + * >=0 : slot number allocated for TR. + * Must be called with preemption disabled. + */ +int ia64_itr_entry(u64 target_mask, u64 va, u64 pte, u64 log_size) +{ + int i, r; + unsigned long psr; + struct ia64_tr_entry *p; + int cpu = smp_processor_id(); + + r = -EINVAL; + /*Check overlap with existing TR entries*/ + if (target_mask & 0x1) { + p = &__per_cpu_idtrs[cpu][0][0]; + for (i = IA64_TR_ALLOC_BASE; i <= per_cpu(ia64_tr_used, cpu); + i++, p++) { + if (p->pte & 0x1) + if (is_tr_overlap(p, va, log_size)) { + printk(KERN_DEBUG "Overlapped Entry" + "Inserted for TR Reigster!!\n"); + goto out; + } + } + } + if (target_mask & 0x2) { + p = &__per_cpu_idtrs[cpu][1][0]; + for (i = IA64_TR_ALLOC_BASE; i <= per_cpu(ia64_tr_used, cpu); + i++, p++) { + if (p->pte & 0x1) + if (is_tr_overlap(p, va, log_size)) { + printk(KERN_DEBUG "Overlapped Entry" + "Inserted for TR Reigster!!\n"); + goto out; + } + } + } + + for (i = IA64_TR_ALLOC_BASE; i < per_cpu(ia64_tr_num, cpu); i++) { + switch (target_mask & 0x3) { + case 1: + if (!(__per_cpu_idtrs[cpu][0][i].pte & 0x1)) + goto found; + continue; + case 2: + if (!(__per_cpu_idtrs[cpu][1][i].pte & 0x1)) + goto found; + continue; + case 3: + if (!(__per_cpu_idtrs[cpu][0][i].pte & 0x1) && + !(__per_cpu_idtrs[cpu][1][i].pte & 0x1)) + goto found; + continue; + default: + r = -EINVAL; + goto out; + } + } +found: + if (i >= per_cpu(ia64_tr_num, cpu)) + return -EBUSY; + + /*Record tr info for mca hander use!*/ + if (i > per_cpu(ia64_tr_used, cpu)) + per_cpu(ia64_tr_used, cpu) = i; + + psr = ia64_clear_ic(); + if (target_mask & 0x1) { + ia64_itr(0x1, i, va, pte, log_size); + ia64_srlz_i(); + p = &__per_cpu_idtrs[cpu][0][i]; + p->ifa = va; + p->pte = pte; + p->itir = log_size << 2; + p->rr = ia64_get_rr(va); + } + if (target_mask & 0x2) { + ia64_itr(0x2, i, va, pte, log_size); + ia64_srlz_i(); + p = &__per_cpu_idtrs[cpu][1][i]; + p->ifa = va; + p->pte = pte; + p->itir = log_size << 2; + p->rr = ia64_get_rr(va); + } + ia64_set_psr(psr); + r = i; +out: + return r; +} +EXPORT_SYMBOL_GPL(ia64_itr_entry); + +/* + * ia64_purge_tr + * + * target_mask: 0x1: purge itr, 0x2 : purge dtr, 0x3 purge idtr. + * slot: slot number to be freed. + * + * Must be called with preemption disabled. + */ +void ia64_ptr_entry(u64 target_mask, int slot) +{ + int cpu = smp_processor_id(); + int i; + struct ia64_tr_entry *p; + + if (slot < IA64_TR_ALLOC_BASE || slot >= per_cpu(ia64_tr_num, cpu)) + return; + + if (target_mask & 0x1) { + p = &__per_cpu_idtrs[cpu][0][slot]; + if ((p->pte&0x1) && is_tr_overlap(p, p->ifa, p->itir>>2)) { + p->pte = 0; + ia64_ptr(0x1, p->ifa, p->itir>>2); + ia64_srlz_i(); + } + } + + if (target_mask & 0x2) { + p = &__per_cpu_idtrs[cpu][1][slot]; + if ((p->pte & 0x1) && is_tr_overlap(p, p->ifa, p->itir>>2)) { + p->pte = 0; + ia64_ptr(0x2, p->ifa, p->itir>>2); + ia64_srlz_i(); + } + } + + for (i = per_cpu(ia64_tr_used, cpu); i >= IA64_TR_ALLOC_BASE; i--) { + if ((__per_cpu_idtrs[cpu][0][i].pte & 0x1) || + (__per_cpu_idtrs[cpu][1][i].pte & 0x1)) + break; + } + per_cpu(ia64_tr_used, cpu) = i; } +EXPORT_SYMBOL_GPL(ia64_ptr_entry); |