diff options
Diffstat (limited to 'arch/sparc64')
-rw-r--r-- | arch/sparc64/defconfig | 16 | ||||
-rw-r--r-- | arch/sparc64/kernel/binfmt_aout32.c | 4 | ||||
-rw-r--r-- | arch/sparc64/kernel/ebus.c | 5 | ||||
-rw-r--r-- | arch/sparc64/kernel/head.S | 35 | ||||
-rw-r--r-- | arch/sparc64/kernel/irq.c | 27 | ||||
-rw-r--r-- | arch/sparc64/kernel/mdesc.c | 6 | ||||
-rw-r--r-- | arch/sparc64/kernel/pci.c | 14 | ||||
-rw-r--r-- | arch/sparc64/kernel/pci_common.c | 4 | ||||
-rw-r--r-- | arch/sparc64/kernel/pci_sun4v.c | 18 | ||||
-rw-r--r-- | arch/sparc64/kernel/prom.c | 9 | ||||
-rw-r--r-- | arch/sparc64/kernel/smp.c | 2 | ||||
-rw-r--r-- | arch/sparc64/kernel/trampoline.S | 33 | ||||
-rw-r--r-- | arch/sparc64/kernel/vio.c | 29 | ||||
-rw-r--r-- | arch/sparc64/lib/NGcopy_from_user.S | 8 | ||||
-rw-r--r-- | arch/sparc64/lib/NGcopy_to_user.S | 8 | ||||
-rw-r--r-- | arch/sparc64/lib/NGmemcpy.S | 371 | ||||
-rw-r--r-- | arch/sparc64/mm/hugetlbpage.c | 2 | ||||
-rw-r--r-- | arch/sparc64/prom/misc.c | 16 |
18 files changed, 377 insertions, 230 deletions
diff --git a/arch/sparc64/defconfig b/arch/sparc64/defconfig index 68338a601f7..7d07297db87 100644 --- a/arch/sparc64/defconfig +++ b/arch/sparc64/defconfig @@ -1,7 +1,7 @@ # # Automatically generated make config: don't edit -# Linux kernel version: 2.6.23-rc1 -# Sun Jul 22 19:24:37 2007 +# Linux kernel version: 2.6.23-rc6 +# Sun Sep 16 09:52:11 2007 # CONFIG_SPARC=y CONFIG_SPARC64=y @@ -32,15 +32,11 @@ CONFIG_HZ=100 CONFIG_DEFCONFIG_LIST="/lib/modules/$UNAME_RELEASE/.config" # -# Code maturity level options +# General setup # CONFIG_EXPERIMENTAL=y CONFIG_BROKEN_ON_SMP=y CONFIG_INIT_ENV_ARG_LIMIT=32 - -# -# General setup -# CONFIG_LOCALVERSION="" # CONFIG_LOCALVERSION_AUTO is not set CONFIG_SWAP=y @@ -555,6 +551,7 @@ CONFIG_E1000_NAPI=y # CONFIG_SIS190 is not set # CONFIG_SKGE is not set # CONFIG_SKY2 is not set +# CONFIG_SK98LIN is not set # CONFIG_VIA_VELOCITY is not set CONFIG_TIGON3=m CONFIG_BNX2=m @@ -809,6 +806,7 @@ CONFIG_HWMON=y # CONFIG_SENSORS_SMSC47M1 is not set # CONFIG_SENSORS_SMSC47M192 is not set # CONFIG_SENSORS_SMSC47B397 is not set +# CONFIG_SENSORS_THMC50 is not set # CONFIG_SENSORS_VIA686A is not set # CONFIG_SENSORS_VT1211 is not set # CONFIG_SENSORS_VT8231 is not set @@ -1162,10 +1160,6 @@ CONFIG_USB_STORAGE=m # CONFIG_MMC is not set # CONFIG_NEW_LEDS is not set # CONFIG_INFINIBAND is not set - -# -# Real Time Clock -# # CONFIG_RTC_CLASS is not set # diff --git a/arch/sparc64/kernel/binfmt_aout32.c b/arch/sparc64/kernel/binfmt_aout32.c index f205fc7cbcd..d208cc7804f 100644 --- a/arch/sparc64/kernel/binfmt_aout32.c +++ b/arch/sparc64/kernel/binfmt_aout32.c @@ -177,7 +177,7 @@ static u32 __user *create_aout32_tables(char __user *p, struct linux_binprm *bpr get_user(c,p++); } while (c); } - put_user(NULL,argv); + put_user(0,argv); current->mm->arg_end = current->mm->env_start = (unsigned long) p; while (envc-->0) { char c; @@ -186,7 +186,7 @@ static u32 __user *create_aout32_tables(char __user *p, struct linux_binprm *bpr get_user(c,p++); } while (c); } - put_user(NULL,envp); + put_user(0,envp); current->mm->env_end = (unsigned long) p; return sp; } diff --git a/arch/sparc64/kernel/ebus.c b/arch/sparc64/kernel/ebus.c index bc9ae36f7a4..04ab81cb4f4 100644 --- a/arch/sparc64/kernel/ebus.c +++ b/arch/sparc64/kernel/ebus.c @@ -375,7 +375,10 @@ static void __init fill_ebus_device(struct device_node *dp, struct linux_ebus_de dev->num_addrs = 0; dev->num_irqs = 0; } else { - (void) of_get_property(dp, "reg", &len); + const int *regs = of_get_property(dp, "reg", &len); + + if (!regs) + len = 0; dev->num_addrs = len / sizeof(struct linux_prom_registers); for (i = 0; i < dev->num_addrs; i++) diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S index 63144ad476f..c4147ad8677 100644 --- a/arch/sparc64/kernel/head.S +++ b/arch/sparc64/kernel/head.S @@ -98,7 +98,7 @@ sparc64_boot: .globl prom_boot_mapped_pc, prom_boot_mapping_mode .globl prom_boot_mapping_phys_high, prom_boot_mapping_phys_low .globl prom_compatible_name, prom_cpu_path, prom_cpu_compatible - .globl is_sun4v, sun4v_chip_type + .globl is_sun4v, sun4v_chip_type, prom_set_trap_table_name prom_peer_name: .asciz "peer" prom_compatible_name: @@ -121,6 +121,8 @@ prom_map_name: .asciz "map" prom_unmap_name: .asciz "unmap" +prom_set_trap_table_name: + .asciz "SUNW,set-trap-table" prom_sun4v_name: .asciz "sun4v" prom_niagara_prefix: @@ -691,15 +693,38 @@ setup_trap_table: sethi %hi(kern_base), %g3 ldx [%g3 + %lo(kern_base)], %g3 add %g2, %g3, %o1 + sethi %hi(sparc64_ttable_tl0), %o0 - call prom_set_trap_table_sun4v - sethi %hi(sparc64_ttable_tl0), %o0 + set prom_set_trap_table_name, %g2 + stx %g2, [%sp + 2047 + 128 + 0x00] + mov 2, %g2 + stx %g2, [%sp + 2047 + 128 + 0x08] + mov 0, %g2 + stx %g2, [%sp + 2047 + 128 + 0x10] + stx %o0, [%sp + 2047 + 128 + 0x18] + stx %o1, [%sp + 2047 + 128 + 0x20] + sethi %hi(p1275buf), %g2 + or %g2, %lo(p1275buf), %g2 + ldx [%g2 + 0x08], %o1 + call %o1 + add %sp, (2047 + 128), %o0 ba,pt %xcc, 2f nop -1: call prom_set_trap_table - sethi %hi(sparc64_ttable_tl0), %o0 +1: sethi %hi(sparc64_ttable_tl0), %o0 + set prom_set_trap_table_name, %g2 + stx %g2, [%sp + 2047 + 128 + 0x00] + mov 1, %g2 + stx %g2, [%sp + 2047 + 128 + 0x08] + mov 0, %g2 + stx %g2, [%sp + 2047 + 128 + 0x10] + stx %o0, [%sp + 2047 + 128 + 0x18] + sethi %hi(p1275buf), %g2 + or %g2, %lo(p1275buf), %g2 + ldx [%g2 + 0x08], %o1 + call %o1 + add %sp, (2047 + 128), %o0 /* Start using proper page size encodings in ctx register. */ 2: sethi %hi(sparc64_kern_pri_context), %g3 diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c index 384abf410cf..23956096b3b 100644 --- a/arch/sparc64/kernel/irq.c +++ b/arch/sparc64/kernel/irq.c @@ -217,8 +217,27 @@ struct irq_handler_data { void (*pre_handler)(unsigned int, void *, void *); void *pre_handler_arg1; void *pre_handler_arg2; + + u32 msi; }; +void sparc64_set_msi(unsigned int virt_irq, u32 msi) +{ + struct irq_handler_data *data = get_irq_chip_data(virt_irq); + + if (data) + data->msi = msi; +} + +u32 sparc64_get_msi(unsigned int virt_irq) +{ + struct irq_handler_data *data = get_irq_chip_data(virt_irq); + + if (data) + return data->msi; + return 0xffffffff; +} + static inline struct ino_bucket *virt_irq_to_bucket(unsigned int virt_irq) { unsigned int real_irq = virt_to_real_irq(virt_irq); @@ -308,7 +327,7 @@ static void sun4u_irq_disable(unsigned int virt_irq) if (likely(data)) { unsigned long imap = data->imap; - u32 tmp = upa_readq(imap); + unsigned long tmp = upa_readq(imap); tmp &= ~IMAP_VALID; upa_writeq(tmp, imap); @@ -741,7 +760,7 @@ unsigned int sun4v_build_msi(u32 devhandle, unsigned int *virt_irq_p, break; } if (devino >= msi_end) - return 0; + return -ENOSPC; sysino = sun4v_devino_to_sysino(devhandle, devino); bucket = &ivector_table[sysino]; @@ -755,8 +774,8 @@ unsigned int sun4v_build_msi(u32 devhandle, unsigned int *virt_irq_p, data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC); if (unlikely(!data)) { - prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n"); - prom_halt(); + virt_irq_free(*virt_irq_p); + return -ENOMEM; } set_irq_chip_data(bucket->virt_irq, data); diff --git a/arch/sparc64/kernel/mdesc.c b/arch/sparc64/kernel/mdesc.c index 9f22e4ff601..856659bb131 100644 --- a/arch/sparc64/kernel/mdesc.c +++ b/arch/sparc64/kernel/mdesc.c @@ -777,8 +777,12 @@ void __devinit mdesc_fill_in_cpu_data(cpumask_t mask) cpuid = *id; #ifdef CONFIG_SMP - if (cpuid >= NR_CPUS) + if (cpuid >= NR_CPUS) { + printk(KERN_WARNING "Ignoring CPU %d which is " + ">= NR_CPUS (%d)\n", + cpuid, NR_CPUS); continue; + } if (!cpu_isset(cpuid, mask)) continue; #else diff --git a/arch/sparc64/kernel/pci.c b/arch/sparc64/kernel/pci.c index 3d93e9203ba..e8dac81d8a0 100644 --- a/arch/sparc64/kernel/pci.c +++ b/arch/sparc64/kernel/pci.c @@ -393,7 +393,6 @@ struct pci_dev *of_create_pci_dev(struct pci_pbm_info *pbm, sd->host_controller = pbm; sd->prom_node = node; sd->op = of_find_device_by_node(node); - sd->msi_num = 0xffffffff; sd = &sd->op->dev.archdata; sd->iommu = pbm->iommu; @@ -745,7 +744,7 @@ static void __devinit pci_of_scan_bus(struct pci_pbm_info *pbm, { struct device_node *child; const u32 *reg; - int reglen, devfn; + int reglen, devfn, prev_devfn; struct pci_dev *dev; if (ofpci_verbose) @@ -753,14 +752,25 @@ static void __devinit pci_of_scan_bus(struct pci_pbm_info *pbm, node->full_name, bus->number); child = NULL; + prev_devfn = -1; while ((child = of_get_next_child(node, child)) != NULL) { if (ofpci_verbose) printk(" * %s\n", child->full_name); reg = of_get_property(child, "reg", ®len); if (reg == NULL || reglen < 20) continue; + devfn = (reg[0] >> 8) & 0xff; + /* This is a workaround for some device trees + * which list PCI devices twice. On the V100 + * for example, device number 3 is listed twice. + * Once as "pm" and once again as "lomp". + */ + if (devfn == prev_devfn) + continue; + prev_devfn = devfn; + /* create a new pci_dev for this device */ dev = of_create_pci_dev(pbm, child, bus, devfn, 0); if (!dev) diff --git a/arch/sparc64/kernel/pci_common.c b/arch/sparc64/kernel/pci_common.c index 2f61c4b1259..c76bfbb7da0 100644 --- a/arch/sparc64/kernel/pci_common.c +++ b/arch/sparc64/kernel/pci_common.c @@ -264,7 +264,7 @@ static int sun4v_read_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn, unsigned int func = PCI_FUNC(devfn); unsigned long ret; - if (bus_dev == pbm->pci_bus && devfn == 0x00) + if (!bus && devfn == 0x00) return pci_host_bridge_read_pci_cfg(bus_dev, devfn, where, size, value); if (config_out_of_range(pbm, bus, devfn, where)) { @@ -300,7 +300,7 @@ static int sun4v_write_pci_cfg(struct pci_bus *bus_dev, unsigned int devfn, unsigned int func = PCI_FUNC(devfn); unsigned long ret; - if (bus_dev == pbm->pci_bus && devfn == 0x00) + if (!bus && devfn == 0x00) return pci_host_bridge_write_pci_cfg(bus_dev, devfn, where, size, value); if (config_out_of_range(pbm, bus, devfn, where)) { diff --git a/arch/sparc64/kernel/pci_sun4v.c b/arch/sparc64/kernel/pci_sun4v.c index 466f4aa8fc8..da724b13e89 100644 --- a/arch/sparc64/kernel/pci_sun4v.c +++ b/arch/sparc64/kernel/pci_sun4v.c @@ -940,13 +940,13 @@ static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p, if (msi_num < 0) return msi_num; - devino = sun4v_build_msi(pbm->devhandle, virt_irq_p, - pbm->msiq_first_devino, - (pbm->msiq_first_devino + - pbm->msiq_num)); - err = -ENOMEM; - if (!devino) + err = sun4v_build_msi(pbm->devhandle, virt_irq_p, + pbm->msiq_first_devino, + (pbm->msiq_first_devino + + pbm->msiq_num)); + if (err < 0) goto out_err; + devino = err; msiqid = ((devino - pbm->msiq_first_devino) + pbm->msiq_first); @@ -971,7 +971,7 @@ static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p, if (pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_VALID)) goto out_err; - pdev->dev.archdata.msi_num = msi_num; + sparc64_set_msi(*virt_irq_p, msi_num); if (entry->msi_attrib.is_64) { msg.address_hi = pbm->msi64_start >> 32; @@ -993,8 +993,6 @@ static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p, out_err: free_msi(pbm, msi_num); - sun4v_destroy_msi(*virt_irq_p); - *virt_irq_p = 0; return err; } @@ -1006,7 +1004,7 @@ static void pci_sun4v_teardown_msi_irq(unsigned int virt_irq, unsigned long msiqid, err; unsigned int msi_num; - msi_num = pdev->dev.archdata.msi_num; + msi_num = sparc64_get_msi(virt_irq); err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi_num, &msiqid); if (err) { printk(KERN_ERR "%s: getmsiq gives error %lu\n", diff --git a/arch/sparc64/kernel/prom.c b/arch/sparc64/kernel/prom.c index d1a78c976ce..a246e962e5a 100644 --- a/arch/sparc64/kernel/prom.c +++ b/arch/sparc64/kernel/prom.c @@ -1046,7 +1046,8 @@ static void __init irq_trans_init(struct device_node *dp) if (!strcmp(dp->name, "fhc") && !strcmp(dp->parent->name, "central")) return central_irq_trans_init(dp); - if (!strcmp(dp->name, "virtual-devices")) + if (!strcmp(dp->name, "virtual-devices") || + !strcmp(dp->name, "niu")) return sun4v_vdev_irq_trans_init(dp); } @@ -1583,8 +1584,12 @@ static void __init of_fill_in_cpu_data(void) ncpus_probed++; #ifdef CONFIG_SMP - if (cpuid >= NR_CPUS) + if (cpuid >= NR_CPUS) { + printk(KERN_WARNING "Ignoring CPU %d which is " + ">= NR_CPUS (%d)\n", + cpuid, NR_CPUS); continue; + } #else /* On uniprocessor we only want the values for the * real physical cpu the kernel booted onto, however diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c index b84c49e3697..c73b7a48b03 100644 --- a/arch/sparc64/kernel/smp.c +++ b/arch/sparc64/kernel/smp.c @@ -353,6 +353,8 @@ static int __devinit smp_boot_one_cpu(unsigned int cpu) int timeout, ret; p = fork_idle(cpu); + if (IS_ERR(p)) + return PTR_ERR(p); callin_flag = 0; cpu_new_thread = task_thread_info(p); diff --git a/arch/sparc64/kernel/trampoline.S b/arch/sparc64/kernel/trampoline.S index 9533a25ce5d..04e81dda13d 100644 --- a/arch/sparc64/kernel/trampoline.S +++ b/arch/sparc64/kernel/trampoline.S @@ -345,7 +345,7 @@ after_lock_tlb: sethi %hi(tramp_stack), %g1 or %g1, %lo(tramp_stack), %g1 add %g1, TRAMP_STACK_SIZE, %g1 - sub %g1, STACKFRAME_SZ + STACK_BIAS, %sp + sub %g1, STACKFRAME_SZ + STACK_BIAS + 256, %sp mov 0, %fp /* Put garbage in these registers to trap any access to them. */ @@ -411,15 +411,38 @@ after_lock_tlb: sethi %hi(kern_base), %g3 ldx [%g3 + %lo(kern_base)], %g3 add %g2, %g3, %o1 + sethi %hi(sparc64_ttable_tl0), %o0 - call prom_set_trap_table_sun4v - sethi %hi(sparc64_ttable_tl0), %o0 + set prom_set_trap_table_name, %g2 + stx %g2, [%sp + 2047 + 128 + 0x00] + mov 2, %g2 + stx %g2, [%sp + 2047 + 128 + 0x08] + mov 0, %g2 + stx %g2, [%sp + 2047 + 128 + 0x10] + stx %o0, [%sp + 2047 + 128 + 0x18] + stx %o1, [%sp + 2047 + 128 + 0x20] + sethi %hi(p1275buf), %g2 + or %g2, %lo(p1275buf), %g2 + ldx [%g2 + 0x08], %o1 + call %o1 + add %sp, (2047 + 128), %o0 ba,pt %xcc, 2f nop -1: call prom_set_trap_table - sethi %hi(sparc64_ttable_tl0), %o0 +1: sethi %hi(sparc64_ttable_tl0), %o0 + set prom_set_trap_table_name, %g2 + stx %g2, [%sp + 2047 + 128 + 0x00] + mov 1, %g2 + stx %g2, [%sp + 2047 + 128 + 0x08] + mov 0, %g2 + stx %g2, [%sp + 2047 + 128 + 0x10] + stx %o0, [%sp + 2047 + 128 + 0x18] + sethi %hi(p1275buf), %g2 + or %g2, %lo(p1275buf), %g2 + ldx [%g2 + 0x08], %o1 + call %o1 + add %sp, (2047 + 128), %o0 2: ldx [%l0], %g6 ldx [%g6 + TI_TASK], %g4 diff --git a/arch/sparc64/kernel/vio.c b/arch/sparc64/kernel/vio.c index 1550ac5673d..0c1ee619d81 100644 --- a/arch/sparc64/kernel/vio.c +++ b/arch/sparc64/kernel/vio.c @@ -292,7 +292,7 @@ static struct vio_dev *vio_create_one(struct mdesc_handle *hp, u64 mp, } vdev->dp = dp; - printk(KERN_ERR "VIO: Adding device %s\n", vdev->dev.bus_id); + printk(KERN_INFO "VIO: Adding device %s\n", vdev->dev.bus_id); err = device_register(&vdev->dev); if (err) { @@ -342,8 +342,33 @@ static struct mdesc_notifier_client vio_device_notifier = { .node_name = "virtual-device-port", }; +/* We are only interested in domain service ports under the + * "domain-services" node. On control nodes there is another port + * under "openboot" that we should not mess with as aparently that is + * reserved exclusively for OBP use. + */ +static void vio_add_ds(struct mdesc_handle *hp, u64 node) +{ + int found; + u64 a; + + found = 0; + mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) { + u64 target = mdesc_arc_target(hp, a); + const char *name = mdesc_node_name(hp, target); + + if (!strcmp(name, "domain-services")) { + found = 1; + break; + } + } + + if (found) + (void) vio_create_one(hp, node, &root_vdev->dev); +} + static struct mdesc_notifier_client vio_ds_notifier = { - .add = vio_add, + .add = vio_add_ds, .remove = vio_remove, .node_name = "domain-services-port", }; diff --git a/arch/sparc64/lib/NGcopy_from_user.S b/arch/sparc64/lib/NGcopy_from_user.S index 2d93456f76d..e7f433f71b4 100644 --- a/arch/sparc64/lib/NGcopy_from_user.S +++ b/arch/sparc64/lib/NGcopy_from_user.S @@ -1,6 +1,6 @@ /* NGcopy_from_user.S: Niagara optimized copy from userspace. * - * Copyright (C) 2006 David S. Miller (davem@davemloft.net) + * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net) */ #define EX_LD(x) \ @@ -8,8 +8,8 @@ .section .fixup; \ .align 4; \ 99: wr %g0, ASI_AIUS, %asi;\ - retl; \ - mov 1, %o0; \ + ret; \ + restore %g0, 1, %o0; \ .section __ex_table,"a";\ .align 4; \ .word 98b, 99b; \ @@ -24,7 +24,7 @@ #define LOAD(type,addr,dest) type##a [addr] ASI_AIUS, dest #define LOAD_TWIN(addr_reg,dest0,dest1) \ ldda [addr_reg] ASI_BLK_INIT_QUAD_LDD_AIUS, dest0 -#define EX_RETVAL(x) 0 +#define EX_RETVAL(x) %g0 #ifdef __KERNEL__ #define PREAMBLE \ diff --git a/arch/sparc64/lib/NGcopy_to_user.S b/arch/sparc64/lib/NGcopy_to_user.S index 34112d5054e..6ea01c5532a 100644 --- a/arch/sparc64/lib/NGcopy_to_user.S +++ b/arch/sparc64/lib/NGcopy_to_user.S @@ -1,6 +1,6 @@ /* NGcopy_to_user.S: Niagara optimized copy to userspace. * - * Copyright (C) 2006 David S. Miller (davem@davemloft.net) + * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net) */ #define EX_ST(x) \ @@ -8,8 +8,8 @@ .section .fixup; \ .align 4; \ 99: wr %g0, ASI_AIUS, %asi;\ - retl; \ - mov 1, %o0; \ + ret; \ + restore %g0, 1, %o0; \ .section __ex_table,"a";\ .align 4; \ .word 98b, 99b; \ @@ -23,7 +23,7 @@ #define FUNC_NAME NGcopy_to_user #define STORE(type,src,addr) type##a src, [addr] ASI_AIUS #define STORE_ASI ASI_BLK_INIT_QUAD_LDD_AIUS -#define EX_RETVAL(x) 0 +#define EX_RETVAL(x) %g0 #ifdef __KERNEL__ /* Writing to %asi is _expensive_ so we hardcode it. diff --git a/arch/sparc64/lib/NGmemcpy.S b/arch/sparc64/lib/NGmemcpy.S index 66063a9a66b..96a14caf696 100644 --- a/arch/sparc64/lib/NGmemcpy.S +++ b/arch/sparc64/lib/NGmemcpy.S @@ -1,6 +1,6 @@ /* NGmemcpy.S: Niagara optimized memcpy. * - * Copyright (C) 2006 David S. Miller (davem@davemloft.net) + * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net) */ #ifdef __KERNEL__ @@ -16,6 +16,12 @@ wr %g0, ASI_PNF, %asi #endif +#ifdef __sparc_v9__ +#define SAVE_AMOUNT 128 +#else +#define SAVE_AMOUNT 64 +#endif + #ifndef STORE_ASI #define STORE_ASI ASI_BLK_INIT_QUAD_LDD_P #endif @@ -50,7 +56,11 @@ #endif #ifndef STORE_INIT +#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA #define STORE_INIT(src,addr) stxa src, [addr] %asi +#else +#define STORE_INIT(src,addr) stx src, [addr + 0x00] +#endif #endif #ifndef FUNC_NAME @@ -73,18 +83,19 @@ .globl FUNC_NAME .type FUNC_NAME,#function -FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ - srlx %o2, 31, %g2 +FUNC_NAME: /* %i0=dst, %i1=src, %i2=len */ + PREAMBLE + save %sp, -SAVE_AMOUNT, %sp + srlx %i2, 31, %g2 cmp %g2, 0 tne %xcc, 5 - PREAMBLE - mov %o0, GLOBAL_SPARE - cmp %o2, 0 + mov %i0, %o0 + cmp %i2, 0 be,pn %XCC, 85f - or %o0, %o1, %o3 - cmp %o2, 16 + or %o0, %i1, %i3 + cmp %i2, 16 blu,a,pn %XCC, 80f - or %o3, %o2, %o3 + or %i3, %i2, %i3 /* 2 blocks (128 bytes) is the minimum we can do the block * copy with. We need to ensure that we'll iterate at least @@ -93,31 +104,31 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ * to (64 - 1) bytes from the length before we perform the * block copy loop. */ - cmp %o2, (2 * 64) + cmp %i2, (2 * 64) blu,pt %XCC, 70f - andcc %o3, 0x7, %g0 + andcc %i3, 0x7, %g0 /* %o0: dst - * %o1: src - * %o2: len (known to be >= 128) + * %i1: src + * %i2: len (known to be >= 128) * - * The block copy loops will use %o4/%o5,%g2/%g3 as + * The block copy loops will use %i4/%i5,%g2/%g3 as * temporaries while copying the data. */ - LOAD(prefetch, %o1, #one_read) + LOAD(prefetch, %i1, #one_read) wr %g0, STORE_ASI, %asi /* Align destination on 64-byte boundary. */ - andcc %o0, (64 - 1), %o4 + andcc %o0, (64 - 1), %i4 be,pt %XCC, 2f - sub %o4, 64, %o4 - sub %g0, %o4, %o4 ! bytes to align dst - sub %o2, %o4, %o2 -1: subcc %o4, 1, %o4 - EX_LD(LOAD(ldub, %o1, %g1)) + sub %i4, 64, %i4 + sub %g0, %i4, %i4 ! bytes to align dst + sub %i2, %i4, %i2 +1: subcc %i4, 1, %i4 + EX_LD(LOAD(ldub, %i1, %g1)) EX_ST(STORE(stb, %g1, %o0)) - add %o1, 1, %o1 + add %i1, 1, %i1 bne,pt %XCC, 1b add %o0, 1, %o0 @@ -136,111 +147,155 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ * aligned store data at a time, this is easy to ensure. */ 2: - andcc %o1, (16 - 1), %o4 - andn %o2, (64 - 1), %g1 ! block copy loop iterator - sub %o2, %g1, %o2 ! final sub-block copy bytes + andcc %i1, (16 - 1), %i4 + andn %i2, (64 - 1), %g1 ! block copy loop iterator be,pt %XCC, 50f - cmp %o4, 8 - be,a,pt %XCC, 10f - sub %o1, 0x8, %o1 + sub %i2, %g1, %i2 ! final sub-block copy bytes + + cmp %i4, 8 + be,pt %XCC, 10f + sub %i1, %i4, %i1 /* Neither 8-byte nor 16-byte aligned, shift and mask. */ - mov %g1, %o4 - and %o1, 0x7, %g1 - sll %g1, 3, %g1 - mov 64, %o3 - andn %o1, 0x7, %o1 - EX_LD(LOAD(ldx, %o1, %g2)) - sub %o3, %g1, %o3 - sllx %g2, %g1, %g2 + and %i4, 0x7, GLOBAL_SPARE + sll GLOBAL_SPARE, 3, GLOBAL_SPARE + mov 64, %i5 + EX_LD(LOAD_TWIN(%i1, %g2, %g3)) + sub %i5, GLOBAL_SPARE, %i5 + mov 16, %o4 + mov 32, %o5 + mov 48, %o7 + mov 64, %i3 + + bg,pn %XCC, 9f + nop -#define SWIVEL_ONE_DWORD(SRC, TMP1, TMP2, PRE_VAL, PRE_SHIFT, POST_SHIFT, DST)\ - EX_LD(LOAD(ldx, SRC, TMP1)); \ - srlx TMP1, PRE_SHIFT, TMP2; \ - or TMP2, PRE_VAL, TMP2; \ - EX_ST(STORE_INIT(TMP2, DST)); \ - sllx TMP1, POST_SHIFT, PRE_VAL; - -1: add %o1, 0x8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x00) - add %o1, 0x8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x08) - add %o1, 0x8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x10) - add %o1, 0x8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x18) - add %o1, 32, %o1 - LOAD(prefetch, %o1, #one_read) - sub %o1, 32 - 8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x20) - add %o1, 8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x28) - add %o1, 8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x30) - add %o1, 8, %o1 - SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x38) - subcc %o4, 64, %o4 - bne,pt %XCC, 1b +#define MIX_THREE_WORDS(WORD1, WORD2, WORD3, PRE_SHIFT, POST_SHIFT, TMP) \ + sllx WORD1, POST_SHIFT, WORD1; \ + srlx WORD2, PRE_SHIFT, TMP; \ + sllx WORD2, POST_SHIFT, WORD2; \ + or WORD1, TMP, WORD1; \ + srlx WORD3, PRE_SHIFT, TMP; \ + or WORD2, TMP, WORD2; + +8: EX_LD(LOAD_TWIN(%i1 + %o4, %o2, %o3)) + MIX_THREE_WORDS(%g2, %g3, %o2, %i5, GLOBAL_SPARE, %o1) + LOAD(prefetch, %i1 + %i3, #one_read) + + EX_ST(STORE_INIT(%g2, %o0 + 0x00)) + EX_ST(STORE_INIT(%g3, %o0 + 0x08)) + + EX_LD(LOAD_TWIN(%i1 + %o5, %g2, %g3)) + MIX_THREE_WORDS(%o2, %o3, %g2, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%o2, %o0 + 0x10)) + EX_ST(STORE_INIT(%o3, %o0 + 0x18)) + + EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) + MIX_THREE_WORDS(%g2, %g3, %o2, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%g2, %o0 + 0x20)) + EX_ST(STORE_INIT(%g3, %o0 + 0x28)) + + EX_LD(LOAD_TWIN(%i1 + %i3, %g2, %g3)) + add %i1, 64, %i1 + MIX_THREE_WORDS(%o2, %o3, %g2, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%o2, %o0 + 0x30)) + EX_ST(STORE_INIT(%o3, %o0 + 0x38)) + + subcc %g1, 64, %g1 + bne,pt %XCC, 8b add %o0, 64, %o0 -#undef SWIVEL_ONE_DWORD + ba,pt %XCC, 60f + add %i1, %i4, %i1 + +9: EX_LD(LOAD_TWIN(%i1 + %o4, %o2, %o3)) + MIX_THREE_WORDS(%g3, %o2, %o3, %i5, GLOBAL_SPARE, %o1) + LOAD(prefetch, %i1 + %i3, #one_read) + + EX_ST(STORE_INIT(%g3, %o0 + 0x00)) + EX_ST(STORE_INIT(%o2, %o0 + 0x08)) + + EX_LD(LOAD_TWIN(%i1 + %o5, %g2, %g3)) + MIX_THREE_WORDS(%o3, %g2, %g3, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%o3, %o0 + 0x10)) + EX_ST(STORE_INIT(%g2, %o0 + 0x18)) + + EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) + MIX_THREE_WORDS(%g3, %o2, %o3, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%g3, %o0 + 0x20)) + EX_ST(STORE_INIT(%o2, %o0 + 0x28)) + + EX_LD(LOAD_TWIN(%i1 + %i3, %g2, %g3)) + add %i1, 64, %i1 + MIX_THREE_WORDS(%o3, %g2, %g3, %i5, GLOBAL_SPARE, %o1) + + EX_ST(STORE_INIT(%o3, %o0 + 0x30)) + EX_ST(STORE_INIT(%g2, %o0 + 0x38)) + + subcc %g1, 64, %g1 + bne,pt %XCC, 9b + add %o0, 64, %o0 - srl %g1, 3, %g1 ba,pt %XCC, 60f - add %o1, %g1, %o1 + add %i1, %i4, %i1 10: /* Destination is 64-byte aligned, source was only 8-byte * aligned but it has been subtracted by 8 and we perform * one twin load ahead, then add 8 back into source when * we finish the loop. */ - EX_LD(LOAD_TWIN(%o1, %o4, %o5)) -1: add %o1, 16, %o1 - EX_LD(LOAD_TWIN(%o1, %g2, %g3)) - add %o1, 16 + 32, %o1 - LOAD(prefetch, %o1, #one_read) - sub %o1, 32, %o1 + EX_LD(LOAD_TWIN(%i1, %o4, %o5)) + mov 16, %o7 + mov 32, %g2 + mov 48, %g3 + mov 64, %o1 +1: EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) + LOAD(prefetch, %i1 + %o1, #one_read) EX_ST(STORE_INIT(%o5, %o0 + 0x00)) ! initializes cache line - EX_ST(STORE_INIT(%g2, %o0 + 0x08)) - EX_LD(LOAD_TWIN(%o1, %o4, %o5)) - add %o1, 16, %o1 - EX_ST(STORE_INIT(%g3, %o0 + 0x10)) + EX_ST(STORE_INIT(%o2, %o0 + 0x08)) + EX_LD(LOAD_TWIN(%i1 + %g2, %o4, %o5)) + EX_ST(STORE_INIT(%o3, %o0 + 0x10)) EX_ST(STORE_INIT(%o4, %o0 + 0x18)) - EX_LD(LOAD_TWIN(%o1, %g2, %g3)) - add %o1, 16, %o1 + EX_LD(LOAD_TWIN(%i1 + %g3, %o2, %o3)) EX_ST(STORE_INIT(%o5, %o0 + 0x20)) - EX_ST(STORE_INIT(%g2, %o0 + 0x28)) - EX_LD(LOAD_TWIN(%o1, %o4, %o5)) - EX_ST(STORE_INIT(%g3, %o0 + 0x30)) + EX_ST(STORE_INIT(%o2, %o0 + 0x28)) + EX_LD(LOAD_TWIN(%i1 + %o1, %o4, %o5)) + add %i1, 64, %i1 + EX_ST(STORE_INIT(%o3, %o0 + 0x30)) EX_ST(STORE_INIT(%o4, %o0 + 0x38)) subcc %g1, 64, %g1 bne,pt %XCC, 1b add %o0, 64, %o0 ba,pt %XCC, 60f - add %o1, 0x8, %o1 + add %i1, 0x8, %i1 50: /* Destination is 64-byte aligned, and source is 16-byte * aligned. */ -1: EX_LD(LOAD_TWIN(%o1, %o4, %o5)) - add %o1, 16, %o1 - EX_LD(LOAD_TWIN(%o1, %g2, %g3)) - add %o1, 16 + 32, %o1 - LOAD(prefetch, %o1, #one_read) - sub %o1, 32, %o1 + mov 16, %o7 + mov 32, %g2 + mov 48, %g3 + mov 64, %o1 +1: EX_LD(LOAD_TWIN(%i1 + %g0, %o4, %o5)) + EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) + LOAD(prefetch, %i1 + %o1, #one_read) EX_ST(STORE_INIT(%o4, %o0 + 0x00)) ! initializes cache line EX_ST(STORE_INIT(%o5, %o0 + 0x08)) - EX_LD(LOAD_TWIN(%o1, %o4, %o5)) - add %o1, 16, %o1 - EX_ST(STORE_INIT(%g2, %o0 + 0x10)) - EX_ST(STORE_INIT(%g3, %o0 + 0x18)) - EX_LD(LOAD_TWIN(%o1, %g2, %g3)) - add %o1, 16, %o1 + EX_LD(LOAD_TWIN(%i1 + %g2, %o4, %o5)) + EX_ST(STORE_INIT(%o2, %o0 + 0x10)) + EX_ST(STORE_INIT(%o3, %o0 + 0x18)) + EX_LD(LOAD_TWIN(%i1 + %g3, %o2, %o3)) + add %i1, 64, %i1 EX_ST(STORE_INIT(%o4, %o0 + 0x20)) EX_ST(STORE_INIT(%o5, %o0 + 0x28)) - EX_ST(STORE_INIT(%g2, %o0 + 0x30)) - EX_ST(STORE_INIT(%g3, %o0 + 0x38)) + EX_ST(STORE_INIT(%o2, %o0 + 0x30)) + EX_ST(STORE_INIT(%o3, %o0 + 0x38)) subcc %g1, 64, %g1 bne,pt %XCC, 1b add %o0, 64, %o0 @@ -249,47 +304,47 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ 60: membar #Sync - /* %o2 contains any final bytes still needed to be copied + /* %i2 contains any final bytes still needed to be copied * over. If anything is left, we copy it one byte at a time. */ - RESTORE_ASI(%o3) - brz,pt %o2, 85f - sub %o0, %o1, %o3 + RESTORE_ASI(%i3) + brz,pt %i2, 85f + sub %o0, %i1, %i3 ba,a,pt %XCC, 90f .align 64 70: /* 16 < len <= 64 */ bne,pn %XCC, 75f - sub %o0, %o1, %o3 + sub %o0, %i1, %i3 72: - andn %o2, 0xf, %o4 - and %o2, 0xf, %o2 -1: subcc %o4, 0x10, %o4 - EX_LD(LOAD(ldx, %o1, %o5)) - add %o1, 0x08, %o1 - EX_LD(LOAD(ldx, %o1, %g1)) - sub %o1, 0x08, %o1 - EX_ST(STORE(stx, %o5, %o1 + %o3)) - add %o1, 0x8, %o1 - EX_ST(STORE(stx, %g1, %o1 + %o3)) + andn %i2, 0xf, %i4 + and %i2, 0xf, %i2 +1: subcc %i4, 0x10, %i4 + EX_LD(LOAD(ldx, %i1, %o4)) + add %i1, 0x08, %i1 + EX_LD(LOAD(ldx, %i1, %g1)) + sub %i1, 0x08, %i1 + EX_ST(STORE(stx, %o4, %i1 + %i3)) + add %i1, 0x8, %i1 + EX_ST(STORE(stx, %g1, %i1 + %i3)) bgu,pt %XCC, 1b - add %o1, 0x8, %o1 -73: andcc %o2, 0x8, %g0 + add %i1, 0x8, %i1 +73: andcc %i2, 0x8, %g0 be,pt %XCC, 1f nop - sub %o2, 0x8, %o2 - EX_LD(LOAD(ldx, %o1, %o5)) - EX_ST(STORE(stx, %o5, %o1 + %o3)) - add %o1, 0x8, %o1 -1: andcc %o2, 0x4, %g0 + sub %i2, 0x8, %i2 + EX_LD(LOAD(ldx, %i1, %o4)) + EX_ST(STORE(stx, %o4, %i1 + %i3)) + add %i1, 0x8, %i1 +1: andcc %i2, 0x4, %g0 be,pt %XCC, 1f nop - sub %o2, 0x4, %o2 - EX_LD(LOAD(lduw, %o1, %o5)) - EX_ST(STORE(stw, %o5, %o1 + %o3)) - add %o1, 0x4, %o1 -1: cmp %o2, 0 + sub %i2, 0x4, %i2 + EX_LD(LOAD(lduw, %i1, %i5)) + EX_ST(STORE(stw, %i5, %i1 + %i3)) + add %i1, 0x4, %i1 +1: cmp %i2, 0 be,pt %XCC, 85f nop ba,pt %xcc, 90f @@ -300,71 +355,71 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ sub %g1, 0x8, %g1 be,pn %icc, 2f sub %g0, %g1, %g1 - sub %o2, %g1, %o2 + sub %i2, %g1, %i2 1: subcc %g1, 1, %g1 - EX_LD(LOAD(ldub, %o1, %o5)) - EX_ST(STORE(stb, %o5, %o1 + %o3)) + EX_LD(LOAD(ldub, %i1, %i5)) + EX_ST(STORE(stb, %i5, %i1 + %i3)) bgu,pt %icc, 1b - add %o1, 1, %o1 + add %i1, 1, %i1 -2: add %o1, %o3, %o0 - andcc %o1, 0x7, %g1 +2: add %i1, %i3, %o0 + andcc %i1, 0x7, %g1 bne,pt %icc, 8f sll %g1, 3, %g1 - cmp %o2, 16 + cmp %i2, 16 bgeu,pt %icc, 72b nop ba,a,pt %xcc, 73b -8: mov 64, %o3 - andn %o1, 0x7, %o1 - EX_LD(LOAD(ldx, %o1, %g2)) - sub %o3, %g1, %o3 - andn %o2, 0x7, %o4 +8: mov 64, %i3 + andn %i1, 0x7, %i1 + EX_LD(LOAD(ldx, %i1, %g2)) + sub %i3, %g1, %i3 + andn %i2, 0x7, %i4 sllx %g2, %g1, %g2 -1: add %o1, 0x8, %o1 - EX_LD(LOAD(ldx, %o1, %g3)) - subcc %o4, 0x8, %o4 - srlx %g3, %o3, %o5 - or %o5, %g2, %o5 - EX_ST(STORE(stx, %o5, %o0)) +1: add %i1, 0x8, %i1 + EX_LD(LOAD(ldx, %i1, %g3)) + subcc %i4, 0x8, %i4 + srlx %g3, %i3, %i5 + or %i5, %g2, %i5 + EX_ST(STORE(stx, %i5, %o0)) add %o0, 0x8, %o0 bgu,pt %icc, 1b sllx %g3, %g1, %g2 srl %g1, 3, %g1 - andcc %o2, 0x7, %o2 + andcc %i2, 0x7, %i2 be,pn %icc, 85f - add %o1, %g1, %o1 + add %i1, %g1, %i1 ba,pt %xcc, 90f - sub %o0, %o1, %o3 + sub %o0, %i1, %i3 .align 64 80: /* 0 < len <= 16 */ - andcc %o3, 0x3, %g0 + andcc %i3, 0x3, %g0 bne,pn %XCC, 90f - sub %o0, %o1, %o3 + sub %o0, %i1, %i3 1: - subcc %o2, 4, %o2 - EX_LD(LOAD(lduw, %o1, %g1)) - EX_ST(STORE(stw, %g1, %o1 + %o3)) + subcc %i2, 4, %i2 + EX_LD(LOAD(lduw, %i1, %g1)) + EX_ST(STORE(stw, %g1, %i1 + %i3)) bgu,pt %XCC, 1b - add %o1, 4, %o1 + add %i1, 4, %i1 -85: retl - mov EX_RETVAL(GLOBAL_SPARE), %o0 +85: ret + restore EX_RETVAL(%i0), %g0, %o0 .align 32 90: - subcc %o2, 1, %o2 - EX_LD(LOAD(ldub, %o1, %g1)) - EX_ST(STORE(stb, %g1, %o1 + %o3)) + subcc %i2, 1, %i2 + EX_LD(LOAD(ldub, %i1, %g1)) + EX_ST(STORE(stb, %g1, %i1 + %i3)) bgu,pt %XCC, 90b - add %o1, 1, %o1 - retl - mov EX_RETVAL(GLOBAL_SPARE), %o0 + add %i1, 1, %i1 + ret + restore EX_RETVAL(%i0), %g0, %o0 .size FUNC_NAME, .-FUNC_NAME diff --git a/arch/sparc64/mm/hugetlbpage.c b/arch/sparc64/mm/hugetlbpage.c index eaba9b70b18..6cfab2e4d34 100644 --- a/arch/sparc64/mm/hugetlbpage.c +++ b/arch/sparc64/mm/hugetlbpage.c @@ -175,7 +175,7 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr, return -ENOMEM; if (flags & MAP_FIXED) { - if (prepare_hugepage_range(addr, len, pgoff)) + if (prepare_hugepage_range(addr, len)) return -EINVAL; return addr; } diff --git a/arch/sparc64/prom/misc.c b/arch/sparc64/prom/misc.c index 68c83ad04ad..bbec7522826 100644 --- a/arch/sparc64/prom/misc.c +++ b/arch/sparc64/prom/misc.c @@ -143,22 +143,6 @@ unsigned char prom_get_idprom(char *idbuf, int num_bytes) return 0xff; } -/* Install Linux trap table so PROM uses that instead of its own. */ -void prom_set_trap_table(unsigned long tba) -{ - p1275_cmd("SUNW,set-trap-table", - (P1275_ARG(0, P1275_ARG_IN_64B) | - P1275_INOUT(1, 0)), tba); -} - -void prom_set_trap_table_sun4v(unsigned long tba, unsigned long mmfsa) -{ - p1275_cmd("SUNW,set-trap-table", - (P1275_ARG(0, P1275_ARG_IN_64B) | - P1275_ARG(1, P1275_ARG_IN_64B) | - P1275_INOUT(2, 0)), tba, mmfsa); -} - int prom_get_mmu_ihandle(void) { int node, ret; |