diff options
Diffstat (limited to 'arch/x86')
-rw-r--r-- | arch/x86/include/asm/amd_nb.h | 2 | ||||
-rw-r--r-- | arch/x86/include/asm/uv/uv_bau.h | 106 | ||||
-rw-r--r-- | arch/x86/kernel/amd_nb.c | 31 | ||||
-rw-r--r-- | arch/x86/kernel/apic/x2apic_uv_x.c | 7 | ||||
-rw-r--r-- | arch/x86/mm/mmap.c | 4 | ||||
-rw-r--r-- | arch/x86/mm/srat.c | 4 | ||||
-rw-r--r-- | arch/x86/pci/Makefile | 3 | ||||
-rw-r--r-- | arch/x86/pci/acpi.c | 18 | ||||
-rw-r--r-- | arch/x86/pci/amd_bus.c | 42 | ||||
-rw-r--r-- | arch/x86/platform/uv/tlb_uv.c | 363 |
10 files changed, 480 insertions, 100 deletions
diff --git a/arch/x86/include/asm/amd_nb.h b/arch/x86/include/asm/amd_nb.h index 8e41071704a5..49ad773f4b9f 100644 --- a/arch/x86/include/asm/amd_nb.h +++ b/arch/x86/include/asm/amd_nb.h @@ -1,6 +1,7 @@ #ifndef _ASM_X86_AMD_NB_H #define _ASM_X86_AMD_NB_H +#include <linux/ioport.h> #include <linux/pci.h> struct amd_nb_bus_dev_range { @@ -13,6 +14,7 @@ extern const struct pci_device_id amd_nb_misc_ids[]; extern const struct amd_nb_bus_dev_range amd_nb_bus_dev_ranges[]; extern bool early_is_amd_nb(u32 value); +extern struct resource *amd_get_mmconfig_range(struct resource *res); extern int amd_cache_northbridges(void); extern void amd_flush_garts(void); extern int amd_numa_init(void); diff --git a/arch/x86/include/asm/uv/uv_bau.h b/arch/x86/include/asm/uv/uv_bau.h index 8e862aaf0d90..1b82f7e87393 100644 --- a/arch/x86/include/asm/uv/uv_bau.h +++ b/arch/x86/include/asm/uv/uv_bau.h @@ -65,7 +65,7 @@ * UV2: Bit 19 selects between * (0): 10 microsecond timebase and * (1): 80 microseconds - * we're using 655us, similar to UV1: 65 units of 10us + * we're using 560us, similar to UV1: 65 units of 10us */ #define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL) #define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (15UL) @@ -167,6 +167,7 @@ #define FLUSH_RETRY_TIMEOUT 2 #define FLUSH_GIVEUP 3 #define FLUSH_COMPLETE 4 +#define FLUSH_RETRY_BUSYBUG 5 /* * tuning the action when the numalink network is extremely delayed @@ -235,10 +236,10 @@ struct bau_msg_payload { /* - * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor) + * UV1 Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor) * see table 4.2.3.0.1 in broacast_assist spec. */ -struct bau_msg_header { +struct uv1_bau_msg_header { unsigned int dest_subnodeid:6; /* must be 0x10, for the LB */ /* bits 5:0 */ unsigned int base_dest_nasid:15; /* nasid of the first bit */ @@ -318,19 +319,87 @@ struct bau_msg_header { }; /* + * UV2 Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor) + * see figure 9-2 of harp_sys.pdf + */ +struct uv2_bau_msg_header { + unsigned int base_dest_nasid:15; /* nasid of the first bit */ + /* bits 14:0 */ /* in uvhub map */ + unsigned int dest_subnodeid:5; /* must be 0x10, for the LB */ + /* bits 19:15 */ + unsigned int rsvd_1:1; /* must be zero */ + /* bit 20 */ + /* Address bits 59:21 */ + /* bits 25:2 of address (44:21) are payload */ + /* these next 24 bits become bytes 12-14 of msg */ + /* bits 28:21 land in byte 12 */ + unsigned int replied_to:1; /* sent as 0 by the source to + byte 12 */ + /* bit 21 */ + unsigned int msg_type:3; /* software type of the + message */ + /* bits 24:22 */ + unsigned int canceled:1; /* message canceled, resource + is to be freed*/ + /* bit 25 */ + unsigned int payload_1:3; /* not currently used */ + /* bits 28:26 */ + + /* bits 36:29 land in byte 13 */ + unsigned int payload_2a:3; /* not currently used */ + unsigned int payload_2b:5; /* not currently used */ + /* bits 36:29 */ + + /* bits 44:37 land in byte 14 */ + unsigned int payload_3:8; /* not currently used */ + /* bits 44:37 */ + + unsigned int rsvd_2:7; /* reserved */ + /* bits 51:45 */ + unsigned int swack_flag:1; /* software acknowledge flag */ + /* bit 52 */ + unsigned int rsvd_3a:3; /* must be zero */ + unsigned int rsvd_3b:8; /* must be zero */ + unsigned int rsvd_3c:8; /* must be zero */ + unsigned int rsvd_3d:3; /* must be zero */ + /* bits 74:53 */ + unsigned int fairness:3; /* usually zero */ + /* bits 77:75 */ + + unsigned int sequence:16; /* message sequence number */ + /* bits 93:78 Suppl_A */ + unsigned int chaining:1; /* next descriptor is part of + this activation*/ + /* bit 94 */ + unsigned int multilevel:1; /* multi-level multicast + format */ + /* bit 95 */ + unsigned int rsvd_4:24; /* ordered / source node / + source subnode / aging + must be zero */ + /* bits 119:96 */ + unsigned int command:8; /* message type */ + /* bits 127:120 */ +}; + +/* * The activation descriptor: * The format of the message to send, plus all accompanying control * Should be 64 bytes */ struct bau_desc { - struct pnmask distribution; + struct pnmask distribution; /* * message template, consisting of header and payload: */ - struct bau_msg_header header; - struct bau_msg_payload payload; + union bau_msg_header { + struct uv1_bau_msg_header uv1_hdr; + struct uv2_bau_msg_header uv2_hdr; + } header; + + struct bau_msg_payload payload; }; -/* +/* UV1: * -payload-- ---------header------ * bytes 0-11 bits 41-56 bits 58-81 * A B (2) C (3) @@ -340,6 +409,16 @@ struct bau_desc { * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector) * ------------payload queue----------- */ +/* UV2: + * -payload-- ---------header------ + * bytes 0-11 bits 70-78 bits 21-44 + * A B (2) C (3) + * + * A/B/C are moved to: + * A C B + * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector) + * ------------payload queue----------- + */ /* * The payload queue on the destination side is an array of these. @@ -385,7 +464,6 @@ struct bau_pq_entry { struct msg_desc { struct bau_pq_entry *msg; int msg_slot; - int swack_slot; struct bau_pq_entry *queue_first; struct bau_pq_entry *queue_last; }; @@ -439,6 +517,9 @@ struct ptc_stats { unsigned long s_retry_messages; /* retry broadcasts */ unsigned long s_bau_reenabled; /* for bau enable/disable */ unsigned long s_bau_disabled; /* for bau enable/disable */ + unsigned long s_uv2_wars; /* uv2 workaround, perm. busy */ + unsigned long s_uv2_wars_hw; /* uv2 workaround, hiwater */ + unsigned long s_uv2_war_waits; /* uv2 workaround, long waits */ /* destination statistics */ unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */ @@ -511,9 +592,12 @@ struct bau_control { short osnode; short uvhub_cpu; short uvhub; + short uvhub_version; short cpus_in_socket; short cpus_in_uvhub; short partition_base_pnode; + short using_desc; /* an index, like uvhub_cpu */ + unsigned int inuse_map; unsigned short message_number; unsigned short uvhub_quiesce; short socket_acknowledge_count[DEST_Q_SIZE]; @@ -531,6 +615,7 @@ struct bau_control { int cong_response_us; int cong_reps; int cong_period; + unsigned long clocks_per_100_usec; cycles_t period_time; long period_requests; struct hub_and_pnode *thp; @@ -591,6 +676,11 @@ static inline void write_mmr_sw_ack(unsigned long mr) uv_write_local_mmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, mr); } +static inline void write_gmmr_sw_ack(int pnode, unsigned long mr) +{ + write_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, mr); +} + static inline unsigned long read_mmr_sw_ack(void) { return read_lmmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE); diff --git a/arch/x86/kernel/amd_nb.c b/arch/x86/kernel/amd_nb.c index 4c39baa8facc..bae1efe6d515 100644 --- a/arch/x86/kernel/amd_nb.c +++ b/arch/x86/kernel/amd_nb.c @@ -119,6 +119,37 @@ bool __init early_is_amd_nb(u32 device) return false; } +struct resource *amd_get_mmconfig_range(struct resource *res) +{ + u32 address; + u64 base, msr; + unsigned segn_busn_bits; + + if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) + return NULL; + + /* assume all cpus from fam10h have mmconfig */ + if (boot_cpu_data.x86 < 0x10) + return NULL; + + address = MSR_FAM10H_MMIO_CONF_BASE; + rdmsrl(address, msr); + + /* mmconfig is not enabled */ + if (!(msr & FAM10H_MMIO_CONF_ENABLE)) + return NULL; + + base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT); + + segn_busn_bits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) & + FAM10H_MMIO_CONF_BUSRANGE_MASK; + + res->flags = IORESOURCE_MEM; + res->start = base; + res->end = base + (1ULL<<(segn_busn_bits + 20)) - 1; + return res; +} + int amd_get_subcaches(int cpu) { struct pci_dev *link = node_to_amd_nb(amd_get_nb_id(cpu))->link; diff --git a/arch/x86/kernel/apic/x2apic_uv_x.c b/arch/x86/kernel/apic/x2apic_uv_x.c index 9d59bbacd4e3..79b05b88aa19 100644 --- a/arch/x86/kernel/apic/x2apic_uv_x.c +++ b/arch/x86/kernel/apic/x2apic_uv_x.c @@ -769,7 +769,12 @@ void __init uv_system_init(void) for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) uv_possible_blades += hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8)); - printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades()); + + /* uv_num_possible_blades() is really the hub count */ + printk(KERN_INFO "UV: Found %d blades, %d hubs\n", + is_uv1_hub() ? uv_num_possible_blades() : + (uv_num_possible_blades() + 1) / 2, + uv_num_possible_blades()); bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades(); uv_blade_info = kzalloc(bytes, GFP_KERNEL); diff --git a/arch/x86/mm/mmap.c b/arch/x86/mm/mmap.c index 4b5ba85eb5c9..845df6835f9f 100644 --- a/arch/x86/mm/mmap.c +++ b/arch/x86/mm/mmap.c @@ -75,9 +75,9 @@ static unsigned long mmap_rnd(void) */ if (current->flags & PF_RANDOMIZE) { if (mmap_is_ia32()) - rnd = (long)get_random_int() % (1<<8); + rnd = get_random_int() % (1<<8); else - rnd = (long)(get_random_int() % (1<<28)); + rnd = get_random_int() % (1<<28); } return rnd << PAGE_SHIFT; } diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c index 81dbfdeb080d..7efd0c615d58 100644 --- a/arch/x86/mm/srat.c +++ b/arch/x86/mm/srat.c @@ -104,6 +104,8 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa) if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0) return; pxm = pa->proximity_domain_lo; + if (acpi_srat_revision >= 2) + pxm |= *((unsigned int*)pa->proximity_domain_hi) << 8; node = setup_node(pxm); if (node < 0) { printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm); @@ -155,6 +157,8 @@ acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma) start = ma->base_address; end = start + ma->length; pxm = ma->proximity_domain; + if (acpi_srat_revision <= 1) + pxm &= 0xff; node = setup_node(pxm); if (node < 0) { printk(KERN_ERR "SRAT: Too many proximity domains.\n"); diff --git a/arch/x86/pci/Makefile b/arch/x86/pci/Makefile index 6b8759f7634e..d24d3da72926 100644 --- a/arch/x86/pci/Makefile +++ b/arch/x86/pci/Makefile @@ -18,8 +18,9 @@ obj-$(CONFIG_X86_NUMAQ) += numaq_32.o obj-$(CONFIG_X86_MRST) += mrst.o obj-y += common.o early.o -obj-y += amd_bus.o bus_numa.o +obj-y += bus_numa.o +obj-$(CONFIG_AMD_NB) += amd_bus.o obj-$(CONFIG_PCI_CNB20LE_QUIRK) += broadcom_bus.o ifeq ($(CONFIG_PCI_DEBUG),y) diff --git a/arch/x86/pci/acpi.c b/arch/x86/pci/acpi.c index 404f21a3ff9e..f8348ab10324 100644 --- a/arch/x86/pci/acpi.c +++ b/arch/x86/pci/acpi.c @@ -149,7 +149,7 @@ setup_resource(struct acpi_resource *acpi_res, void *data) struct acpi_resource_address64 addr; acpi_status status; unsigned long flags; - u64 start, end; + u64 start, orig_end, end; status = resource_to_addr(acpi_res, &addr); if (!ACPI_SUCCESS(status)) @@ -165,7 +165,21 @@ setup_resource(struct acpi_resource *acpi_res, void *data) return AE_OK; start = addr.minimum + addr.translation_offset; - end = addr.maximum + addr.translation_offset; + orig_end = end = addr.maximum + addr.translation_offset; + + /* Exclude non-addressable range or non-addressable portion of range */ + end = min(end, (u64)iomem_resource.end); + if (end <= start) { + dev_info(&info->bridge->dev, + "host bridge window [%#llx-%#llx] " + "(ignored, not CPU addressable)\n", start, orig_end); + return AE_OK; + } else if (orig_end != end) { + dev_info(&info->bridge->dev, + "host bridge window [%#llx-%#llx] " + "([%#llx-%#llx] ignored, not CPU addressable)\n", + start, orig_end, end + 1, orig_end); + } res = &info->res[info->res_num]; res->name = info->name; diff --git a/arch/x86/pci/amd_bus.c b/arch/x86/pci/amd_bus.c index 026e4931d162..385a940b5422 100644 --- a/arch/x86/pci/amd_bus.c +++ b/arch/x86/pci/amd_bus.c @@ -30,34 +30,6 @@ static struct pci_hostbridge_probe pci_probes[] __initdata = { { 0, 0x18, PCI_VENDOR_ID_AMD, 0x1300 }, }; -static u64 __initdata fam10h_mmconf_start; -static u64 __initdata fam10h_mmconf_end; -static void __init get_pci_mmcfg_amd_fam10h_range(void) -{ - u32 address; - u64 base, msr; - unsigned segn_busn_bits; - - /* assume all cpus from fam10h have mmconf */ - if (boot_cpu_data.x86 < 0x10) - return; - - address = MSR_FAM10H_MMIO_CONF_BASE; - rdmsrl(address, msr); - - /* mmconfig is not enable */ - if (!(msr & FAM10H_MMIO_CONF_ENABLE)) - return; - - base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT); - - segn_busn_bits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) & - FAM10H_MMIO_CONF_BUSRANGE_MASK; - - fam10h_mmconf_start = base; - fam10h_mmconf_end = base + (1ULL<<(segn_busn_bits + 20)) - 1; -} - #define RANGE_NUM 16 /** @@ -85,6 +57,9 @@ static int __init early_fill_mp_bus_info(void) u64 val; u32 address; bool found; + struct resource fam10h_mmconf_res, *fam10h_mmconf; + u64 fam10h_mmconf_start; + u64 fam10h_mmconf_end; if (!early_pci_allowed()) return -1; @@ -211,12 +186,17 @@ static int __init early_fill_mp_bus_info(void) subtract_range(range, RANGE_NUM, 0, end); /* get mmconfig */ - get_pci_mmcfg_amd_fam10h_range(); + fam10h_mmconf = amd_get_mmconfig_range(&fam10h_mmconf_res); /* need to take out mmconf range */ - if (fam10h_mmconf_end) { - printk(KERN_DEBUG "Fam 10h mmconf [%llx, %llx]\n", fam10h_mmconf_start, fam10h_mmconf_end); + if (fam10h_mmconf) { + printk(KERN_DEBUG "Fam 10h mmconf %pR\n", fam10h_mmconf); + fam10h_mmconf_start = fam10h_mmconf->start; + fam10h_mmconf_end = fam10h_mmconf->end; subtract_range(range, RANGE_NUM, fam10h_mmconf_start, fam10h_mmconf_end + 1); + } else { + fam10h_mmconf_start = 0; + fam10h_mmconf_end = 0; } /* mmio resource */ diff --git a/arch/x86/platform/uv/tlb_uv.c b/arch/x86/platform/uv/tlb_uv.c index 5b552198f774..9010ca715c03 100644 --- a/arch/x86/platform/uv/tlb_uv.c +++ b/arch/x86/platform/uv/tlb_uv.c @@ -157,13 +157,14 @@ static int __init uvhub_to_first_apicid(int uvhub) * clear of the Timeout bit (as well) will free the resource. No reply will * be sent (the hardware will only do one reply per message). */ -static void reply_to_message(struct msg_desc *mdp, struct bau_control *bcp) +static void reply_to_message(struct msg_desc *mdp, struct bau_control *bcp, + int do_acknowledge) { unsigned long dw; struct bau_pq_entry *msg; msg = mdp->msg; - if (!msg->canceled) { + if (!msg->canceled && do_acknowledge) { dw = (msg->swack_vec << UV_SW_ACK_NPENDING) | msg->swack_vec; write_mmr_sw_ack(dw); } @@ -212,8 +213,8 @@ static void bau_process_retry_msg(struct msg_desc *mdp, if (mmr & (msg_res << UV_SW_ACK_NPENDING)) { unsigned long mr; /* - * is the resource timed out? - * make everyone ignore the cancelled message. + * Is the resource timed out? + * Make everyone ignore the cancelled message. */ msg2->canceled = 1; stat->d_canceled++; @@ -231,8 +232,8 @@ static void bau_process_retry_msg(struct msg_desc *mdp, * Do all the things a cpu should do for a TLB shootdown message. * Other cpu's may come here at the same time for this message. */ -static void bau_process_message(struct msg_desc *mdp, - struct bau_control *bcp) +static void bau_process_message(struct msg_desc *mdp, struct bau_control *bcp, + int do_acknowledge) { short socket_ack_count = 0; short *sp; @@ -284,8 +285,9 @@ static void bau_process_message(struct msg_desc *mdp, if (msg_ack_count == bcp->cpus_in_uvhub) { /* * All cpus in uvhub saw it; reply + * (unless we are in the UV2 workaround) */ - reply_to_message(mdp, bcp); + reply_to_message(mdp, bcp, do_acknowledge); } } @@ -491,27 +493,138 @@ static int uv1_wait_completion(struct bau_desc *bau_desc, /* * UV2 has an extra bit of status in the ACTIVATION_STATUS_2 register. */ -static unsigned long uv2_read_status(unsigned long offset, int rshft, int cpu) +static unsigned long uv2_read_status(unsigned long offset, int rshft, int desc) { unsigned long descriptor_status; unsigned long descriptor_status2; descriptor_status = ((read_lmmr(offset) >> rshft) & UV_ACT_STATUS_MASK); - descriptor_status2 = (read_mmr_uv2_status() >> cpu) & 0x1UL; + descriptor_status2 = (read_mmr_uv2_status() >> desc) & 0x1UL; descriptor_status = (descriptor_status << 1) | descriptor_status2; return descriptor_status; } +/* + * Return whether the status of the descriptor that is normally used for this + * cpu (the one indexed by its hub-relative cpu number) is busy. + * The status of the original 32 descriptors is always reflected in the 64 + * bits of UVH_LB_BAU_SB_ACTIVATION_STATUS_0. + * The bit provided by the activation_status_2 register is irrelevant to + * the status if it is only being tested for busy or not busy. + */ +int normal_busy(struct bau_control *bcp) +{ + int cpu = bcp->uvhub_cpu; + int mmr_offset; + int right_shift; + + mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0; + right_shift = cpu * UV_ACT_STATUS_SIZE; + return (((((read_lmmr(mmr_offset) >> right_shift) & + UV_ACT_STATUS_MASK)) << 1) == UV2H_DESC_BUSY); +} + +/* + * Entered when a bau descriptor has gone into a permanent busy wait because + * of a hardware bug. + * Workaround the bug. + */ +int handle_uv2_busy(struct bau_control *bcp) +{ + int busy_one = bcp->using_desc; + int normal = bcp->uvhub_cpu; + int selected = -1; + int i; + unsigned long descriptor_status; + unsigned long status; + int mmr_offset; + struct bau_desc *bau_desc_old; + struct bau_desc *bau_desc_new; + struct bau_control *hmaster = bcp->uvhub_master; + struct ptc_stats *stat = bcp->statp; + cycles_t ttm; + + stat->s_uv2_wars++; + spin_lock(&hmaster->uvhub_lock); + /* try for the original first */ + if (busy_one != normal) { + if (!normal_busy(bcp)) + selected = normal; + } + if (selected < 0) { + /* can't use the normal, select an alternate */ + mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1; + descriptor_status = read_lmmr(mmr_offset); + + /* scan available descriptors 32-63 */ + for (i = 0; i < UV_CPUS_PER_AS; i++) { + if ((hmaster->inuse_map & (1 << i)) == 0) { + status = ((descriptor_status >> + (i * UV_ACT_STATUS_SIZE)) & + UV_ACT_STATUS_MASK) << 1; + if (status != UV2H_DESC_BUSY) { + selected = i + UV_CPUS_PER_AS; + break; + } + } + } + } + + if (busy_one != normal) + /* mark the busy alternate as not in-use */ + hmaster->inuse_map &= ~(1 << (busy_one - UV_CPUS_PER_AS)); + + if (selected >= 0) { + /* switch to the selected descriptor */ + if (selected != normal) { + /* set the selected alternate as in-use */ + hmaster->inuse_map |= + (1 << (selected - UV_CPUS_PER_AS)); + if (selected > stat->s_uv2_wars_hw) + stat->s_uv2_wars_hw = selected; + } + bau_desc_old = bcp->descriptor_base; + bau_desc_old += (ITEMS_PER_DESC * busy_one); + bcp->using_desc = selected; + bau_desc_new = bcp->descriptor_base; + bau_desc_new += (ITEMS_PER_DESC * selected); + *bau_desc_new = *bau_desc_old; + } else { + /* + * All are busy. Wait for the normal one for this cpu to + * free up. + */ + stat->s_uv2_war_waits++; + spin_unlock(&hmaster->uvhub_lock); + ttm = get_cycles(); + do { + cpu_relax(); + } while (normal_busy(bcp)); + spin_lock(&hmaster->uvhub_lock); + /* switch to the original descriptor */ + bcp->using_desc = normal; + bau_desc_old = bcp->descriptor_base; + bau_desc_old += (ITEMS_PER_DESC * bcp->using_desc); + bcp->using_desc = (ITEMS_PER_DESC * normal); + bau_desc_new = bcp->descriptor_base; + bau_desc_new += (ITEMS_PER_DESC * normal); + *bau_desc_new = *bau_desc_old; /* copy the entire descriptor */ + } + spin_unlock(&hmaster->uvhub_lock); + return FLUSH_RETRY_BUSYBUG; +} + static int uv2_wait_completion(struct bau_desc *bau_desc, unsigned long mmr_offset, int right_shift, struct bau_control *bcp, long try) { unsigned long descriptor_stat; cycles_t ttm; - int cpu = bcp->uvhub_cpu; + int desc = bcp->using_desc; + long busy_reps = 0; struct ptc_stats *stat = bcp->statp; - descriptor_stat = uv2_read_status(mmr_offset, right_shift, cpu); + descriptor_stat = uv2_read_status(mmr_offset, right_shift, desc); /* spin on the status MMR, waiting for it to go idle */ while (descriptor_stat != UV2H_DESC_IDLE) { @@ -542,12 +655,23 @@ static int uv2_wait_completion(struct bau_desc *bau_desc, bcp->conseccompletes = 0; return FLUSH_RETRY_TIMEOUT; } else { + busy_reps++; + if (busy_reps > 1000000) { + /* not to hammer on the clock */ + busy_reps = 0; + ttm = get_cycles(); + if ((ttm - bcp->send_message) > + (bcp->clocks_per_100_usec)) { + return handle_uv2_busy(bcp); + } + } /* * descriptor_stat is still BUSY */ cpu_relax(); } - descriptor_stat = uv2_read_status(mmr_offset, right_shift, cpu); + descriptor_stat = uv2_read_status(mmr_offset, right_shift, + desc); } bcp->conseccompletes++; return FLUSH_COMPLETE; @@ -563,17 +687,17 @@ static int wait_completion(struct bau_desc *bau_desc, { int right_shift; unsigned long mmr_offset; - int cpu = bcp->uvhub_cpu; + int desc = bcp->using_desc; - if (cpu < UV_CPUS_PER_AS) { + if (desc < UV_CPUS_PER_AS) { mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0; - right_shift = cpu * UV_ACT_STATUS_SIZE; + right_shift = desc * UV_ACT_STATUS_SIZE; } else { mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1; - right_shift = ((cpu - UV_CPUS_PER_AS) * UV_ACT_STATUS_SIZE); + right_shift = ((desc - UV_CPUS_PER_AS) * UV_ACT_STATUS_SIZE); } - if (is_uv1_hub()) + if (bcp->uvhub_version == 1) return uv1_wait_completion(bau_desc, mmr_offset, right_shift, bcp, try); else @@ -752,19 +876,22 @@ static void handle_cmplt(int completion_status, struct bau_desc *bau_desc, * Returns 1 if it gives up entirely and the original cpu mask is to be * returned to the kernel. */ -int uv_flush_send_and_wait(struct bau_desc *bau_desc, - struct cpumask *flush_mask, struct bau_control *bcp) +int uv_flush_send_and_wait(struct cpumask *flush_mask, struct bau_control *bcp) { int seq_number = 0; int completion_stat = 0; + int uv1 = 0; long try = 0; unsigned long index; cycles_t time1; cycles_t time2; struct ptc_stats *stat = bcp->statp; struct bau_control *hmaster = bcp->uvhub_master; + struct uv1_bau_msg_header *uv1_hdr = NULL; + struct uv2_bau_msg_header *uv2_hdr = NULL; + struct bau_desc *bau_desc; - if (is_uv1_hub()) + if (bcp->uvhub_version == 1) uv1_throttle(hmaster, stat); while (hmaster->uvhub_quiesce) @@ -772,22 +899,39 @@ int uv_flush_send_and_wait(struct bau_desc *bau_desc, time1 = get_cycles(); do { - if (try == 0) { - bau_desc->header.msg_type = MSG_REGULAR; + bau_desc = bcp->descriptor_base; + bau_desc += (ITEMS_PER_DESC * bcp->using_desc); + if (bcp->uvhub_version == 1) { + uv1 = 1; + uv1_hdr = &bau_desc->header.uv1_hdr; + } else + uv2_hdr = &bau_desc->header.uv2_hdr; + if ((try == 0) || (completion_stat == FLUSH_RETRY_BUSYBUG)) { + if (uv1) + uv1_hdr->msg_type = MSG_REGULAR; + else + uv2_hdr->msg_type = MSG_REGULAR; seq_number = bcp->message_number++; } else { - bau_desc->header.msg_type = MSG_RETRY; + if (uv1) + uv1_hdr->msg_type = MSG_RETRY; + else + uv2_hdr->msg_type = MSG_RETRY; stat->s_retry_messages++; } - bau_desc->header.sequence = seq_number; - index = (1UL << AS_PUSH_SHIFT) | bcp->uvhub_cpu; + if (uv1) + uv1_hdr->sequence = seq_number; + else + uv2_hdr->sequence = seq_number; + index = (1UL << AS_PUSH_SHIFT) | bcp->using_desc; bcp->send_message = get_cycles(); write_mmr_activation(index); try++; completion_stat = wait_completion(bau_desc, bcp, try); + /* UV2: wait_completion() may change the bcp->using_desc */ handle_cmplt(completion_stat, bau_desc, bcp, hmaster, stat); @@ -798,6 +942,7 @@ int uv_flush_send_and_wait(struct bau_desc *bau_desc, } cpu_relax(); } while ((completion_stat == FLUSH_RETRY_PLUGGED) || + (completion_stat == FLUSH_RETRY_BUSYBUG) || (completion_stat == FLUSH_RETRY_TIMEOUT)); time2 = get_cycles(); @@ -812,6 +957,7 @@ int uv_flush_send_and_wait(struct bau_desc *bau_desc, record_send_stats(time1, time2, bcp, stat, completion_stat, try); if (completion_stat == FLUSH_GIVEUP) + /* FLUSH_GIVEUP will fall back to using IPI's for tlb flush */ return 1; return 0; } @@ -967,7 +1113,7 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, stat->s_ntargself++; bau_desc = bcp->descriptor_base; - bau_desc += ITEMS_PER_DESC * bcp->uvhub_cpu; + bau_desc += (ITEMS_PER_DESC * bcp->using_desc); bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE); if (set_distrib_bits(flush_mask, bcp, bau_desc, &locals, &remotes)) return NULL; @@ -980,13 +1126,86 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, * uv_flush_send_and_wait returns 0 if all cpu's were messaged, * or 1 if it gave up and the original cpumask should be returned. */ - if (!uv_flush_send_and_wait(bau_desc, flush_mask, bcp)) + if (!uv_flush_send_and_wait(flush_mask, bcp)) return NULL; else return cpumask; } /* + * Search the message queue for any 'other' message with the same software + * acknowledge resource bit vector. + */ +struct bau_pq_entry *find_another_by_swack(struct bau_pq_entry *msg, + struct bau_control *bcp, unsigned char swack_vec) +{ + struct bau_pq_entry *msg_next = msg + 1; + + if (msg_next > bcp->queue_last) + msg_next = bcp->queue_first; + while ((msg_next->swack_vec != 0) && (msg_next != msg)) { + if (msg_next->swack_vec == swack_vec) + return msg_next; + msg_next++; + if (msg_next > bcp->queue_last) + msg_next = bcp->queue_first; + } + return NULL; +} + +/* + * UV2 needs to work around a bug in which an arriving message has not + * set a bit in the UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE register. + * Such a message must be ignored. + */ +void process_uv2_message(struct msg_desc *mdp, struct bau_control *bcp) +{ + unsigned long mmr_image; + unsigned char swack_vec; + struct bau_pq_entry *msg = mdp->msg; + struct bau_pq_entry *other_msg; + + mmr_image = read_mmr_sw_ack(); + swack_vec = msg->swack_vec; + + if ((swack_vec & mmr_image) == 0) { + /* + * This message was assigned a swack resource, but no + * reserved acknowlegment is pending. + * The bug has prevented this message from setting the MMR. + * And no other message has used the same sw_ack resource. + * Do the requested shootdown but do not reply to the msg. + * (the 0 means make no acknowledge) + */ + bau_process_message(mdp, bcp, 0); + return; + } + + /* + * Some message has set the MMR 'pending' bit; it might have been + * another message. Look for that message. + */ + other_msg = find_another_by_swack(msg, bcp, msg->swack_vec); + if (other_msg) { + /* There is another. Do not ack the current one. */ + bau_process_message(mdp, bcp, 0); + /* + * Let the natural processing of that message acknowledge + * it. Don't get the processing of sw_ack's out of order. + */ + return; + } + + /* + * There is no other message using this sw_ack, so it is safe to + * acknowledge it. + */ + bau_process_message(mdp, bcp, 1); + + return; +} + +/* * The BAU message interrupt comes here. (registered by set_intr_gate) * See entry_64.S * @@ -1022,9 +1241,11 @@ void uv_bau_message_interrupt(struct pt_regs *regs) count++; msgdesc.msg_slot = msg - msgdesc.queue_first; - msgdesc.swack_slot = ffs(msg->swack_vec) - 1; msgdesc.msg = msg; - bau_process_message(&msgdesc, bcp); + if (bcp->uvhub_version == 2) + process_uv2_message(&msgdesc, bcp); + else + bau_process_message(&msgdesc, bcp, 1); msg++; if (msg > msgdesc.queue_last) @@ -1083,7 +1304,7 @@ static void __init enable_timeouts(void) */ mmr_image |= (1L << SOFTACK_MSHIFT); if (is_uv2_hub()) { - mmr_image |= (1L << UV2_LEG_SHFT); + mmr_image &= ~(1L << UV2_LEG_SHFT); mmr_image |= (1L << UV2_EXT_SHFT); } write_mmr_misc_control(pnode, mmr_image); @@ -1142,7 +1363,7 @@ static int ptc_seq_show(struct seq_file *file, void *data) seq_printf(file, "all one mult none retry canc nocan reset rcan "); seq_printf(file, - "disable enable\n"); + "disable enable wars warshw warwaits\n"); } if (cpu < num_possible_cpus() && cpu_online(cpu)) { stat = &per_cpu(ptcstats, cpu); @@ -1173,8 +1394,10 @@ static int ptc_seq_show(struct seq_file *file, void *data) stat->d_nomsg, stat->d_retries, stat->d_canceled, stat->d_nocanceled, stat->d_resets, stat->d_rcanceled); - seq_printf(file, "%ld %ld\n", - stat->s_bau_disabled, stat->s_bau_reenabled); + seq_printf(file, "%ld %ld %ld %ld %ld\n", + stat->s_bau_disabled, stat->s_bau_reenabled, + stat->s_uv2_wars, stat->s_uv2_wars_hw, + stat->s_uv2_war_waits); } return 0; } @@ -1432,12 +1655,15 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode) { int i; int cpu; + int uv1 = 0; unsigned long gpa; unsigned long m; unsigned long n; size_t dsize; struct bau_desc *bau_desc; struct bau_desc *bd2; + struct uv1_bau_msg_header *uv1_hdr; + struct uv2_bau_msg_header *uv2_hdr; struct bau_control *bcp; /* @@ -1451,6 +1677,8 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode) gpa = uv_gpa(bau_desc); n = uv_gpa_to_gnode(gpa); m = uv_gpa_to_offset(gpa); + if (is_uv1_hub()) + uv1 = 1; /* the 14-bit pnode */ write_mmr_descriptor_base(pnode, (n << UV_DESC_PSHIFT | m)); @@ -1461,21 +1689,33 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode) */ for (i = 0, bd2 = bau_desc; i < (ADP_SZ * ITEMS_PER_DESC); i++, bd2++) { memset(bd2, 0, sizeof(struct bau_desc)); - bd2->header.swack_flag = 1; - /* - * The base_dest_nasid set in the message header is the nasid - * of the first uvhub in the partition. The bit map will - * indicate destination pnode numbers relative to that base. - * They may not be consecutive if nasid striding is being used. - */ - bd2->header.base_dest_nasid = UV_PNODE_TO_NASID(base_pnode); - bd2->header.dest_subnodeid = UV_LB_SUBNODEID; - bd2->header.command = UV_NET_ENDPOINT_INTD; - bd2->header.int_both = 1; - /* - * all others need to be set to zero: - * fairness chaining multilevel count replied_to - */ + if (uv1) { + uv1_hdr = &bd2->header.uv1_hdr; + uv1_hdr->swack_flag = 1; + /* + * The base_dest_nasid set in the message header + * is the nasid of the first uvhub in the partition. + * The bit map will indicate destination pnode numbers + * relative to that base. They may not be consecutive + * if nasid striding is being used. + */ + uv1_hdr->base_dest_nasid = + UV_PNODE_TO_NASID(base_pnode); + uv1_hdr->dest_subnodeid = UV_LB_SUBNODEID; + uv1_hdr->command = UV_NET_ENDPOINT_INTD; + uv1_hdr->int_both = 1; + /* + * all others need to be set to zero: + * fairness chaining multilevel count replied_to + */ + } else { + uv2_hdr = &bd2->header.uv2_hdr; + uv2_hdr->swack_flag = 1; + uv2_hdr->base_dest_nasid = + UV_PNODE_TO_NASID(base_pnode); + uv2_hdr->dest_subnodeid = UV_LB_SUBNODEID; + uv2_hdr->command = UV_NET_ENDPOINT_INTD; + } } for_each_present_cpu(cpu) { if (pnode != uv_blade_to_pnode(uv_cpu_to_blade_id(cpu))) @@ -1531,6 +1771,7 @@ static void pq_init(int node, int pnode) write_mmr_payload_first(pnode, pn_first); write_mmr_payload_tail(pnode, first); write_mmr_payload_last(pnode, last); + write_gmmr_sw_ack(pnode, 0xffffUL); /* in effect, all msg_type's are set to MSG_NOOP */ memset(pqp, 0, sizeof(struct bau_pq_entry) * DEST_Q_SIZE); @@ -1584,14 +1825,14 @@ static int calculate_destination_timeout(void) ts_ns = base * mult1 * mult2; ret = ts_ns / 1000; } else { - /* 4 bits 0/1 for 10/80us, 3 bits of multiplier */ - mmr_image = uv_read_local_mmr(UVH_AGING_PRESCALE_SEL); + /* 4 bits 0/1 for 10/80us base, 3 bits of multiplier */ + mmr_image = uv_read_local_mmr(UVH_LB_BAU_MISC_CONTROL); mmr_image = (mmr_image & UV_SA_MASK) >> UV_SA_SHFT; if (mmr_image & (1L << UV2_ACK_UNITS_SHFT)) - mult1 = 80; + base = 80; else - mult1 = 10; - base = mmr_image & UV2_ACK_MASK; + base = 10; + mult1 = mmr_image & UV2_ACK_MASK; ret = mult1 * base; } return ret; @@ -1618,6 +1859,7 @@ static void __init init_per_cpu_tunables(void) bcp->cong_response_us = congested_respns_us; bcp->cong_reps = congested_reps; bcp->cong_period = congested_period; + bcp->clocks_per_100_usec = usec_2_cycles(100); } } @@ -1728,8 +1970,17 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp, bcp->cpus_in_socket = sdp->num_cpus; bcp->socket_master = *smasterp; bcp->uvhub = bdp->uvhub; + if (is_uv1_hub()) + bcp->uvhub_version = 1; + else if (is_uv2_hub()) + bcp->uvhub_version = 2; + else { + printk(KERN_EMERG "uvhub version not 1 or 2\n"); + return 1; + } bcp->uvhub_master = *hmasterp; bcp->uvhub_cpu = uv_cpu_hub_info(cpu)->blade_processor_id; + bcp->using_desc = bcp->uvhub_cpu; if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) { printk(KERN_EMERG "%d cpus per uvhub invalid\n", bcp->uvhub_cpu); @@ -1845,6 +2096,8 @@ static int __init uv_bau_init(void) uv_base_pnode = uv_blade_to_pnode(uvhub); } + enable_timeouts(); + if (init_per_cpu(nuvhubs, uv_base_pnode)) { nobau = 1; return 0; @@ -1855,7 +2108,6 @@ static int __init uv_bau_init(void) if (uv_blade_nr_possible_cpus(uvhub)) init_uvhub(uvhub, vector, uv_base_pnode); - enable_timeouts(); alloc_intr_gate(vector, uv_bau_message_intr1); for_each_possible_blade(uvhub) { @@ -1867,7 +2119,8 @@ static int __init uv_bau_init(void) val = 1L << 63; write_gmmr_activation(pnode, val); mmr = 1; /* should be 1 to broadcast to both sockets */ - write_mmr_data_broadcast(pnode, mmr); + if (!is_uv1_hub()) + write_mmr_data_broadcast(pnode, mmr); } } |