diff options
Diffstat (limited to 'arch/x86/platform/uv')
| -rw-r--r-- | arch/x86/platform/uv/Makefile | 2 | ||||
| -rw-r--r-- | arch/x86/platform/uv/bios_uv.c | 3 | ||||
| -rw-r--r-- | arch/x86/platform/uv/tlb_uv.c | 1948 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_irq.c | 25 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_nmi.c | 727 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_sysfs.c | 2 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_time.c | 38 |
7 files changed, 1991 insertions, 754 deletions
diff --git a/arch/x86/platform/uv/Makefile b/arch/x86/platform/uv/Makefile index 6c40995fefb..52079bebd01 100644 --- a/arch/x86/platform/uv/Makefile +++ b/arch/x86/platform/uv/Makefile @@ -1 +1 @@ -obj-$(CONFIG_X86_UV) += tlb_uv.o bios_uv.o uv_irq.o uv_sysfs.o uv_time.o +obj-$(CONFIG_X86_UV) += tlb_uv.o bios_uv.o uv_irq.o uv_sysfs.o uv_time.o uv_nmi.o diff --git a/arch/x86/platform/uv/bios_uv.c b/arch/x86/platform/uv/bios_uv.c index 8bc57baaa9a..1584cbed0dc 100644 --- a/arch/x86/platform/uv/bios_uv.c +++ b/arch/x86/platform/uv/bios_uv.c @@ -20,6 +20,7 @@ */ #include <linux/efi.h> +#include <linux/export.h> #include <asm/efi.h> #include <linux/io.h> #include <asm/uv/bios.h> @@ -38,7 +39,7 @@ s64 uv_bios_call(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3, u64 a4, u64 a5) */ return BIOS_STATUS_UNIMPLEMENTED; - ret = efi_call6((void *)__va(tab->function), (u64)which, + ret = efi_call((void *)__va(tab->function), (u64)which, a1, a2, a3, a4, a5); return ret; } diff --git a/arch/x86/platform/uv/tlb_uv.c b/arch/x86/platform/uv/tlb_uv.c index a318194002b..dfe605ac1bc 100644 --- a/arch/x86/platform/uv/tlb_uv.c +++ b/arch/x86/platform/uv/tlb_uv.c @@ -1,7 +1,7 @@ /* * SGI UltraViolet TLB flush routines. * - * (c) 2008-2010 Cliff Wickman <cpw@sgi.com>, SGI. + * (c) 2008-2012 Cliff Wickman <cpw@sgi.com>, SGI. * * This code is released under the GNU General Public License version 2 or * later. @@ -11,6 +11,7 @@ #include <linux/debugfs.h> #include <linux/kernel.h> #include <linux/slab.h> +#include <linux/delay.h> #include <asm/mmu_context.h> #include <asm/uv/uv.h> @@ -34,27 +35,79 @@ static int timeout_base_ns[] = { 5242880, 167772160 }; + static int timeout_us; static int nobau; -static int baudisabled; -static spinlock_t disable_lock; +static int nobau_perm; static cycles_t congested_cycles; /* tunables: */ -static int max_bau_concurrent = MAX_BAU_CONCURRENT; -static int max_bau_concurrent_constant = MAX_BAU_CONCURRENT; -static int plugged_delay = PLUGGED_DELAY; -static int plugsb4reset = PLUGSB4RESET; -static int timeoutsb4reset = TIMEOUTSB4RESET; -static int ipi_reset_limit = IPI_RESET_LIMIT; -static int complete_threshold = COMPLETE_THRESHOLD; -static int congested_response_us = CONGESTED_RESPONSE_US; -static int congested_reps = CONGESTED_REPS; -static int congested_period = CONGESTED_PERIOD; +static int max_concurr = MAX_BAU_CONCURRENT; +static int max_concurr_const = MAX_BAU_CONCURRENT; +static int plugged_delay = PLUGGED_DELAY; +static int plugsb4reset = PLUGSB4RESET; +static int giveup_limit = GIVEUP_LIMIT; +static int timeoutsb4reset = TIMEOUTSB4RESET; +static int ipi_reset_limit = IPI_RESET_LIMIT; +static int complete_threshold = COMPLETE_THRESHOLD; +static int congested_respns_us = CONGESTED_RESPONSE_US; +static int congested_reps = CONGESTED_REPS; +static int disabled_period = DISABLED_PERIOD; + +static struct tunables tunables[] = { + {&max_concurr, MAX_BAU_CONCURRENT}, /* must be [0] */ + {&plugged_delay, PLUGGED_DELAY}, + {&plugsb4reset, PLUGSB4RESET}, + {&timeoutsb4reset, TIMEOUTSB4RESET}, + {&ipi_reset_limit, IPI_RESET_LIMIT}, + {&complete_threshold, COMPLETE_THRESHOLD}, + {&congested_respns_us, CONGESTED_RESPONSE_US}, + {&congested_reps, CONGESTED_REPS}, + {&disabled_period, DISABLED_PERIOD}, + {&giveup_limit, GIVEUP_LIMIT} +}; + static struct dentry *tunables_dir; static struct dentry *tunables_file; -static int __init setup_nobau(char *arg) +/* these correspond to the statistics printed by ptc_seq_show() */ +static char *stat_description[] = { + "sent: number of shootdown messages sent", + "stime: time spent sending messages", + "numuvhubs: number of hubs targeted with shootdown", + "numuvhubs16: number times 16 or more hubs targeted", + "numuvhubs8: number times 8 or more hubs targeted", + "numuvhubs4: number times 4 or more hubs targeted", + "numuvhubs2: number times 2 or more hubs targeted", + "numuvhubs1: number times 1 hub targeted", + "numcpus: number of cpus targeted with shootdown", + "dto: number of destination timeouts", + "retries: destination timeout retries sent", + "rok: : destination timeouts successfully retried", + "resetp: ipi-style resource resets for plugs", + "resett: ipi-style resource resets for timeouts", + "giveup: fall-backs to ipi-style shootdowns", + "sto: number of source timeouts", + "bz: number of stay-busy's", + "throt: number times spun in throttle", + "swack: image of UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE", + "recv: shootdown messages received", + "rtime: time spent processing messages", + "all: shootdown all-tlb messages", + "one: shootdown one-tlb messages", + "mult: interrupts that found multiple messages", + "none: interrupts that found no messages", + "retry: number of retry messages processed", + "canc: number messages canceled by retries", + "nocan: number retries that found nothing to cancel", + "reset: number of ipi-style reset requests processed", + "rcan: number messages canceled by reset requests", + "disable: number times use of the BAU was disabled", + "enable: number times use of the BAU was re-enabled" +}; + +static int __init +setup_nobau(char *arg) { nobau = 1; return 0; @@ -62,15 +115,46 @@ static int __init setup_nobau(char *arg) early_param("nobau", setup_nobau); /* base pnode in this partition */ -static int uv_partition_base_pnode __read_mostly; -/* position of pnode (which is nasid>>1): */ -static int uv_nshift __read_mostly; -static unsigned long uv_mmask __read_mostly; +static int uv_base_pnode __read_mostly; static DEFINE_PER_CPU(struct ptc_stats, ptcstats); static DEFINE_PER_CPU(struct bau_control, bau_control); static DEFINE_PER_CPU(cpumask_var_t, uv_flush_tlb_mask); +static void +set_bau_on(void) +{ + int cpu; + struct bau_control *bcp; + + if (nobau_perm) { + pr_info("BAU not initialized; cannot be turned on\n"); + return; + } + nobau = 0; + for_each_present_cpu(cpu) { + bcp = &per_cpu(bau_control, cpu); + bcp->nobau = 0; + } + pr_info("BAU turned on\n"); + return; +} + +static void +set_bau_off(void) +{ + int cpu; + struct bau_control *bcp; + + nobau = 1; + for_each_present_cpu(cpu) { + bcp = &per_cpu(bau_control, cpu); + bcp->nobau = 1; + } + pr_info("BAU turned off\n"); + return; +} + /* * Determine the first node on a uvhub. 'Nodes' are used for kernel * memory allocation. @@ -108,60 +192,53 @@ 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 inline void uv_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_payload_queue_entry *msg; + struct bau_pq_entry *msg; msg = mdp->msg; - if (!msg->canceled) { - dw = (msg->sw_ack_vector << UV_SW_ACK_NPENDING) | - msg->sw_ack_vector; - uv_write_local_mmr( - UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, dw); + if (!msg->canceled && do_acknowledge) { + dw = (msg->swack_vec << UV_SW_ACK_NPENDING) | msg->swack_vec; + write_mmr_sw_ack(dw); } msg->replied_to = 1; - msg->sw_ack_vector = 0; + msg->swack_vec = 0; } /* * Process the receipt of a RETRY message */ -static inline void uv_bau_process_retry_msg(struct msg_desc *mdp, - struct bau_control *bcp) +static void bau_process_retry_msg(struct msg_desc *mdp, + struct bau_control *bcp) { int i; int cancel_count = 0; - int slot2; unsigned long msg_res; unsigned long mmr = 0; - struct bau_payload_queue_entry *msg; - struct bau_payload_queue_entry *msg2; - struct ptc_stats *stat; + struct bau_pq_entry *msg = mdp->msg; + struct bau_pq_entry *msg2; + struct ptc_stats *stat = bcp->statp; - msg = mdp->msg; - stat = bcp->statp; stat->d_retries++; /* * cancel any message from msg+1 to the retry itself */ for (msg2 = msg+1, i = 0; i < DEST_Q_SIZE; msg2++, i++) { - if (msg2 > mdp->va_queue_last) - msg2 = mdp->va_queue_first; + if (msg2 > mdp->queue_last) + msg2 = mdp->queue_first; if (msg2 == msg) break; - /* same conditions for cancellation as uv_do_reset */ + /* same conditions for cancellation as do_reset */ if ((msg2->replied_to == 0) && (msg2->canceled == 0) && - (msg2->sw_ack_vector) && ((msg2->sw_ack_vector & - msg->sw_ack_vector) == 0) && + (msg2->swack_vec) && ((msg2->swack_vec & + msg->swack_vec) == 0) && (msg2->sending_cpu == msg->sending_cpu) && (msg2->msg_type != MSG_NOOP)) { - slot2 = msg2 - mdp->va_queue_first; - mmr = uv_read_local_mmr - (UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE); - msg_res = msg2->sw_ack_vector; + mmr = read_mmr_sw_ack(); + msg_res = msg2->swack_vec; /* * This is a message retry; clear the resources held * by the previous message only if they timed out. @@ -169,17 +246,16 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp, * situation to report. */ 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++; cancel_count++; - uv_write_local_mmr( - UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, - (msg_res << UV_SW_ACK_NPENDING) | - msg_res); + mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res; + write_mmr_sw_ack(mr); } } } @@ -191,20 +267,19 @@ static inline void uv_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 uv_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) { - int msg_ack_count; short socket_ack_count = 0; - struct ptc_stats *stat; - struct bau_payload_queue_entry *msg; + short *sp; + struct atomic_short *asp; + struct ptc_stats *stat = bcp->statp; + struct bau_pq_entry *msg = mdp->msg; struct bau_control *smaster = bcp->socket_master; /* * This must be a normal message, or retry of a normal message */ - msg = mdp->msg; - stat = bcp->statp; if (msg->address == TLB_FLUSH_ALL) { local_flush_tlb(); stat->d_alltlb++; @@ -221,30 +296,33 @@ static void uv_bau_process_message(struct msg_desc *mdp, * cpu number. */ if (msg->msg_type == MSG_RETRY && bcp == bcp->uvhub_master) - uv_bau_process_retry_msg(mdp, bcp); + bau_process_retry_msg(mdp, bcp); /* - * This is a sw_ack message, so we have to reply to it. + * This is a swack message, so we have to reply to it. * Count each responding cpu on the socket. This avoids * pinging the count's cache line back and forth between * the sockets. */ - socket_ack_count = atomic_add_short_return(1, (struct atomic_short *) - &smaster->socket_acknowledge_count[mdp->msg_slot]); + sp = &smaster->socket_acknowledge_count[mdp->msg_slot]; + asp = (struct atomic_short *)sp; + socket_ack_count = atom_asr(1, asp); if (socket_ack_count == bcp->cpus_in_socket) { + int msg_ack_count; /* * Both sockets dump their completed count total into * the message's count. */ - smaster->socket_acknowledge_count[mdp->msg_slot] = 0; - msg_ack_count = atomic_add_short_return(socket_ack_count, - (struct atomic_short *)&msg->acknowledge_count); + *sp = 0; + asp = (struct atomic_short *)&msg->acknowledge_count; + msg_ack_count = atom_asr(socket_ack_count, asp); if (msg_ack_count == bcp->cpus_in_uvhub) { /* * All cpus in uvhub saw it; reply + * (unless we are in the UV2 workaround) */ - uv_reply_to_message(mdp, bcp); + reply_to_message(mdp, bcp, do_acknowledge); } } @@ -252,14 +330,18 @@ static void uv_bau_process_message(struct msg_desc *mdp, } /* - * Determine the first cpu on a uvhub. + * Determine the first cpu on a pnode. */ -static int uvhub_to_first_cpu(int uvhub) +static int pnode_to_first_cpu(int pnode, struct bau_control *smaster) { int cpu; - for_each_present_cpu(cpu) - if (uvhub == uv_cpu_to_blade_id(cpu)) + struct hub_and_pnode *hpp; + + for_each_present_cpu(cpu) { + hpp = &smaster->thp[cpu]; + if (pnode == hpp->pnode) return cpu; + } return -1; } @@ -267,62 +349,51 @@ static int uvhub_to_first_cpu(int uvhub) * Last resort when we get a large number of destination timeouts is * to clear resources held by a given cpu. * Do this with IPI so that all messages in the BAU message queue - * can be identified by their nonzero sw_ack_vector field. + * can be identified by their nonzero swack_vec field. * * This is entered for a single cpu on the uvhub. * The sender want's this uvhub to free a specific message's - * sw_ack resources. + * swack resources. */ -static void -uv_do_reset(void *ptr) +static void do_reset(void *ptr) { int i; - int slot; - int count = 0; - unsigned long mmr; - unsigned long msg_res; - struct bau_control *bcp; - struct reset_args *rap; - struct bau_payload_queue_entry *msg; - struct ptc_stats *stat; + struct bau_control *bcp = &per_cpu(bau_control, smp_processor_id()); + struct reset_args *rap = (struct reset_args *)ptr; + struct bau_pq_entry *msg; + struct ptc_stats *stat = bcp->statp; - bcp = &per_cpu(bau_control, smp_processor_id()); - rap = (struct reset_args *)ptr; - stat = bcp->statp; stat->d_resets++; - /* * We're looking for the given sender, and - * will free its sw_ack resource. + * will free its swack resource. * If all cpu's finally responded after the timeout, its * message 'replied_to' was set. */ - for (msg = bcp->va_queue_first, i = 0; i < DEST_Q_SIZE; msg++, i++) { - /* uv_do_reset: same conditions for cancellation as - uv_bau_process_retry_msg() */ + for (msg = bcp->queue_first, i = 0; i < DEST_Q_SIZE; msg++, i++) { + unsigned long msg_res; + /* do_reset: same conditions for cancellation as + bau_process_retry_msg() */ if ((msg->replied_to == 0) && (msg->canceled == 0) && (msg->sending_cpu == rap->sender) && - (msg->sw_ack_vector) && + (msg->swack_vec) && (msg->msg_type != MSG_NOOP)) { + unsigned long mmr; + unsigned long mr; /* * make everyone else ignore this message */ msg->canceled = 1; - slot = msg - bcp->va_queue_first; - count++; /* * only reset the resource if it is still pending */ - mmr = uv_read_local_mmr - (UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE); - msg_res = msg->sw_ack_vector; + mmr = read_mmr_sw_ack(); + msg_res = msg->swack_vec; + mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res; if (mmr & msg_res) { stat->d_rcanceled++; - uv_write_local_mmr( - UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, - (msg_res << UV_SW_ACK_NPENDING) | - msg_res); + write_mmr_sw_ack(mr); } } } @@ -333,41 +404,78 @@ uv_do_reset(void *ptr) * Use IPI to get all target uvhubs to release resources held by * a given sending cpu number. */ -static void uv_reset_with_ipi(struct bau_target_uvhubmask *distribution, - int sender) +static void reset_with_ipi(struct pnmask *distribution, struct bau_control *bcp) { - int uvhub; - int cpu; - cpumask_t mask; + int pnode; + int apnode; + int maskbits; + int sender = bcp->cpu; + cpumask_t *mask = bcp->uvhub_master->cpumask; + struct bau_control *smaster = bcp->socket_master; struct reset_args reset_args; reset_args.sender = sender; - - cpus_clear(mask); + cpus_clear(*mask); /* find a single cpu for each uvhub in this distribution mask */ - for (uvhub = 0; - uvhub < sizeof(struct bau_target_uvhubmask) * BITSPERBYTE; - uvhub++) { - if (!bau_uvhub_isset(uvhub, distribution)) + maskbits = sizeof(struct pnmask) * BITSPERBYTE; + /* each bit is a pnode relative to the partition base pnode */ + for (pnode = 0; pnode < maskbits; pnode++) { + int cpu; + if (!bau_uvhub_isset(pnode, distribution)) continue; - /* find a cpu for this uvhub */ - cpu = uvhub_to_first_cpu(uvhub); - cpu_set(cpu, mask); + apnode = pnode + bcp->partition_base_pnode; + cpu = pnode_to_first_cpu(apnode, smaster); + cpu_set(cpu, *mask); } - /* IPI all cpus; Preemption is already disabled */ - smp_call_function_many(&mask, uv_do_reset, (void *)&reset_args, 1); + + /* IPI all cpus; preemption is already disabled */ + smp_call_function_many(mask, do_reset, (void *)&reset_args, 1); return; } -static inline unsigned long -cycles_2_us(unsigned long long cyc) +/* + * Not to be confused with cycles_2_ns() from tsc.c; this gives a relative + * number, not an absolute. It converts a duration in cycles to a duration in + * ns. + */ +static inline unsigned long long cycles_2_ns(unsigned long long cyc) { + struct cyc2ns_data *data = cyc2ns_read_begin(); unsigned long long ns; - unsigned long us; - ns = (cyc * per_cpu(cyc2ns, smp_processor_id())) - >> CYC2NS_SCALE_FACTOR; - us = ns / 1000; - return us; + + ns = mul_u64_u32_shr(cyc, data->cyc2ns_mul, data->cyc2ns_shift); + + cyc2ns_read_end(data); + return ns; +} + +/* + * The reverse of the above; converts a duration in ns to a duration in cycles. + */ +static inline unsigned long long ns_2_cycles(unsigned long long ns) +{ + struct cyc2ns_data *data = cyc2ns_read_begin(); + unsigned long long cyc; + + cyc = (ns << data->cyc2ns_shift) / data->cyc2ns_mul; + + cyc2ns_read_end(data); + return cyc; +} + +static inline unsigned long cycles_2_us(unsigned long long cyc) +{ + return cycles_2_ns(cyc) / NSEC_PER_USEC; +} + +static inline cycles_t sec_2_cycles(unsigned long sec) +{ + return ns_2_cycles(sec * NSEC_PER_SEC); +} + +static inline unsigned long long usec_2_cycles(unsigned long usec) +{ + return ns_2_cycles(usec * NSEC_PER_USEC); } /* @@ -375,56 +483,56 @@ cycles_2_us(unsigned long long cyc) * leaves uvhub_quiesce set so that no new broadcasts are started by * bau_flush_send_and_wait() */ -static inline void -quiesce_local_uvhub(struct bau_control *hmaster) +static inline void quiesce_local_uvhub(struct bau_control *hmaster) { - atomic_add_short_return(1, (struct atomic_short *) - &hmaster->uvhub_quiesce); + atom_asr(1, (struct atomic_short *)&hmaster->uvhub_quiesce); } /* * mark this quiet-requestor as done */ -static inline void -end_uvhub_quiesce(struct bau_control *hmaster) +static inline void end_uvhub_quiesce(struct bau_control *hmaster) { - atomic_add_short_return(-1, (struct atomic_short *) - &hmaster->uvhub_quiesce); + atom_asr(-1, (struct atomic_short *)&hmaster->uvhub_quiesce); +} + +static unsigned long uv1_read_status(unsigned long mmr_offset, int right_shift) +{ + unsigned long descriptor_status; + + descriptor_status = uv_read_local_mmr(mmr_offset); + descriptor_status >>= right_shift; + descriptor_status &= UV_ACT_STATUS_MASK; + return descriptor_status; } /* * Wait for completion of a broadcast software ack message * return COMPLETE, RETRY(PLUGGED or TIMEOUT) or GIVEUP */ -static int uv_wait_completion(struct bau_desc *bau_desc, - unsigned long mmr_offset, int right_shift, int this_cpu, - struct bau_control *bcp, struct bau_control *smaster, long try) +static int uv1_wait_completion(struct bau_desc *bau_desc, + unsigned long mmr_offset, int right_shift, + struct bau_control *bcp, long try) { unsigned long descriptor_status; - cycles_t ttime; + cycles_t ttm; struct ptc_stats *stat = bcp->statp; - struct bau_control *hmaster; - - hmaster = bcp->uvhub_master; + descriptor_status = uv1_read_status(mmr_offset, right_shift); /* spin on the status MMR, waiting for it to go idle */ - while ((descriptor_status = (((unsigned long) - uv_read_local_mmr(mmr_offset) >> - right_shift) & UV_ACT_STATUS_MASK)) != - DESC_STATUS_IDLE) { + while ((descriptor_status != DS_IDLE)) { /* - * Our software ack messages may be blocked because there are - * no swack resources available. As long as none of them - * has timed out hardware will NACK our message and its - * state will stay IDLE. + * Our software ack messages may be blocked because + * there are no swack resources available. As long + * as none of them has timed out hardware will NACK + * our message and its state will stay IDLE. */ - if (descriptor_status == DESC_STATUS_SOURCE_TIMEOUT) { + if (descriptor_status == DS_SOURCE_TIMEOUT) { stat->s_stimeout++; return FLUSH_GIVEUP; - } else if (descriptor_status == - DESC_STATUS_DESTINATION_TIMEOUT) { + } else if (descriptor_status == DS_DESTINATION_TIMEOUT) { stat->s_dtimeout++; - ttime = get_cycles(); + ttm = get_cycles(); /* * Our retries may be blocked by all destination @@ -432,8 +540,7 @@ static int uv_wait_completion(struct bau_desc *bau_desc, * pending. In that case hardware returns the * ERROR that looks like a destination timeout. */ - if (cycles_2_us(ttime - bcp->send_message) < - timeout_us) { + if (cycles_2_us(ttm - bcp->send_message) < timeout_us) { bcp->conseccompletes = 0; return FLUSH_RETRY_PLUGGED; } @@ -446,117 +553,319 @@ static int uv_wait_completion(struct bau_desc *bau_desc, */ cpu_relax(); } + descriptor_status = uv1_read_status(mmr_offset, right_shift); } bcp->conseccompletes++; return FLUSH_COMPLETE; } -static inline cycles_t -sec_2_cycles(unsigned long sec) +/* + * UV2 could have an extra bit of status in the ACTIVATION_STATUS_2 register. + * But not currently used. + */ +static unsigned long uv2_read_status(unsigned long offset, int rshft, int desc) { - unsigned long ns; - cycles_t cyc; + unsigned long descriptor_status; - ns = sec * 1000000000; - cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id())); - return cyc; + descriptor_status = + ((read_lmmr(offset) >> rshft) & UV_ACT_STATUS_MASK) << 1; + return descriptor_status; } /* - * conditionally add 1 to *v, unless *v is >= u - * return 0 if we cannot add 1 to *v because it is >= u - * return 1 if we can add 1 to *v because it is < u - * the add is atomic - * - * This is close to atomic_add_unless(), but this allows the 'u' value - * to be lowered below the current 'v'. atomic_add_unless can only stop - * on equal. + * 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. */ -static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u) +int normal_busy(struct bau_control *bcp) { - spin_lock(lock); - if (atomic_read(v) >= u) { - spin_unlock(lock); - return 0; + 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) +{ + struct ptc_stats *stat = bcp->statp; + + stat->s_uv2_wars++; + bcp->busy = 1; + return FLUSH_GIVEUP; +} + +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 desc = bcp->uvhub_cpu; + long busy_reps = 0; + struct ptc_stats *stat = bcp->statp; + + 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) { + if ((descriptor_stat == UV2H_DESC_SOURCE_TIMEOUT)) { + /* + * A h/w bug on the destination side may + * have prevented the message being marked + * pending, thus it doesn't get replied to + * and gets continually nacked until it times + * out with a SOURCE_TIMEOUT. + */ + stat->s_stimeout++; + return FLUSH_GIVEUP; + } else if (descriptor_stat == UV2H_DESC_DEST_TIMEOUT) { + ttm = get_cycles(); + + /* + * Our retries may be blocked by all destination + * swack resources being consumed, and a timeout + * pending. In that case hardware returns the + * ERROR that looks like a destination timeout. + * Without using the extended status we have to + * deduce from the short time that this was a + * strong nack. + */ + if (cycles_2_us(ttm - bcp->send_message) < timeout_us) { + bcp->conseccompletes = 0; + stat->s_plugged++; + /* FLUSH_RETRY_PLUGGED causes hang on boot */ + return FLUSH_GIVEUP; + } + stat->s_dtimeout++; + bcp->conseccompletes = 0; + /* FLUSH_RETRY_TIMEOUT causes hang on boot */ + return FLUSH_GIVEUP; + } 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->timeout_interval) + return handle_uv2_busy(bcp); + } + /* + * descriptor_stat is still BUSY + */ + cpu_relax(); + } + descriptor_stat = uv2_read_status(mmr_offset, right_shift, + desc); } - atomic_inc(v); - spin_unlock(lock); - return 1; + bcp->conseccompletes++; + return FLUSH_COMPLETE; +} + +/* + * There are 2 status registers; each and array[32] of 2 bits. Set up for + * which register to read and position in that register based on cpu in + * current hub. + */ +static int wait_completion(struct bau_desc *bau_desc, + struct bau_control *bcp, long try) +{ + int right_shift; + unsigned long mmr_offset; + int desc = bcp->uvhub_cpu; + + if (desc < UV_CPUS_PER_AS) { + mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0; + right_shift = desc * UV_ACT_STATUS_SIZE; + } else { + mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1; + right_shift = ((desc - UV_CPUS_PER_AS) * UV_ACT_STATUS_SIZE); + } + + if (bcp->uvhub_version == 1) + return uv1_wait_completion(bau_desc, mmr_offset, right_shift, + bcp, try); + else + return uv2_wait_completion(bau_desc, mmr_offset, right_shift, + bcp, try); } /* - * Our retries are blocked by all destination swack resources being + * Our retries are blocked by all destination sw ack resources being * in use, and a timeout is pending. In that case hardware immediately * returns the ERROR that looks like a destination timeout. */ -static void -destination_plugged(struct bau_desc *bau_desc, struct bau_control *bcp, +static void destination_plugged(struct bau_desc *bau_desc, + struct bau_control *bcp, struct bau_control *hmaster, struct ptc_stats *stat) { udelay(bcp->plugged_delay); bcp->plugged_tries++; + if (bcp->plugged_tries >= bcp->plugsb4reset) { bcp->plugged_tries = 0; + quiesce_local_uvhub(hmaster); + spin_lock(&hmaster->queue_lock); - uv_reset_with_ipi(&bau_desc->distribution, bcp->cpu); + reset_with_ipi(&bau_desc->distribution, bcp); spin_unlock(&hmaster->queue_lock); + end_uvhub_quiesce(hmaster); + bcp->ipi_attempts++; stat->s_resets_plug++; } } -static void -destination_timeout(struct bau_desc *bau_desc, struct bau_control *bcp, - struct bau_control *hmaster, struct ptc_stats *stat) +static void destination_timeout(struct bau_desc *bau_desc, + struct bau_control *bcp, struct bau_control *hmaster, + struct ptc_stats *stat) { - hmaster->max_bau_concurrent = 1; + hmaster->max_concurr = 1; bcp->timeout_tries++; if (bcp->timeout_tries >= bcp->timeoutsb4reset) { bcp->timeout_tries = 0; + quiesce_local_uvhub(hmaster); + spin_lock(&hmaster->queue_lock); - uv_reset_with_ipi(&bau_desc->distribution, bcp->cpu); + reset_with_ipi(&bau_desc->distribution, bcp); spin_unlock(&hmaster->queue_lock); + end_uvhub_quiesce(hmaster); + bcp->ipi_attempts++; stat->s_resets_timeout++; } } /* - * Completions are taking a very long time due to a congested numalink - * network. + * Stop all cpus on a uvhub from using the BAU for a period of time. + * This is reversed by check_enable. */ -static void -disable_for_congestion(struct bau_control *bcp, struct ptc_stats *stat) +static void disable_for_period(struct bau_control *bcp, struct ptc_stats *stat) { int tcpu; struct bau_control *tbcp; + struct bau_control *hmaster; + cycles_t tm1; - /* let only one cpu do this disabling */ - spin_lock(&disable_lock); - if (!baudisabled && bcp->period_requests && - ((bcp->period_time / bcp->period_requests) > congested_cycles)) { - /* it becomes this cpu's job to turn on the use of the - BAU again */ - baudisabled = 1; - bcp->set_bau_off = 1; - bcp->set_bau_on_time = get_cycles() + - sec_2_cycles(bcp->congested_period); + hmaster = bcp->uvhub_master; + spin_lock(&hmaster->disable_lock); + if (!bcp->baudisabled) { stat->s_bau_disabled++; + tm1 = get_cycles(); for_each_present_cpu(tcpu) { tbcp = &per_cpu(bau_control, tcpu); + if (tbcp->uvhub_master == hmaster) { tbcp->baudisabled = 1; + tbcp->set_bau_on_time = + tm1 + bcp->disabled_period; + } } } - spin_unlock(&disable_lock); + spin_unlock(&hmaster->disable_lock); } -/** - * uv_flush_send_and_wait - * +static void count_max_concurr(int stat, struct bau_control *bcp, + struct bau_control *hmaster) +{ + bcp->plugged_tries = 0; + bcp->timeout_tries = 0; + if (stat != FLUSH_COMPLETE) + return; + if (bcp->conseccompletes <= bcp->complete_threshold) + return; + if (hmaster->max_concurr >= hmaster->max_concurr_const) + return; + hmaster->max_concurr++; +} + +static void record_send_stats(cycles_t time1, cycles_t time2, + struct bau_control *bcp, struct ptc_stats *stat, + int completion_status, int try) +{ + cycles_t elapsed; + + if (time2 > time1) { + elapsed = time2 - time1; + stat->s_time += elapsed; + + if ((completion_status == FLUSH_COMPLETE) && (try == 1)) { + bcp->period_requests++; + bcp->period_time += elapsed; + if ((elapsed > congested_cycles) && + (bcp->period_requests > bcp->cong_reps) && + ((bcp->period_time / bcp->period_requests) > + congested_cycles)) { + stat->s_congested++; + disable_for_period(bcp, stat); + } + } + } else + stat->s_requestor--; + + if (completion_status == FLUSH_COMPLETE && try > 1) + stat->s_retriesok++; + else if (completion_status == FLUSH_GIVEUP) { + stat->s_giveup++; + if (get_cycles() > bcp->period_end) + bcp->period_giveups = 0; + bcp->period_giveups++; + if (bcp->period_giveups == 1) + bcp->period_end = get_cycles() + bcp->disabled_period; + if (bcp->period_giveups > bcp->giveup_limit) { + disable_for_period(bcp, stat); + stat->s_giveuplimit++; + } + } +} + +/* + * Because of a uv1 hardware bug only a limited number of concurrent + * requests can be made. + */ +static void uv1_throttle(struct bau_control *hmaster, struct ptc_stats *stat) +{ + spinlock_t *lock = &hmaster->uvhub_lock; + atomic_t *v; + + v = &hmaster->active_descriptor_count; + if (!atomic_inc_unless_ge(lock, v, hmaster->max_concurr)) { + stat->s_throttles++; + do { + cpu_relax(); + } while (!atomic_inc_unless_ge(lock, v, hmaster->max_concurr)); + } +} + +/* + * Handle the completion status of a message send. + */ +static void handle_cmplt(int completion_status, struct bau_desc *bau_desc, + struct bau_control *bcp, struct bau_control *hmaster, + struct ptc_stats *stat) +{ + if (completion_status == FLUSH_RETRY_PLUGGED) + destination_plugged(bau_desc, bcp, hmaster, stat); + else if (completion_status == FLUSH_RETRY_TIMEOUT) + destination_timeout(bau_desc, bcp, hmaster, stat); +} + +/* * Send a broadcast and wait for it to complete. * * The flush_mask contains the cpus the broadcast is to be sent to including @@ -566,115 +875,191 @@ disable_for_congestion(struct bau_control *bcp, struct ptc_stats *stat) * 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, + struct bau_desc *bau_desc) { - int right_shift; - int completion_status = 0; int seq_number = 0; + int completion_stat = 0; + int uv1 = 0; long try = 0; - int cpu = bcp->uvhub_cpu; - int this_cpu = bcp->cpu; - unsigned long mmr_offset; unsigned long index; cycles_t time1; cycles_t time2; - cycles_t elapsed; struct ptc_stats *stat = bcp->statp; - struct bau_control *smaster = bcp->socket_master; struct bau_control *hmaster = bcp->uvhub_master; + struct uv1_bau_msg_header *uv1_hdr = NULL; + struct uv2_bau_msg_header *uv2_hdr = NULL; - if (!atomic_inc_unless_ge(&hmaster->uvhub_lock, - &hmaster->active_descriptor_count, - hmaster->max_bau_concurrent)) { - stat->s_throttles++; - do { - cpu_relax(); - } while (!atomic_inc_unless_ge(&hmaster->uvhub_lock, - &hmaster->active_descriptor_count, - hmaster->max_bau_concurrent)); + if (bcp->uvhub_version == 1) { + uv1 = 1; + uv1_throttle(hmaster, stat); } + while (hmaster->uvhub_quiesce) cpu_relax(); - if (cpu < UV_CPUS_PER_ACT_STATUS) { - mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_0; - right_shift = cpu * UV_ACT_STATUS_SIZE; - } else { - mmr_offset = UVH_LB_BAU_SB_ACTIVATION_STATUS_1; - right_shift = - ((cpu - UV_CPUS_PER_ACT_STATUS) * UV_ACT_STATUS_SIZE); - } time1 = get_cycles(); + if (uv1) + uv1_hdr = &bau_desc->header.uv1_hdr; + else + uv2_hdr = &bau_desc->header.uv2_hdr; + do { if (try == 0) { - bau_desc->header.msg_type = MSG_REGULAR; + 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 << UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT) | - bcp->uvhub_cpu; + + if (uv1) + uv1_hdr->sequence = seq_number; + else + uv2_hdr->sequence = seq_number; + index = (1UL << AS_PUSH_SHIFT) | bcp->uvhub_cpu; bcp->send_message = get_cycles(); - uv_write_local_mmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index); + + write_mmr_activation(index); + try++; - completion_status = uv_wait_completion(bau_desc, mmr_offset, - right_shift, this_cpu, bcp, smaster, try); + completion_stat = wait_completion(bau_desc, bcp, try); + + handle_cmplt(completion_stat, bau_desc, bcp, hmaster, stat); - if (completion_status == FLUSH_RETRY_PLUGGED) { - destination_plugged(bau_desc, bcp, hmaster, stat); - } else if (completion_status == FLUSH_RETRY_TIMEOUT) { - destination_timeout(bau_desc, bcp, hmaster, stat); - } if (bcp->ipi_attempts >= bcp->ipi_reset_limit) { bcp->ipi_attempts = 0; - completion_status = FLUSH_GIVEUP; + stat->s_overipilimit++; + completion_stat = FLUSH_GIVEUP; break; } cpu_relax(); - } while ((completion_status == FLUSH_RETRY_PLUGGED) || - (completion_status == FLUSH_RETRY_TIMEOUT)); + } while ((completion_stat == FLUSH_RETRY_PLUGGED) || + (completion_stat == FLUSH_RETRY_TIMEOUT)); + time2 = get_cycles(); - bcp->plugged_tries = 0; - bcp->timeout_tries = 0; - if ((completion_status == FLUSH_COMPLETE) && - (bcp->conseccompletes > bcp->complete_threshold) && - (hmaster->max_bau_concurrent < - hmaster->max_bau_concurrent_constant)) - hmaster->max_bau_concurrent++; + + count_max_concurr(completion_stat, bcp, hmaster); + while (hmaster->uvhub_quiesce) cpu_relax(); + atomic_dec(&hmaster->active_descriptor_count); - if (time2 > time1) { - elapsed = time2 - time1; - stat->s_time += elapsed; - if ((completion_status == FLUSH_COMPLETE) && (try == 1)) { - bcp->period_requests++; - bcp->period_time += elapsed; - if ((elapsed > congested_cycles) && - (bcp->period_requests > bcp->congested_reps)) { - disable_for_congestion(bcp, stat); + + 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; +} + +/* + * The BAU is disabled for this uvhub. When the disabled time period has + * expired re-enable it. + * Return 0 if it is re-enabled for all cpus on this uvhub. + */ +static int check_enable(struct bau_control *bcp, struct ptc_stats *stat) +{ + int tcpu; + struct bau_control *tbcp; + struct bau_control *hmaster; + + hmaster = bcp->uvhub_master; + spin_lock(&hmaster->disable_lock); + if (bcp->baudisabled && (get_cycles() >= bcp->set_bau_on_time)) { + stat->s_bau_reenabled++; + for_each_present_cpu(tcpu) { + tbcp = &per_cpu(bau_control, tcpu); + if (tbcp->uvhub_master == hmaster) { + tbcp->baudisabled = 0; + tbcp->period_requests = 0; + tbcp->period_time = 0; + tbcp->period_giveups = 0; } } + spin_unlock(&hmaster->disable_lock); + return 0; + } + spin_unlock(&hmaster->disable_lock); + return -1; +} + +static void record_send_statistics(struct ptc_stats *stat, int locals, int hubs, + int remotes, struct bau_desc *bau_desc) +{ + stat->s_requestor++; + stat->s_ntargcpu += remotes + locals; + stat->s_ntargremotes += remotes; + stat->s_ntarglocals += locals; + + /* uvhub statistics */ + hubs = bau_uvhub_weight(&bau_desc->distribution); + if (locals) { + stat->s_ntarglocaluvhub++; + stat->s_ntargremoteuvhub += (hubs - 1); } else - stat->s_requestor--; - if (completion_status == FLUSH_COMPLETE && try > 1) - stat->s_retriesok++; - else if (completion_status == FLUSH_GIVEUP) { - stat->s_giveup++; - return 1; + stat->s_ntargremoteuvhub += hubs; + + stat->s_ntarguvhub += hubs; + + if (hubs >= 16) + stat->s_ntarguvhub16++; + else if (hubs >= 8) + stat->s_ntarguvhub8++; + else if (hubs >= 4) + stat->s_ntarguvhub4++; + else if (hubs >= 2) + stat->s_ntarguvhub2++; + else + stat->s_ntarguvhub1++; +} + +/* + * Translate a cpu mask to the uvhub distribution mask in the BAU + * activation descriptor. + */ +static int set_distrib_bits(struct cpumask *flush_mask, struct bau_control *bcp, + struct bau_desc *bau_desc, int *localsp, int *remotesp) +{ + int cpu; + int pnode; + int cnt = 0; + struct hub_and_pnode *hpp; + + for_each_cpu(cpu, flush_mask) { + /* + * The distribution vector is a bit map of pnodes, relative + * to the partition base pnode (and the partition base nasid + * in the header). + * Translate cpu to pnode and hub using a local memory array. + */ + hpp = &bcp->socket_master->thp[cpu]; + pnode = hpp->pnode - bcp->partition_base_pnode; + bau_uvhub_set(pnode, &bau_desc->distribution); + cnt++; + if (hpp->uvhub == bcp->uvhub) + (*localsp)++; + else + (*remotesp)++; } + if (!cnt) + return 1; return 0; } -/** - * uv_flush_tlb_others - globally purge translation cache of a virtual - * address or all TLB's +/* + * globally purge translation cache of a virtual address or all TLB's * @cpumask: mask of all cpu's in which the address is to be removed * @mm: mm_struct containing virtual address range - * @va: virtual address to be removed (or TLB_FLUSH_ALL for all TLB's on cpu) + * @start: start virtual address to be removed from TLB + * @end: end virtual address to be remove from TLB * @cpu: the current cpu * * This is the entry point for initiating any UV global TLB shootdown. @@ -695,11 +1080,9 @@ int uv_flush_send_and_wait(struct bau_desc *bau_desc, * done. The returned pointer is valid till preemption is re-enabled. */ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, - struct mm_struct *mm, - unsigned long va, unsigned int cpu) + struct mm_struct *mm, unsigned long start, + unsigned long end, unsigned int cpu) { - int tcpu; - int uvhub; int locals = 0; int remotes = 0; int hubs = 0; @@ -707,31 +1090,33 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, struct cpumask *flush_mask; struct ptc_stats *stat; struct bau_control *bcp; - struct bau_control *tbcp; + unsigned long descriptor_status; + unsigned long status; + + bcp = &per_cpu(bau_control, cpu); - /* kernel was booted 'nobau' */ - if (nobau) + if (bcp->nobau) return cpumask; - bcp = &per_cpu(bau_control, cpu); stat = bcp->statp; + stat->s_enters++; + + if (bcp->busy) { + descriptor_status = + read_lmmr(UVH_LB_BAU_SB_ACTIVATION_STATUS_0); + status = ((descriptor_status >> (bcp->uvhub_cpu * + UV_ACT_STATUS_SIZE)) & UV_ACT_STATUS_MASK) << 1; + if (status == UV2H_DESC_BUSY) + return cpumask; + bcp->busy = 0; + } /* bau was disabled due to slow response */ if (bcp->baudisabled) { - /* the cpu that disabled it must re-enable it */ - if (bcp->set_bau_off) { - if (get_cycles() >= bcp->set_bau_on_time) { - stat->s_bau_reenabled++; - baudisabled = 0; - for_each_present_cpu(tcpu) { - tbcp = &per_cpu(bau_control, tcpu); - tbcp->baudisabled = 0; - tbcp->period_requests = 0; - tbcp->period_time = 0; - } - } + if (check_enable(bcp, stat)) { + stat->s_ipifordisabled++; + return cpumask; } - return cpumask; } /* @@ -742,63 +1127,107 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, flush_mask = (struct cpumask *)per_cpu(uv_flush_tlb_mask, cpu); /* don't actually do a shootdown of the local cpu */ cpumask_andnot(flush_mask, cpumask, cpumask_of(cpu)); + if (cpu_isset(cpu, *cpumask)) stat->s_ntargself++; bau_desc = bcp->descriptor_base; - bau_desc += UV_ITEMS_PER_DESCRIPTOR * bcp->uvhub_cpu; + bau_desc += (ITEMS_PER_DESC * bcp->uvhub_cpu); bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE); - - /* cpu statistics */ - for_each_cpu(tcpu, flush_mask) { - uvhub = uv_cpu_to_blade_id(tcpu); - bau_uvhub_set(uvhub, &bau_desc->distribution); - if (uvhub == bcp->uvhub) - locals++; - else - remotes++; - } - if ((locals + remotes) == 0) + if (set_distrib_bits(flush_mask, bcp, bau_desc, &locals, &remotes)) return NULL; - stat->s_requestor++; - stat->s_ntargcpu += remotes + locals; - stat->s_ntargremotes += remotes; - stat->s_ntarglocals += locals; - remotes = bau_uvhub_weight(&bau_desc->distribution); - /* uvhub statistics */ - hubs = bau_uvhub_weight(&bau_desc->distribution); - if (locals) { - stat->s_ntarglocaluvhub++; - stat->s_ntargremoteuvhub += (hubs - 1); - } else - stat->s_ntargremoteuvhub += hubs; - stat->s_ntarguvhub += hubs; - if (hubs >= 16) - stat->s_ntarguvhub16++; - else if (hubs >= 8) - stat->s_ntarguvhub8++; - else if (hubs >= 4) - stat->s_ntarguvhub4++; - else if (hubs >= 2) - stat->s_ntarguvhub2++; - else - stat->s_ntarguvhub1++; + record_send_statistics(stat, locals, hubs, remotes, bau_desc); - bau_desc->payload.address = va; + if (!end || (end - start) <= PAGE_SIZE) + bau_desc->payload.address = start; + else + bau_desc->payload.address = TLB_FLUSH_ALL; bau_desc->payload.sending_cpu = cpu; - /* * 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, bau_desc)) return NULL; else return cpumask; } /* + * Search the message queue for any 'other' unprocessed message with the + * same software acknowledge resource bit vector as the 'msg' message. + */ +struct bau_pq_entry *find_another_by_swack(struct bau_pq_entry *msg, + struct bau_control *bcp) +{ + struct bau_pq_entry *msg_next = msg + 1; + unsigned char swack_vec = msg->swack_vec; + + if (msg_next > bcp->queue_last) + msg_next = bcp->queue_first; + while (msg_next != msg) { + if ((msg_next->canceled == 0) && (msg_next->replied_to == 0) && + (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. + */ + /* + * 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); + if (other_msg) { + /* + * There is another. Process this one but do not + * ack it. + */ + bau_process_message(mdp, bcp, 0); + /* + * Let the natural processing of that other message + * acknowledge it. Don't get the processing of sw_ack's + * out of order. + */ + return; + } + } + + /* + * Either the MMR shows this one pending a reply or 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 * @@ -816,26 +1245,34 @@ void uv_bau_message_interrupt(struct pt_regs *regs) { int count = 0; cycles_t time_start; - struct bau_payload_queue_entry *msg; + struct bau_pq_entry *msg; struct bau_control *bcp; struct ptc_stats *stat; struct msg_desc msgdesc; + ack_APIC_irq(); time_start = get_cycles(); + bcp = &per_cpu(bau_control, smp_processor_id()); stat = bcp->statp; - msgdesc.va_queue_first = bcp->va_queue_first; - msgdesc.va_queue_last = bcp->va_queue_last; + + msgdesc.queue_first = bcp->queue_first; + msgdesc.queue_last = bcp->queue_last; + msg = bcp->bau_msg_head; - while (msg->sw_ack_vector) { + while (msg->swack_vec) { count++; - msgdesc.msg_slot = msg - msgdesc.va_queue_first; - msgdesc.sw_ack_slot = ffs(msg->sw_ack_vector) - 1; + + msgdesc.msg_slot = msg - msgdesc.queue_first; msgdesc.msg = msg; - uv_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.va_queue_last) - msg = msgdesc.va_queue_first; + if (msg > msgdesc.queue_last) + msg = msgdesc.queue_first; bcp->bau_msg_head = msg; } stat->d_time += (get_cycles() - time_start); @@ -843,18 +1280,15 @@ void uv_bau_message_interrupt(struct pt_regs *regs) stat->d_nomsg++; else if (count > 1) stat->d_multmsg++; - ack_APIC_irq(); } /* - * uv_enable_timeouts - * - * Each target uvhub (i.e. a uvhub that has no cpu's) needs to have + * Each target uvhub (i.e. a uvhub that has cpu's) needs to have * shootdown message timeouts enabled. The timeout does not cause * an interrupt, but causes an error message to be returned to * the sender. */ -static void uv_enable_timeouts(void) +static void __init enable_timeouts(void) { int uvhub; int nuvhubs; @@ -868,47 +1302,44 @@ static void uv_enable_timeouts(void) continue; pnode = uv_blade_to_pnode(uvhub); - mmr_image = - uv_read_global_mmr64(pnode, UVH_LB_BAU_MISC_CONTROL); + mmr_image = read_mmr_misc_control(pnode); /* * Set the timeout period and then lock it in, in three * steps; captures and locks in the period. * * To program the period, the SOFT_ACK_MODE must be off. */ - mmr_image &= ~((unsigned long)1 << - UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT); - uv_write_global_mmr64 - (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image); + mmr_image &= ~(1L << SOFTACK_MSHIFT); + write_mmr_misc_control(pnode, mmr_image); /* * Set the 4-bit period. */ - mmr_image &= ~((unsigned long)0xf << - UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT); - mmr_image |= (UV_INTD_SOFT_ACK_TIMEOUT_PERIOD << - UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT); - uv_write_global_mmr64 - (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image); + mmr_image &= ~((unsigned long)0xf << SOFTACK_PSHIFT); + mmr_image |= (SOFTACK_TIMEOUT_PERIOD << SOFTACK_PSHIFT); + write_mmr_misc_control(pnode, mmr_image); /* + * UV1: * Subsequent reversals of the timebase bit (3) cause an * immediate timeout of one or all INTD resources as * indicated in bits 2:0 (7 causes all of them to timeout). */ - mmr_image |= ((unsigned long)1 << - UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT); - uv_write_global_mmr64 - (pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image); + mmr_image |= (1L << SOFTACK_MSHIFT); + if (is_uv2_hub()) { + /* hw bug workaround; do not use extended status */ + mmr_image &= ~(1L << UV2_EXT_SHFT); + } + write_mmr_misc_control(pnode, mmr_image); } } -static void *uv_ptc_seq_start(struct seq_file *file, loff_t *offset) +static void *ptc_seq_start(struct seq_file *file, loff_t *offset) { if (*offset < num_possible_cpus()) return offset; return NULL; } -static void *uv_ptc_seq_next(struct seq_file *file, void *data, loff_t *offset) +static void *ptc_seq_next(struct seq_file *file, void *data, loff_t *offset) { (*offset)++; if (*offset < num_possible_cpus()) @@ -916,82 +1347,77 @@ static void *uv_ptc_seq_next(struct seq_file *file, void *data, loff_t *offset) return NULL; } -static void uv_ptc_seq_stop(struct seq_file *file, void *data) -{ -} - -static inline unsigned long long -microsec_2_cycles(unsigned long microsec) +static void ptc_seq_stop(struct seq_file *file, void *data) { - unsigned long ns; - unsigned long long cyc; - - ns = microsec * 1000; - cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id())); - return cyc; } /* - * Display the statistics thru /proc. + * Display the statistics thru /proc/sgi_uv/ptc_statistics * 'data' points to the cpu number + * Note: see the descriptions in stat_description[]. */ -static int uv_ptc_seq_show(struct seq_file *file, void *data) +static int ptc_seq_show(struct seq_file *file, void *data) { struct ptc_stats *stat; + struct bau_control *bcp; int cpu; cpu = *(loff_t *)data; - if (!cpu) { seq_printf(file, - "# cpu sent stime self locals remotes ncpus localhub "); + "# cpu bauoff sent stime self locals remotes ncpus localhub "); seq_printf(file, "remotehub numuvhubs numuvhubs16 numuvhubs8 "); seq_printf(file, - "numuvhubs4 numuvhubs2 numuvhubs1 dto "); + "numuvhubs4 numuvhubs2 numuvhubs1 dto snacks retries "); seq_printf(file, - "retries rok resetp resett giveup sto bz throt "); + "rok resetp resett giveup sto bz throt disable "); seq_printf(file, - "sw_ack recv rtime all "); + "enable wars warshw warwaits enters ipidis plugged "); seq_printf(file, - "one mult none retry canc nocan reset rcan "); + "ipiover glim cong swack recv rtime all one mult "); seq_printf(file, - "disable enable\n"); + "none retry canc nocan reset rcan\n"); } if (cpu < num_possible_cpus() && cpu_online(cpu)) { - stat = &per_cpu(ptcstats, cpu); + bcp = &per_cpu(bau_control, cpu); + stat = bcp->statp; /* source side statistics */ seq_printf(file, - "cpu %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ", - cpu, stat->s_requestor, cycles_2_us(stat->s_time), + "cpu %d %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ", + cpu, bcp->nobau, stat->s_requestor, + cycles_2_us(stat->s_time), stat->s_ntargself, stat->s_ntarglocals, stat->s_ntargremotes, stat->s_ntargcpu, stat->s_ntarglocaluvhub, stat->s_ntargremoteuvhub, stat->s_ntarguvhub, stat->s_ntarguvhub16); - seq_printf(file, "%ld %ld %ld %ld %ld ", + seq_printf(file, "%ld %ld %ld %ld %ld %ld ", stat->s_ntarguvhub8, stat->s_ntarguvhub4, stat->s_ntarguvhub2, stat->s_ntarguvhub1, - stat->s_dtimeout); + stat->s_dtimeout, stat->s_strongnacks); seq_printf(file, "%ld %ld %ld %ld %ld %ld %ld %ld ", stat->s_retry_messages, stat->s_retriesok, stat->s_resets_plug, stat->s_resets_timeout, stat->s_giveup, stat->s_stimeout, stat->s_busy, stat->s_throttles); + seq_printf(file, "%ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ", + stat->s_bau_disabled, stat->s_bau_reenabled, + stat->s_uv2_wars, stat->s_uv2_wars_hw, + stat->s_uv2_war_waits, stat->s_enters, + stat->s_ipifordisabled, stat->s_plugged, + stat->s_overipilimit, stat->s_giveuplimit, + stat->s_congested); /* destination side statistics */ seq_printf(file, - "%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ", - uv_read_global_mmr64(uv_cpu_to_pnode(cpu), - UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE), + "%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n", + read_gmmr_sw_ack(uv_cpu_to_pnode(cpu)), stat->d_requestee, cycles_2_us(stat->d_time), stat->d_alltlb, stat->d_onetlb, stat->d_multmsg, 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); } - return 0; } @@ -999,18 +1425,19 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data) * Display the tunables thru debugfs */ static ssize_t tunables_read(struct file *file, char __user *userbuf, - size_t count, loff_t *ppos) + size_t count, loff_t *ppos) { char *buf; int ret; - buf = kasprintf(GFP_KERNEL, "%s %s %s\n%d %d %d %d %d %d %d %d %d\n", - "max_bau_concurrent plugged_delay plugsb4reset", - "timeoutsb4reset ipi_reset_limit complete_threshold", - "congested_response_us congested_reps congested_period", - max_bau_concurrent, plugged_delay, plugsb4reset, + buf = kasprintf(GFP_KERNEL, "%s %s %s\n%d %d %d %d %d %d %d %d %d %d\n", + "max_concur plugged_delay plugsb4reset timeoutsb4reset", + "ipi_reset_limit complete_threshold congested_response_us", + "congested_reps disabled_period giveup_limit", + max_concurr, plugged_delay, plugsb4reset, timeoutsb4reset, ipi_reset_limit, complete_threshold, - congested_response_us, congested_reps, congested_period); + congested_respns_us, congested_reps, disabled_period, + giveup_limit); if (!buf) return -ENOMEM; @@ -1021,13 +1448,16 @@ static ssize_t tunables_read(struct file *file, char __user *userbuf, } /* - * -1: resetf the statistics + * handle a write to /proc/sgi_uv/ptc_statistics + * -1: reset the statistics * 0: display meaning of the statistics */ -static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user, - size_t count, loff_t *data) +static ssize_t ptc_proc_write(struct file *file, const char __user *user, + size_t count, loff_t *data) { int cpu; + int i; + int elements; long input_arg; char optstr[64]; struct ptc_stats *stat; @@ -1037,79 +1467,26 @@ static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user, if (copy_from_user(optstr, user, count)) return -EFAULT; optstr[count - 1] = '\0'; + + if (!strcmp(optstr, "on")) { + set_bau_on(); + return count; + } else if (!strcmp(optstr, "off")) { + set_bau_off(); + return count; + } + if (strict_strtol(optstr, 10, &input_arg) < 0) { printk(KERN_DEBUG "%s is invalid\n", optstr); return -EINVAL; } if (input_arg == 0) { + elements = ARRAY_SIZE(stat_description); printk(KERN_DEBUG "# cpu: cpu number\n"); printk(KERN_DEBUG "Sender statistics:\n"); - printk(KERN_DEBUG - "sent: number of shootdown messages sent\n"); - printk(KERN_DEBUG - "stime: time spent sending messages\n"); - printk(KERN_DEBUG - "numuvhubs: number of hubs targeted with shootdown\n"); - printk(KERN_DEBUG - "numuvhubs16: number times 16 or more hubs targeted\n"); - printk(KERN_DEBUG - "numuvhubs8: number times 8 or more hubs targeted\n"); - printk(KERN_DEBUG - "numuvhubs4: number times 4 or more hubs targeted\n"); - printk(KERN_DEBUG - "numuvhubs2: number times 2 or more hubs targeted\n"); - printk(KERN_DEBUG - "numuvhubs1: number times 1 hub targeted\n"); - printk(KERN_DEBUG - "numcpus: number of cpus targeted with shootdown\n"); - printk(KERN_DEBUG - "dto: number of destination timeouts\n"); - printk(KERN_DEBUG - "retries: destination timeout retries sent\n"); - printk(KERN_DEBUG - "rok: : destination timeouts successfully retried\n"); - printk(KERN_DEBUG - "resetp: ipi-style resource resets for plugs\n"); - printk(KERN_DEBUG - "resett: ipi-style resource resets for timeouts\n"); - printk(KERN_DEBUG - "giveup: fall-backs to ipi-style shootdowns\n"); - printk(KERN_DEBUG - "sto: number of source timeouts\n"); - printk(KERN_DEBUG - "bz: number of stay-busy's\n"); - printk(KERN_DEBUG - "throt: number times spun in throttle\n"); - printk(KERN_DEBUG "Destination side statistics:\n"); - printk(KERN_DEBUG - "sw_ack: image of UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE\n"); - printk(KERN_DEBUG - "recv: shootdown messages received\n"); - printk(KERN_DEBUG - "rtime: time spent processing messages\n"); - printk(KERN_DEBUG - "all: shootdown all-tlb messages\n"); - printk(KERN_DEBUG - "one: shootdown one-tlb messages\n"); - printk(KERN_DEBUG - "mult: interrupts that found multiple messages\n"); - printk(KERN_DEBUG - "none: interrupts that found no messages\n"); - printk(KERN_DEBUG - "retry: number of retry messages processed\n"); - printk(KERN_DEBUG - "canc: number messages canceled by retries\n"); - printk(KERN_DEBUG - "nocan: number retries that found nothing to cancel\n"); - printk(KERN_DEBUG - "reset: number of ipi-style reset requests processed\n"); - printk(KERN_DEBUG - "rcan: number messages canceled by reset requests\n"); - printk(KERN_DEBUG - "disable: number times use of the BAU was disabled\n"); - printk(KERN_DEBUG - "enable: number times use of the BAU was re-enabled\n"); + for (i = 0; i < elements; i++) + printk(KERN_DEBUG "%s\n", stat_description[i]); } else if (input_arg == -1) { for_each_present_cpu(cpu) { stat = &per_cpu(ptcstats, cpu); @@ -1136,27 +1513,18 @@ static int local_atoi(const char *name) } /* - * set the tunables - * 0 values reset them to defaults + * Parse the values written to /sys/kernel/debug/sgi_uv/bau_tunables. + * Zero values reset them to defaults. */ -static ssize_t tunables_write(struct file *file, const char __user *user, - size_t count, loff_t *data) +static int parse_tunables_write(struct bau_control *bcp, char *instr, + int count) { - int cpu; - int cnt = 0; - int val; char *p; char *q; - char instr[64]; - struct bau_control *bcp; - - if (count == 0 || count > sizeof(instr)-1) - return -EINVAL; - if (copy_from_user(instr, user, count)) - return -EFAULT; + int cnt = 0; + int val; + int e = ARRAY_SIZE(tunables); - instr[count] = '\0'; - /* count the fields */ p = instr + strspn(instr, WHITESPACE); q = p; for (; *p; p = q + strspn(q, WHITESPACE)) { @@ -1165,8 +1533,8 @@ static ssize_t tunables_write(struct file *file, const char __user *user, if (q == p) break; } - if (cnt != 9) { - printk(KERN_INFO "bau tunable error: should be 9 numbers\n"); + if (cnt != e) { + printk(KERN_INFO "bau tunable error: should be %d values\n", e); return -EINVAL; } @@ -1178,97 +1546,82 @@ static ssize_t tunables_write(struct file *file, const char __user *user, switch (cnt) { case 0: if (val == 0) { - max_bau_concurrent = MAX_BAU_CONCURRENT; - max_bau_concurrent_constant = - MAX_BAU_CONCURRENT; + max_concurr = MAX_BAU_CONCURRENT; + max_concurr_const = MAX_BAU_CONCURRENT; continue; } - bcp = &per_cpu(bau_control, smp_processor_id()); if (val < 1 || val > bcp->cpus_in_uvhub) { printk(KERN_DEBUG "Error: BAU max concurrent %d is invalid\n", val); return -EINVAL; } - max_bau_concurrent = val; - max_bau_concurrent_constant = val; - continue; - case 1: - if (val == 0) - plugged_delay = PLUGGED_DELAY; - else - plugged_delay = val; - continue; - case 2: - if (val == 0) - plugsb4reset = PLUGSB4RESET; - else - plugsb4reset = val; - continue; - case 3: - if (val == 0) - timeoutsb4reset = TIMEOUTSB4RESET; - else - timeoutsb4reset = val; - continue; - case 4: - if (val == 0) - ipi_reset_limit = IPI_RESET_LIMIT; - else - ipi_reset_limit = val; + max_concurr = val; + max_concurr_const = val; continue; - case 5: - if (val == 0) - complete_threshold = COMPLETE_THRESHOLD; - else - complete_threshold = val; - continue; - case 6: - if (val == 0) - congested_response_us = CONGESTED_RESPONSE_US; - else - congested_response_us = val; - continue; - case 7: - if (val == 0) - congested_reps = CONGESTED_REPS; - else - congested_reps = val; - continue; - case 8: + default: if (val == 0) - congested_period = CONGESTED_PERIOD; + *tunables[cnt].tunp = tunables[cnt].deflt; else - congested_period = val; + *tunables[cnt].tunp = val; continue; } if (q == p) break; } + return 0; +} + +/* + * Handle a write to debugfs. (/sys/kernel/debug/sgi_uv/bau_tunables) + */ +static ssize_t tunables_write(struct file *file, const char __user *user, + size_t count, loff_t *data) +{ + int cpu; + int ret; + char instr[100]; + struct bau_control *bcp; + + if (count == 0 || count > sizeof(instr)-1) + return -EINVAL; + if (copy_from_user(instr, user, count)) + return -EFAULT; + + instr[count] = '\0'; + + cpu = get_cpu(); + bcp = &per_cpu(bau_control, cpu); + ret = parse_tunables_write(bcp, instr, count); + put_cpu(); + if (ret) + return ret; + for_each_present_cpu(cpu) { bcp = &per_cpu(bau_control, cpu); - bcp->max_bau_concurrent = max_bau_concurrent; - bcp->max_bau_concurrent_constant = max_bau_concurrent; - bcp->plugged_delay = plugged_delay; - bcp->plugsb4reset = plugsb4reset; - bcp->timeoutsb4reset = timeoutsb4reset; - bcp->ipi_reset_limit = ipi_reset_limit; - bcp->complete_threshold = complete_threshold; - bcp->congested_response_us = congested_response_us; - bcp->congested_reps = congested_reps; - bcp->congested_period = congested_period; + bcp->max_concurr = max_concurr; + bcp->max_concurr_const = max_concurr; + bcp->plugged_delay = plugged_delay; + bcp->plugsb4reset = plugsb4reset; + bcp->timeoutsb4reset = timeoutsb4reset; + bcp->ipi_reset_limit = ipi_reset_limit; + bcp->complete_threshold = complete_threshold; + bcp->cong_response_us = congested_respns_us; + bcp->cong_reps = congested_reps; + bcp->disabled_period = sec_2_cycles(disabled_period); + bcp->giveup_limit = giveup_limit; } return count; } static const struct seq_operations uv_ptc_seq_ops = { - .start = uv_ptc_seq_start, - .next = uv_ptc_seq_next, - .stop = uv_ptc_seq_stop, - .show = uv_ptc_seq_show + .start = ptc_seq_start, + .next = ptc_seq_next, + .stop = ptc_seq_stop, + .show = ptc_seq_show }; -static int uv_ptc_proc_open(struct inode *inode, struct file *file) +static int ptc_proc_open(struct inode *inode, struct file *file) { return seq_open(file, &uv_ptc_seq_ops); } @@ -1279,9 +1632,9 @@ static int tunables_open(struct inode *inode, struct file *file) } static const struct file_operations proc_uv_ptc_operations = { - .open = uv_ptc_proc_open, + .open = ptc_proc_open, .read = seq_read, - .write = uv_ptc_proc_write, + .write = ptc_proc_write, .llseek = seq_lseek, .release = seq_release, }; @@ -1315,7 +1668,7 @@ static int __init uv_ptc_init(void) return -EINVAL; } tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600, - tunables_dir, NULL, &tunables_fops); + tunables_dir, NULL, &tunables_fops); if (!tunables_file) { printk(KERN_ERR "unable to create debugfs file %s\n", UV_BAU_TUNABLES_FILE); @@ -1325,57 +1678,77 @@ static int __init uv_ptc_init(void) } /* - * initialize the sending side's sending buffers + * Initialize the sending side's sending buffers. */ -static void -uv_activation_descriptor_init(int node, int pnode) +static void activation_descriptor_init(int node, int pnode, int base_pnode) { int i; int cpu; - unsigned long pa; + 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; /* - * each bau_desc is 64 bytes; there are 8 (UV_ITEMS_PER_DESCRIPTOR) - * per cpu; and up to 32 (UV_ADP_SIZE) cpu's per uvhub + * each bau_desc is 64 bytes; there are 8 (ITEMS_PER_DESC) + * per cpu; and one per cpu on the uvhub (ADP_SZ) */ - bau_desc = kmalloc_node(sizeof(struct bau_desc) * UV_ADP_SIZE - * UV_ITEMS_PER_DESCRIPTOR, GFP_KERNEL, node); + dsize = sizeof(struct bau_desc) * ADP_SZ * ITEMS_PER_DESC; + bau_desc = kmalloc_node(dsize, GFP_KERNEL, node); BUG_ON(!bau_desc); - pa = uv_gpa(bau_desc); /* need the real nasid*/ - n = pa >> uv_nshift; - m = pa & uv_mmask; - - uv_write_global_mmr64(pnode, UVH_LB_BAU_SB_DESCRIPTOR_BASE, - (n << UV_DESC_BASE_PNODE_SHIFT | m)); + 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)); /* - * initializing all 8 (UV_ITEMS_PER_DESCRIPTOR) descriptors for each + * Initializing all 8 (ITEMS_PER_DESC) descriptors for each * cpu even though we only use the first one; one descriptor can * describe a broadcast to 256 uv hubs. */ - for (i = 0, bd2 = bau_desc; i < (UV_ADP_SIZE*UV_ITEMS_PER_DESCRIPTOR); - i++, bd2++) { + for (i = 0, bd2 = bau_desc; i < (ADP_SZ * ITEMS_PER_DESC); i++, bd2++) { memset(bd2, 0, sizeof(struct bau_desc)); - bd2->header.sw_ack_flag = 1; - /* - * base_dest_nodeid is the nasid (pnode<<1) of the first uvhub - * in the partition. The bit map will indicate uvhub numbers, - * which are 0-N in a partition. Pnodes are unique system-wide. - */ - bd2->header.base_dest_nodeid = uv_partition_base_pnode << 1; - bd2->header.dest_subnodeid = 0x10; /* the LB */ - 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 { + /* + * BIOS uses legacy mode, but UV2 hardware always + * uses native mode for selective broadcasts. + */ + 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))) @@ -1391,57 +1764,56 @@ uv_activation_descriptor_init(int node, int pnode) * - node is first node (kernel memory notion) on the uvhub * - pnode is the uvhub's physical identifier */ -static void -uv_payload_queue_init(int node, int pnode) +static void pq_init(int node, int pnode) { - int pn; int cpu; + size_t plsize; char *cp; - unsigned long pa; - struct bau_payload_queue_entry *pqp; - struct bau_payload_queue_entry *pqp_malloc; + void *vp; + unsigned long pn; + unsigned long first; + unsigned long pn_first; + unsigned long last; + struct bau_pq_entry *pqp; struct bau_control *bcp; - pqp = kmalloc_node((DEST_Q_SIZE + 1) - * sizeof(struct bau_payload_queue_entry), - GFP_KERNEL, node); + plsize = (DEST_Q_SIZE + 1) * sizeof(struct bau_pq_entry); + vp = kmalloc_node(plsize, GFP_KERNEL, node); + pqp = (struct bau_pq_entry *)vp; BUG_ON(!pqp); - pqp_malloc = pqp; cp = (char *)pqp + 31; - pqp = (struct bau_payload_queue_entry *)(((unsigned long)cp >> 5) << 5); + pqp = (struct bau_pq_entry *)(((unsigned long)cp >> 5) << 5); for_each_present_cpu(cpu) { if (pnode != uv_cpu_to_pnode(cpu)) continue; /* for every cpu on this pnode: */ bcp = &per_cpu(bau_control, cpu); - bcp->va_queue_first = pqp; - bcp->bau_msg_head = pqp; - bcp->va_queue_last = pqp + (DEST_Q_SIZE - 1); + bcp->queue_first = pqp; + bcp->bau_msg_head = pqp; + bcp->queue_last = pqp + (DEST_Q_SIZE - 1); } /* - * need the pnode of where the memory was really allocated + * need the gnode of where the memory was really allocated */ - pa = uv_gpa(pqp); - pn = pa >> uv_nshift; - uv_write_global_mmr64(pnode, - UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, - ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) | - uv_physnodeaddr(pqp)); - uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL, - uv_physnodeaddr(pqp)); - uv_write_global_mmr64(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST, - (unsigned long) - uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1))); + pn = uv_gpa_to_gnode(uv_gpa(pqp)); + first = uv_physnodeaddr(pqp); + pn_first = ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) | first; + last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1)); + 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_payload_queue_entry) * DEST_Q_SIZE); + memset(pqp, 0, sizeof(struct bau_pq_entry) * DEST_Q_SIZE); } /* * Initialization of each UV hub's structures */ -static void __init uv_init_uvhub(int uvhub, int vector) +static void __init init_uvhub(int uvhub, int vector, int base_pnode) { int node; int pnode; @@ -1449,24 +1821,24 @@ static void __init uv_init_uvhub(int uvhub, int vector) node = uvhub_to_first_node(uvhub); pnode = uv_blade_to_pnode(uvhub); - uv_activation_descriptor_init(node, pnode); - uv_payload_queue_init(node, pnode); + + activation_descriptor_init(node, pnode, base_pnode); + + pq_init(node, pnode); /* - * the below initialization can't be in firmware because the - * messaging IRQ will be determined by the OS + * The below initialization can't be in firmware because the + * messaging IRQ will be determined by the OS. */ - apicid = uvhub_to_first_apicid(uvhub); - uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, - ((apicid << 32) | vector)); + apicid = uvhub_to_first_apicid(uvhub) | uv_apicid_hibits; + write_mmr_data_config(pnode, ((apicid << 32) | vector)); } /* * We will set BAU_MISC_CONTROL with a timeout period. * But the BIOS has set UVH_AGING_PRESCALE_SEL and UVH_TRANSACTION_TIMEOUT. - * So the destination timeout period has be be calculated from them. + * So the destination timeout period has to be calculated from them. */ -static int -calculate_destination_timeout(void) +static int calculate_destination_timeout(void) { unsigned long mmr_image; int mult1; @@ -1476,125 +1848,254 @@ calculate_destination_timeout(void) int ret; unsigned long ts_ns; - mult1 = UV_INTD_SOFT_ACK_TIMEOUT_PERIOD & BAU_MISC_CONTROL_MULT_MASK; - mmr_image = uv_read_local_mmr(UVH_AGING_PRESCALE_SEL); - index = (mmr_image >> BAU_URGENCY_7_SHIFT) & BAU_URGENCY_7_MASK; - mmr_image = uv_read_local_mmr(UVH_TRANSACTION_TIMEOUT); - mult2 = (mmr_image >> BAU_TRANS_SHIFT) & BAU_TRANS_MASK; - base = timeout_base_ns[index]; - ts_ns = base * mult1 * mult2; - ret = ts_ns / 1000; + if (is_uv1_hub()) { + mult1 = SOFTACK_TIMEOUT_PERIOD & BAU_MISC_CONTROL_MULT_MASK; + mmr_image = uv_read_local_mmr(UVH_AGING_PRESCALE_SEL); + index = (mmr_image >> BAU_URGENCY_7_SHIFT) & BAU_URGENCY_7_MASK; + mmr_image = uv_read_local_mmr(UVH_TRANSACTION_TIMEOUT); + mult2 = (mmr_image >> BAU_TRANS_SHIFT) & BAU_TRANS_MASK; + ts_ns = timeout_base_ns[index]; + ts_ns *= (mult1 * mult2); + ret = ts_ns / 1000; + } else { + /* 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)) + base = 80; + else + base = 10; + mult1 = mmr_image & UV2_ACK_MASK; + ret = mult1 * base; + } return ret; } +static void __init init_per_cpu_tunables(void) +{ + int cpu; + struct bau_control *bcp; + + for_each_present_cpu(cpu) { + bcp = &per_cpu(bau_control, cpu); + bcp->baudisabled = 0; + if (nobau) + bcp->nobau = 1; + bcp->statp = &per_cpu(ptcstats, cpu); + /* time interval to catch a hardware stay-busy bug */ + bcp->timeout_interval = usec_2_cycles(2*timeout_us); + bcp->max_concurr = max_concurr; + bcp->max_concurr_const = max_concurr; + bcp->plugged_delay = plugged_delay; + bcp->plugsb4reset = plugsb4reset; + bcp->timeoutsb4reset = timeoutsb4reset; + bcp->ipi_reset_limit = ipi_reset_limit; + bcp->complete_threshold = complete_threshold; + bcp->cong_response_us = congested_respns_us; + bcp->cong_reps = congested_reps; + bcp->disabled_period = sec_2_cycles(disabled_period); + bcp->giveup_limit = giveup_limit; + spin_lock_init(&bcp->queue_lock); + spin_lock_init(&bcp->uvhub_lock); + spin_lock_init(&bcp->disable_lock); + } +} + /* - * initialize the bau_control structure for each cpu + * Scan all cpus to collect blade and socket summaries. */ -static void __init uv_init_per_cpu(int nuvhubs) +static int __init get_cpu_topology(int base_pnode, + struct uvhub_desc *uvhub_descs, + unsigned char *uvhub_mask) { - int i; int cpu; int pnode; int uvhub; - int have_hmaster; - short socket = 0; - unsigned short socket_mask; - unsigned char *uvhub_mask; + int socket; struct bau_control *bcp; struct uvhub_desc *bdp; struct socket_desc *sdp; - struct bau_control *hmaster = NULL; - struct bau_control *smaster = NULL; - struct socket_desc { - short num_cpus; - short cpu_number[16]; - }; - struct uvhub_desc { - unsigned short socket_mask; - short num_cpus; - short uvhub; - short pnode; - struct socket_desc socket[2]; - }; - struct uvhub_desc *uvhub_descs; - - timeout_us = calculate_destination_timeout(); - uvhub_descs = kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL); - memset(uvhub_descs, 0, nuvhubs * sizeof(struct uvhub_desc)); - uvhub_mask = kzalloc((nuvhubs+7)/8, GFP_KERNEL); for_each_present_cpu(cpu) { bcp = &per_cpu(bau_control, cpu); + memset(bcp, 0, sizeof(struct bau_control)); + pnode = uv_cpu_hub_info(cpu)->pnode; + if ((pnode - base_pnode) >= UV_DISTRIBUTION_SIZE) { + printk(KERN_EMERG + "cpu %d pnode %d-%d beyond %d; BAU disabled\n", + cpu, pnode, base_pnode, UV_DISTRIBUTION_SIZE); + return 1; + } + + bcp->osnode = cpu_to_node(cpu); + bcp->partition_base_pnode = base_pnode; + uvhub = uv_cpu_hub_info(cpu)->numa_blade_id; *(uvhub_mask + (uvhub/8)) |= (1 << (uvhub%8)); bdp = &uvhub_descs[uvhub]; + bdp->num_cpus++; bdp->uvhub = uvhub; bdp->pnode = pnode; + /* kludge: 'assuming' one node per socket, and assuming that disabling a socket just leaves a gap in node numbers */ - socket = (cpu_to_node(cpu) & 1); + socket = bcp->osnode & 1; bdp->socket_mask |= (1 << socket); sdp = &bdp->socket[socket]; sdp->cpu_number[sdp->num_cpus] = cpu; sdp->num_cpus++; + if (sdp->num_cpus > MAX_CPUS_PER_SOCKET) { + printk(KERN_EMERG "%d cpus per socket invalid\n", + sdp->num_cpus); + return 1; + } } + return 0; +} + +/* + * Each socket is to get a local array of pnodes/hubs. + */ +static void make_per_cpu_thp(struct bau_control *smaster) +{ + int cpu; + size_t hpsz = sizeof(struct hub_and_pnode) * num_possible_cpus(); + + smaster->thp = kmalloc_node(hpsz, GFP_KERNEL, smaster->osnode); + memset(smaster->thp, 0, hpsz); + for_each_present_cpu(cpu) { + smaster->thp[cpu].pnode = uv_cpu_hub_info(cpu)->pnode; + smaster->thp[cpu].uvhub = uv_cpu_hub_info(cpu)->numa_blade_id; + } +} + +/* + * Each uvhub is to get a local cpumask. + */ +static void make_per_hub_cpumask(struct bau_control *hmaster) +{ + int sz = sizeof(cpumask_t); + + hmaster->cpumask = kzalloc_node(sz, GFP_KERNEL, hmaster->osnode); +} + +/* + * Initialize all the per_cpu information for the cpu's on a given socket, + * given what has been gathered into the socket_desc struct. + * And reports the chosen hub and socket masters back to the caller. + */ +static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp, + struct bau_control **smasterp, + struct bau_control **hmasterp) +{ + int i; + int cpu; + struct bau_control *bcp; + + for (i = 0; i < sdp->num_cpus; i++) { + cpu = sdp->cpu_number[i]; + bcp = &per_cpu(bau_control, cpu); + bcp->cpu = cpu; + if (i == 0) { + *smasterp = bcp; + if (!(*hmasterp)) + *hmasterp = bcp; + } + bcp->cpus_in_uvhub = bdp->num_cpus; + 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; + if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) { + printk(KERN_EMERG "%d cpus per uvhub invalid\n", + bcp->uvhub_cpu); + return 1; + } + } + return 0; +} + +/* + * Summarize the blade and socket topology into the per_cpu structures. + */ +static int __init summarize_uvhub_sockets(int nuvhubs, + struct uvhub_desc *uvhub_descs, + unsigned char *uvhub_mask) +{ + int socket; + int uvhub; + unsigned short socket_mask; + for (uvhub = 0; uvhub < nuvhubs; uvhub++) { + struct uvhub_desc *bdp; + struct bau_control *smaster = NULL; + struct bau_control *hmaster = NULL; + if (!(*(uvhub_mask + (uvhub/8)) & (1 << (uvhub%8)))) continue; - have_hmaster = 0; + bdp = &uvhub_descs[uvhub]; socket_mask = bdp->socket_mask; socket = 0; while (socket_mask) { - if (!(socket_mask & 1)) - goto nextsocket; - sdp = &bdp->socket[socket]; - for (i = 0; i < sdp->num_cpus; i++) { - cpu = sdp->cpu_number[i]; - bcp = &per_cpu(bau_control, cpu); - bcp->cpu = cpu; - if (i == 0) { - smaster = bcp; - if (!have_hmaster) { - have_hmaster++; - hmaster = bcp; - } - } - bcp->cpus_in_uvhub = bdp->num_cpus; - bcp->cpus_in_socket = sdp->num_cpus; - bcp->socket_master = smaster; - bcp->uvhub = bdp->uvhub; - bcp->uvhub_master = hmaster; - bcp->uvhub_cpu = uv_cpu_hub_info(cpu)-> - blade_processor_id; + struct socket_desc *sdp; + if ((socket_mask & 1)) { + sdp = &bdp->socket[socket]; + if (scan_sock(sdp, bdp, &smaster, &hmaster)) + return 1; + make_per_cpu_thp(smaster); } -nextsocket: socket++; socket_mask = (socket_mask >> 1); } + make_per_hub_cpumask(hmaster); } + return 0; +} + +/* + * initialize the bau_control structure for each cpu + */ +static int __init init_per_cpu(int nuvhubs, int base_part_pnode) +{ + unsigned char *uvhub_mask; + void *vp; + struct uvhub_desc *uvhub_descs; + + timeout_us = calculate_destination_timeout(); + + vp = kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL); + uvhub_descs = (struct uvhub_desc *)vp; + memset(uvhub_descs, 0, nuvhubs * sizeof(struct uvhub_desc)); + uvhub_mask = kzalloc((nuvhubs+7)/8, GFP_KERNEL); + + if (get_cpu_topology(base_part_pnode, uvhub_descs, uvhub_mask)) + goto fail; + + if (summarize_uvhub_sockets(nuvhubs, uvhub_descs, uvhub_mask)) + goto fail; + kfree(uvhub_descs); kfree(uvhub_mask); - for_each_present_cpu(cpu) { - bcp = &per_cpu(bau_control, cpu); - bcp->baudisabled = 0; - bcp->statp = &per_cpu(ptcstats, cpu); - /* time interval to catch a hardware stay-busy bug */ - bcp->timeout_interval = microsec_2_cycles(2*timeout_us); - bcp->max_bau_concurrent = max_bau_concurrent; - bcp->max_bau_concurrent_constant = max_bau_concurrent; - bcp->plugged_delay = plugged_delay; - bcp->plugsb4reset = plugsb4reset; - bcp->timeoutsb4reset = timeoutsb4reset; - bcp->ipi_reset_limit = ipi_reset_limit; - bcp->complete_threshold = complete_threshold; - bcp->congested_response_us = congested_response_us; - bcp->congested_reps = congested_reps; - bcp->congested_period = congested_period; - } + init_per_cpu_tunables(); + return 0; + +fail: + kfree(uvhub_descs); + kfree(uvhub_mask); + return 1; } /* @@ -1606,51 +2107,54 @@ static int __init uv_bau_init(void) int pnode; int nuvhubs; int cur_cpu; + int cpus; int vector; - unsigned long mmr; + cpumask_var_t *mask; if (!is_uv_system()) return 0; - if (nobau) - return 0; - - for_each_possible_cpu(cur_cpu) - zalloc_cpumask_var_node(&per_cpu(uv_flush_tlb_mask, cur_cpu), - GFP_KERNEL, cpu_to_node(cur_cpu)); + for_each_possible_cpu(cur_cpu) { + mask = &per_cpu(uv_flush_tlb_mask, cur_cpu); + zalloc_cpumask_var_node(mask, GFP_KERNEL, cpu_to_node(cur_cpu)); + } - uv_nshift = uv_hub_info->m_val; - uv_mmask = (1UL << uv_hub_info->m_val) - 1; nuvhubs = uv_num_possible_blades(); - spin_lock_init(&disable_lock); - congested_cycles = microsec_2_cycles(congested_response_us); + congested_cycles = usec_2_cycles(congested_respns_us); - uv_init_per_cpu(nuvhubs); + uv_base_pnode = 0x7fffffff; + for (uvhub = 0; uvhub < nuvhubs; uvhub++) { + cpus = uv_blade_nr_possible_cpus(uvhub); + if (cpus && (uv_blade_to_pnode(uvhub) < uv_base_pnode)) + uv_base_pnode = uv_blade_to_pnode(uvhub); + } + + enable_timeouts(); - uv_partition_base_pnode = 0x7fffffff; - for (uvhub = 0; uvhub < nuvhubs; uvhub++) - if (uv_blade_nr_possible_cpus(uvhub) && - (uv_blade_to_pnode(uvhub) < uv_partition_base_pnode)) - uv_partition_base_pnode = uv_blade_to_pnode(uvhub); + if (init_per_cpu(nuvhubs, uv_base_pnode)) { + set_bau_off(); + nobau_perm = 1; + return 0; + } vector = UV_BAU_MESSAGE; for_each_possible_blade(uvhub) if (uv_blade_nr_possible_cpus(uvhub)) - uv_init_uvhub(uvhub, vector); + init_uvhub(uvhub, vector, uv_base_pnode); - uv_enable_timeouts(); alloc_intr_gate(vector, uv_bau_message_intr1); for_each_possible_blade(uvhub) { if (uv_blade_nr_possible_cpus(uvhub)) { + unsigned long val; + unsigned long mmr; pnode = uv_blade_to_pnode(uvhub); /* INIT the bau */ - uv_write_global_mmr64(pnode, - UVH_LB_BAU_SB_ACTIVATION_CONTROL, - ((unsigned long)1 << 63)); + val = 1L << 63; + write_gmmr_activation(pnode, val); mmr = 1; /* should be 1 to broadcast to both sockets */ - uv_write_global_mmr64(pnode, UVH_BAU_DATA_BROADCAST, - mmr); + if (!is_uv1_hub()) + write_mmr_data_broadcast(pnode, mmr); } } diff --git a/arch/x86/platform/uv/uv_irq.c b/arch/x86/platform/uv/uv_irq.c index 7b24460917d..b233681af4d 100644 --- a/arch/x86/platform/uv/uv_irq.c +++ b/arch/x86/platform/uv/uv_irq.c @@ -25,7 +25,7 @@ struct uv_irq_2_mmr_pnode{ int irq; }; -static spinlock_t uv_irq_lock; +static DEFINE_SPINLOCK(uv_irq_lock); static struct rb_root uv_irq_root; static int uv_set_irq_affinity(struct irq_data *, const struct cpumask *, bool); @@ -131,10 +131,11 @@ arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade, unsigned long mmr_offset, int limit) { const struct cpumask *eligible_cpu = cpumask_of(cpu); - struct irq_cfg *cfg = get_irq_chip_data(irq); + struct irq_cfg *cfg = irq_get_chip_data(irq); unsigned long mmr_value; struct uv_IO_APIC_route_entry *entry; int mmr_pnode, err; + unsigned int dest; BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long)); @@ -143,12 +144,16 @@ arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade, if (err != 0) return err; + err = apic->cpu_mask_to_apicid_and(eligible_cpu, eligible_cpu, &dest); + if (err != 0) + return err; + if (limit == UV_AFFINITY_CPU) irq_set_status_flags(irq, IRQ_NO_BALANCING); else irq_set_status_flags(irq, IRQ_MOVE_PCNTXT); - set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq, + irq_set_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq, irq_name); mmr_value = 0; @@ -159,7 +164,7 @@ arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade, entry->polarity = 0; entry->trigger = 0; entry->mask = 0; - entry->dest = apic->cpu_mask_to_apicid(eligible_cpu); + entry->dest = dest; mmr_pnode = uv_blade_to_pnode(mmr_blade); uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value); @@ -222,7 +227,7 @@ uv_set_irq_affinity(struct irq_data *data, const struct cpumask *mask, if (cfg->move_in_progress) send_cleanup_vector(cfg); - return 0; + return IRQ_SET_MASK_OK_NOCOPY; } /* @@ -233,11 +238,9 @@ uv_set_irq_affinity(struct irq_data *data, const struct cpumask *mask, int uv_setup_irq(char *irq_name, int cpu, int mmr_blade, unsigned long mmr_offset, int limit) { - int irq, ret; - - irq = create_irq_nr(NR_IRQS_LEGACY, uv_blade_to_memory_nid(mmr_blade)); + int ret, irq = irq_alloc_hwirq(uv_blade_to_memory_nid(mmr_blade)); - if (irq <= 0) + if (!irq) return -EBUSY; ret = arch_enable_uv_irq(irq_name, irq, cpu, mmr_blade, mmr_offset, @@ -245,7 +248,7 @@ int uv_setup_irq(char *irq_name, int cpu, int mmr_blade, if (ret == irq) uv_set_irq_2_mmr_info(irq, mmr_offset, mmr_blade); else - destroy_irq(irq); + irq_free_hwirq(irq); return ret; } @@ -280,6 +283,6 @@ void uv_teardown_irq(unsigned int irq) n = n->rb_right; } spin_unlock_irqrestore(&uv_irq_lock, irqflags); - destroy_irq(irq); + irq_free_hwirq(irq); } EXPORT_SYMBOL_GPL(uv_teardown_irq); diff --git a/arch/x86/platform/uv/uv_nmi.c b/arch/x86/platform/uv/uv_nmi.c new file mode 100644 index 00000000000..c89c93320c1 --- /dev/null +++ b/arch/x86/platform/uv/uv_nmi.c @@ -0,0 +1,727 @@ +/* + * SGI NMI support routines + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + * Copyright (c) 2009-2013 Silicon Graphics, Inc. All Rights Reserved. + * Copyright (c) Mike Travis + */ + +#include <linux/cpu.h> +#include <linux/delay.h> +#include <linux/kdb.h> +#include <linux/kexec.h> +#include <linux/kgdb.h> +#include <linux/module.h> +#include <linux/nmi.h> +#include <linux/sched.h> +#include <linux/slab.h> + +#include <asm/apic.h> +#include <asm/current.h> +#include <asm/kdebug.h> +#include <asm/local64.h> +#include <asm/nmi.h> +#include <asm/traps.h> +#include <asm/uv/uv.h> +#include <asm/uv/uv_hub.h> +#include <asm/uv/uv_mmrs.h> + +/* + * UV handler for NMI + * + * Handle system-wide NMI events generated by the global 'power nmi' command. + * + * Basic operation is to field the NMI interrupt on each cpu and wait + * until all cpus have arrived into the nmi handler. If some cpus do not + * make it into the handler, try and force them in with the IPI(NMI) signal. + * + * We also have to lessen UV Hub MMR accesses as much as possible as this + * disrupts the UV Hub's primary mission of directing NumaLink traffic and + * can cause system problems to occur. + * + * To do this we register our primary NMI notifier on the NMI_UNKNOWN + * chain. This reduces the number of false NMI calls when the perf + * tools are running which generate an enormous number of NMIs per + * second (~4M/s for 1024 cpu threads). Our secondary NMI handler is + * very short as it only checks that if it has been "pinged" with the + * IPI(NMI) signal as mentioned above, and does not read the UV Hub's MMR. + * + */ + +static struct uv_hub_nmi_s **uv_hub_nmi_list; + +DEFINE_PER_CPU(struct uv_cpu_nmi_s, __uv_cpu_nmi); +EXPORT_PER_CPU_SYMBOL_GPL(__uv_cpu_nmi); + +static unsigned long nmi_mmr; +static unsigned long nmi_mmr_clear; +static unsigned long nmi_mmr_pending; + +static atomic_t uv_in_nmi; +static atomic_t uv_nmi_cpu = ATOMIC_INIT(-1); +static atomic_t uv_nmi_cpus_in_nmi = ATOMIC_INIT(-1); +static atomic_t uv_nmi_slave_continue; +static cpumask_var_t uv_nmi_cpu_mask; + +/* Values for uv_nmi_slave_continue */ +#define SLAVE_CLEAR 0 +#define SLAVE_CONTINUE 1 +#define SLAVE_EXIT 2 + +/* + * Default is all stack dumps go to the console and buffer. + * Lower level to send to log buffer only. + */ +static int uv_nmi_loglevel = CONSOLE_LOGLEVEL_DEFAULT; +module_param_named(dump_loglevel, uv_nmi_loglevel, int, 0644); + +/* + * The following values show statistics on how perf events are affecting + * this system. + */ +static int param_get_local64(char *buffer, const struct kernel_param *kp) +{ + return sprintf(buffer, "%lu\n", local64_read((local64_t *)kp->arg)); +} + +static int param_set_local64(const char *val, const struct kernel_param *kp) +{ + /* clear on any write */ + local64_set((local64_t *)kp->arg, 0); + return 0; +} + +static struct kernel_param_ops param_ops_local64 = { + .get = param_get_local64, + .set = param_set_local64, +}; +#define param_check_local64(name, p) __param_check(name, p, local64_t) + +static local64_t uv_nmi_count; +module_param_named(nmi_count, uv_nmi_count, local64, 0644); + +static local64_t uv_nmi_misses; +module_param_named(nmi_misses, uv_nmi_misses, local64, 0644); + +static local64_t uv_nmi_ping_count; +module_param_named(ping_count, uv_nmi_ping_count, local64, 0644); + +static local64_t uv_nmi_ping_misses; +module_param_named(ping_misses, uv_nmi_ping_misses, local64, 0644); + +/* + * Following values allow tuning for large systems under heavy loading + */ +static int uv_nmi_initial_delay = 100; +module_param_named(initial_delay, uv_nmi_initial_delay, int, 0644); + +static int uv_nmi_slave_delay = 100; +module_param_named(slave_delay, uv_nmi_slave_delay, int, 0644); + +static int uv_nmi_loop_delay = 100; +module_param_named(loop_delay, uv_nmi_loop_delay, int, 0644); + +static int uv_nmi_trigger_delay = 10000; +module_param_named(trigger_delay, uv_nmi_trigger_delay, int, 0644); + +static int uv_nmi_wait_count = 100; +module_param_named(wait_count, uv_nmi_wait_count, int, 0644); + +static int uv_nmi_retry_count = 500; +module_param_named(retry_count, uv_nmi_retry_count, int, 0644); + +/* + * Valid NMI Actions: + * "dump" - dump process stack for each cpu + * "ips" - dump IP info for each cpu + * "kdump" - do crash dump + * "kdb" - enter KDB (default) + * "kgdb" - enter KGDB + */ +static char uv_nmi_action[8] = "kdb"; +module_param_string(action, uv_nmi_action, sizeof(uv_nmi_action), 0644); + +static inline bool uv_nmi_action_is(const char *action) +{ + return (strncmp(uv_nmi_action, action, strlen(action)) == 0); +} + +/* Setup which NMI support is present in system */ +static void uv_nmi_setup_mmrs(void) +{ + if (uv_read_local_mmr(UVH_NMI_MMRX_SUPPORTED)) { + uv_write_local_mmr(UVH_NMI_MMRX_REQ, + 1UL << UVH_NMI_MMRX_REQ_SHIFT); + nmi_mmr = UVH_NMI_MMRX; + nmi_mmr_clear = UVH_NMI_MMRX_CLEAR; + nmi_mmr_pending = 1UL << UVH_NMI_MMRX_SHIFT; + pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMRX_TYPE); + } else { + nmi_mmr = UVH_NMI_MMR; + nmi_mmr_clear = UVH_NMI_MMR_CLEAR; + nmi_mmr_pending = 1UL << UVH_NMI_MMR_SHIFT; + pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMR_TYPE); + } +} + +/* Read NMI MMR and check if NMI flag was set by BMC. */ +static inline int uv_nmi_test_mmr(struct uv_hub_nmi_s *hub_nmi) +{ + hub_nmi->nmi_value = uv_read_local_mmr(nmi_mmr); + atomic_inc(&hub_nmi->read_mmr_count); + return !!(hub_nmi->nmi_value & nmi_mmr_pending); +} + +static inline void uv_local_mmr_clear_nmi(void) +{ + uv_write_local_mmr(nmi_mmr_clear, nmi_mmr_pending); +} + +/* + * If first cpu in on this hub, set hub_nmi "in_nmi" and "owner" values and + * return true. If first cpu in on the system, set global "in_nmi" flag. + */ +static int uv_set_in_nmi(int cpu, struct uv_hub_nmi_s *hub_nmi) +{ + int first = atomic_add_unless(&hub_nmi->in_nmi, 1, 1); + + if (first) { + atomic_set(&hub_nmi->cpu_owner, cpu); + if (atomic_add_unless(&uv_in_nmi, 1, 1)) + atomic_set(&uv_nmi_cpu, cpu); + + atomic_inc(&hub_nmi->nmi_count); + } + return first; +} + +/* Check if this is a system NMI event */ +static int uv_check_nmi(struct uv_hub_nmi_s *hub_nmi) +{ + int cpu = smp_processor_id(); + int nmi = 0; + + local64_inc(&uv_nmi_count); + uv_cpu_nmi.queries++; + + do { + nmi = atomic_read(&hub_nmi->in_nmi); + if (nmi) + break; + + if (raw_spin_trylock(&hub_nmi->nmi_lock)) { + + /* check hub MMR NMI flag */ + if (uv_nmi_test_mmr(hub_nmi)) { + uv_set_in_nmi(cpu, hub_nmi); + nmi = 1; + break; + } + + /* MMR NMI flag is clear */ + raw_spin_unlock(&hub_nmi->nmi_lock); + + } else { + /* wait a moment for the hub nmi locker to set flag */ + cpu_relax(); + udelay(uv_nmi_slave_delay); + + /* re-check hub in_nmi flag */ + nmi = atomic_read(&hub_nmi->in_nmi); + if (nmi) + break; + } + + /* check if this BMC missed setting the MMR NMI flag */ + if (!nmi) { + nmi = atomic_read(&uv_in_nmi); + if (nmi) + uv_set_in_nmi(cpu, hub_nmi); + } + + } while (0); + + if (!nmi) + local64_inc(&uv_nmi_misses); + + return nmi; +} + +/* Need to reset the NMI MMR register, but only once per hub. */ +static inline void uv_clear_nmi(int cpu) +{ + struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi; + + if (cpu == atomic_read(&hub_nmi->cpu_owner)) { + atomic_set(&hub_nmi->cpu_owner, -1); + atomic_set(&hub_nmi->in_nmi, 0); + uv_local_mmr_clear_nmi(); + raw_spin_unlock(&hub_nmi->nmi_lock); + } +} + +/* Print non-responding cpus */ +static void uv_nmi_nr_cpus_pr(char *fmt) +{ + static char cpu_list[1024]; + int len = sizeof(cpu_list); + int c = cpumask_weight(uv_nmi_cpu_mask); + int n = cpulist_scnprintf(cpu_list, len, uv_nmi_cpu_mask); + + if (n >= len-1) + strcpy(&cpu_list[len - 6], "...\n"); + + printk(fmt, c, cpu_list); +} + +/* Ping non-responding cpus attemping to force them into the NMI handler */ +static void uv_nmi_nr_cpus_ping(void) +{ + int cpu; + + for_each_cpu(cpu, uv_nmi_cpu_mask) + atomic_set(&uv_cpu_nmi_per(cpu).pinging, 1); + + apic->send_IPI_mask(uv_nmi_cpu_mask, APIC_DM_NMI); +} + +/* Clean up flags for cpus that ignored both NMI and ping */ +static void uv_nmi_cleanup_mask(void) +{ + int cpu; + + for_each_cpu(cpu, uv_nmi_cpu_mask) { + atomic_set(&uv_cpu_nmi_per(cpu).pinging, 0); + atomic_set(&uv_cpu_nmi_per(cpu).state, UV_NMI_STATE_OUT); + cpumask_clear_cpu(cpu, uv_nmi_cpu_mask); + } +} + +/* Loop waiting as cpus enter nmi handler */ +static int uv_nmi_wait_cpus(int first) +{ + int i, j, k, n = num_online_cpus(); + int last_k = 0, waiting = 0; + + if (first) { + cpumask_copy(uv_nmi_cpu_mask, cpu_online_mask); + k = 0; + } else { + k = n - cpumask_weight(uv_nmi_cpu_mask); + } + + udelay(uv_nmi_initial_delay); + for (i = 0; i < uv_nmi_retry_count; i++) { + int loop_delay = uv_nmi_loop_delay; + + for_each_cpu(j, uv_nmi_cpu_mask) { + if (atomic_read(&uv_cpu_nmi_per(j).state)) { + cpumask_clear_cpu(j, uv_nmi_cpu_mask); + if (++k >= n) + break; + } + } + if (k >= n) { /* all in? */ + k = n; + break; + } + if (last_k != k) { /* abort if no new cpus coming in */ + last_k = k; + waiting = 0; + } else if (++waiting > uv_nmi_wait_count) + break; + + /* extend delay if waiting only for cpu 0 */ + if (waiting && (n - k) == 1 && + cpumask_test_cpu(0, uv_nmi_cpu_mask)) + loop_delay *= 100; + + udelay(loop_delay); + } + atomic_set(&uv_nmi_cpus_in_nmi, k); + return n - k; +} + +/* Wait until all slave cpus have entered UV NMI handler */ +static void uv_nmi_wait(int master) +{ + /* indicate this cpu is in */ + atomic_set(&uv_cpu_nmi.state, UV_NMI_STATE_IN); + + /* if not the first cpu in (the master), then we are a slave cpu */ + if (!master) + return; + + do { + /* wait for all other cpus to gather here */ + if (!uv_nmi_wait_cpus(1)) + break; + + /* if not all made it in, send IPI NMI to them */ + uv_nmi_nr_cpus_pr(KERN_ALERT + "UV: Sending NMI IPI to %d non-responding CPUs: %s\n"); + uv_nmi_nr_cpus_ping(); + + /* if all cpus are in, then done */ + if (!uv_nmi_wait_cpus(0)) + break; + + uv_nmi_nr_cpus_pr(KERN_ALERT + "UV: %d CPUs not in NMI loop: %s\n"); + } while (0); + + pr_alert("UV: %d of %d CPUs in NMI\n", + atomic_read(&uv_nmi_cpus_in_nmi), num_online_cpus()); +} + +static void uv_nmi_dump_cpu_ip_hdr(void) +{ + printk(KERN_DEFAULT + "\nUV: %4s %6s %-32s %s (Note: PID 0 not listed)\n", + "CPU", "PID", "COMMAND", "IP"); +} + +static void uv_nmi_dump_cpu_ip(int cpu, struct pt_regs *regs) +{ + printk(KERN_DEFAULT "UV: %4d %6d %-32.32s ", + cpu, current->pid, current->comm); + + printk_address(regs->ip); +} + +/* Dump this cpu's state */ +static void uv_nmi_dump_state_cpu(int cpu, struct pt_regs *regs) +{ + const char *dots = " ................................. "; + + if (uv_nmi_action_is("ips")) { + if (cpu == 0) + uv_nmi_dump_cpu_ip_hdr(); + + if (current->pid != 0) + uv_nmi_dump_cpu_ip(cpu, regs); + + } else if (uv_nmi_action_is("dump")) { + printk(KERN_DEFAULT + "UV:%sNMI process trace for CPU %d\n", dots, cpu); + show_regs(regs); + } + atomic_set(&uv_cpu_nmi.state, UV_NMI_STATE_DUMP_DONE); +} + +/* Trigger a slave cpu to dump it's state */ +static void uv_nmi_trigger_dump(int cpu) +{ + int retry = uv_nmi_trigger_delay; + + if (atomic_read(&uv_cpu_nmi_per(cpu).state) != UV_NMI_STATE_IN) + return; + + atomic_set(&uv_cpu_nmi_per(cpu).state, UV_NMI_STATE_DUMP); + do { + cpu_relax(); + udelay(10); + if (atomic_read(&uv_cpu_nmi_per(cpu).state) + != UV_NMI_STATE_DUMP) + return; + } while (--retry > 0); + + pr_crit("UV: CPU %d stuck in process dump function\n", cpu); + atomic_set(&uv_cpu_nmi_per(cpu).state, UV_NMI_STATE_DUMP_DONE); +} + +/* Wait until all cpus ready to exit */ +static void uv_nmi_sync_exit(int master) +{ + atomic_dec(&uv_nmi_cpus_in_nmi); + if (master) { + while (atomic_read(&uv_nmi_cpus_in_nmi) > 0) + cpu_relax(); + atomic_set(&uv_nmi_slave_continue, SLAVE_CLEAR); + } else { + while (atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + } +} + +/* Walk through cpu list and dump state of each */ +static void uv_nmi_dump_state(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int tcpu; + int ignored = 0; + int saved_console_loglevel = console_loglevel; + + pr_alert("UV: tracing %s for %d CPUs from CPU %d\n", + uv_nmi_action_is("ips") ? "IPs" : "processes", + atomic_read(&uv_nmi_cpus_in_nmi), cpu); + + console_loglevel = uv_nmi_loglevel; + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + for_each_online_cpu(tcpu) { + if (cpumask_test_cpu(tcpu, uv_nmi_cpu_mask)) + ignored++; + else if (tcpu == cpu) + uv_nmi_dump_state_cpu(tcpu, regs); + else + uv_nmi_trigger_dump(tcpu); + } + if (ignored) + printk(KERN_DEFAULT "UV: %d CPUs ignored NMI\n", + ignored); + + console_loglevel = saved_console_loglevel; + pr_alert("UV: process trace complete\n"); + } else { + while (!atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + while (atomic_read(&uv_cpu_nmi.state) != UV_NMI_STATE_DUMP) + cpu_relax(); + uv_nmi_dump_state_cpu(cpu, regs); + } + uv_nmi_sync_exit(master); +} + +static void uv_nmi_touch_watchdogs(void) +{ + touch_softlockup_watchdog_sync(); + clocksource_touch_watchdog(); + rcu_cpu_stall_reset(); + touch_nmi_watchdog(); +} + +#if defined(CONFIG_KEXEC) +static atomic_t uv_nmi_kexec_failed; +static void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs) +{ + /* Call crash to dump system state */ + if (master) { + pr_emerg("UV: NMI executing crash_kexec on CPU%d\n", cpu); + crash_kexec(regs); + + pr_emerg("UV: crash_kexec unexpectedly returned, "); + if (!kexec_crash_image) { + pr_cont("crash kernel not loaded\n"); + atomic_set(&uv_nmi_kexec_failed, 1); + uv_nmi_sync_exit(1); + return; + } + pr_cont("kexec busy, stalling cpus while waiting\n"); + } + + /* If crash exec fails the slaves should return, otherwise stall */ + while (atomic_read(&uv_nmi_kexec_failed) == 0) + mdelay(10); + + /* Crash kernel most likely not loaded, return in an orderly fashion */ + uv_nmi_sync_exit(0); +} + +#else /* !CONFIG_KEXEC */ +static inline void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs) +{ + if (master) + pr_err("UV: NMI kdump: KEXEC not supported in this kernel\n"); +} +#endif /* !CONFIG_KEXEC */ + +#ifdef CONFIG_KGDB +#ifdef CONFIG_KGDB_KDB +static inline int uv_nmi_kdb_reason(void) +{ + return KDB_REASON_SYSTEM_NMI; +} +#else /* !CONFIG_KGDB_KDB */ +static inline int uv_nmi_kdb_reason(void) +{ + /* Insure user is expecting to attach gdb remote */ + if (uv_nmi_action_is("kgdb")) + return 0; + + pr_err("UV: NMI error: KDB is not enabled in this kernel\n"); + return -1; +} +#endif /* CONFIG_KGDB_KDB */ + +/* + * Call KGDB/KDB from NMI handler + * + * Note that if both KGDB and KDB are configured, then the action of 'kgdb' or + * 'kdb' has no affect on which is used. See the KGDB documention for further + * information. + */ +static void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int reason = uv_nmi_kdb_reason(); + int ret; + + if (reason < 0) + return; + + /* call KGDB NMI handler as MASTER */ + ret = kgdb_nmicallin(cpu, X86_TRAP_NMI, regs, reason, + &uv_nmi_slave_continue); + if (ret) { + pr_alert("KGDB returned error, is kgdboc set?\n"); + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + } + } else { + /* wait for KGDB signal that it's ready for slaves to enter */ + int sig; + + do { + cpu_relax(); + sig = atomic_read(&uv_nmi_slave_continue); + } while (!sig); + + /* call KGDB as slave */ + if (sig == SLAVE_CONTINUE) + kgdb_nmicallback(cpu, regs); + } + uv_nmi_sync_exit(master); +} + +#else /* !CONFIG_KGDB */ +static inline void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master) +{ + pr_err("UV: NMI error: KGDB is not enabled in this kernel\n"); +} +#endif /* !CONFIG_KGDB */ + +/* + * UV NMI handler + */ +int uv_handle_nmi(unsigned int reason, struct pt_regs *regs) +{ + struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi; + int cpu = smp_processor_id(); + int master = 0; + unsigned long flags; + + local_irq_save(flags); + + /* If not a UV System NMI, ignore */ + if (!atomic_read(&uv_cpu_nmi.pinging) && !uv_check_nmi(hub_nmi)) { + local_irq_restore(flags); + return NMI_DONE; + } + + /* Indicate we are the first CPU into the NMI handler */ + master = (atomic_read(&uv_nmi_cpu) == cpu); + + /* If NMI action is "kdump", then attempt to do it */ + if (uv_nmi_action_is("kdump")) + uv_nmi_kdump(cpu, master, regs); + + /* Pause as all cpus enter the NMI handler */ + uv_nmi_wait(master); + + /* Dump state of each cpu */ + if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump")) + uv_nmi_dump_state(cpu, regs, master); + + /* Call KGDB/KDB if enabled */ + else if (uv_nmi_action_is("kdb") || uv_nmi_action_is("kgdb")) + uv_call_kgdb_kdb(cpu, regs, master); + + /* Clear per_cpu "in nmi" flag */ + atomic_set(&uv_cpu_nmi.state, UV_NMI_STATE_OUT); + + /* Clear MMR NMI flag on each hub */ + uv_clear_nmi(cpu); + + /* Clear global flags */ + if (master) { + if (cpumask_weight(uv_nmi_cpu_mask)) + uv_nmi_cleanup_mask(); + atomic_set(&uv_nmi_cpus_in_nmi, -1); + atomic_set(&uv_nmi_cpu, -1); + atomic_set(&uv_in_nmi, 0); + } + + uv_nmi_touch_watchdogs(); + local_irq_restore(flags); + + return NMI_HANDLED; +} + +/* + * NMI handler for pulling in CPUs when perf events are grabbing our NMI + */ +static int uv_handle_nmi_ping(unsigned int reason, struct pt_regs *regs) +{ + int ret; + + uv_cpu_nmi.queries++; + if (!atomic_read(&uv_cpu_nmi.pinging)) { + local64_inc(&uv_nmi_ping_misses); + return NMI_DONE; + } + + uv_cpu_nmi.pings++; + local64_inc(&uv_nmi_ping_count); + ret = uv_handle_nmi(reason, regs); + atomic_set(&uv_cpu_nmi.pinging, 0); + return ret; +} + +static void uv_register_nmi_notifier(void) +{ + if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv")) + pr_warn("UV: NMI handler failed to register\n"); + + if (register_nmi_handler(NMI_LOCAL, uv_handle_nmi_ping, 0, "uvping")) + pr_warn("UV: PING NMI handler failed to register\n"); +} + +void uv_nmi_init(void) +{ + unsigned int value; + + /* + * Unmask NMI on all cpus + */ + value = apic_read(APIC_LVT1) | APIC_DM_NMI; + value &= ~APIC_LVT_MASKED; + apic_write(APIC_LVT1, value); +} + +void uv_nmi_setup(void) +{ + int size = sizeof(void *) * (1 << NODES_SHIFT); + int cpu, nid; + + /* Setup hub nmi info */ + uv_nmi_setup_mmrs(); + uv_hub_nmi_list = kzalloc(size, GFP_KERNEL); + pr_info("UV: NMI hub list @ 0x%p (%d)\n", uv_hub_nmi_list, size); + BUG_ON(!uv_hub_nmi_list); + size = sizeof(struct uv_hub_nmi_s); + for_each_present_cpu(cpu) { + nid = cpu_to_node(cpu); + if (uv_hub_nmi_list[nid] == NULL) { + uv_hub_nmi_list[nid] = kzalloc_node(size, + GFP_KERNEL, nid); + BUG_ON(!uv_hub_nmi_list[nid]); + raw_spin_lock_init(&(uv_hub_nmi_list[nid]->nmi_lock)); + atomic_set(&uv_hub_nmi_list[nid]->cpu_owner, -1); + } + uv_hub_nmi_per(cpu) = uv_hub_nmi_list[nid]; + } + BUG_ON(!alloc_cpumask_var(&uv_nmi_cpu_mask, GFP_KERNEL)); + uv_register_nmi_notifier(); +} diff --git a/arch/x86/platform/uv/uv_sysfs.c b/arch/x86/platform/uv/uv_sysfs.c index 309c70fb775..5d4ba301e77 100644 --- a/arch/x86/platform/uv/uv_sysfs.c +++ b/arch/x86/platform/uv/uv_sysfs.c @@ -19,7 +19,7 @@ * Copyright (c) Russ Anderson */ -#include <linux/sysdev.h> +#include <linux/device.h> #include <asm/uv/bios.h> #include <asm/uv/uv.h> diff --git a/arch/x86/platform/uv/uv_time.c b/arch/x86/platform/uv/uv_time.c index 56e421bc379..5c86786bbfd 100644 --- a/arch/x86/platform/uv/uv_time.c +++ b/arch/x86/platform/uv/uv_time.c @@ -15,7 +15,7 @@ * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * Copyright (c) 2009 Silicon Graphics, Inc. All Rights Reserved. + * Copyright (c) 2009-2013 Silicon Graphics, Inc. All Rights Reserved. * Copyright (c) Dimitri Sivanich */ #include <linux/clockchips.h> @@ -37,10 +37,9 @@ static void uv_rtc_timer_setup(enum clock_event_mode, static struct clocksource clocksource_uv = { .name = RTC_NAME, - .rating = 400, + .rating = 299, .read = uv_read_rtc, .mask = (cycle_t)UVH_RTC_REAL_TIME_CLOCK_MASK, - .shift = 10, .flags = CLOCK_SOURCE_IS_CONTINUOUS, }; @@ -89,6 +88,7 @@ static void uv_rtc_send_IPI(int cpu) apicid = cpu_physical_id(cpu); pnode = uv_apicid_to_pnode(apicid); + apicid |= uv_apicid_hibits; val = (1UL << UVH_IPI_INT_SEND_SHFT) | (apicid << UVH_IPI_INT_APIC_ID_SHFT) | (X86_PLATFORM_IPI_VECTOR << UVH_IPI_INT_VECTOR_SHFT); @@ -99,25 +99,35 @@ static void uv_rtc_send_IPI(int cpu) /* Check for an RTC interrupt pending */ static int uv_intr_pending(int pnode) { - return uv_read_global_mmr64(pnode, UVH_EVENT_OCCURRED0) & - UVH_EVENT_OCCURRED0_RTC1_MASK; + if (is_uv1_hub()) + return uv_read_global_mmr64(pnode, UVH_EVENT_OCCURRED0) & + UV1H_EVENT_OCCURRED0_RTC1_MASK; + else if (is_uvx_hub()) + return uv_read_global_mmr64(pnode, UVXH_EVENT_OCCURRED2) & + UVXH_EVENT_OCCURRED2_RTC_1_MASK; + return 0; } /* Setup interrupt and return non-zero if early expiration occurred. */ static int uv_setup_intr(int cpu, u64 expires) { u64 val; + unsigned long apicid = cpu_physical_id(cpu) | uv_apicid_hibits; int pnode = uv_cpu_to_pnode(cpu); uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, UVH_RTC1_INT_CONFIG_M_MASK); uv_write_global_mmr64(pnode, UVH_INT_CMPB, -1L); - uv_write_global_mmr64(pnode, UVH_EVENT_OCCURRED0_ALIAS, - UVH_EVENT_OCCURRED0_RTC1_MASK); + if (is_uv1_hub()) + uv_write_global_mmr64(pnode, UVH_EVENT_OCCURRED0_ALIAS, + UV1H_EVENT_OCCURRED0_RTC1_MASK); + else + uv_write_global_mmr64(pnode, UVXH_EVENT_OCCURRED2_ALIAS, + UVXH_EVENT_OCCURRED2_RTC_1_MASK); val = (X86_PLATFORM_IPI_VECTOR << UVH_RTC1_INT_CONFIG_VECTOR_SHFT) | - ((u64)cpu_physical_id(cpu) << UVH_RTC1_INT_CONFIG_APIC_ID_SHFT); + ((u64)apicid << UVH_RTC1_INT_CONFIG_APIC_ID_SHFT); /* Set configuration */ uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, val); @@ -149,10 +159,9 @@ static __init int uv_rtc_allocate_timers(void) { int cpu; - blade_info = kmalloc(uv_possible_blades * sizeof(void *), GFP_KERNEL); + blade_info = kzalloc(uv_possible_blades * sizeof(void *), GFP_KERNEL); if (!blade_info) return -ENOMEM; - memset(blade_info, 0, uv_possible_blades * sizeof(void *)); for_each_present_cpu(cpu) { int nid = cpu_to_node(cpu); @@ -370,14 +379,7 @@ static __init int uv_rtc_setup_clock(void) if (!is_uv_system()) return -ENODEV; - clocksource_uv.mult = clocksource_hz2mult(sn_rtc_cycles_per_second, - clocksource_uv.shift); - - /* If single blade, prefer tsc */ - if (uv_num_possible_blades() == 1) - clocksource_uv.rating = 250; - - rc = clocksource_register(&clocksource_uv); + rc = clocksource_register_hz(&clocksource_uv, sn_rtc_cycles_per_second); if (rc) printk(KERN_INFO "UV RTC clocksource failed rc %d\n", rc); else |