Skip to content

L
linux
Commit 36450e9c authored by Linus Torvalds's avatar Linus Torvalds
Browse files
Options

Merge branch 'x86-uv-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip 

* 'x86-uv-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86, UV: Initialize BAU MMRs only on hubs with cpus
  x86, UV: Modularize BAU send and wait
  x86, UV: BAU broadcast to the local hub
  x86, UV: Correct BAU regular message type
  x86, UV: Remove BAU check for stay-busy
  x86, UV: Correct BAU discovery of hubs and sockets
  x86, UV: Correct BAU software acknowledge
  x86, UV: BAU structure rearranging
  x86, UV: Shorten access to BAU statistics structure
  x86, UV: Disable BAU on network congestion
  x86, UV: BAU tunables into a debugfs file
  x86, UV: Calculate BAU destination timeout
parents 2f2c7795 93a7ca0c
Hide whitespace changes
Inline Side-by-side
Showing with 608 additions and 299 deletions
arch/x86/include/asm/uv/uv_bau.h
View file @ 36450e9c
...@@ -34,6 +34,7 @@ ...@@ -34,6 +34,7 @@
*/ */
#define UV_ITEMS_PER_DESCRIPTOR 8 #define UV_ITEMS_PER_DESCRIPTOR 8
/* the 'throttle' to prevent the hardware stay-busy bug */
#define MAX_BAU_CONCURRENT 3 #define MAX_BAU_CONCURRENT 3
#define UV_CPUS_PER_ACT_STATUS 32 #define UV_CPUS_PER_ACT_STATUS 32
#define UV_ACT_STATUS_MASK 0x3 #define UV_ACT_STATUS_MASK 0x3
...@@ -45,10 +46,26 @@ ...@@ -45,10 +46,26 @@
#define UV_DESC_BASE_PNODE_SHIFT 49 #define UV_DESC_BASE_PNODE_SHIFT 49
#define UV_PAYLOADQ_PNODE_SHIFT 49 #define UV_PAYLOADQ_PNODE_SHIFT 49
#define UV_PTC_BASENAME "sgi_uv/ptc_statistics" #define UV_PTC_BASENAME "sgi_uv/ptc_statistics"
#define UV_BAU_BASENAME "sgi_uv/bau_tunables"
#define UV_BAU_TUNABLES_DIR "sgi_uv"
#define UV_BAU_TUNABLES_FILE "bau_tunables"
#define WHITESPACE " \t\n"
#define uv_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask)) #define uv_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask))
#define UV_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15 #define UV_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15
#define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHIFT 16 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHIFT 16
#define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x000000000bUL #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x0000000009UL
/* [19:16] SOFT_ACK timeout period 19: 1 is urgency 7 17:16 1 is multiplier */
#define BAU_MISC_CONTROL_MULT_MASK 3
#define UVH_AGING_PRESCALE_SEL 0x000000b000UL
/* [30:28] URGENCY_7 an index into a table of times */
#define BAU_URGENCY_7_SHIFT 28
#define BAU_URGENCY_7_MASK 7
#define UVH_TRANSACTION_TIMEOUT 0x000000b200UL
/* [45:40] BAU - BAU transaction timeout select - a multiplier */
#define BAU_TRANS_SHIFT 40
#define BAU_TRANS_MASK 0x3f
/* /*
* bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
...@@ -59,24 +76,21 @@ ...@@ -59,24 +76,21 @@
#define DESC_STATUS_SOURCE_TIMEOUT 3 #define DESC_STATUS_SOURCE_TIMEOUT 3
/* /*
* source side threshholds at which message retries print a warning * delay for 'plugged' timeout retries, in microseconds
*/
#define SOURCE_TIMEOUT_LIMIT 20
#define DESTINATION_TIMEOUT_LIMIT 20
/*
* misc. delays, in microseconds
*/ */
#define THROTTLE_DELAY 10 #define PLUGGED_DELAY 10
#define TIMEOUT_DELAY 10
#define BIOS_TO 1000
/* BIOS is assumed to set the destination timeout to 1003520 nanoseconds */
/* /*
* threshholds at which to use IPI to free resources * threshholds at which to use IPI to free resources
*/ */
/* after this # consecutive 'plugged' timeouts, use IPI to release resources */
#define PLUGSB4RESET 100 #define PLUGSB4RESET 100
#define TIMEOUTSB4RESET 100 /* after this many consecutive timeouts, use IPI to release resources */
#define TIMEOUTSB4RESET 1
/* at this number uses of IPI to release resources, giveup the request */
#define IPI_RESET_LIMIT 1
/* after this # consecutive successes, bump up the throttle if it was lowered */
#define COMPLETE_THRESHOLD 5
/* /*
* number of entries in the destination side payload queue * number of entries in the destination side payload queue
...@@ -95,6 +109,13 @@ ...@@ -95,6 +109,13 @@
#define FLUSH_GIVEUP 3 #define FLUSH_GIVEUP 3
#define FLUSH_COMPLETE 4 #define FLUSH_COMPLETE 4
/*
* tuning the action when the numalink network is extremely delayed
*/
#define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in microseconds */
#define CONGESTED_REPS 10 /* long delays averaged over this many broadcasts */
#define CONGESTED_PERIOD 30 /* time for the bau to be disabled, in seconds */
/* /*
* Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor) * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
* If the 'multilevel' flag in the header portion of the descriptor * If the 'multilevel' flag in the header portion of the descriptor
...@@ -300,37 +321,16 @@ struct bau_payload_queue_entry { ...@@ -300,37 +321,16 @@ struct bau_payload_queue_entry {
/* bytes 24-31 */ /* bytes 24-31 */
}; };
/* struct msg_desc {
* one per-cpu; to locate the software tables struct bau_payload_queue_entry *msg;
*/ int msg_slot;
struct bau_control { int sw_ack_slot;
struct bau_desc *descriptor_base;
struct bau_payload_queue_entry *va_queue_first; struct bau_payload_queue_entry *va_queue_first;
struct bau_payload_queue_entry *va_queue_last; struct bau_payload_queue_entry *va_queue_last;
struct bau_payload_queue_entry *bau_msg_head; };
struct bau_control *uvhub_master;
struct bau_control *socket_master; struct reset_args {
unsigned long timeout_interval; int sender;
atomic_t active_descriptor_count;
int max_concurrent;
int max_concurrent_constant;
int retry_message_scans;
int plugged_tries;
int timeout_tries;
int ipi_attempts;
int conseccompletes;
short cpu;
short uvhub_cpu;
short uvhub;
short cpus_in_socket;
short cpus_in_uvhub;
unsigned short message_number;
unsigned short uvhub_quiesce;
short socket_acknowledge_count[DEST_Q_SIZE];
cycles_t send_message;
spinlock_t masks_lock;
spinlock_t uvhub_lock;
spinlock_t queue_lock;
}; };
/* /*
...@@ -344,18 +344,25 @@ struct ptc_stats { ...@@ -344,18 +344,25 @@ struct ptc_stats {
unsigned long s_dtimeout; /* destination side timeouts */ unsigned long s_dtimeout; /* destination side timeouts */
unsigned long s_time; /* time spent in sending side */ unsigned long s_time; /* time spent in sending side */
unsigned long s_retriesok; /* successful retries */ unsigned long s_retriesok; /* successful retries */
unsigned long s_ntargcpu; /* number of cpus targeted */ unsigned long s_ntargcpu; /* total number of cpu's targeted */
unsigned long s_ntarguvhub; /* number of uvhubs targeted */ unsigned long s_ntargself; /* times the sending cpu was targeted */
unsigned long s_ntarguvhub16; /* number of times >= 16 target hubs */ unsigned long s_ntarglocals; /* targets of cpus on the local blade */
unsigned long s_ntarguvhub8; /* number of times >= 8 target hubs */ unsigned long s_ntargremotes; /* targets of cpus on remote blades */
unsigned long s_ntarguvhub4; /* number of times >= 4 target hubs */ unsigned long s_ntarglocaluvhub; /* targets of the local hub */
unsigned long s_ntarguvhub2; /* number of times >= 2 target hubs */ unsigned long s_ntargremoteuvhub; /* remotes hubs targeted */
unsigned long s_ntarguvhub1; /* number of times == 1 target hub */ unsigned long s_ntarguvhub; /* total number of uvhubs targeted */
unsigned long s_ntarguvhub16; /* number of times target hubs >= 16*/
unsigned long s_ntarguvhub8; /* number of times target hubs >= 8 */
unsigned long s_ntarguvhub4; /* number of times target hubs >= 4 */
unsigned long s_ntarguvhub2; /* number of times target hubs >= 2 */
unsigned long s_ntarguvhub1; /* number of times target hubs == 1 */
unsigned long s_resets_plug; /* ipi-style resets from plug state */ unsigned long s_resets_plug; /* ipi-style resets from plug state */
unsigned long s_resets_timeout; /* ipi-style resets from timeouts */ unsigned long s_resets_timeout; /* ipi-style resets from timeouts */
unsigned long s_busy; /* status stayed busy past s/w timer */ unsigned long s_busy; /* status stayed busy past s/w timer */
unsigned long s_throttles; /* waits in throttle */ unsigned long s_throttles; /* waits in throttle */
unsigned long s_retry_messages; /* retry broadcasts */ unsigned long s_retry_messages; /* retry broadcasts */
unsigned long s_bau_reenabled; /* for bau enable/disable */
unsigned long s_bau_disabled; /* for bau enable/disable */
/* destination statistics */ /* destination statistics */
unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */ unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */
unsigned long d_onetlb; /* times just one tlb on this cpu was flushed */ unsigned long d_onetlb; /* times just one tlb on this cpu was flushed */
...@@ -370,6 +377,52 @@ struct ptc_stats { ...@@ -370,6 +377,52 @@ struct ptc_stats {
unsigned long d_rcanceled; /* number of messages canceled by resets */ unsigned long d_rcanceled; /* number of messages canceled by resets */
}; };
/*
* one per-cpu; to locate the software tables
*/
struct bau_control {
struct bau_desc *descriptor_base;
struct bau_payload_queue_entry *va_queue_first;
struct bau_payload_queue_entry *va_queue_last;
struct bau_payload_queue_entry *bau_msg_head;
struct bau_control *uvhub_master;
struct bau_control *socket_master;
struct ptc_stats *statp;
unsigned long timeout_interval;
unsigned long set_bau_on_time;
atomic_t active_descriptor_count;
int plugged_tries;
int timeout_tries;
int ipi_attempts;
int conseccompletes;
int baudisabled;
int set_bau_off;
short cpu;
short uvhub_cpu;
short uvhub;
short cpus_in_socket;
short cpus_in_uvhub;
unsigned short message_number;
unsigned short uvhub_quiesce;
short socket_acknowledge_count[DEST_Q_SIZE];
cycles_t send_message;
spinlock_t uvhub_lock;
spinlock_t queue_lock;
/* tunables */
int max_bau_concurrent;
int max_bau_concurrent_constant;
int plugged_delay;
int plugsb4reset;
int timeoutsb4reset;
int ipi_reset_limit;
int complete_threshold;
int congested_response_us;
int congested_reps;
int congested_period;
cycles_t period_time;
long period_requests;
};
static inline int bau_uvhub_isset(int uvhub, struct bau_target_uvhubmask *dstp) static inline int bau_uvhub_isset(int uvhub, struct bau_target_uvhubmask *dstp)
{ {
return constant_test_bit(uvhub, &dstp->bits[0]); return constant_test_bit(uvhub, &dstp->bits[0]);
......
arch/x86/kernel/tlb_uv.c
View file @ 36450e9c
...@@ -8,6 +8,7 @@ ...@@ -8,6 +8,7 @@
*/ */
#include <linux/seq_file.h> #include <linux/seq_file.h>
#include <linux/proc_fs.h> #include <linux/proc_fs.h>
#include <linux/debugfs.h>
#include <linux/kernel.h> #include <linux/kernel.h>
#include <linux/slab.h> #include <linux/slab.h>
...@@ -22,19 +23,37 @@ ...@@ -22,19 +23,37 @@
#include <asm/irq_vectors.h> #include <asm/irq_vectors.h>
#include <asm/timer.h> #include <asm/timer.h>
struct msg_desc { /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
struct bau_payload_queue_entry *msg; static int timeout_base_ns[] = {
int msg_slot; 20,
int sw_ack_slot; 160,
struct bau_payload_queue_entry *va_queue_first; 1280,
struct bau_payload_queue_entry *va_queue_last; 10240,
81920,
655360,
5242880,
167772160
}; };
static int timeout_us;
#define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x000000000bUL
static int uv_bau_max_concurrent __read_mostly;
static int nobau; static int nobau;
static int baudisabled;
static spinlock_t disable_lock;
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 struct dentry *tunables_dir;
static struct dentry *tunables_file;
static int __init setup_nobau(char *arg) static int __init setup_nobau(char *arg)
{ {
nobau = 1; nobau = 1;
...@@ -52,10 +71,6 @@ static DEFINE_PER_CPU(struct ptc_stats, ptcstats); ...@@ -52,10 +71,6 @@ static DEFINE_PER_CPU(struct ptc_stats, ptcstats);
static DEFINE_PER_CPU(struct bau_control, bau_control); static DEFINE_PER_CPU(struct bau_control, bau_control);
static DEFINE_PER_CPU(cpumask_var_t, uv_flush_tlb_mask); static DEFINE_PER_CPU(cpumask_var_t, uv_flush_tlb_mask);
struct reset_args {
int sender;
};
/* /*
* Determine the first node on a uvhub. 'Nodes' are used for kernel * Determine the first node on a uvhub. 'Nodes' are used for kernel
* memory allocation. * memory allocation.
...@@ -126,7 +141,7 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp, ...@@ -126,7 +141,7 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp,
struct ptc_stats *stat; struct ptc_stats *stat;
msg = mdp->msg; msg = mdp->msg;
stat = &per_cpu(ptcstats, bcp->cpu); stat = bcp->statp;
stat->d_retries++; stat->d_retries++;
/* /*
* cancel any message from msg+1 to the retry itself * cancel any message from msg+1 to the retry itself
...@@ -146,15 +161,14 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp, ...@@ -146,15 +161,14 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp,
slot2 = msg2 - mdp->va_queue_first; slot2 = msg2 - mdp->va_queue_first;
mmr = uv_read_local_mmr mmr = uv_read_local_mmr
(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE); (UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
msg_res = ((msg2->sw_ack_vector << 8) | msg_res = msg2->sw_ack_vector;
msg2->sw_ack_vector);
/* /*
* This is a message retry; clear the resources held * This is a message retry; clear the resources held
* by the previous message only if they timed out. * by the previous message only if they timed out.
* If it has not timed out we have an unexpected * If it has not timed out we have an unexpected
* situation to report. * situation to report.
*/ */
if (mmr & (msg_res << 8)) { if (mmr & (msg_res << UV_SW_ACK_NPENDING)) {
/* /*
* is the resource timed out? * is the resource timed out?
* make everyone ignore the cancelled message. * make everyone ignore the cancelled message.
...@@ -164,9 +178,9 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp, ...@@ -164,9 +178,9 @@ static inline void uv_bau_process_retry_msg(struct msg_desc *mdp,
cancel_count++; cancel_count++;
uv_write_local_mmr( uv_write_local_mmr(
UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS,
(msg_res << 8) | msg_res); (msg_res << UV_SW_ACK_NPENDING) |
} else msg_res);
printk(KERN_INFO "note bau retry: no effect\n"); }
} }
} }
if (!cancel_count) if (!cancel_count)
...@@ -190,7 +204,7 @@ static void uv_bau_process_message(struct msg_desc *mdp, ...@@ -190,7 +204,7 @@ static void uv_bau_process_message(struct msg_desc *mdp,
* This must be a normal message, or retry of a normal message * This must be a normal message, or retry of a normal message
*/ */
msg = mdp->msg; msg = mdp->msg;
stat = &per_cpu(ptcstats, bcp->cpu); stat = bcp->statp;
if (msg->address == TLB_FLUSH_ALL) { if (msg->address == TLB_FLUSH_ALL) {
local_flush_tlb(); local_flush_tlb();
stat->d_alltlb++; stat->d_alltlb++;
...@@ -274,7 +288,7 @@ uv_do_reset(void *ptr) ...@@ -274,7 +288,7 @@ uv_do_reset(void *ptr)
bcp = &per_cpu(bau_control, smp_processor_id()); bcp = &per_cpu(bau_control, smp_processor_id());
rap = (struct reset_args *)ptr; rap = (struct reset_args *)ptr;
stat = &per_cpu(ptcstats, bcp->cpu); stat = bcp->statp;
stat->d_resets++; stat->d_resets++;
/* /*
...@@ -302,13 +316,13 @@ uv_do_reset(void *ptr) ...@@ -302,13 +316,13 @@ uv_do_reset(void *ptr)
*/ */
mmr = uv_read_local_mmr mmr = uv_read_local_mmr
(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE); (UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
msg_res = ((msg->sw_ack_vector << 8) | msg_res = msg->sw_ack_vector;
msg->sw_ack_vector);
if (mmr & msg_res) { if (mmr & msg_res) {
stat->d_rcanceled++; stat->d_rcanceled++;
uv_write_local_mmr( uv_write_local_mmr(
UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS,
msg_res); (msg_res << UV_SW_ACK_NPENDING) |
msg_res);
} }
} }
} }
...@@ -386,17 +400,12 @@ static int uv_wait_completion(struct bau_desc *bau_desc, ...@@ -386,17 +400,12 @@ static int uv_wait_completion(struct bau_desc *bau_desc,
unsigned long mmr_offset, int right_shift, int this_cpu, unsigned long mmr_offset, int right_shift, int this_cpu,
struct bau_control *bcp, struct bau_control *smaster, long try) struct bau_control *bcp, struct bau_control *smaster, long try)
{ {
int relaxes = 0;
unsigned long descriptor_status; unsigned long descriptor_status;
unsigned long mmr;
unsigned long mask;
cycles_t ttime; cycles_t ttime;
cycles_t timeout_time; struct ptc_stats *stat = bcp->statp;
struct ptc_stats *stat = &per_cpu(ptcstats, this_cpu);
struct bau_control *hmaster; struct bau_control *hmaster;
hmaster = bcp->uvhub_master; hmaster = bcp->uvhub_master;
timeout_time = get_cycles() + bcp->timeout_interval;
/* spin on the status MMR, waiting for it to go idle */ /* spin on the status MMR, waiting for it to go idle */
while ((descriptor_status = (((unsigned long) while ((descriptor_status = (((unsigned long)
...@@ -423,7 +432,8 @@ static int uv_wait_completion(struct bau_desc *bau_desc, ...@@ -423,7 +432,8 @@ static int uv_wait_completion(struct bau_desc *bau_desc,
* pending. In that case hardware returns the * pending. In that case hardware returns the
* ERROR that looks like a destination timeout. * ERROR that looks like a destination timeout.
*/ */
if (cycles_2_us(ttime - bcp->send_message) < BIOS_TO) { if (cycles_2_us(ttime - bcp->send_message) <
timeout_us) {
bcp->conseccompletes = 0; bcp->conseccompletes = 0;
return FLUSH_RETRY_PLUGGED; return FLUSH_RETRY_PLUGGED;
} }
...@@ -435,26 +445,6 @@ static int uv_wait_completion(struct bau_desc *bau_desc, ...@@ -435,26 +445,6 @@ static int uv_wait_completion(struct bau_desc *bau_desc,
* descriptor_status is still BUSY * descriptor_status is still BUSY
*/ */
cpu_relax(); cpu_relax();
relaxes++;
if (relaxes >= 10000) {
relaxes = 0;
if (get_cycles() > timeout_time) {
quiesce_local_uvhub(hmaster);
/* single-thread the register change */
spin_lock(&hmaster->masks_lock);
mmr = uv_read_local_mmr(mmr_offset);
mask = 0UL;
mask |= (3UL < right_shift);
mask = ~mask;
mmr &= mask;
uv_write_local_mmr(mmr_offset, mmr);
spin_unlock(&hmaster->masks_lock);
end_uvhub_quiesce(hmaster);
stat->s_busy++;
return FLUSH_GIVEUP;
}
}
} }
} }
bcp->conseccompletes++; bcp->conseccompletes++;
...@@ -494,56 +484,116 @@ static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u) ...@@ -494,56 +484,116 @@ static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u)
return 1; return 1;
} }
/*
* Our retries are blocked by all destination swack 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,
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);
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)
{
hmaster->max_bau_concurrent = 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);
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.
*/
static void
disable_for_congestion(struct bau_control *bcp, struct ptc_stats *stat)
{
int tcpu;
struct bau_control *tbcp;
/* 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);
stat->s_bau_disabled++;
for_each_present_cpu(tcpu) {
tbcp = &per_cpu(bau_control, tcpu);
tbcp->baudisabled = 1;
}
}
spin_unlock(&disable_lock);
}
/** /**
* uv_flush_send_and_wait * uv_flush_send_and_wait
* *
* Send a broadcast and wait for it to complete. * Send a broadcast and wait for it to complete.
* *
* The flush_mask contains the cpus the broadcast is to be sent to, plus * The flush_mask contains the cpus the broadcast is to be sent to including
* cpus that are on the local uvhub. * cpus that are on the local uvhub.
* *
* Returns NULL if all flushing represented in the mask was done. The mask * Returns 0 if all flushing represented in the mask was done.
* is zeroed. * Returns 1 if it gives up entirely and the original cpu mask is to be
* Returns @flush_mask if some remote flushing remains to be done. The * returned to the kernel.
* mask will have some bits still set, representing any cpus on the local
* uvhub (not current cpu) and any on remote uvhubs if the broadcast failed.
*/ */
const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc, int uv_flush_send_and_wait(struct bau_desc *bau_desc,
struct cpumask *flush_mask, struct cpumask *flush_mask, struct bau_control *bcp)
struct bau_control *bcp)
{ {
int right_shift; int right_shift;
int uvhub;
int bit;
int completion_status = 0; int completion_status = 0;
int seq_number = 0; int seq_number = 0;
long try = 0; long try = 0;
int cpu = bcp->uvhub_cpu; int cpu = bcp->uvhub_cpu;
int this_cpu = bcp->cpu; int this_cpu = bcp->cpu;
int this_uvhub = bcp->uvhub;
unsigned long mmr_offset; unsigned long mmr_offset;
unsigned long index; unsigned long index;
cycles_t time1; cycles_t time1;
cycles_t time2; cycles_t time2;
struct ptc_stats *stat = &per_cpu(ptcstats, bcp->cpu); cycles_t elapsed;
struct ptc_stats *stat = bcp->statp;
struct bau_control *smaster = bcp->socket_master; struct bau_control *smaster = bcp->socket_master;
struct bau_control *hmaster = bcp->uvhub_master; struct bau_control *hmaster = bcp->uvhub_master;
/*
* Spin here while there are hmaster->max_concurrent or more active
* descriptors. This is the per-uvhub 'throttle'.
*/
if (!atomic_inc_unless_ge(&hmaster->uvhub_lock, if (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
&hmaster->active_descriptor_count, &hmaster->active_descriptor_count,
hmaster->max_concurrent)) { hmaster->max_bau_concurrent)) {
stat->s_throttles++; stat->s_throttles++;
do { do {
cpu_relax(); cpu_relax();
} while (!atomic_inc_unless_ge(&hmaster->uvhub_lock, } while (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
&hmaster->active_descriptor_count, &hmaster->active_descriptor_count,
hmaster->max_concurrent)); hmaster->max_bau_concurrent));
} }
while (hmaster->uvhub_quiesce) while (hmaster->uvhub_quiesce)
cpu_relax(); cpu_relax();
...@@ -557,23 +607,10 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc, ...@@ -557,23 +607,10 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc,
} }
time1 = get_cycles(); time1 = get_cycles();
do { do {
/*
* Every message from any given cpu gets a unique message
* sequence number. But retries use that same number.
* Our message may have timed out at the destination because
* all sw-ack resources are in use and there is a timeout
* pending there. In that case, our last send never got
* placed into the queue and we need to persist until it
* does.
*
* Make any retry a type MSG_RETRY so that the destination will
* free any resource held by a previous message from this cpu.
*/
if (try == 0) { if (try == 0) {
/* use message type set by the caller the first time */ bau_desc->header.msg_type = MSG_REGULAR;
seq_number = bcp->message_number++; seq_number = bcp->message_number++;
} else { } else {
/* use RETRY type on all the rest; same sequence */
bau_desc->header.msg_type = MSG_RETRY; bau_desc->header.msg_type = MSG_RETRY;
stat->s_retry_messages++; stat->s_retry_messages++;
} }
...@@ -581,50 +618,17 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc, ...@@ -581,50 +618,17 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc,
index = (1UL << UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT) | index = (1UL << UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT) |
bcp->uvhub_cpu; bcp->uvhub_cpu;
bcp->send_message = get_cycles(); bcp->send_message = get_cycles();
uv_write_local_mmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index); uv_write_local_mmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index);
try++; try++;
completion_status = uv_wait_completion(bau_desc, mmr_offset, completion_status = uv_wait_completion(bau_desc, mmr_offset,
right_shift, this_cpu, bcp, smaster, try); right_shift, this_cpu, bcp, smaster, try);
if (completion_status == FLUSH_RETRY_PLUGGED) { if (completion_status == FLUSH_RETRY_PLUGGED) {
/* destination_plugged(bau_desc, bcp, hmaster, stat);
* Our retries may be blocked by all destination swack
* resources being consumed, and a timeout pending. In
* that case hardware immediately returns the ERROR
* that looks like a destination timeout.
*/
udelay(TIMEOUT_DELAY);
bcp->plugged_tries++;
if (bcp->plugged_tries >= PLUGSB4RESET) {
bcp->plugged_tries = 0;
quiesce_local_uvhub(hmaster);
spin_lock(&hmaster->queue_lock);
uv_reset_with_ipi(&bau_desc->distribution,
this_cpu);
spin_unlock(&hmaster->queue_lock);
end_uvhub_quiesce(hmaster);
bcp->ipi_attempts++;
stat->s_resets_plug++;
}
} else if (completion_status == FLUSH_RETRY_TIMEOUT) { } else if (completion_status == FLUSH_RETRY_TIMEOUT) {
hmaster->max_concurrent = 1; destination_timeout(bau_desc, bcp, hmaster, stat);
bcp->timeout_tries++;
udelay(TIMEOUT_DELAY);
if (bcp->timeout_tries >= TIMEOUTSB4RESET) {
bcp->timeout_tries = 0;
quiesce_local_uvhub(hmaster);
spin_lock(&hmaster->queue_lock);
uv_reset_with_ipi(&bau_desc->distribution,
this_cpu);
spin_unlock(&hmaster->queue_lock);
end_uvhub_quiesce(hmaster);
bcp->ipi_attempts++;
stat->s_resets_timeout++;
}
} }
if (bcp->ipi_attempts >= 3) { if (bcp->ipi_attempts >= bcp->ipi_reset_limit) {
bcp->ipi_attempts = 0; bcp->ipi_attempts = 0;
completion_status = FLUSH_GIVEUP; completion_status = FLUSH_GIVEUP;
break; break;
...@@ -633,49 +637,36 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc, ...@@ -633,49 +637,36 @@ const struct cpumask *uv_flush_send_and_wait(struct bau_desc *bau_desc,
} while ((completion_status == FLUSH_RETRY_PLUGGED) || } while ((completion_status == FLUSH_RETRY_PLUGGED) ||
(completion_status == FLUSH_RETRY_TIMEOUT)); (completion_status == FLUSH_RETRY_TIMEOUT));
time2 = get_cycles(); time2 = get_cycles();
bcp->plugged_tries = 0;
if ((completion_status == FLUSH_COMPLETE) && (bcp->conseccompletes > 5) bcp->timeout_tries = 0;
&& (hmaster->max_concurrent < hmaster->max_concurrent_constant)) if ((completion_status == FLUSH_COMPLETE) &&
hmaster->max_concurrent++; (bcp->conseccompletes > bcp->complete_threshold) &&
(hmaster->max_bau_concurrent <
/* hmaster->max_bau_concurrent_constant))
* hold any cpu not timing out here; no other cpu currently held by hmaster->max_bau_concurrent++;
* the 'throttle' should enter the activation code
*/
while (hmaster->uvhub_quiesce) while (hmaster->uvhub_quiesce)
cpu_relax(); cpu_relax();
atomic_dec(&hmaster->active_descriptor_count); atomic_dec(&hmaster->active_descriptor_count);
if (time2 > time1) {
/* guard against cycles wrap */ elapsed = time2 - time1;
if (time2 > time1) stat->s_time += elapsed;
stat->s_time += (time2 - time1); if ((completion_status == FLUSH_COMPLETE) && (try == 1)) {
else bcp->period_requests++;
stat->s_requestor--; /* don't count this one */ bcp->period_time += elapsed;
if ((elapsed > congested_cycles) &&
(bcp->period_requests > bcp->congested_reps)) {
disable_for_congestion(bcp, stat);
}
}
} else
stat->s_requestor--;
if (completion_status == FLUSH_COMPLETE && try > 1) if (completion_status == FLUSH_COMPLETE && try > 1)
stat->s_retriesok++; stat->s_retriesok++;
else if (completion_status == FLUSH_GIVEUP) { else if (completion_status == FLUSH_GIVEUP) {
/*
* Cause the caller to do an IPI-style TLB shootdown on
* the target cpu's, all of which are still in the mask.
*/
stat->s_giveup++; stat->s_giveup++;
return flush_mask; return 1;
}
/*
* Success, so clear the remote cpu's from the mask so we don't
* use the IPI method of shootdown on them.
*/
for_each_cpu(bit, flush_mask) {
uvhub = uv_cpu_to_blade_id(bit);
if (uvhub == this_uvhub)
continue;
cpumask_clear_cpu(bit, flush_mask);
} }
if (!cpumask_empty(flush_mask)) return 0;
return flush_mask;
return NULL;
} }
/** /**
...@@ -707,70 +698,89 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, ...@@ -707,70 +698,89 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask,
struct mm_struct *mm, struct mm_struct *mm,
unsigned long va, unsigned int cpu) unsigned long va, unsigned int cpu)
{ {
int remotes;
int tcpu; int tcpu;
int uvhub; int uvhub;
int locals = 0; int locals = 0;
int remotes = 0;
int hubs = 0;
struct bau_desc *bau_desc; struct bau_desc *bau_desc;
struct cpumask *flush_mask; struct cpumask *flush_mask;
struct ptc_stats *stat; struct ptc_stats *stat;
struct bau_control *bcp; struct bau_control *bcp;
struct bau_control *tbcp;
/* kernel was booted 'nobau' */
if (nobau) if (nobau)
return cpumask; return cpumask;
bcp = &per_cpu(bau_control, cpu); bcp = &per_cpu(bau_control, cpu);
stat = bcp->statp;
/* 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;
}
}
}
return cpumask;
}
/* /*
* Each sending cpu has a per-cpu mask which it fills from the caller's * Each sending cpu has a per-cpu mask which it fills from the caller's
* cpu mask. Only remote cpus are converted to uvhubs and copied. * cpu mask. All cpus are converted to uvhubs and copied to the
* activation descriptor.
*/ */
flush_mask = (struct cpumask *)per_cpu(uv_flush_tlb_mask, cpu); flush_mask = (struct cpumask *)per_cpu(uv_flush_tlb_mask, cpu);
/* /* don't actually do a shootdown of the local cpu */
* copy cpumask to flush_mask, removing current cpu
* (current cpu should already have been flushed by the caller and
* should never be returned if we return flush_mask)
*/
cpumask_andnot(flush_mask, cpumask, cpumask_of(cpu)); cpumask_andnot(flush_mask, cpumask, cpumask_of(cpu));
if (cpu_isset(cpu, *cpumask)) if (cpu_isset(cpu, *cpumask))
locals++; /* current cpu was targeted */ stat->s_ntargself++;
bau_desc = bcp->descriptor_base; bau_desc = bcp->descriptor_base;
bau_desc += UV_ITEMS_PER_DESCRIPTOR * bcp->uvhub_cpu; bau_desc += UV_ITEMS_PER_DESCRIPTOR * bcp->uvhub_cpu;
bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE); bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE);
remotes = 0;
/* cpu statistics */
for_each_cpu(tcpu, flush_mask) { for_each_cpu(tcpu, flush_mask) {
uvhub = uv_cpu_to_blade_id(tcpu); uvhub = uv_cpu_to_blade_id(tcpu);
if (uvhub == bcp->uvhub) {
locals++;
continue;
}
bau_uvhub_set(uvhub, &bau_desc->distribution); bau_uvhub_set(uvhub, &bau_desc->distribution);
remotes++; if (uvhub == bcp->uvhub)
} locals++;
if (remotes == 0) {
/*
* No off_hub flushing; return status for local hub.
* Return the caller's mask if all were local (the current
* cpu may be in that mask).
*/
if (locals)
return cpumask;
else else
return NULL; remotes++;
} }
stat = &per_cpu(ptcstats, cpu); if ((locals + remotes) == 0)
return NULL;
stat->s_requestor++; stat->s_requestor++;
stat->s_ntargcpu += remotes; stat->s_ntargcpu += remotes + locals;
stat->s_ntargremotes += remotes;
stat->s_ntarglocals += locals;
remotes = bau_uvhub_weight(&bau_desc->distribution); remotes = bau_uvhub_weight(&bau_desc->distribution);
stat->s_ntarguvhub += remotes;
if (remotes >= 16) /* 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++; stat->s_ntarguvhub16++;
else if (remotes >= 8) else if (hubs >= 8)
stat->s_ntarguvhub8++; stat->s_ntarguvhub8++;
else if (remotes >= 4) else if (hubs >= 4)
stat->s_ntarguvhub4++; stat->s_ntarguvhub4++;
else if (remotes >= 2) else if (hubs >= 2)
stat->s_ntarguvhub2++; stat->s_ntarguvhub2++;
else else
stat->s_ntarguvhub1++; stat->s_ntarguvhub1++;
...@@ -779,10 +789,13 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask, ...@@ -779,10 +789,13 @@ const struct cpumask *uv_flush_tlb_others(const struct cpumask *cpumask,
bau_desc->payload.sending_cpu = cpu; bau_desc->payload.sending_cpu = cpu;
/* /*
* uv_flush_send_and_wait returns null if all cpu's were messaged, or * uv_flush_send_and_wait returns 0 if all cpu's were messaged,
* the adjusted flush_mask if any cpu's were not messaged. * or 1 if it gave up and the original cpumask should be returned.
*/ */
return uv_flush_send_and_wait(bau_desc, flush_mask, bcp); if (!uv_flush_send_and_wait(bau_desc, flush_mask, bcp))
return NULL;
else
return cpumask;
} }
/* /*
...@@ -810,7 +823,7 @@ void uv_bau_message_interrupt(struct pt_regs *regs) ...@@ -810,7 +823,7 @@ void uv_bau_message_interrupt(struct pt_regs *regs)
time_start = get_cycles(); time_start = get_cycles();
bcp = &per_cpu(bau_control, smp_processor_id()); bcp = &per_cpu(bau_control, smp_processor_id());
stat = &per_cpu(ptcstats, smp_processor_id()); stat = bcp->statp;
msgdesc.va_queue_first = bcp->va_queue_first; msgdesc.va_queue_first = bcp->va_queue_first;
msgdesc.va_queue_last = bcp->va_queue_last; msgdesc.va_queue_last = bcp->va_queue_last;
msg = bcp->bau_msg_head; msg = bcp->bau_msg_head;
...@@ -908,12 +921,12 @@ static void uv_ptc_seq_stop(struct seq_file *file, void *data) ...@@ -908,12 +921,12 @@ static void uv_ptc_seq_stop(struct seq_file *file, void *data)
} }
static inline unsigned long long static inline unsigned long long
millisec_2_cycles(unsigned long millisec) microsec_2_cycles(unsigned long microsec)
{ {
unsigned long ns; unsigned long ns;
unsigned long long cyc; unsigned long long cyc;
ns = millisec * 1000; ns = microsec * 1000;
cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id())); cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id()));
return cyc; return cyc;
} }
...@@ -931,15 +944,19 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data) ...@@ -931,15 +944,19 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data)
if (!cpu) { if (!cpu) {
seq_printf(file, seq_printf(file,
"# cpu sent stime numuvhubs numuvhubs16 numuvhubs8 "); "# cpu sent stime self locals remotes ncpus localhub ");
seq_printf(file, seq_printf(file,
"numuvhubs4 numuvhubs2 numuvhubs1 numcpus dto "); "remotehub numuvhubs numuvhubs16 numuvhubs8 ");
seq_printf(file,
"numuvhubs4 numuvhubs2 numuvhubs1 dto ");
seq_printf(file, seq_printf(file,
"retries rok resetp resett giveup sto bz throt "); "retries rok resetp resett giveup sto bz throt ");
seq_printf(file, seq_printf(file,
"sw_ack recv rtime all "); "sw_ack recv rtime all ");
seq_printf(file, seq_printf(file,
"one mult none retry canc nocan reset rcan\n"); "one mult none retry canc nocan reset rcan ");
seq_printf(file,
"disable enable\n");
} }
if (cpu < num_possible_cpus() && cpu_online(cpu)) { if (cpu < num_possible_cpus() && cpu_online(cpu)) {
stat = &per_cpu(ptcstats, cpu); stat = &per_cpu(ptcstats, cpu);
...@@ -947,18 +964,23 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data) ...@@ -947,18 +964,23 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data)
seq_printf(file, seq_printf(file,
"cpu %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ", "cpu %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ",
cpu, stat->s_requestor, cycles_2_us(stat->s_time), cpu, stat->s_requestor, cycles_2_us(stat->s_time),
stat->s_ntarguvhub, stat->s_ntarguvhub16, 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 ",
stat->s_ntarguvhub8, stat->s_ntarguvhub4, stat->s_ntarguvhub8, stat->s_ntarguvhub4,
stat->s_ntarguvhub2, stat->s_ntarguvhub1, stat->s_ntarguvhub2, stat->s_ntarguvhub1,
stat->s_ntargcpu, stat->s_dtimeout); stat->s_dtimeout);
seq_printf(file, "%ld %ld %ld %ld %ld %ld %ld %ld ", seq_printf(file, "%ld %ld %ld %ld %ld %ld %ld %ld ",
stat->s_retry_messages, stat->s_retriesok, stat->s_retry_messages, stat->s_retriesok,
stat->s_resets_plug, stat->s_resets_timeout, stat->s_resets_plug, stat->s_resets_timeout,
stat->s_giveup, stat->s_stimeout, stat->s_giveup, stat->s_stimeout,
stat->s_busy, stat->s_throttles); stat->s_busy, stat->s_throttles);
/* destination side statistics */ /* destination side statistics */
seq_printf(file, seq_printf(file,
"%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n", "%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld ",
uv_read_global_mmr64(uv_cpu_to_pnode(cpu), uv_read_global_mmr64(uv_cpu_to_pnode(cpu),
UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE), UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE),
stat->d_requestee, cycles_2_us(stat->d_time), stat->d_requestee, cycles_2_us(stat->d_time),
...@@ -966,15 +988,36 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data) ...@@ -966,15 +988,36 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data)
stat->d_nomsg, stat->d_retries, stat->d_canceled, stat->d_nomsg, stat->d_retries, stat->d_canceled,
stat->d_nocanceled, stat->d_resets, stat->d_nocanceled, stat->d_resets,
stat->d_rcanceled); stat->d_rcanceled);
seq_printf(file, "%ld %ld\n",
stat->s_bau_disabled, stat->s_bau_reenabled);
} }
return 0; return 0;
} }
/*
* Display the tunables thru debugfs
*/
static ssize_t tunables_read(struct file *file, char __user *userbuf,
size_t count, loff_t *ppos)
{
char buf[300];
int ret;
ret = snprintf(buf, 300, "%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,
timeoutsb4reset, ipi_reset_limit, complete_threshold,
congested_response_us, congested_reps, congested_period);
return simple_read_from_buffer(userbuf, count, ppos, buf, ret);
}
/* /*
* -1: resetf the statistics * -1: resetf the statistics
* 0: display meaning of the statistics * 0: display meaning of the statistics
* >0: maximum concurrent active descriptors per uvhub (throttle)
*/ */
static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user, static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user,
size_t count, loff_t *data) size_t count, loff_t *data)
...@@ -983,7 +1026,6 @@ static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user, ...@@ -983,7 +1026,6 @@ static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user,
long input_arg; long input_arg;
char optstr[64]; char optstr[64];
struct ptc_stats *stat; struct ptc_stats *stat;
struct bau_control *bcp;
if (count == 0 || count > sizeof(optstr)) if (count == 0 || count > sizeof(optstr))
return -EINVAL; return -EINVAL;
...@@ -1059,29 +1101,158 @@ static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user, ...@@ -1059,29 +1101,158 @@ static ssize_t uv_ptc_proc_write(struct file *file, const char __user *user,
"reset: number of ipi-style reset requests processed\n"); "reset: number of ipi-style reset requests processed\n");
printk(KERN_DEBUG printk(KERN_DEBUG
"rcan: number messages canceled by reset requests\n"); "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");
} else if (input_arg == -1) { } else if (input_arg == -1) {
for_each_present_cpu(cpu) { for_each_present_cpu(cpu) {
stat = &per_cpu(ptcstats, cpu); stat = &per_cpu(ptcstats, cpu);
memset(stat, 0, sizeof(struct ptc_stats)); memset(stat, 0, sizeof(struct ptc_stats));
} }
} else { }
uv_bau_max_concurrent = input_arg;
bcp = &per_cpu(bau_control, smp_processor_id()); return count;
if (uv_bau_max_concurrent < 1 || }
uv_bau_max_concurrent > bcp->cpus_in_uvhub) {
printk(KERN_DEBUG static int local_atoi(const char *name)
"Error: BAU max concurrent %d; %d is invalid\n", {
bcp->max_concurrent, uv_bau_max_concurrent); int val = 0;
return -EINVAL;
} for (;; name++) {
printk(KERN_DEBUG "Set BAU max concurrent:%d\n", switch (*name) {
uv_bau_max_concurrent); case '0' ... '9':
for_each_present_cpu(cpu) { val = 10*val+(*name-'0');
bcp = &per_cpu(bau_control, cpu); break;
bcp->max_concurrent = uv_bau_max_concurrent; default:
return val;
} }
} }
}
/*
* set the tunables
* 0 values reset them to defaults
*/
static ssize_t tunables_write(struct file *file, const char __user *user,
size_t count, loff_t *data)
{
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;
instr[count] = '\0';
/* count the fields */
p = instr + strspn(instr, WHITESPACE);
q = p;
for (; *p; p = q + strspn(q, WHITESPACE)) {
q = p + strcspn(p, WHITESPACE);
cnt++;
if (q == p)
break;
}
if (cnt != 9) {
printk(KERN_INFO "bau tunable error: should be 9 numbers\n");
return -EINVAL;
}
p = instr + strspn(instr, WHITESPACE);
q = p;
for (cnt = 0; *p; p = q + strspn(q, WHITESPACE), cnt++) {
q = p + strcspn(p, WHITESPACE);
val = local_atoi(p);
switch (cnt) {
case 0:
if (val == 0) {
max_bau_concurrent = MAX_BAU_CONCURRENT;
max_bau_concurrent_constant =
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;
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:
if (val == 0)
congested_period = CONGESTED_PERIOD;
else
congested_period = val;
continue;
}
if (q == p)
break;
}
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;
}
return count; return count;
} }
...@@ -1097,6 +1268,11 @@ static int uv_ptc_proc_open(struct inode *inode, struct file *file) ...@@ -1097,6 +1268,11 @@ static int uv_ptc_proc_open(struct inode *inode, struct file *file)
return seq_open(file, &uv_ptc_seq_ops); return seq_open(file, &uv_ptc_seq_ops);
} }
static int tunables_open(struct inode *inode, struct file *file)
{
return 0;
}
static const struct file_operations proc_uv_ptc_operations = { static const struct file_operations proc_uv_ptc_operations = {
.open = uv_ptc_proc_open, .open = uv_ptc_proc_open,
.read = seq_read, .read = seq_read,
...@@ -1105,6 +1281,12 @@ static const struct file_operations proc_uv_ptc_operations = { ...@@ -1105,6 +1281,12 @@ static const struct file_operations proc_uv_ptc_operations = {
.release = seq_release, .release = seq_release,
}; };
static const struct file_operations tunables_fops = {
.open = tunables_open,
.read = tunables_read,
.write = tunables_write,
};
static int __init uv_ptc_init(void) static int __init uv_ptc_init(void)
{ {
struct proc_dir_entry *proc_uv_ptc; struct proc_dir_entry *proc_uv_ptc;
...@@ -1119,6 +1301,20 @@ static int __init uv_ptc_init(void) ...@@ -1119,6 +1301,20 @@ static int __init uv_ptc_init(void)
UV_PTC_BASENAME); UV_PTC_BASENAME);
return -EINVAL; return -EINVAL;
} }
tunables_dir = debugfs_create_dir(UV_BAU_TUNABLES_DIR, NULL);
if (!tunables_dir) {
printk(KERN_ERR "unable to create debugfs directory %s\n",
UV_BAU_TUNABLES_DIR);
return -EINVAL;
}
tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600,
tunables_dir, NULL, &tunables_fops);
if (!tunables_file) {
printk(KERN_ERR "unable to create debugfs file %s\n",
UV_BAU_TUNABLES_FILE);
return -EINVAL;
}
return 0; return 0;
} }
...@@ -1258,16 +1454,45 @@ static void __init uv_init_uvhub(int uvhub, int vector) ...@@ -1258,16 +1454,45 @@ static void __init uv_init_uvhub(int uvhub, int vector)
((apicid << 32) | vector)); ((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.
*/
static int
calculate_destination_timeout(void)
{
unsigned long mmr_image;
int mult1;
int mult2;
int index;
int base;
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;
return ret;
}
/* /*
* initialize the bau_control structure for each cpu * initialize the bau_control structure for each cpu
*/ */
static void uv_init_per_cpu(int nuvhubs) static void uv_init_per_cpu(int nuvhubs)
{ {
int i, j, k; int i;
int cpu; int cpu;
int pnode; int pnode;
int uvhub; int uvhub;
short socket = 0; short socket = 0;
unsigned short socket_mask;
unsigned int uvhub_mask;
struct bau_control *bcp; struct bau_control *bcp;
struct uvhub_desc *bdp; struct uvhub_desc *bdp;
struct socket_desc *sdp; struct socket_desc *sdp;
...@@ -1278,7 +1503,7 @@ static void uv_init_per_cpu(int nuvhubs) ...@@ -1278,7 +1503,7 @@ static void uv_init_per_cpu(int nuvhubs)
short cpu_number[16]; short cpu_number[16];
}; };
struct uvhub_desc { struct uvhub_desc {
short num_sockets; unsigned short socket_mask;
short num_cpus; short num_cpus;
short uvhub; short uvhub;
short pnode; short pnode;
...@@ -1286,57 +1511,83 @@ static void uv_init_per_cpu(int nuvhubs) ...@@ -1286,57 +1511,83 @@ static void uv_init_per_cpu(int nuvhubs)
}; };
struct uvhub_desc *uvhub_descs; struct uvhub_desc *uvhub_descs;
timeout_us = calculate_destination_timeout();
uvhub_descs = (struct uvhub_desc *) uvhub_descs = (struct uvhub_desc *)
kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL); kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL);
memset(uvhub_descs, 0, nuvhubs * sizeof(struct uvhub_desc)); memset(uvhub_descs, 0, nuvhubs * sizeof(struct uvhub_desc));
for_each_present_cpu(cpu) { for_each_present_cpu(cpu) {
bcp = &per_cpu(bau_control, cpu); bcp = &per_cpu(bau_control, cpu);
memset(bcp, 0, sizeof(struct bau_control)); memset(bcp, 0, sizeof(struct bau_control));
spin_lock_init(&bcp->masks_lock);
bcp->max_concurrent = uv_bau_max_concurrent;
pnode = uv_cpu_hub_info(cpu)->pnode; pnode = uv_cpu_hub_info(cpu)->pnode;
uvhub = uv_cpu_hub_info(cpu)->numa_blade_id; uvhub = uv_cpu_hub_info(cpu)->numa_blade_id;
uvhub_mask |= (1 << uvhub);
bdp = &uvhub_descs[uvhub]; bdp = &uvhub_descs[uvhub];
bdp->num_cpus++; bdp->num_cpus++;
bdp->uvhub = uvhub; bdp->uvhub = uvhub;
bdp->pnode = pnode; bdp->pnode = pnode;
/* time interval to catch a hardware stay-busy bug */ /* kludge: 'assuming' one node per socket, and assuming that
bcp->timeout_interval = millisec_2_cycles(3); disabling a socket just leaves a gap in node numbers */
/* kludge: assume uv_hub.h is constant */ socket = (cpu_to_node(cpu) & 1);;
socket = (cpu_physical_id(cpu)>>5)&1; bdp->socket_mask |= (1 << socket);
if (socket >= bdp->num_sockets)
bdp->num_sockets = socket+1;
sdp = &bdp->socket[socket]; sdp = &bdp->socket[socket];
sdp->cpu_number[sdp->num_cpus] = cpu; sdp->cpu_number[sdp->num_cpus] = cpu;
sdp->num_cpus++; sdp->num_cpus++;
} }
socket = 0; uvhub = 0;
for_each_possible_blade(uvhub) { while (uvhub_mask) {
if (!(uvhub_mask & 1))
goto nexthub;
bdp = &uvhub_descs[uvhub]; bdp = &uvhub_descs[uvhub];
for (i = 0; i < bdp->num_sockets; i++) { socket_mask = bdp->socket_mask;
sdp = &bdp->socket[i]; socket = 0;
for (j = 0; j < sdp->num_cpus; j++) { while (socket_mask) {
cpu = sdp->cpu_number[j]; 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 = &per_cpu(bau_control, cpu);
bcp->cpu = cpu; bcp->cpu = cpu;
if (j == 0) { if (i == 0) {
smaster = bcp; smaster = bcp;
if (i == 0) if (socket == 0)
hmaster = bcp; hmaster = bcp;
} }
bcp->cpus_in_uvhub = bdp->num_cpus; bcp->cpus_in_uvhub = bdp->num_cpus;
bcp->cpus_in_socket = sdp->num_cpus; bcp->cpus_in_socket = sdp->num_cpus;
bcp->socket_master = smaster; bcp->socket_master = smaster;
bcp->uvhub = bdp->uvhub;
bcp->uvhub_master = hmaster; bcp->uvhub_master = hmaster;
for (k = 0; k < DEST_Q_SIZE; k++) bcp->uvhub_cpu = uv_cpu_hub_info(cpu)->
bcp->socket_acknowledge_count[k] = 0; blade_processor_id;
bcp->uvhub_cpu =
uv_cpu_hub_info(cpu)->blade_processor_id;
} }
nextsocket:
socket++; socket++;
socket_mask = (socket_mask >> 1);
} }
nexthub:
uvhub++;
uvhub_mask = (uvhub_mask >> 1);
} }
kfree(uvhub_descs); kfree(uvhub_descs);
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;
}
} }
/* /*
...@@ -1361,10 +1612,11 @@ static int __init uv_bau_init(void) ...@@ -1361,10 +1612,11 @@ static int __init uv_bau_init(void)
zalloc_cpumask_var_node(&per_cpu(uv_flush_tlb_mask, cur_cpu), zalloc_cpumask_var_node(&per_cpu(uv_flush_tlb_mask, cur_cpu),
GFP_KERNEL, cpu_to_node(cur_cpu)); GFP_KERNEL, cpu_to_node(cur_cpu));
uv_bau_max_concurrent = MAX_BAU_CONCURRENT;
uv_nshift = uv_hub_info->m_val; uv_nshift = uv_hub_info->m_val;
uv_mmask = (1UL << uv_hub_info->m_val) - 1; uv_mmask = (1UL << uv_hub_info->m_val) - 1;
nuvhubs = uv_num_possible_blades(); nuvhubs = uv_num_possible_blades();
spin_lock_init(&disable_lock);
congested_cycles = microsec_2_cycles(congested_response_us);
uv_init_per_cpu(nuvhubs); uv_init_per_cpu(nuvhubs);
...@@ -1383,15 +1635,19 @@ static int __init uv_bau_init(void) ...@@ -1383,15 +1635,19 @@ static int __init uv_bau_init(void)
alloc_intr_gate(vector, uv_bau_message_intr1); alloc_intr_gate(vector, uv_bau_message_intr1);
for_each_possible_blade(uvhub) { for_each_possible_blade(uvhub) {
pnode = uv_blade_to_pnode(uvhub); if (uv_blade_nr_possible_cpus(uvhub)) {
/* INIT the bau */ pnode = uv_blade_to_pnode(uvhub);
uv_write_global_mmr64(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, /* INIT the bau */
((unsigned long)1 << 63)); uv_write_global_mmr64(pnode,
mmr = 1; /* should be 1 to broadcast to both sockets */ UVH_LB_BAU_SB_ACTIVATION_CONTROL,
uv_write_global_mmr64(pnode, UVH_BAU_DATA_BROADCAST, mmr); ((unsigned long)1 << 63));
mmr = 1; /* should be 1 to broadcast to both sockets */
uv_write_global_mmr64(pnode, UVH_BAU_DATA_BROADCAST,
mmr);
}
} }
return 0; return 0;
} }
core_initcall(uv_bau_init); core_initcall(uv_bau_init);
core_initcall(uv_ptc_init); fs_initcall(uv_ptc_init);
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7