#define UV_DESC_BASE_PNODE_SHIFT 49
#define UV_PAYLOADQ_PNODE_SHIFT 49
#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_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15
#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 DESC_STATUS_DESTINATION_TIMEOUT 2
#define DESC_STATUS_SOURCE_TIMEOUT 3
+#define TIMEOUT_DELAY 10
/*
- * source side threshholds at which message retries print a warning
- */
-#define SOURCE_TIMEOUT_LIMIT 20
-#define DESTINATION_TIMEOUT_LIMIT 20
-
-/*
- * misc. delays, in microseconds
+ * delay for 'plugged' timeout retries, in microseconds
*/
-#define THROTTLE_DELAY 10
-#define TIMEOUT_DELAY 10
-#define BIOS_TO 1000
-/* BIOS is assumed to set the destination timeout to 1003520 nanoseconds */
+#define PLUGGED_DELAY 10
/*
* threshholds at which to use IPI to free resources
*/
+/* after this # consecutive 'plugged' timeouts, use IPI to release resources */
#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
#define FLUSH_GIVEUP 3
#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)
* If the 'multilevel' flag in the header portion of the descriptor
struct bau_control *uvhub_master;
struct bau_control *socket_master;
unsigned long timeout_interval;
+ unsigned long set_bau_on_time;
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;
+ int set_bau_off;
short cpu;
short uvhub_cpu;
short uvhub;
spinlock_t masks_lock;
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;
};
/*
*/
#include <linux/seq_file.h>
#include <linux/proc_fs.h>
+#include <linux/debugfs.h>
#include <linux/kernel.h>
#include <linux/slab.h>
167772160
};
static int timeout_us;
+static int nobau;
-#define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x000000000bUL
-
-static int uv_bau_max_concurrent __read_mostly;
+/* 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 nobau;
static int __init setup_nobau(char *arg)
{
nobau = 1;
unsigned long index;
cycles_t time1;
cycles_t time2;
+ cycles_t elapsed;
struct ptc_stats *stat = &per_cpu(ptcstats, bcp->cpu);
struct bau_control *smaster = bcp->socket_master;
struct bau_control *hmaster = bcp->uvhub_master;
/*
- * Spin here while there are hmaster->max_concurrent or more active
+ * Spin here while there are hmaster->max_bau_concurrent or more active
* descriptors. This is the per-uvhub 'throttle'.
*/
if (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
&hmaster->active_descriptor_count,
- hmaster->max_concurrent)) {
+ hmaster->max_bau_concurrent)) {
stat->s_throttles++;
do {
cpu_relax();
} while (!atomic_inc_unless_ge(&hmaster->uvhub_lock,
&hmaster->active_descriptor_count,
- hmaster->max_concurrent));
+ hmaster->max_bau_concurrent));
}
while (hmaster->uvhub_quiesce)
* that case hardware immediately returns the ERROR
* that looks like a destination timeout.
*/
- udelay(TIMEOUT_DELAY);
+ udelay(bcp->plugged_delay);
bcp->plugged_tries++;
- if (bcp->plugged_tries >= PLUGSB4RESET) {
+ if (bcp->plugged_tries >= bcp->plugsb4reset) {
bcp->plugged_tries = 0;
quiesce_local_uvhub(hmaster);
spin_lock(&hmaster->queue_lock);
stat->s_resets_plug++;
}
} else if (completion_status == FLUSH_RETRY_TIMEOUT) {
- hmaster->max_concurrent = 1;
+ hmaster->max_bau_concurrent = 1;
bcp->timeout_tries++;
udelay(TIMEOUT_DELAY);
- if (bcp->timeout_tries >= TIMEOUTSB4RESET) {
+ if (bcp->timeout_tries >= bcp->timeoutsb4reset) {
bcp->timeout_tries = 0;
quiesce_local_uvhub(hmaster);
spin_lock(&hmaster->queue_lock);
stat->s_resets_timeout++;
}
}
- if (bcp->ipi_attempts >= 3) {
+ if (bcp->ipi_attempts >= bcp->ipi_reset_limit) {
bcp->ipi_attempts = 0;
completion_status = FLUSH_GIVEUP;
break;
(completion_status == FLUSH_RETRY_TIMEOUT));
time2 = get_cycles();
- if ((completion_status == FLUSH_COMPLETE) && (bcp->conseccompletes > 5)
- && (hmaster->max_concurrent < hmaster->max_concurrent_constant))
- hmaster->max_concurrent++;
+ 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++;
/*
* hold any cpu not timing out here; no other cpu currently held by
atomic_dec(&hmaster->active_descriptor_count);
/* guard against cycles wrap */
- if (time2 > time1)
- stat->s_time += (time2 - time1);
- else
+ if (time2 > time1) {
+ elapsed = time2 - time1;
+ stat->s_time += elapsed;
+ } else
stat->s_requestor--; /* don't count this one */
if (completion_status == FLUSH_COMPLETE && try > 1)
stat->s_retriesok++;
struct ptc_stats *stat;
struct bau_control *bcp;
+ /* kernel was booted 'nobau' */
if (nobau)
return cpumask;
bcp = &per_cpu(bau_control, cpu);
+
/*
* 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.
stat->s_resets_plug, stat->s_resets_timeout,
stat->s_giveup, stat->s_stimeout,
stat->s_busy, stat->s_throttles);
+
/* destination side statistics */
seq_printf(file,
"%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
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
* 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,
size_t count, loff_t *data)
long input_arg;
char optstr[64];
struct ptc_stats *stat;
- struct bau_control *bcp;
if (count == 0 || count > sizeof(optstr))
return -EINVAL;
stat = &per_cpu(ptcstats, cpu);
memset(stat, 0, sizeof(struct ptc_stats));
}
- } else {
- uv_bau_max_concurrent = input_arg;
- bcp = &per_cpu(bau_control, smp_processor_id());
- if (uv_bau_max_concurrent < 1 ||
- uv_bau_max_concurrent > bcp->cpus_in_uvhub) {
- printk(KERN_DEBUG
- "Error: BAU max concurrent %d; %d is invalid\n",
- bcp->max_concurrent, uv_bau_max_concurrent);
- return -EINVAL;
- }
- printk(KERN_DEBUG "Set BAU max concurrent:%d\n",
- uv_bau_max_concurrent);
- for_each_present_cpu(cpu) {
- bcp = &per_cpu(bau_control, cpu);
- bcp->max_concurrent = uv_bau_max_concurrent;
+ }
+
+ return count;
+}
+
+static int local_atoi(const char *name)
+{
+ int val = 0;
+
+ for (;; name++) {
+ switch (*name) {
+ case '0' ... '9':
+ val = 10*val+(*name-'0');
+ break;
+ 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 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 = {
.open = uv_ptc_proc_open,
.read = seq_read,
.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)
{
struct proc_dir_entry *proc_uv_ptc;
UV_PTC_BASENAME);
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;
}
bcp = &per_cpu(bau_control, cpu);
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;
uvhub = uv_cpu_hub_info(cpu)->numa_blade_id;
bdp = &uvhub_descs[uvhub];
bdp->num_cpus++;
bdp->uvhub = uvhub;
bdp->pnode = pnode;
- /* time interval to catch a hardware stay-busy bug */
- bcp->timeout_interval = microsec_2_cycles(2*timeout_us);
/* kludge: assume uv_hub.h is constant */
socket = (cpu_physical_id(cpu)>>5)&1;
if (socket >= bdp->num_sockets)
}
}
kfree(uvhub_descs);
+ for_each_present_cpu(cpu) {
+ bcp = &per_cpu(bau_control, 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;
+ }
}
/*
zalloc_cpumask_var_node(&per_cpu(uv_flush_tlb_mask, cur_cpu),
GFP_KERNEL, cpu_to_node(cur_cpu));
- uv_bau_max_concurrent = MAX_BAU_CONCURRENT;
+ max_bau_concurrent = MAX_BAU_CONCURRENT;
uv_nshift = uv_hub_info->m_val;
uv_mmask = (1UL << uv_hub_info->m_val) - 1;
nuvhubs = uv_num_possible_blades();
return 0;
}
core_initcall(uv_bau_init);
-core_initcall(uv_ptc_init);
+fs_initcall(uv_ptc_init);
projects / karo-tx-linux.git / commitdiff