From 10ebd8556b7990499c896a550e3d416b444211e6 Mon Sep 17 00:00:00 2001
From: hc <hc@nodka.com>
Date: Fri, 10 May 2024 02:23:07 +0000
Subject: [PATCH] add led
---
kernel/arch/s390/kernel/perf_cpum_sf.c | 534 ++++++++++++++++++++++++++++++++++++++---------------------
1 files changed, 344 insertions(+), 190 deletions(-)
diff --git a/kernel/arch/s390/kernel/perf_cpum_sf.c b/kernel/arch/s390/kernel/perf_cpum_sf.c
index c8e1e32..bcd31e0 100644
--- a/kernel/arch/s390/kernel/perf_cpum_sf.c
+++ b/kernel/arch/s390/kernel/perf_cpum_sf.c
@@ -156,21 +156,22 @@
}
}
- debug_sprintf_event(sfdbg, 5,
- "free_sampling_buffer: freed sdbt=%p\n", sfb->sdbt);
+ debug_sprintf_event(sfdbg, 5, "%s: freed sdbt %#lx\n", __func__,
+ (unsigned long)sfb->sdbt);
memset(sfb, 0, sizeof(*sfb));
}
static int alloc_sample_data_block(unsigned long *sdbt, gfp_t gfp_flags)
{
- unsigned long sdb, *trailer;
+ struct hws_trailer_entry *te;
+ unsigned long sdb;
/* Allocate and initialize sample-data-block */
sdb = get_zeroed_page(gfp_flags);
if (!sdb)
return -ENOMEM;
- trailer = trailer_entry_ptr(sdb);
- *trailer = SDB_TE_ALERT_REQ_MASK;
+ te = (struct hws_trailer_entry *)trailer_entry_ptr(sdb);
+ te->header.a = 1;
/* Link SDB into the sample-data-block-table */
*sdbt = sdb;
@@ -212,10 +213,11 @@
* the sampling buffer origin.
*/
if (sfb->sdbt != get_next_sdbt(tail)) {
- debug_sprintf_event(sfdbg, 3, "realloc_sampling_buffer: "
- "sampling buffer is not linked: origin=%p"
- "tail=%p\n",
- (void *) sfb->sdbt, (void *) tail);
+ debug_sprintf_event(sfdbg, 3, "%s: "
+ "sampling buffer is not linked: origin %#lx"
+ " tail %#lx\n", __func__,
+ (unsigned long)sfb->sdbt,
+ (unsigned long)tail);
return -EINVAL;
}
@@ -264,8 +266,8 @@
*tail = (unsigned long) sfb->sdbt + 1;
sfb->tail = tail;
- debug_sprintf_event(sfdbg, 4, "realloc_sampling_buffer: new buffer"
- " settings: sdbt=%lu sdb=%lu\n",
+ debug_sprintf_event(sfdbg, 4, "%s: new buffer"
+ " settings: sdbt %lu sdb %lu\n", __func__,
sfb->num_sdbt, sfb->num_sdb);
return rc;
}
@@ -305,12 +307,13 @@
rc = realloc_sampling_buffer(sfb, num_sdb, GFP_KERNEL);
if (rc) {
free_sampling_buffer(sfb);
- debug_sprintf_event(sfdbg, 4, "alloc_sampling_buffer: "
- "realloc_sampling_buffer failed with rc=%i\n", rc);
+ debug_sprintf_event(sfdbg, 4, "%s: "
+ "realloc_sampling_buffer failed with rc %i\n",
+ __func__, rc);
} else
debug_sprintf_event(sfdbg, 4,
- "alloc_sampling_buffer: tear=%p dear=%p\n",
- sfb->sdbt, (void *) *sfb->sdbt);
+ "%s: tear %#lx dear %#lx\n", __func__,
+ (unsigned long)sfb->sdbt, (unsigned long)*sfb->sdbt);
return rc;
}
@@ -370,28 +373,33 @@
static int allocate_buffers(struct cpu_hw_sf *cpuhw, struct hw_perf_event *hwc)
{
- unsigned long n_sdb, freq, factor;
+ unsigned long n_sdb, freq;
size_t sample_size;
/* Calculate sampling buffers using 4K pages
*
- * 1. Determine the sample data size which depends on the used
- * sampling functions, for example, basic-sampling or
- * basic-sampling with diagnostic-sampling.
+ * 1. The sampling size is 32 bytes for basic sampling. This size
+ * is the same for all machine types. Diagnostic
+ * sampling uses auxlilary data buffer setup which provides the
+ * memory for SDBs using linux common code auxiliary trace
+ * setup.
*
- * 2. Use the sampling frequency as input. The sampling buffer is
- * designed for almost one second. This can be adjusted through
- * the "factor" variable.
- * In any case, alloc_sampling_buffer() sets the Alert Request
+ * 2. Function alloc_sampling_buffer() sets the Alert Request
* Control indicator to trigger a measurement-alert to harvest
- * sample-data-blocks (sdb).
+ * sample-data-blocks (SDB). This is done per SDB. This
+ * measurement alert interrupt fires quick enough to handle
+ * one SDB, on very high frequency and work loads there might
+ * be 2 to 3 SBDs available for sample processing.
+ * Currently there is no need for setup alert request on every
+ * n-th page. This is counterproductive as one IRQ triggers
+ * a very high number of samples to be processed at one IRQ.
*
- * 3. Compute the number of sample-data-blocks and ensure a minimum
- * of CPUM_SF_MIN_SDB. Also ensure the upper limit does not
- * exceed a "calculated" maximum. The symbolic maximum is
- * designed for basic-sampling only and needs to be increased if
- * diagnostic-sampling is active.
- * See also the remarks for these symbolic constants.
+ * 3. Use the sampling frequency as input.
+ * Compute the number of SDBs and ensure a minimum
+ * of CPUM_SF_MIN_SDB. Depending on frequency add some more
+ * SDBs to handle a higher sampling rate.
+ * Use a minimum of CPUM_SF_MIN_SDB and allow for 100 samples
+ * (one SDB) for every 10000 HZ frequency increment.
*
* 4. Compute the number of sample-data-block-tables (SDBT) and
* ensure a minimum of CPUM_SF_MIN_SDBT (one table can manage up
@@ -399,10 +407,7 @@
*/
sample_size = sizeof(struct hws_basic_entry);
freq = sample_rate_to_freq(&cpuhw->qsi, SAMPL_RATE(hwc));
- factor = 1;
- n_sdb = DIV_ROUND_UP(freq, factor * ((PAGE_SIZE-64) / sample_size));
- if (n_sdb < CPUM_SF_MIN_SDB)
- n_sdb = CPUM_SF_MIN_SDB;
+ n_sdb = CPUM_SF_MIN_SDB + DIV_ROUND_UP(freq, 10000);
/* If there is already a sampling buffer allocated, it is very likely
* that the sampling facility is enabled too. If the event to be
@@ -417,8 +422,8 @@
return 0;
debug_sprintf_event(sfdbg, 3,
- "allocate_buffers: rate=%lu f=%lu sdb=%lu/%lu"
- " sample_size=%lu cpuhw=%p\n",
+ "%s: rate %lu f %lu sdb %lu/%lu"
+ " sample_size %lu cpuhw %p\n", __func__,
SAMPL_RATE(hwc), freq, n_sdb, sfb_max_limit(hwc),
sample_size, cpuhw);
@@ -478,8 +483,8 @@
if (num)
sfb_account_allocs(num, hwc);
- debug_sprintf_event(sfdbg, 5, "sfb: overflow: overflow=%llu ratio=%lu"
- " num=%lu\n", OVERFLOW_REG(hwc), ratio, num);
+ debug_sprintf_event(sfdbg, 5, "%s: overflow %llu ratio %lu num %lu\n",
+ __func__, OVERFLOW_REG(hwc), ratio, num);
OVERFLOW_REG(hwc) = 0;
}
@@ -517,16 +522,15 @@
*/
rc = realloc_sampling_buffer(sfb, num, GFP_ATOMIC);
if (rc)
- debug_sprintf_event(sfdbg, 5, "sfb: extend: realloc "
- "failed with rc=%i\n", rc);
+ debug_sprintf_event(sfdbg, 5, "%s: realloc failed with rc %i\n",
+ __func__, rc);
if (sfb_has_pending_allocs(sfb, hwc))
- debug_sprintf_event(sfdbg, 5, "sfb: extend: "
- "req=%lu alloc=%lu remaining=%lu\n",
- num, sfb->num_sdb - num_old,
+ debug_sprintf_event(sfdbg, 5, "%s: "
+ "req %lu alloc %lu remaining %lu\n",
+ __func__, num, sfb->num_sdb - num_old,
sfb_pending_allocs(sfb, hwc));
}
-
/* Number of perf events counting hardware events */
static atomic_t num_events;
@@ -552,20 +556,22 @@
err = sf_disable();
if (err)
pr_err("Switching off the sampling facility failed "
- "with rc=%i\n", err);
+ "with rc %i\n", err);
debug_sprintf_event(sfdbg, 5,
- "setup_pmc_cpu: initialized: cpuhw=%p\n", cpusf);
+ "%s: initialized: cpuhw %p\n", __func__,
+ cpusf);
break;
case PMC_RELEASE:
cpusf->flags &= ~PMU_F_RESERVED;
err = sf_disable();
if (err) {
pr_err("Switching off the sampling facility failed "
- "with rc=%i\n", err);
+ "with rc %i\n", err);
} else
deallocate_buffers(cpusf);
debug_sprintf_event(sfdbg, 5,
- "setup_pmc_cpu: released: cpuhw=%p\n", cpusf);
+ "%s: released: cpuhw %p\n", __func__,
+ cpusf);
break;
}
if (err)
@@ -610,13 +616,6 @@
hwc->sample_period = period;
hwc->last_period = hwc->sample_period;
local64_set(&hwc->period_left, hwc->sample_period);
-}
-
-static void hw_reset_registers(struct hw_perf_event *hwc,
- unsigned long *sdbt_origin)
-{
- /* (Re)set to first sample-data-block-table */
- TEAR_REG(hwc) = (unsigned long) sdbt_origin;
}
static unsigned long hw_limit_rate(const struct hws_qsi_info_block *si,
@@ -674,7 +673,7 @@
rcu_read_lock();
perf_prepare_sample(&header, data, event, regs);
- if (perf_output_begin(&handle, event, header.size))
+ if (perf_output_begin(&handle, data, event, header.size))
goto out;
/* Update the process ID (see also kernel/events/core.c) */
@@ -687,13 +686,88 @@
rcu_read_unlock();
}
+static unsigned long getrate(bool freq, unsigned long sample,
+ struct hws_qsi_info_block *si)
+{
+ unsigned long rate;
+
+ if (freq) {
+ rate = freq_to_sample_rate(si, sample);
+ rate = hw_limit_rate(si, rate);
+ } else {
+ /* The min/max sampling rates specifies the valid range
+ * of sample periods. If the specified sample period is
+ * out of range, limit the period to the range boundary.
+ */
+ rate = hw_limit_rate(si, sample);
+
+ /* The perf core maintains a maximum sample rate that is
+ * configurable through the sysctl interface. Ensure the
+ * sampling rate does not exceed this value. This also helps
+ * to avoid throttling when pushing samples with
+ * perf_event_overflow().
+ */
+ if (sample_rate_to_freq(si, rate) >
+ sysctl_perf_event_sample_rate) {
+ debug_sprintf_event(sfdbg, 1, "%s: "
+ "Sampling rate exceeds maximum "
+ "perf sample rate\n", __func__);
+ rate = 0;
+ }
+ }
+ return rate;
+}
+
+/* The sampling information (si) contains information about the
+ * min/max sampling intervals and the CPU speed. So calculate the
+ * correct sampling interval and avoid the whole period adjust
+ * feedback loop.
+ *
+ * Since the CPU Measurement sampling facility can not handle frequency
+ * calculate the sampling interval when frequency is specified using
+ * this formula:
+ * interval := cpu_speed * 1000000 / sample_freq
+ *
+ * Returns errno on bad input and zero on success with parameter interval
+ * set to the correct sampling rate.
+ *
+ * Note: This function turns off freq bit to avoid calling function
+ * perf_adjust_period(). This causes frequency adjustment in the common
+ * code part which causes tremendous variations in the counter values.
+ */
+static int __hw_perf_event_init_rate(struct perf_event *event,
+ struct hws_qsi_info_block *si)
+{
+ struct perf_event_attr *attr = &event->attr;
+ struct hw_perf_event *hwc = &event->hw;
+ unsigned long rate;
+
+ if (attr->freq) {
+ if (!attr->sample_freq)
+ return -EINVAL;
+ rate = getrate(attr->freq, attr->sample_freq, si);
+ attr->freq = 0; /* Don't call perf_adjust_period() */
+ SAMPL_FLAGS(hwc) |= PERF_CPUM_SF_FREQ_MODE;
+ } else {
+ rate = getrate(attr->freq, attr->sample_period, si);
+ if (!rate)
+ return -EINVAL;
+ }
+ attr->sample_period = rate;
+ SAMPL_RATE(hwc) = rate;
+ hw_init_period(hwc, SAMPL_RATE(hwc));
+ debug_sprintf_event(sfdbg, 4, "%s: cpu %d period %#llx freq %d,%#lx\n",
+ __func__, event->cpu, event->attr.sample_period,
+ event->attr.freq, SAMPLE_FREQ_MODE(hwc));
+ return 0;
+}
+
static int __hw_perf_event_init(struct perf_event *event)
{
struct cpu_hw_sf *cpuhw;
struct hws_qsi_info_block si;
struct perf_event_attr *attr = &event->attr;
struct hw_perf_event *hwc = &event->hw;
- unsigned long rate;
int cpu, err;
/* Reserve CPU-measurement sampling facility */
@@ -741,6 +815,12 @@
goto out;
}
+ if (si.ribm & CPU_MF_SF_RIBM_NOTAV) {
+ pr_warn("CPU Measurement Facility sampling is temporarily not available\n");
+ err = -EBUSY;
+ goto out;
+ }
+
/* Always enable basic sampling */
SAMPL_FLAGS(hwc) = PERF_CPUM_SF_BASIC_MODE;
@@ -759,43 +839,9 @@
if (attr->config1 & PERF_CPUM_SF_FULL_BLOCKS)
SAMPL_FLAGS(hwc) |= PERF_CPUM_SF_FULL_BLOCKS;
- /* The sampling information (si) contains information about the
- * min/max sampling intervals and the CPU speed. So calculate the
- * correct sampling interval and avoid the whole period adjust
- * feedback loop.
- */
- rate = 0;
- if (attr->freq) {
- if (!attr->sample_freq) {
- err = -EINVAL;
- goto out;
- }
- rate = freq_to_sample_rate(&si, attr->sample_freq);
- rate = hw_limit_rate(&si, rate);
- attr->freq = 0;
- attr->sample_period = rate;
- } else {
- /* The min/max sampling rates specifies the valid range
- * of sample periods. If the specified sample period is
- * out of range, limit the period to the range boundary.
- */
- rate = hw_limit_rate(&si, hwc->sample_period);
-
- /* The perf core maintains a maximum sample rate that is
- * configurable through the sysctl interface. Ensure the
- * sampling rate does not exceed this value. This also helps
- * to avoid throttling when pushing samples with
- * perf_event_overflow().
- */
- if (sample_rate_to_freq(&si, rate) >
- sysctl_perf_event_sample_rate) {
- err = -EINVAL;
- debug_sprintf_event(sfdbg, 1, "Sampling rate exceeds maximum perf sample rate\n");
- goto out;
- }
- }
- SAMPL_RATE(hwc) = rate;
- hw_init_period(hwc, SAMPL_RATE(hwc));
+ err = __hw_perf_event_init_rate(event, &si);
+ if (err)
+ goto out;
/* Initialize sample data overflow accounting */
hwc->extra_reg.reg = REG_OVERFLOW;
@@ -836,12 +882,21 @@
return err;
}
+static bool is_callchain_event(struct perf_event *event)
+{
+ u64 sample_type = event->attr.sample_type;
+
+ return sample_type & (PERF_SAMPLE_CALLCHAIN | PERF_SAMPLE_REGS_USER |
+ PERF_SAMPLE_STACK_USER);
+}
+
static int cpumsf_pmu_event_init(struct perf_event *event)
{
int err;
/* No support for taken branch sampling */
- if (has_branch_stack(event))
+ /* No support for callchain, stacks and registers */
+ if (has_branch_stack(event) || is_callchain_event(event))
return -EOPNOTSUPP;
switch (event->attr.type) {
@@ -867,7 +922,7 @@
/* Check online status of the CPU to which the event is pinned */
if (event->cpu >= 0 && !cpu_online(event->cpu))
- return -ENODEV;
+ return -ENODEV;
/* Force reset of idle/hv excludes regardless of what the
* user requested.
@@ -915,9 +970,10 @@
* buffer extents
*/
sfb_account_overflows(cpuhw, hwc);
- if (sfb_has_pending_allocs(&cpuhw->sfb, hwc))
- extend_sampling_buffer(&cpuhw->sfb, hwc);
+ extend_sampling_buffer(&cpuhw->sfb, hwc);
}
+ /* Rate may be adjusted with ioctl() */
+ cpuhw->lsctl.interval = SAMPL_RATE(&cpuhw->event->hw);
}
/* (Re)enable the PMU and sampling facility */
@@ -927,7 +983,7 @@
err = lsctl(&cpuhw->lsctl);
if (err) {
cpuhw->flags &= ~PMU_F_ENABLED;
- pr_err("Loading sampling controls failed: op=%i err=%i\n",
+ pr_err("Loading sampling controls failed: op %i err %i\n",
1, err);
return;
}
@@ -935,10 +991,11 @@
/* Load current program parameter */
lpp(&S390_lowcore.lpp);
- debug_sprintf_event(sfdbg, 6, "pmu_enable: es=%i cs=%i ed=%i cd=%i "
- "tear=%p dear=%p\n", cpuhw->lsctl.es, cpuhw->lsctl.cs,
- cpuhw->lsctl.ed, cpuhw->lsctl.cd,
- (void *) cpuhw->lsctl.tear, (void *) cpuhw->lsctl.dear);
+ debug_sprintf_event(sfdbg, 6, "%s: es %i cs %i ed %i cd %i "
+ "interval %#lx tear %#lx dear %#lx\n", __func__,
+ cpuhw->lsctl.es, cpuhw->lsctl.cs, cpuhw->lsctl.ed,
+ cpuhw->lsctl.cd, cpuhw->lsctl.interval,
+ cpuhw->lsctl.tear, cpuhw->lsctl.dear);
}
static void cpumsf_pmu_disable(struct pmu *pmu)
@@ -961,13 +1018,14 @@
err = lsctl(&inactive);
if (err) {
- pr_err("Loading sampling controls failed: op=%i err=%i\n",
+ pr_err("Loading sampling controls failed: op %i err %i\n",
2, err);
return;
}
/* Save state of TEAR and DEAR register contents */
- if (!qsi(&si)) {
+ err = qsi(&si);
+ if (!err) {
/* TEAR/DEAR values are valid only if the sampling facility is
* enabled. Note that cpumsf_pmu_disable() might be called even
* for a disabled sampling facility because cpumsf_pmu_enable()
@@ -978,8 +1036,8 @@
cpuhw->lsctl.dear = si.dear;
}
} else
- debug_sprintf_event(sfdbg, 3, "cpumsf_pmu_disable: "
- "qsi() failed with err=%i\n", err);
+ debug_sprintf_event(sfdbg, 3, "%s: qsi() failed with err %i\n",
+ __func__, err);
cpuhw->flags &= ~PMU_F_ENABLED;
}
@@ -1092,14 +1150,6 @@
local64_add(count, &event->count);
}
-static void debug_sample_entry(struct hws_basic_entry *sample,
- struct hws_trailer_entry *te)
-{
- debug_sprintf_event(sfdbg, 4, "hw_collect_samples: Found unknown "
- "sampling data entry: te->f=%i basic.def=%04x (%p)\n",
- te->f, sample->def, sample);
-}
-
/* hw_collect_samples() - Walk through a sample-data-block and collect samples
* @event: The perf event
* @sdbt: Sample-data-block table
@@ -1153,7 +1203,11 @@
/* Count discarded samples */
*overflow += 1;
} else {
- debug_sample_entry(sample, te);
+ debug_sprintf_event(sfdbg, 4,
+ "%s: Found unknown"
+ " sampling data entry: te->f %i"
+ " basic.def %#4x (%p)\n", __func__,
+ te->header.f, sample->def, sample);
/* Sample slot is not yet written or other record.
*
* This condition can occur if the buffer was reused
@@ -1164,7 +1218,7 @@
* that are not full. Stop processing if the first
* invalid format was detected.
*/
- if (!te->f)
+ if (!te->header.f)
break;
}
@@ -1172,6 +1226,16 @@
sample->def = 0;
sample++;
}
+}
+
+static inline __uint128_t __cdsg(__uint128_t *ptr, __uint128_t old, __uint128_t new)
+{
+ asm volatile(
+ " cdsg %[old],%[new],%[ptr]\n"
+ : [old] "+d" (old), [ptr] "+QS" (*ptr)
+ : [new] "d" (new)
+ : "memory", "cc");
+ return old;
}
/* hw_perf_event_update() - Process sampling buffer
@@ -1190,10 +1254,11 @@
*/
static void hw_perf_event_update(struct perf_event *event, int flush_all)
{
+ unsigned long long event_overflow, sampl_overflow, num_sdb;
+ union hws_trailer_header old, prev, new;
struct hw_perf_event *hwc = &event->hw;
struct hws_trailer_entry *te;
unsigned long *sdbt;
- unsigned long long event_overflow, sampl_overflow, num_sdb, te_flags;
int done;
/*
@@ -1213,25 +1278,25 @@
te = (struct hws_trailer_entry *) trailer_entry_ptr(*sdbt);
/* Leave loop if no more work to do (block full indicator) */
- if (!te->f) {
+ if (!te->header.f) {
done = 1;
if (!flush_all)
break;
}
/* Check the sample overflow count */
- if (te->overflow)
+ if (te->header.overflow)
/* Account sample overflows and, if a particular limit
* is reached, extend the sampling buffer.
* For details, see sfb_account_overflows().
*/
- sampl_overflow += te->overflow;
+ sampl_overflow += te->header.overflow;
/* Timestamps are valid for full sample-data-blocks only */
- debug_sprintf_event(sfdbg, 6, "hw_perf_event_update: sdbt=%p "
- "overflow=%llu timestamp=0x%llx\n",
- sdbt, te->overflow,
- (te->f) ? trailer_timestamp(te) : 0ULL);
+ debug_sprintf_event(sfdbg, 6, "%s: sdbt %#lx "
+ "overflow %llu timestamp %#llx\n",
+ __func__, (unsigned long)sdbt, te->header.overflow,
+ (te->header.f) ? trailer_timestamp(te) : 0ULL);
/* Collect all samples from a single sample-data-block and
* flag if an (perf) event overflow happened. If so, the PMU
@@ -1241,12 +1306,16 @@
num_sdb++;
/* Reset trailer (using compare-double-and-swap) */
+ /* READ_ONCE() 16 byte header */
+ prev.val = __cdsg(&te->header.val, 0, 0);
do {
- te_flags = te->flags & ~SDB_TE_BUFFER_FULL_MASK;
- te_flags |= SDB_TE_ALERT_REQ_MASK;
- } while (!cmpxchg_double(&te->flags, &te->overflow,
- te->flags, te->overflow,
- te_flags, 0ULL));
+ old.val = prev.val;
+ new.val = prev.val;
+ new.f = 0;
+ new.a = 1;
+ new.overflow = 0;
+ prev.val = __cdsg(&te->header.val, old.val, new.val);
+ } while (prev.val != old.val);
/* Advance to next sample-data-block */
sdbt++;
@@ -1284,9 +1353,11 @@
}
if (sampl_overflow || event_overflow)
- debug_sprintf_event(sfdbg, 4, "hw_perf_event_update: "
- "overflow stats: sample=%llu event=%llu\n",
- sampl_overflow, event_overflow);
+ debug_sprintf_event(sfdbg, 4, "%s: "
+ "overflows: sample %llu event %llu"
+ " total %llu num_sdb %llu\n",
+ __func__, sampl_overflow, event_overflow,
+ OVERFLOW_REG(hwc), num_sdb);
}
#define AUX_SDB_INDEX(aux, i) ((i) % aux->sfb.num_sdb)
@@ -1329,7 +1400,7 @@
range_scan = AUX_SDB_NUM_ALERT(aux);
for (i = 0, idx = aux->head; i < range_scan; i++, idx++) {
te = aux_sdb_trailer(aux, idx);
- if (!(te->flags & SDB_TE_BUFFER_FULL_MASK))
+ if (!te->header.f)
break;
}
/* i is num of SDBs which are full */
@@ -1337,9 +1408,10 @@
/* Remove alert indicators in the buffer */
te = aux_sdb_trailer(aux, aux->alert_mark);
- te->flags &= ~SDB_TE_ALERT_REQ_MASK;
+ te->header.a = 0;
- debug_sprintf_event(sfdbg, 6, "aux_output_end: collect %lx SDBs\n", i);
+ debug_sprintf_event(sfdbg, 6, "%s: SDBs %ld range %ld head %ld\n",
+ __func__, i, range_scan, aux->head);
}
/*
@@ -1372,14 +1444,18 @@
* SDBs between aux->head and aux->empty_mark are already ready
* for new data. range_scan is num of SDBs not within them.
*/
+ debug_sprintf_event(sfdbg, 6,
+ "%s: range %ld head %ld alert %ld empty %ld\n",
+ __func__, range, aux->head, aux->alert_mark,
+ aux->empty_mark);
if (range > AUX_SDB_NUM_EMPTY(aux)) {
range_scan = range - AUX_SDB_NUM_EMPTY(aux);
idx = aux->empty_mark + 1;
for (i = 0; i < range_scan; i++, idx++) {
te = aux_sdb_trailer(aux, idx);
- te->flags &= ~(SDB_TE_BUFFER_FULL_MASK |
- SDB_TE_ALERT_REQ_MASK);
- te->overflow = 0;
+ te->header.f = 0;
+ te->header.a = 0;
+ te->header.overflow = 0;
}
/* Save the position of empty SDBs */
aux->empty_mark = aux->head + range - 1;
@@ -1388,7 +1464,7 @@
/* Set alert indicator */
aux->alert_mark = aux->head + range/2 - 1;
te = aux_sdb_trailer(aux, aux->alert_mark);
- te->flags = te->flags | SDB_TE_ALERT_REQ_MASK;
+ te->header.a = 1;
/* Reset hardware buffer head */
head = AUX_SDB_INDEX(aux, aux->head);
@@ -1397,15 +1473,11 @@
cpuhw->lsctl.tear = base + offset * sizeof(unsigned long);
cpuhw->lsctl.dear = aux->sdb_index[head];
- debug_sprintf_event(sfdbg, 6, "aux_output_begin: "
- "head->alert_mark->empty_mark (num_alert, range)"
- "[%lx -> %lx -> %lx] (%lx, %lx) "
- "tear index %lx, tear %lx dear %lx\n",
+ debug_sprintf_event(sfdbg, 6, "%s: head %ld alert %ld empty %ld "
+ "index %ld tear %#lx dear %#lx\n", __func__,
aux->head, aux->alert_mark, aux->empty_mark,
- AUX_SDB_NUM_ALERT(aux), range,
head / CPUM_SF_SDB_PER_TABLE,
- cpuhw->lsctl.tear,
- cpuhw->lsctl.dear);
+ cpuhw->lsctl.tear, cpuhw->lsctl.dear);
return 0;
}
@@ -1419,14 +1491,17 @@
static bool aux_set_alert(struct aux_buffer *aux, unsigned long alert_index,
unsigned long long *overflow)
{
- unsigned long long orig_overflow, orig_flags, new_flags;
+ union hws_trailer_header old, prev, new;
struct hws_trailer_entry *te;
te = aux_sdb_trailer(aux, alert_index);
+ /* READ_ONCE() 16 byte header */
+ prev.val = __cdsg(&te->header.val, 0, 0);
do {
- orig_flags = te->flags;
- *overflow = orig_overflow = te->overflow;
- if (orig_flags & SDB_TE_BUFFER_FULL_MASK) {
+ old.val = prev.val;
+ new.val = prev.val;
+ *overflow = old.overflow;
+ if (old.f) {
/*
* SDB is already set by hardware.
* Abort and try to set somewhere
@@ -1434,10 +1509,10 @@
*/
return false;
}
- new_flags = orig_flags | SDB_TE_ALERT_REQ_MASK;
- } while (!cmpxchg_double(&te->flags, &te->overflow,
- orig_flags, orig_overflow,
- new_flags, 0ULL));
+ new.a = 1;
+ new.overflow = 0;
+ prev.val = __cdsg(&te->header.val, old.val, new.val);
+ } while (prev.val != old.val);
return true;
}
@@ -1466,10 +1541,14 @@
static bool aux_reset_buffer(struct aux_buffer *aux, unsigned long range,
unsigned long long *overflow)
{
- unsigned long long orig_overflow, orig_flags, new_flags;
- unsigned long i, range_scan, idx;
+ unsigned long i, range_scan, idx, idx_old;
+ union hws_trailer_header old, prev, new;
+ unsigned long long orig_overflow;
struct hws_trailer_entry *te;
+ debug_sprintf_event(sfdbg, 6, "%s: range %ld head %ld alert %ld "
+ "empty %ld\n", __func__, range, aux->head,
+ aux->alert_mark, aux->empty_mark);
if (range <= AUX_SDB_NUM_EMPTY(aux))
/*
* No need to scan. All SDBs in range are marked as empty.
@@ -1492,26 +1571,32 @@
* indicator fall into this range, set it.
*/
range_scan = range - AUX_SDB_NUM_EMPTY(aux);
- idx = aux->empty_mark + 1;
+ idx_old = idx = aux->empty_mark + 1;
for (i = 0; i < range_scan; i++, idx++) {
te = aux_sdb_trailer(aux, idx);
+ /* READ_ONCE() 16 byte header */
+ prev.val = __cdsg(&te->header.val, 0, 0);
do {
- orig_flags = te->flags;
- orig_overflow = te->overflow;
- new_flags = orig_flags & ~SDB_TE_BUFFER_FULL_MASK;
+ old.val = prev.val;
+ new.val = prev.val;
+ orig_overflow = old.overflow;
+ new.f = 0;
+ new.overflow = 0;
if (idx == aux->alert_mark)
- new_flags |= SDB_TE_ALERT_REQ_MASK;
+ new.a = 1;
else
- new_flags &= ~SDB_TE_ALERT_REQ_MASK;
- } while (!cmpxchg_double(&te->flags, &te->overflow,
- orig_flags, orig_overflow,
- new_flags, 0ULL));
+ new.a = 0;
+ prev.val = __cdsg(&te->header.val, old.val, new.val);
+ } while (prev.val != old.val);
*overflow += orig_overflow;
}
/* Update empty_mark to new position */
aux->empty_mark = aux->head + range - 1;
+ debug_sprintf_event(sfdbg, 6, "%s: range_scan %ld idx %ld..%ld "
+ "empty %ld\n", __func__, range_scan, idx_old,
+ idx - 1, aux->empty_mark);
return true;
}
@@ -1533,8 +1618,9 @@
/* Inform user space new data arrived */
size = AUX_SDB_NUM_ALERT(aux) << PAGE_SHIFT;
+ debug_sprintf_event(sfdbg, 6, "%s: #alert %ld\n", __func__,
+ size >> PAGE_SHIFT);
perf_aux_output_end(handle, size);
- num_sdb = aux->sfb.num_sdb;
num_sdb = aux->sfb.num_sdb;
while (!done) {
@@ -1544,7 +1630,9 @@
pr_err("The AUX buffer with %lu pages for the "
"diagnostic-sampling mode is full\n",
num_sdb);
- debug_sprintf_event(sfdbg, 1, "AUX buffer used up\n");
+ debug_sprintf_event(sfdbg, 1,
+ "%s: AUX buffer used up\n",
+ __func__);
break;
}
if (WARN_ON_ONCE(!aux))
@@ -1566,24 +1654,24 @@
size = range << PAGE_SHIFT;
perf_aux_output_end(&cpuhw->handle, size);
pr_err("Sample data caused the AUX buffer with %lu "
- "pages to overflow\n", num_sdb);
- debug_sprintf_event(sfdbg, 1, "head %lx range %lx "
- "overflow %llx\n",
+ "pages to overflow\n", aux->sfb.num_sdb);
+ debug_sprintf_event(sfdbg, 1, "%s: head %ld range %ld "
+ "overflow %lld\n", __func__,
aux->head, range, overflow);
} else {
size = AUX_SDB_NUM_ALERT(aux) << PAGE_SHIFT;
perf_aux_output_end(&cpuhw->handle, size);
- debug_sprintf_event(sfdbg, 6, "head %lx alert %lx "
+ debug_sprintf_event(sfdbg, 6, "%s: head %ld alert %ld "
"already full, try another\n",
+ __func__,
aux->head, aux->alert_mark);
}
}
if (done)
- debug_sprintf_event(sfdbg, 6, "aux_reset_buffer: "
- "[%lx -> %lx -> %lx] (%lx, %lx)\n",
- aux->head, aux->alert_mark, aux->empty_mark,
- AUX_SDB_NUM_ALERT(aux), range);
+ debug_sprintf_event(sfdbg, 6, "%s: head %ld alert %ld "
+ "empty %ld\n", __func__, aux->head,
+ aux->alert_mark, aux->empty_mark);
}
/*
@@ -1606,8 +1694,7 @@
kfree(aux->sdb_index);
kfree(aux);
- debug_sprintf_event(sfdbg, 4, "aux_buffer_free: free "
- "%lu SDBTs\n", num_sdbt);
+ debug_sprintf_event(sfdbg, 4, "%s: SDBTs %lu\n", __func__, num_sdbt);
}
static void aux_sdb_init(unsigned long sdb)
@@ -1665,7 +1752,7 @@
sfb = &aux->sfb;
/* Allocate sdbt_index for fast reference */
- n_sdbt = (nr_pages + CPUM_SF_SDB_PER_TABLE - 1) / CPUM_SF_SDB_PER_TABLE;
+ n_sdbt = DIV_ROUND_UP(nr_pages, CPUM_SF_SDB_PER_TABLE);
aux->sdbt_index = kmalloc_array(n_sdbt, sizeof(void *), GFP_KERNEL);
if (!aux->sdbt_index)
goto no_sdbt_index;
@@ -1715,8 +1802,7 @@
*/
aux->empty_mark = sfb->num_sdb - 1;
- debug_sprintf_event(sfdbg, 4, "aux_buffer_setup: setup %lu SDBTs"
- " and %lu SDBs\n",
+ debug_sprintf_event(sfdbg, 4, "%s: SDBTs %lu SDBs %lu\n", __func__,
sfb->num_sdbt, sfb->num_sdb);
return aux;
@@ -1737,6 +1823,44 @@
static void cpumsf_pmu_read(struct perf_event *event)
{
/* Nothing to do ... updates are interrupt-driven */
+}
+
+/* Check if the new sampling period/freqeuncy is appropriate.
+ *
+ * Return non-zero on error and zero on passed checks.
+ */
+static int cpumsf_pmu_check_period(struct perf_event *event, u64 value)
+{
+ struct hws_qsi_info_block si;
+ unsigned long rate;
+ bool do_freq;
+
+ memset(&si, 0, sizeof(si));
+ if (event->cpu == -1) {
+ if (qsi(&si))
+ return -ENODEV;
+ } else {
+ /* Event is pinned to a particular CPU, retrieve the per-CPU
+ * sampling structure for accessing the CPU-specific QSI.
+ */
+ struct cpu_hw_sf *cpuhw = &per_cpu(cpu_hw_sf, event->cpu);
+
+ si = cpuhw->qsi;
+ }
+
+ do_freq = !!SAMPLE_FREQ_MODE(&event->hw);
+ rate = getrate(do_freq, value, &si);
+ if (!rate)
+ return -EINVAL;
+
+ event->attr.sample_period = rate;
+ SAMPL_RATE(&event->hw) = rate;
+ hw_init_period(&event->hw, SAMPL_RATE(&event->hw));
+ debug_sprintf_event(sfdbg, 4, "%s:"
+ " cpu %d value %#llx period %#llx freq %d\n",
+ __func__, event->cpu, value,
+ event->attr.sample_period, do_freq);
+ return 0;
}
/* Activate sampling control.
@@ -1810,7 +1934,7 @@
if (!SAMPL_DIAG_MODE(&event->hw)) {
cpuhw->lsctl.tear = (unsigned long) cpuhw->sfb.sdbt;
cpuhw->lsctl.dear = *(unsigned long *) cpuhw->sfb.sdbt;
- hw_reset_registers(&event->hw, cpuhw->sfb.sdbt);
+ TEAR_REG(&event->hw) = (unsigned long) cpuhw->sfb.sdbt;
}
/* Ensure sampling functions are in the disabled state. If disabled,
@@ -1865,10 +1989,30 @@
CPUMF_EVENT_ATTR(SF, SF_CYCLES_BASIC, PERF_EVENT_CPUM_SF);
CPUMF_EVENT_ATTR(SF, SF_CYCLES_BASIC_DIAG, PERF_EVENT_CPUM_SF_DIAG);
-static struct attribute *cpumsf_pmu_events_attr[] = {
- CPUMF_EVENT_PTR(SF, SF_CYCLES_BASIC),
- NULL,
- NULL,
+/* Attribute list for CPU_SF.
+ *
+ * The availablitiy depends on the CPU_MF sampling facility authorization
+ * for basic + diagnositic samples. This is determined at initialization
+ * time by the sampling facility device driver.
+ * If the authorization for basic samples is turned off, it should be
+ * also turned off for diagnostic sampling.
+ *
+ * During initialization of the device driver, check the authorization
+ * level for diagnostic sampling and installs the attribute
+ * file for diagnostic sampling if necessary.
+ *
+ * For now install a placeholder to reference all possible attributes:
+ * SF_CYCLES_BASIC and SF_CYCLES_BASIC_DIAG.
+ * Add another entry for the final NULL pointer.
+ */
+enum {
+ SF_CYCLES_BASIC_ATTR_IDX = 0,
+ SF_CYCLES_BASIC_DIAG_ATTR_IDX,
+ SF_CYCLES_ATTR_MAX
+};
+
+static struct attribute *cpumsf_pmu_events_attr[SF_CYCLES_ATTR_MAX + 1] = {
+ [SF_CYCLES_BASIC_ATTR_IDX] = CPUMF_EVENT_PTR(SF, SF_CYCLES_BASIC)
};
PMU_FORMAT_ATTR(event, "config:0-63");
@@ -1882,10 +2026,12 @@
.name = "events",
.attrs = cpumsf_pmu_events_attr,
};
+
static struct attribute_group cpumsf_pmu_format_group = {
.name = "format",
.attrs = cpumsf_pmu_format_attr,
};
+
static const struct attribute_group *cpumsf_pmu_attr_groups[] = {
&cpumsf_pmu_events_group,
&cpumsf_pmu_format_group,
@@ -1908,6 +2054,8 @@
.setup_aux = aux_buffer_setup,
.free_aux = aux_buffer_free,
+
+ .check_period = cpumsf_pmu_check_period,
};
static void cpumf_measurement_alert(struct ext_code ext_code,
@@ -1941,7 +2089,8 @@
/* Report measurement alerts only for non-PRA codes */
if (alert != CPU_MF_INT_SF_PRA)
- debug_sprintf_event(sfdbg, 6, "measurement alert: 0x%x\n", alert);
+ debug_sprintf_event(sfdbg, 6, "%s: alert %#x\n", __func__,
+ alert);
/* Sampling authorization change request */
if (alert & CPU_MF_INT_SF_SACA)
@@ -1962,6 +2111,7 @@
sf_disable();
}
}
+
static int cpusf_pmu_setup(unsigned int cpu, int flags)
{
/* Ignore the notification if no events are scheduled on the PMU.
@@ -2018,7 +2168,7 @@
sfb_set_limits(min, max);
pr_info("The sampling buffer limits have changed to: "
- "min=%lu max=%lu (diag=x%lu)\n",
+ "min %lu max %lu (diag %lu)\n",
CPUM_SF_MIN_SDB, CPUM_SF_MAX_SDB, CPUM_SF_SDB_DIAG_FACTOR);
return 0;
}
@@ -2036,7 +2186,7 @@
static void __init pr_cpumsf_err(unsigned int reason)
{
pr_err("Sampling facility support for perf is not available: "
- "reason=%04x\n", reason);
+ "reason %#x\n", reason);
}
static int __init init_cpum_sampling_pmu(void)
@@ -2063,7 +2213,10 @@
if (si.ad) {
sfb_set_limits(CPUM_SF_MIN_SDB, CPUM_SF_MAX_SDB);
- cpumsf_pmu_events_attr[1] =
+ /* Sampling of diagnostic data authorized,
+ * install event into attribute list of PMU device.
+ */
+ cpumsf_pmu_events_attr[SF_CYCLES_BASIC_DIAG_ATTR_IDX] =
CPUMF_EVENT_PTR(SF, SF_CYCLES_BASIC_DIAG);
}
@@ -2096,5 +2249,6 @@
out:
return err;
}
+
arch_initcall(init_cpum_sampling_pmu);
core_param(cpum_sfb_size, CPUM_SF_MAX_SDB, sfb_size, 0644);
--
Gitblit v1.6.2