With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com --- .../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) { - int trace_id; + int trace_id, ret = 0;
- /* - * This will allocate a trace ID to the cpu, - * or return the one currently allocated. - * The trace id function has its own lock - */ - trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu); - if (IS_VALID_CS_TRACE_ID(trace_id)) - drvdata->trcid = (u8)trace_id; - else - dev_err(&drvdata->csdev->dev, - "Failed to allocate trace ID for %s on CPU%d\n", - dev_name(&drvdata->csdev->dev), drvdata->cpu); - return trace_id; + if (!drvdata->trcid) { + /* + * This will allocate a trace ID to the cpu, + * or return the one currently allocated. + * The trace id function has its own lock + */ + trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu); + if (IS_VALID_CS_TRACE_ID(trace_id)) + drvdata->trcid = (u8)trace_id; + else { + ret = -EINVAL; + dev_err(&drvdata->csdev->dev, + "Failed to allocate trace ID for %s on CPU%d\n", + dev_name(&drvdata->csdev->dev), drvdata->cpu); + } + } else + ret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid); + + return ret; }
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) { + drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu); }
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + int val; + struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent); + + val = drvdata->trcid; + return sysfs_emit(buf, "0x%x\n", val); +} + +static ssize_t traceid_store(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t size) +{ + int ret; + unsigned long val; + struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent); + + ret = kstrtoul(buf, 16, &val); + if (ret) + return ret; + + if (!drvdata->csdev->enable) + drvdata->trcid = val; + + return size; +} +static DEVICE_ATTR_RW(traceid); + static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr, + &dev_attr_traceid.attr, NULL, };
diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{ + int traceid, ret = 0; + + if (!drvdata->traceid) { + traceid = coresight_trace_id_get_system_id(); + if (traceid < 0) + return traceid; + + drvdata->traceid = traceid; + } else + ret = coresight_trace_id_set_system_id(drvdata->traceid); + + return ret; + +} + static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val; + int ret; + struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
if (mode != CS_MODE_SYSFS) @@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
+ ret = stm_configure_trace_id(drvdata); + if (ret) + return ret; + pm_runtime_get_sync(csdev->dev.parent);
spin_lock(&drvdata->spinlock); @@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
+ + coresight_trace_id_put_system_id(drvdata->traceid); /* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without @@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock); + drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);
@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid); + +static ssize_t traceid_store(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t size) +{ + int ret; + unsigned long val; + struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent); + + ret = kstrtoul(buf, 16, &val); + if (ret) + return ret; + + if (!IS_VALID_CS_TRACE_ID(val)) { + dev_err(&drvdata->csdev->dev, "Invalid trace id\n"); + return -EINVAL; + } + + if (!drvdata->csdev->enable) + drvdata->traceid = val; + else + dev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n", + __func__); + + return size; +} +static DEVICE_ATTR_RW(traceid);
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) { - int ret, trace_id; + int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL; @@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
- trace_id = coresight_trace_id_get_system_id(); - if (trace_id < 0) { - ret = trace_id; - goto cs_unregister; - } - drvdata->traceid = (u8)trace_id; - pm_runtime_put(&adev->dev);
dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;
-cs_unregister: - coresight_unregister(drvdata->csdev); - stm_unregister: stm_unregister_device(&drvdata->stm); return ret; @@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
- coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev);
stm_unregister_device(&drvdata->stm); diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{ + int traceid, ret; + /* + * TPDA must has a unique atid. This atid can uniquely + * identify the TPDM trace source connected to the TPDA. + * The TPDMs which are connected to same TPDA share the + * same trace-id. When TPDA does packetization, different + * port will have unique channel number for decoding. + */ + if (!drvdata->traceid) { + traceid = coresight_trace_id_get_system_id(); + if (traceid < 0) + return traceid; + + drvdata->traceid = traceid; + } else + ret = coresight_trace_id_set_system_id(drvdata->traceid); + + return ret; +} + /* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID; - val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid); + val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR); + }
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port) @@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base); + }
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); + int ret; + + ret = tpda_configure_trace_id(drvdata); + if (ret) + return ret;
spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0) @@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
+ coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock); + + drvdata->traceid = 0; + if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);
@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev, + struct device_attribute *attr, char *buf) { - int atid; - /* - * TPDA must has a unique atid. This atid can uniquely - * identify the TPDM trace source connected to the TPDA. - * The TPDMs which are connected to same TPDA share the - * same trace-id. When TPDA does packetization, different - * port will have unique channel number for decoding. - */ - atid = coresight_trace_id_get_system_id(); - if (atid < 0) - return atid; + int val; + struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
- drvdata->atid = atid; - return 0; + val = drvdata->traceid; + return sysfs_emit(buf, "0x%x\n", val); }
-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t size) { int ret; + unsigned long val; + struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent); + + ret = kstrtoul(buf, 16, &val); + if (ret) + return ret; + + if (!IS_VALID_CS_TRACE_ID(val)) { + dev_err(drvdata->dev, "Invalid trace id\n"); + return -EINVAL; + } + + if (!drvdata->csdev->enable) + drvdata->traceid = val; + else + dev_err(drvdata->dev, "Device must be enabled! %s\n", __func__); + + return size; +} +static DEVICE_ATTR_RW(traceid); + +static struct attribute *coresight_tpda_attrs[] = { + &dev_attr_traceid.attr, + NULL, +}; + +static const struct attribute_group coresight_tpda_group = { + .attrs = coresight_tpda_attrs, +}; + +static const struct attribute_group *coresight_tpda_groups[] = { + &coresight_tpda_group, + NULL, +}; + +static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{ + drvdata->traceid = 0; +} + +static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
- ret = tpda_init_default_data(drvdata); - if (ret) - return ret; + tpda_init_default_data(drvdata);
desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name) @@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base); + desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev); @@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
- coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev); }
diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@ * @dev: The device entity associated to this component. * @csdev: component vitals needed by the framework. * @spinlock: lock for the drvdata value. - * @enable: enable status of the component. + * @traceid: the atid value of TPDA. */ struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock; - u8 atid; + u8 traceid; };
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */ diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{ + if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) + return -EINVAL; + if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) + return -EINVAL; + set_bit(id, id_map->used_ids); + + DUMP_ID_MAP(id_map); + + return 0; +} + static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) @@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{ + unsigned long flags; + + spin_lock_irqsave(&id_map_lock, flags); + + if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) { + spin_unlock_irqrestore(&id_map_lock, flags); + return -EINVAL; + } + + if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) { + spin_unlock_irqrestore(&id_map_lock, flags); + return -EINVAL; + } + + set_bit(id, id_map->used_ids); + + /* allocate the new id to the cpu */ + atomic_set(&per_cpu(cpu_id, cpu), id); + + cpumask_clear_cpu(cpu, &cpu_id_release_pending); + clear_bit(id, id_map->pend_rel_ids); + + spin_unlock_irqrestore(&id_map_lock, flags); + DUMP_ID_CPU(cpu, id); + DUMP_ID_MAP(id_map); + + return 0; +} + static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{ + return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default); +} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id); + int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{ + return coresight_trace_id_set(id, &id_map_default); +} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id); + int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
+ /* Allocate and release IDs for a single default trace ID map */
+/** + * Set the CoreSight Trace Id for the CPU. + * + * Set CoreSight Trace Id associated with the CPU. + * + * @cpu: The CPU index for the id. + * @id: Coresight Trace ID value. + * + * return: 0 if set successfully or -EINVAL if fail to set. + */ +int coresight_trace_id_set_cpu_id(int cpu, int id); + /** * Read and optionally allocate a CoreSight trace ID and associate with a CPU. * @@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/** + * Set trace id for a system component. + * + * Set the trace id if system component needs a static id for the trace. + * + * @id: value of trace ID. + * + * return: 0 if set successfully or -EINVAL if fail to set. + */ +int coresight_trace_id_set_system_id(int id); + /** * Allocate a CoreSight trace ID for a system component. *
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;}
This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);}
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,};
diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);
Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);stm_unregister: stm_unregister_device(&drvdata->stm); return ret; @@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);
redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port) @@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf){
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);}
-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size){ int ret;
unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);}
diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@
- @dev: The device entity associated to this component.
- @csdev: component vitals needed by the framework.
- @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
*/
- @traceid: the atid value of TPDA.
struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */ diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) @@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
/* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
/**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
/**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
Hi Mike,
On 4/11/2023 10:09 PM, Mike Leach wrote:
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
This change is to add trace id support for the following two cases: 1. There are more than 127 sources like there are 200 sources which need the trace id. then 127 is not enough to allocate the ids for all the trace sources. 2. HW component that has a specific trace id hard-coded into the HW that cannot be changed.
To support cases above, I think we need to make following changes: Each driver can call trace id set funnction (coresight_trace_id_set_system_id/coresight_trace_id_set_cpu_id) to set the id. Set the id when enable the source and release the id when disable the source.
Please help to provide your comments.
Thanks Jinlong Mao
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);- stm_unregister: stm_unregister_device(&drvdata->stm); return ret;
@@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
@@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@
- @dev: The device entity associated to this component.
- @csdev: component vitals needed by the framework.
- @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
*/ struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
- @traceid: the atid value of TPDA.
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */
diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
- static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))
@@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
- /* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
- /**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
- /**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
Hi
On Wed, 12 Apr 2023 at 03:53, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/11/2023 10:09 PM, Mike Leach wrote:
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
This change is to add trace id support for the following two cases:
- There are more than 127 sources like there are 200 sources which need
the trace id. then 127 is not enough to allocate the ids for all the trace sources.
Agreed - and that is precisely why we introduced the dynamic allocation patch set.
(The actual number is slightly less than 127 as IDs 0x0 and 0x70 to 0x7F are reserved.)
However, there is no need for your changes in this case. The dynamic allocator will allocate up to the maximum IDs, then will return an error. The design is such that the allocator controls the available IDs and the requestor must always use the assigned ID. Devices should never be setting their own ID (other than the case of trace Id hardwired into a device below).
IDs are returned to the pool either:-
a) in the perf case, once the perf session is complete. We must maintain a CPU / Trace ID relationship throughout the perf session to correctly associate trace data in the perf file to the relevant CPU for decode. perf can cause multiple enable / disable events per CPU during a single trace session.
b) in the sysfs case, once the sysfs device is reset. (in the ETM4 case for example when the sysfs reset file is written). The reason for this is that the user can enable the ETM via sysfs, which will then allocate an ID. This may be read by the user who will assume it is constant throughout the lifetime of a trace session (which again may be multiple enable / disable events). This maintains the original functionality of the previous static allocation mechanism. The only change with the dynamic mechanism is that the ID is released on reset.
Your change merely replicates the functionality of the allocator that assigns an ID to the CPU, and always maintains the ID for that CPU throughout the trace session. You are calling the read/allocate call, then setting the same ID each time we enable - this is precisely the exisiting functionality - with the added race condition that you may be setting an ID already in use if you have previously released it. This is not desireable.
Moreover, this potentially breaks the perf case which requries that the IDs remain constant for a given CPU for the duration of the perf sessions, which may result in many enable/disable cycles for a CPU/ETM. Any allocation system must work for both the perf and sysfs cases.
So any trace session that has sources in access of the number of available IDs, must choose carefully the sources to be involved in the trace session. This is an unavoidable limitation of trace where multiple sources trace into a single trace buffer. This problem is significantly reduced in v9 / ETE cases where each ETE / cpu combination has its own TRBE buffer and does not need a trace ID.
- HW component that has a specific trace id hard-coded into the HW that
cannot be changed.
Then this must be managed with by allocating an ID at the device probe time, and release once the driver is unloaded. Otherwise there could be a potential clash if the hard coded value is taken by another device while the hard coded device is not in use, or is enabled before the hard coded device.
If this is actually needed on your system, then the preferred method would be a coresight_reserve_id(id) API function, to ensure that the hardcoded ID is removed from the pool of IDs as soon as possible., and unreserved on device unload.
It is not advisable to have many hard coded devices in a system as this will deplete the pool of available trace IDs.
To support cases above, I think we need to make following changes: Each driver can call trace id set funnction (coresight_trace_id_set_system_id/coresight_trace_id_set_cpu_id) to set the id. Set the id when enable the source and release the id when disable the source.
See my explanation above - IDs are only released when the trace session is complete, as some sessions can result in multiple enable/disable cycles where the ID/ CPU is maintained as constant.
Best Regards
Mike
Please help to provide your comments.
Thanks Jinlong Mao
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);- stm_unregister: stm_unregister_device(&drvdata->stm); return ret;
@@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
@@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@
- @dev: The device entity associated to this component.
- @csdev: component vitals needed by the framework.
- @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
*/ struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
- @traceid: the atid value of TPDA.
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */
diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
- static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))
@@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
- /* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
- /**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
- /**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
Hi Mike,
On 4/12/2023 4:45 PM, Mike Leach wrote:
Hi
On Wed, 12 Apr 2023 at 03:53, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/11/2023 10:09 PM, Mike Leach wrote:
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
This change is to add trace id support for the following two cases:
- There are more than 127 sources like there are 200 sources which need
the trace id. then 127 is not enough to allocate the ids for all the trace sources.
Agreed - and that is precisely why we introduced the dynamic allocation patch set.
(The actual number is slightly less than 127 as IDs 0x0 and 0x70 to 0x7F are reserved.)
However, there is no need for your changes in this case. The dynamic allocator will allocate up to the maximum IDs, then will return an error. The design is such that the allocator controls the available IDs and the requestor must always use the assigned ID. Devices should never be setting their own ID (other than the case of trace Id hardwired into a device below).
IDs are returned to the pool either:-
a) in the perf case, once the perf session is complete. We must maintain a CPU / Trace ID relationship throughout the perf session to correctly associate trace data in the perf file to the relevant CPU for decode. perf can cause multiple enable / disable events per CPU during a single trace session.
b) in the sysfs case, once the sysfs device is reset. (in the ETM4 case for example when the sysfs reset file is written). The reason for this is that the user can enable the ETM via sysfs, which will then allocate an ID. This may be read by the user who will assume it is constant throughout the lifetime of a trace session (which again may be multiple enable / disable events). This maintains the original functionality of the previous static allocation mechanism. The only change with the dynamic mechanism is that the ID is released on reset.
Your change merely replicates the functionality of the allocator that assigns an ID to the CPU, and always maintains the ID for that CPU throughout the trace session. You are calling the read/allocate call, then setting the same ID each time we enable - this is precisely the exisiting functionality - with the added race condition that you may be setting an ID already in use if you have previously released it. This is not desireable.
Moreover, this potentially breaks the perf case which requries that the IDs remain constant for a given CPU for the duration of the perf sessions, which may result in many enable/disable cycles for a CPU/ETM. Any allocation system must work for both the perf and sysfs cases.
So any trace session that has sources in access of the number of available IDs, must choose carefully the sources to be involved in the trace session. This is an unavoidable limitation of trace where multiple sources trace into a single trace buffer. This problem is significantly reduced in v9 / ETE cases where each ETE / cpu combination has its own TRBE buffer and does not need a trace ID.
- HW component that has a specific trace id hard-coded into the HW that
cannot be changed.
Then this must be managed with by allocating an ID at the device probe time, and release once the driver is unloaded. Otherwise there could be a potential clash if the hard coded value is taken by another device while the hard coded device is not in use, or is enabled before the hard coded device.
If this is actually needed on your system, then the preferred method would be a coresight_reserve_id(id) API function, to ensure that the hardcoded ID is removed from the pool of IDs as soon as possible., and unreserved on device unload.
It is not advisable to have many hard coded devices in a system as this will deplete the pool of available trace IDs.
To support cases above, I think we need to make following changes: Each driver can call trace id set funnction (coresight_trace_id_set_system_id/coresight_trace_id_set_cpu_id) to set the id. Set the id when enable the source and release the id when disable the source.
See my explanation above - IDs are only released when the trace session is complete, as some sessions can result in multiple enable/disable cycles where the ID/ CPU is maintained as constant.
Best Regards
Mike
Thanks for your clarification.
We also have case that decode tool can set the trace id of source by sysfs node so that tool can control the source's traceid.
Best Regards Jinlong Mao
Please help to provide your comments.
Thanks Jinlong Mao
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);- stm_unregister: stm_unregister_device(&drvdata->stm); return ret;
@@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
@@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@ * @dev: The device entity associated to this component. * @csdev: component vitals needed by the framework. * @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
- @traceid: the atid value of TPDA. */
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */
diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
- static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))
@@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
- /* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
- /**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
- /**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
Hi,
On Wed, 12 Apr 2023 at 10:30, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/12/2023 4:45 PM, Mike Leach wrote:
Hi
On Wed, 12 Apr 2023 at 03:53, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/11/2023 10:09 PM, Mike Leach wrote:
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
This change is to add trace id support for the following two cases:
- There are more than 127 sources like there are 200 sources which need
the trace id. then 127 is not enough to allocate the ids for all the trace sources.
Agreed - and that is precisely why we introduced the dynamic allocation patch set.
(The actual number is slightly less than 127 as IDs 0x0 and 0x70 to 0x7F are reserved.)
However, there is no need for your changes in this case. The dynamic allocator will allocate up to the maximum IDs, then will return an error. The design is such that the allocator controls the available IDs and the requestor must always use the assigned ID. Devices should never be setting their own ID (other than the case of trace Id hardwired into a device below).
IDs are returned to the pool either:-
a) in the perf case, once the perf session is complete. We must maintain a CPU / Trace ID relationship throughout the perf session to correctly associate trace data in the perf file to the relevant CPU for decode. perf can cause multiple enable / disable events per CPU during a single trace session.
b) in the sysfs case, once the sysfs device is reset. (in the ETM4 case for example when the sysfs reset file is written). The reason for this is that the user can enable the ETM via sysfs, which will then allocate an ID. This may be read by the user who will assume it is constant throughout the lifetime of a trace session (which again may be multiple enable / disable events). This maintains the original functionality of the previous static allocation mechanism. The only change with the dynamic mechanism is that the ID is released on reset.
Your change merely replicates the functionality of the allocator that assigns an ID to the CPU, and always maintains the ID for that CPU throughout the trace session. You are calling the read/allocate call, then setting the same ID each time we enable - this is precisely the exisiting functionality - with the added race condition that you may be setting an ID already in use if you have previously released it. This is not desireable.
Moreover, this potentially breaks the perf case which requries that the IDs remain constant for a given CPU for the duration of the perf sessions, which may result in many enable/disable cycles for a CPU/ETM. Any allocation system must work for both the perf and sysfs cases.
So any trace session that has sources in access of the number of available IDs, must choose carefully the sources to be involved in the trace session. This is an unavoidable limitation of trace where multiple sources trace into a single trace buffer. This problem is significantly reduced in v9 / ETE cases where each ETE / cpu combination has its own TRBE buffer and does not need a trace ID.
- HW component that has a specific trace id hard-coded into the HW that
cannot be changed.
Then this must be managed with by allocating an ID at the device probe time, and release once the driver is unloaded. Otherwise there could be a potential clash if the hard coded value is taken by another device while the hard coded device is not in use, or is enabled before the hard coded device.
If this is actually needed on your system, then the preferred method would be a coresight_reserve_id(id) API function, to ensure that the hardcoded ID is removed from the pool of IDs as soon as possible., and unreserved on device unload.
It is not advisable to have many hard coded devices in a system as this will deplete the pool of available trace IDs.
To support cases above, I think we need to make following changes: Each driver can call trace id set funnction (coresight_trace_id_set_system_id/coresight_trace_id_set_cpu_id) to set the id. Set the id when enable the source and release the id when disable the source.
See my explanation above - IDs are only released when the trace session is complete, as some sessions can result in multiple enable/disable cycles where the ID/ CPU is maintained as constant.
Best Regards
Mike
Thanks for your clarification.
We also have case that decode tool can set the trace id of source by sysfs node so that tool can control the source's traceid.
Unfortunately this is not compatible with the allocator mechanisms detailed above - due to potential races introduced by any API to allow the devices to set their own Trace IDs.
Your tool needs to read the IDs allocated to devices via sysfs.
Regards
Mike
Best Regards Jinlong Mao
Please help to provide your comments.
Thanks Jinlong Mao
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr, @@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr, @@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);- stm_unregister: stm_unregister_device(&drvdata->stm); return ret;
@@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) { @@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
@@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@ * @dev: The device entity associated to this component. * @csdev: component vitals needed by the framework. * @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
- @traceid: the atid value of TPDA. */
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */
diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
- static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))
@@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags; @@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */
+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default); @@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default); diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
- /* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
- /**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
- /**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
Hi Mike,
On 4/12/2023 5:50 PM, Mike Leach wrote:
Hi,
On Wed, 12 Apr 2023 at 10:30, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/12/2023 4:45 PM, Mike Leach wrote:
Hi
On Wed, 12 Apr 2023 at 03:53, Jinlong Mao quic_jinlmao@quicinc.com wrote:
Hi Mike,
On 4/11/2023 10:09 PM, Mike Leach wrote:
Hi,
On Mon, 10 Apr 2023 at 14:40, Mao Jinlong quic_jinlmao@quicinc.com wrote:
With a growing number of trace sources, the number of source could be greater than 127 in future. There will be lack of ids. There could be also trace id reserved for some trace HW. With this change, trace id will be only configured when enable the source. Trace id will be dynamically allocated when traceid of driver data is not set when enable source. Once traceid of driver data is set when enable source, the traceid will be set as the specific value.
Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com
This change is to add trace id support for the following two cases:
- There are more than 127 sources like there are 200 sources which need
the trace id. then 127 is not enough to allocate the ids for all the trace sources.
Agreed - and that is precisely why we introduced the dynamic allocation patch set.
(The actual number is slightly less than 127 as IDs 0x0 and 0x70 to 0x7F are reserved.)
However, there is no need for your changes in this case. The dynamic allocator will allocate up to the maximum IDs, then will return an error. The design is such that the allocator controls the available IDs and the requestor must always use the assigned ID. Devices should never be setting their own ID (other than the case of trace Id hardwired into a device below).
IDs are returned to the pool either:-
a) in the perf case, once the perf session is complete. We must maintain a CPU / Trace ID relationship throughout the perf session to correctly associate trace data in the perf file to the relevant CPU for decode. perf can cause multiple enable / disable events per CPU during a single trace session.
b) in the sysfs case, once the sysfs device is reset. (in the ETM4 case for example when the sysfs reset file is written). The reason for this is that the user can enable the ETM via sysfs, which will then allocate an ID. This may be read by the user who will assume it is constant throughout the lifetime of a trace session (which again may be multiple enable / disable events). This maintains the original functionality of the previous static allocation mechanism. The only change with the dynamic mechanism is that the ID is released on reset.
Your change merely replicates the functionality of the allocator that assigns an ID to the CPU, and always maintains the ID for that CPU throughout the trace session. You are calling the read/allocate call, then setting the same ID each time we enable - this is precisely the exisiting functionality - with the added race condition that you may be setting an ID already in use if you have previously released it. This is not desireable.
Moreover, this potentially breaks the perf case which requries that the IDs remain constant for a given CPU for the duration of the perf sessions, which may result in many enable/disable cycles for a CPU/ETM. Any allocation system must work for both the perf and sysfs cases.
So any trace session that has sources in access of the number of available IDs, must choose carefully the sources to be involved in the trace session. This is an unavoidable limitation of trace where multiple sources trace into a single trace buffer. This problem is significantly reduced in v9 / ETE cases where each ETE / cpu combination has its own TRBE buffer and does not need a trace ID.
- HW component that has a specific trace id hard-coded into the HW that
cannot be changed.
Then this must be managed with by allocating an ID at the device probe time, and release once the driver is unloaded. Otherwise there could be a potential clash if the hard coded value is taken by another device while the hard coded device is not in use, or is enabled before the hard coded device.
If this is actually needed on your system, then the preferred method would be a coresight_reserve_id(id) API function, to ensure that the hardcoded ID is removed from the pool of IDs as soon as possible., and unreserved on device unload.
It is not advisable to have many hard coded devices in a system as this will deplete the pool of available trace IDs.
To support cases above, I think we need to make following changes: Each driver can call trace id set funnction (coresight_trace_id_set_system_id/coresight_trace_id_set_cpu_id) to set the id. Set the id when enable the source and release the id when disable the source.
See my explanation above - IDs are only released when the trace session is complete, as some sessions can result in multiple enable/disable cycles where the ID/ CPU is maintained as constant.
Best Regards
Mike
Thanks for your clarification.
We also have case that decode tool can set the trace id of source by sysfs node so that tool can control the source's traceid.
Unfortunately this is not compatible with the allocator mechanisms detailed above - due to potential races introduced by any API to allow the devices to set their own Trace IDs.
Your tool needs to read the IDs allocated to devices via sysfs.
Regards
Mike
It is possible that user loses the trace id info during a long time test with lots of enbaling/disabling source and device reboot. We can use a boot up service to configure the trace id of some special source for test. Then the trace id of the source will be static during enabling/disable and reboot.
Allocating the id dynamically in code and assigning the trace id by sysfs node can be coexisting.
Best Regards Jinlong Mao
Best Regards Jinlong Mao
Please help to provide your comments.
Thanks Jinlong Mao
.../coresight/coresight-etm4x-core.c | 35 +++--- .../coresight/coresight-etm4x-sysfs.c | 30 +++++ drivers/hwtracing/coresight/coresight-stm.c | 67 ++++++++--- drivers/hwtracing/coresight/coresight-tpda.c | 107 ++++++++++++++---- drivers/hwtracing/coresight/coresight-tpda.h | 4 +- .../hwtracing/coresight/coresight-trace-id.c | 56 +++++++++ .../hwtracing/coresight/coresight-trace-id.h | 24 ++++ 7 files changed, 274 insertions(+), 49 deletions(-)diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 1ea8f173cca0..8f2e4d2b0fc2 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -233,25 +233,32 @@ static int etm4_cpu_id(struct coresight_device *csdev)
int etm4_read_alloc_trace_id(struct etmv4_drvdata *drvdata) {
int trace_id;
int trace_id, ret = 0;
/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;elsedev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);return trace_id;
if (!drvdata->trcid) {/** This will allocate a trace ID to the cpu,* or return the one currently allocated.* The trace id function has its own lock*/trace_id = coresight_trace_id_get_cpu_id(drvdata->cpu);if (IS_VALID_CS_TRACE_ID(trace_id))drvdata->trcid = (u8)trace_id;else {ret = -EINVAL;dev_err(&drvdata->csdev->dev,"Failed to allocate trace ID for %s on CPU%d\n",dev_name(&drvdata->csdev->dev), drvdata->cpu);}} elseret = coresight_trace_id_set_cpu_id(drvdata->cpu, drvdata->trcid);return ret;This change is redundant. coresight_trace_id_get_cpu_id() will allocate a trace id to the cpu if none is currently mapped, otherwise it will return the currently mapped trace id. There is no need to remap the id to the cpu every time.
void etm4_release_trace_id(struct etmv4_drvdata *drvdata) {
drvdata->trcid = 0; coresight_trace_id_put_cpu_id(drvdata->cpu);diff --git a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c index 5e62aa40ecd0..bd342e63868c 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-sysfs.c @@ -2335,6 +2335,35 @@ static ssize_t ts_source_show(struct device *dev, } static DEVICE_ATTR_RO(ts_source);
+static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)+{
int val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);val = drvdata->trcid;return sysfs_emit(buf, "0x%x\n", val);+}
This is also redundant. It replicates the trctraceid_show function
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct etmv4_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!drvdata->csdev->enable)drvdata->trcid = val;return size;+} +static DEVICE_ATTR_RW(traceid);
Users are not permitted to set trace id values manually in this way. This will potentially set an ID that is already used for another device.
static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_nr_pe_cmp.attr, &dev_attr_nr_addr_cmp.attr,@@ -2390,6 +2419,7 @@ static struct attribute *coresight_etmv4_attrs[] = { &dev_attr_vmid_masks.attr, &dev_attr_cpu.attr, &dev_attr_ts_source.attr,
&dev_attr_traceid.attr, NULL,diff --git a/drivers/hwtracing/coresight/coresight-stm.c b/drivers/hwtracing/coresight/coresight-stm.c index 66a614c5492c..1291f5f39ab1 100644 --- a/drivers/hwtracing/coresight/coresight-stm.c +++ b/drivers/hwtracing/coresight/coresight-stm.c @@ -192,10 +192,29 @@ static void stm_enable_hw(struct stm_drvdata *drvdata) CS_LOCK(drvdata->base); }
+static int stm_configure_trace_id(struct stm_drvdata *drvdata) +{
int traceid, ret = 0;if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);return ret;+}
again the mapping function remain with the id reserved till the put id is called, so no need to set the id every time once allocated.
static int stm_enable(struct coresight_device *csdev, struct perf_event *event, u32 mode) { u32 val;
int ret;struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); if (mode != CS_MODE_SYSFS)@@ -207,6 +226,10 @@ static int stm_enable(struct coresight_device *csdev, if (val) return -EBUSY;
ret = stm_configure_trace_id(drvdata);if (ret)return ret;pm_runtime_get_sync(csdev->dev.parent); spin_lock(&drvdata->spinlock);@@ -261,6 +284,8 @@ static void stm_disable(struct coresight_device *csdev, struct stm_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent); struct csdev_access *csa = &csdev->access;
coresight_trace_id_put_system_id(drvdata->traceid);Enabling and releasing trace Ids in this way is potentially problematical if the device is enabled and disabled multiple times during a single trace recording session. Differnt IDs could be used.
We enable this ID at probe time as we expect there to be only a single STM on any system - so allocating an ID for the lifetime of the driver is reasonable. If this change is required, then the sysfs model of allovate on read or enable, release on sysfs reset should be followed.
/* * For as long as the tracer isn't disabled another entity can't * change its status. As such we can read the status here without@@ -268,6 +293,7 @@ static void stm_disable(struct coresight_device *csdev, */ if (local_read(&drvdata->mode) == CS_MODE_SYSFS) { spin_lock(&drvdata->spinlock);
drvdata->traceid = 0; stm_disable_hw(drvdata); spin_unlock(&drvdata->spinlock);@@ -608,7 +634,33 @@ static ssize_t traceid_show(struct device *dev, val = drvdata->traceid; return sprintf(buf, "%#lx\n", val); } -static DEVICE_ATTR_RO(traceid);
+static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)+{
int ret;unsigned long val;struct stm_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(&drvdata->csdev->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(&drvdata->csdev->dev, "Device must be enabled! %s\n",__func__);return size;+} +static DEVICE_ATTR_RW(traceid);
Again do not do this - you are overriding the trace id mapping. Users are not permitted to set arbitrary trace id values in this way
static struct attribute *coresight_stm_attrs[] = { &dev_attr_hwevent_enable.attr,@@ -806,7 +858,7 @@ static void stm_init_generic_data(struct stm_drvdata *drvdata,
static int stm_probe(struct amba_device *adev, const struct amba_id *id) {
int ret, trace_id;
int ret; void __iomem *base; struct device *dev = &adev->dev; struct coresight_platform_data *pdata = NULL;@@ -890,22 +942,12 @@ static int stm_probe(struct amba_device *adev, const struct amba_id *id) goto stm_unregister; }
trace_id = coresight_trace_id_get_system_id();if (trace_id < 0) {ret = trace_id;goto cs_unregister;}drvdata->traceid = (u8)trace_id;pm_runtime_put(&adev->dev); dev_info(&drvdata->csdev->dev, "%s initialized\n", (char *)coresight_get_uci_data(id)); return 0;-cs_unregister:
coresight_unregister(drvdata->csdev);- stm_unregister: stm_unregister_device(&drvdata->stm); return ret;
@@ -915,7 +957,6 @@ static void stm_remove(struct amba_device *adev) { struct stm_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->traceid); coresight_unregister(drvdata->csdev); stm_unregister_device(&drvdata->stm);diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c index f712e112ecff..41f83a5de3f2 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.c +++ b/drivers/hwtracing/coresight/coresight-tpda.c @@ -21,6 +21,28 @@
DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");+static int tpda_configure_trace_id(struct tpda_drvdata *drvdata) +{
int traceid, ret;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/if (!drvdata->traceid) {traceid = coresight_trace_id_get_system_id();if (traceid < 0)return traceid;drvdata->traceid = traceid;} elseret = coresight_trace_id_set_system_id(drvdata->traceid);redundant call - system ids are reserved till put system id is called.
return ret;+}
See above comments get_system_id is sufficient. If you choose to do this on enable, then simply do not call the put till the trace session is finished.
You have a potential race here between devices using the system id calls trying to re-claim an id previously released.
/* Settings pre enabling port control register */ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata) {@@ -28,8 +50,9 @@ static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
val = readl_relaxed(drvdata->base + TPDA_CR); val &= ~TPDA_CR_ATID;
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->atid);
val |= FIELD_PREP(TPDA_CR_ATID, drvdata->traceid); writel_relaxed(val, drvdata->base + TPDA_CR);}
static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
@@ -52,11 +75,17 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port) tpda_enable_port(drvdata, port);
CS_LOCK(drvdata->base);
}
static int tpda_enable(struct coresight_device *csdev, int inport, int outport) { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
int ret;ret = tpda_configure_trace_id(drvdata);if (ret)return ret; spin_lock(&drvdata->spinlock); if (atomic_read(&csdev->refcnt[inport]) == 0)@@ -87,7 +116,11 @@ static void tpda_disable(struct coresight_device *csdev, int inport, { struct tpda_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
coresight_trace_id_put_system_id(drvdata->traceid); spin_lock(&drvdata->spinlock);drvdata->traceid = 0;if (atomic_dec_return(&csdev->refcnt[inport]) == 0) __tpda_disable(drvdata, inport);@@ -105,27 +138,63 @@ static const struct coresight_ops tpda_cs_ops = { .link_ops = &tpda_link_ops, };
-static int tpda_init_default_data(struct tpda_drvdata *drvdata) +static ssize_t traceid_show(struct device *dev,
struct device_attribute *attr, char *buf)
int atid;/** TPDA must has a unique atid. This atid can uniquely* identify the TPDM trace source connected to the TPDA.* The TPDMs which are connected to same TPDA share the* same trace-id. When TPDA does packetization, different* port will have unique channel number for decoding.*/atid = coresight_trace_id_get_system_id();if (atid < 0)return atid;
int val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);
drvdata->atid = atid;return 0;
val = drvdata->traceid;return sysfs_emit(buf, "0x%x\n", val);-static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +static ssize_t traceid_store(struct device *dev,
struct device_attribute *attr,const char *buf, size_t size)unsigned long val;struct tpda_drvdata *drvdata = dev_get_drvdata(dev->parent);ret = kstrtoul(buf, 16, &val);if (ret)return ret;if (!IS_VALID_CS_TRACE_ID(val)) {dev_err(drvdata->dev, "Invalid trace id\n");return -EINVAL;}if (!drvdata->csdev->enable)drvdata->traceid = val;elsedev_err(drvdata->dev, "Device must be enabled! %s\n", __func__);return size;+} +static DEVICE_ATTR_RW(traceid);
This may override the mapping from the trace id API - there is no guarantee that the value here will be available. In general setting trace IDs via sysfs is prohibited.
The CPU / ETM4 sysfs will allocate on traceid read, or enable, and then release the ID on sysfs reset on the device. This ensures that teh user can read a sysfs id value that will be retained over the trace session and viewable before or after the session is complete
+static struct attribute *coresight_tpda_attrs[] = {
&dev_attr_traceid.attr,NULL,+};
+static const struct attribute_group coresight_tpda_group = {
.attrs = coresight_tpda_attrs,+};
+static const struct attribute_group *coresight_tpda_groups[] = {
&coresight_tpda_group,NULL,+};
+static void tpda_init_default_data(struct tpda_drvdata *drvdata) +{
drvdata->traceid = 0;+}
+static int tpda_probe(struct amba_device *adev, const struct amba_id *id) +{ struct device *dev = &adev->dev; struct coresight_platform_data *pdata; struct tpda_drvdata *drvdata; @@ -151,9 +220,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id)
spin_lock_init(&drvdata->spinlock);
ret = tpda_init_default_data(drvdata);if (ret)return ret;
tpda_init_default_data(drvdata); desc.name = coresight_alloc_device_name(&tpda_devs, dev); if (!desc.name)@@ -164,6 +231,7 @@ static int tpda_probe(struct amba_device *adev, const struct amba_id *id) desc.pdata = adev->dev.platform_data; desc.dev = &adev->dev; desc.access = CSDEV_ACCESS_IOMEM(base);
desc.groups = coresight_tpda_groups; drvdata->csdev = coresight_register(&desc); if (IS_ERR(drvdata->csdev)) return PTR_ERR(drvdata->csdev);@@ -178,7 +246,6 @@ static void tpda_remove(struct amba_device *adev) { struct tpda_drvdata *drvdata = dev_get_drvdata(&adev->dev);
coresight_trace_id_put_system_id(drvdata->atid); coresight_unregister(drvdata->csdev);diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h index 0399678df312..5de6dc92f450 100644 --- a/drivers/hwtracing/coresight/coresight-tpda.h +++ b/drivers/hwtracing/coresight/coresight-tpda.h @@ -22,14 +22,14 @@ * @dev: The device entity associated to this component. * @csdev: component vitals needed by the framework. * @spinlock: lock for the drvdata value.
- @enable: enable status of the component.
- @traceid: the atid value of TPDA. */ struct tpda_drvdata { void __iomem *base; struct device *dev; struct coresight_device *csdev; spinlock_t spinlock;
u8 atid;
u8 traceid;};
#endif /* _CORESIGHT_CORESIGHT_TPDA_H */
diff --git a/drivers/hwtracing/coresight/coresight-trace-id.c b/drivers/hwtracing/coresight/coresight-trace-id.c index af5b4ef59cea..192626efcb01 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.c +++ b/drivers/hwtracing/coresight/coresight-trace-id.c @@ -110,6 +110,19 @@ static int coresight_trace_id_alloc_new_id(struct coresight_trace_id_map *id_map return id; }
+static int coresight_trace_id_set(int id, struct coresight_trace_id_map *id_map) +{
if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))return -EINVAL;if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id))return -EINVAL;set_bit(id, id_map->used_ids);DUMP_ID_MAP(id_map);return 0;+}
- static void coresight_trace_id_free(int id, struct coresight_trace_id_map *id_map) { if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id))
@@ -195,6 +208,37 @@ static int coresight_trace_id_map_get_cpu_id(int cpu, struct coresight_trace_id_ return id; }
+static int coresight_trace_id_map_set_cpu_id(int cpu, int id, struct coresight_trace_id_map *id_map) +{
unsigned long flags;spin_lock_irqsave(&id_map_lock, flags);if (WARN(!IS_VALID_CS_TRACE_ID(id), "Invalid Trace ID %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}if (WARN(test_bit(id, id_map->used_ids), "ID is already used: %d\n", id)) {spin_unlock_irqrestore(&id_map_lock, flags);return -EINVAL;}set_bit(id, id_map->used_ids);/* allocate the new id to the cpu */atomic_set(&per_cpu(cpu_id, cpu), id);cpumask_clear_cpu(cpu, &cpu_id_release_pending);clear_bit(id, id_map->pend_rel_ids);spin_unlock_irqrestore(&id_map_lock, flags);DUMP_ID_CPU(cpu, id);DUMP_ID_MAP(id_map);return 0;+}
This is redundant and simply replicates the existing code. cpu mapped IDs are retained for the full trace session, across multiple enable/disable operations. For perf, the IDs will be retained for the entire perf trace session, during which the etm can be enabled and disabled multiple times. For sysfs, the ids will be allocated on enable and retained till the user writes to the sysfs reset function in the ETM4X.
static void coresight_trace_id_map_put_cpu_id(int cpu, struct coresight_trace_id_map *id_map) { unsigned long flags;@@ -251,6 +295,12 @@ static void coresight_trace_id_map_put_system_id(struct coresight_trace_id_map *
/* API functions */+int coresight_trace_id_set_cpu_id(int cpu, int id) +{
return coresight_trace_id_map_set_cpu_id(cpu, id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_cpu_id);
Redundant API, Ids are allocated and persistent as described above.
int coresight_trace_id_get_cpu_id(int cpu) { return coresight_trace_id_map_get_cpu_id(cpu, &id_map_default);@@ -269,6 +319,12 @@ int coresight_trace_id_read_cpu_id(int cpu) } EXPORT_SYMBOL_GPL(coresight_trace_id_read_cpu_id);
+int coresight_trace_id_set_system_id(int id) +{
return coresight_trace_id_set(id, &id_map_default);+} +EXPORT_SYMBOL_GPL(coresight_trace_id_set_system_id);
This is unnecessary unless you have a device that has a specific trace id hardcoded into the device that cannot be changed.
If there is such a case then we will have to implement a reservation mechanism that reserves an ID to that specific device. To prevent race conditions this would have to occur at device probe time and be retained throughout the lifetime of the device driver.
If you need to retain system Ids across multiple enable / disable cycles for sysfs then follow the model in the ETM4x driver.
Regards
Mike
int coresight_trace_id_get_system_id(void) { return coresight_trace_id_map_get_system_id(&id_map_default);diff --git a/drivers/hwtracing/coresight/coresight-trace-id.h b/drivers/hwtracing/coresight/coresight-trace-id.h index 3797777d367e..5dab9a473266 100644 --- a/drivers/hwtracing/coresight/coresight-trace-id.h +++ b/drivers/hwtracing/coresight/coresight-trace-id.h @@ -61,8 +61,21 @@ struct coresight_trace_id_map { DECLARE_BITMAP(pend_rel_ids, CORESIGHT_TRACE_IDS_MAX); };
- /* Allocate and release IDs for a single default trace ID map */
+/**
- Set the CoreSight Trace Id for the CPU.
- Set CoreSight Trace Id associated with the CPU.
- @cpu: The CPU index for the id.
- @id: Coresight Trace ID value.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_cpu_id(int cpu, int id);
- /**
- Read and optionally allocate a CoreSight trace ID and associate with a CPU.
@@ -111,6 +124,17 @@ void coresight_trace_id_put_cpu_id(int cpu); */ int coresight_trace_id_read_cpu_id(int cpu);
+/**
- Set trace id for a system component.
- Set the trace id if system component needs a static id for the trace.
- @id: value of trace ID.
- return: 0 if set successfully or -EINVAL if fail to set.
- */
+int coresight_trace_id_set_system_id(int id);
- /**
- Allocate a CoreSight trace ID for a system component.
-- 2.39.0
-
Jinlong Mao -
Mao Jinlong
-
Mike Leach