coresight: tmc-etf: Add support for CPU-wide trace scenarios
authorMathieu Poirier <mathieu.poirier@linaro.org>
Thu, 25 Apr 2019 19:53:09 +0000 (13:53 -0600)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 25 Apr 2019 20:00:18 +0000 (22:00 +0200)
This patch adds support for CPU-wide trace scenarios by making sure that
only the sources monitoring the same process have access to a common sink.
Because the sink is shared between sources, the first source to use the
sink switches it on while the last one does the cleanup.  Any attempt to
modify the HW is overlooked for as long as more than one source is using
a sink.

Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Tested-by: Leo Yan <leo.yan@linaro.org>
Tested-by: Robert Walker <robert.walker@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/hwtracing/coresight/coresight-tmc-etf.c

index 1df1f8fade71b161cba16d25d5659a2488d195a5..2527b5d3b65e96203f7a7c0387a34d4175ccb6c9 100644 (file)
@@ -223,6 +223,7 @@ out:
 static int tmc_enable_etf_sink_perf(struct coresight_device *csdev, void *data)
 {
        int ret = 0;
+       pid_t pid;
        unsigned long flags;
        struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
        struct perf_output_handle *handle = data;
@@ -233,18 +234,39 @@ static int tmc_enable_etf_sink_perf(struct coresight_device *csdev, void *data)
                if (drvdata->reading)
                        break;
                /*
-                * In Perf mode there can be only one writer per sink.  There
-                * is also no need to continue if the ETB/ETF is already
-                * operated from sysFS.
+                * No need to continue if the ETB/ETF is already operated
+                * from sysFS.
                 */
-               if (drvdata->mode != CS_MODE_DISABLED)
+               if (drvdata->mode == CS_MODE_SYSFS) {
+                       ret = -EBUSY;
                        break;
+               }
+
+               /* Get a handle on the pid of the process to monitor */
+               pid = task_pid_nr(handle->event->owner);
+
+               if (drvdata->pid != -1 && drvdata->pid != pid) {
+                       ret = -EBUSY;
+                       break;
+               }
 
                ret = tmc_set_etf_buffer(csdev, handle);
                if (ret)
                        break;
+
+               /*
+                * No HW configuration is needed if the sink is already in
+                * use for this session.
+                */
+               if (drvdata->pid == pid) {
+                       atomic_inc(csdev->refcnt);
+                       break;
+               }
+
                ret  = tmc_etb_enable_hw(drvdata);
                if (!ret) {
+                       /* Associate with monitored process. */
+                       drvdata->pid = pid;
                        drvdata->mode = CS_MODE_PERF;
                        atomic_inc(csdev->refcnt);
                }
@@ -300,6 +322,8 @@ static int tmc_disable_etf_sink(struct coresight_device *csdev)
        /* Complain if we (somehow) got out of sync */
        WARN_ON_ONCE(drvdata->mode == CS_MODE_DISABLED);
        tmc_etb_disable_hw(drvdata);
+       /* Dissociate from monitored process. */
+       drvdata->pid = -1;
        drvdata->mode = CS_MODE_DISABLED;
 
        spin_unlock_irqrestore(&drvdata->spinlock, flags);
@@ -414,7 +438,7 @@ static unsigned long tmc_update_etf_buffer(struct coresight_device *csdev,
        u32 *buf_ptr;
        u64 read_ptr, write_ptr;
        u32 status;
-       unsigned long offset, to_read, flags;
+       unsigned long offset, to_read = 0, flags;
        struct cs_buffers *buf = sink_config;
        struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
 
@@ -426,6 +450,11 @@ static unsigned long tmc_update_etf_buffer(struct coresight_device *csdev,
                return 0;
 
        spin_lock_irqsave(&drvdata->spinlock, flags);
+
+       /* Don't do anything if another tracer is using this sink */
+       if (atomic_read(csdev->refcnt) != 1)
+               goto out;
+
        CS_UNLOCK(drvdata->base);
 
        tmc_flush_and_stop(drvdata);
@@ -519,6 +548,7 @@ static unsigned long tmc_update_etf_buffer(struct coresight_device *csdev,
                to_read = buf->nr_pages << PAGE_SHIFT;
        }
        CS_LOCK(drvdata->base);
+out:
        spin_unlock_irqrestore(&drvdata->spinlock, flags);
 
        return to_read;