Hi Suzuki,

在 3/23/2023 7:51 PM, Suzuki K Poulose 写道:
On 23/03/2023 06:03, Tao Zhang wrote:
Read the DSB element size from the device tree. Set the register
bit that controls the DSB element size of the corresponding port.

Signed-off-by: Tao Zhang <quic_taozha@quicinc.com>
---
  drivers/hwtracing/coresight/coresight-tpda.c | 58 ++++++++++++++++++++++++++++
  drivers/hwtracing/coresight/coresight-tpda.h |  4 ++
  2 files changed, 62 insertions(+)

diff --git a/drivers/hwtracing/coresight/coresight-tpda.c b/drivers/hwtracing/coresight/coresight-tpda.c
index f712e11..8dcfc4a 100644
--- a/drivers/hwtracing/coresight/coresight-tpda.c
+++ b/drivers/hwtracing/coresight/coresight-tpda.c
@@ -21,6 +21,47 @@
    DEFINE_CORESIGHT_DEVLIST(tpda_devs, "tpda");
  +/* Search and read element data size from the TPDM node in
+ * the devicetree. Each input port of TPDA is connected to
+ * a TPDM. Different TPDM supports different types of dataset,
+ * and some may support more than one type of dataset.
+ * Parameter "inport" is used to pass in the input port number
+ * of TPDA, and it is set to 0 in the recursize call.
+ * Parameter "parent" is used to pass in the original call.
+ */

I am still not clear why we need to do this recursively ?

Some TPDMs are not directly output connected to the TPDAs. So here I

use a recursive method to check from the TPDA input port until I find

the connected TPDM.

Do you have a better suggestion besides a recursive method?


+static int tpda_set_element_size(struct tpda_drvdata *drvdata,
+               struct coresight_device *csdev, int inport, bool parent)

Please could we renamse csdev => tpda_dev

Since this is a recursively called function, this Coresight device is not

necessarily TPDA, it can be other Coresight device.


+{
+    static int nr_inport;
+    int i;
+    struct coresight_device *in_csdev;

similarly tpdm_dev ?
Same as above, this variable may not necessarily be a TPDM.

Could we not add a check here to see if the dsb_esize[inport] is already
set and then bail out, reading this over and over ?

I will update this in the next patch series.
+
+    if (inport > (TPDA_MAX_INPORTS - 1))
+        return -EINVAL;
+
+    if (parent)
+        nr_inport = inport;
+
+    for (i = 0; i < csdev->pdata->nr_inconns; i++) {
+        in_csdev = csdev->pdata->in_conns[i].remote_dev;

Please note, the names of the structure field might change in the
next version of James' series
Got it. I will keep an eye out for the James' patch series.

+        if (!in_csdev)
+            break;
+
+        if (parent)
+            if (csdev->pdata->in_conns[i].port != inport)
+                continue;
+
+        if (in_csdev && strstr(dev_name(&in_csdev->dev), "tpdm")) {

Isn't there a better way to distinguish a device to be TPDM ? May be we
could even add a source_sub_type - SOURCE_TPDM instead of using
SOURCE_OTHERS ? Do you expect other sources to be connected to TPDA?
e.g., STMs ?

I can add "SOURCE_TPDM" as a source_sub_type, but SOURCE_OTHERS needs

to be kept since the other Coresight component we will upstream later may

need it.


+            of_property_read_u32(in_csdev->dev.parent->of_node,
+                    "qcom,dsb-element-size", &drvdata->dsb_esize[nr_inport]);
+            break;
+        }
+        tpda_set_element_size(drvdata, in_csdev, 0, false);

What is the point of this ? Is this for covering the a TPDA connected to
another TPDA ?

e.g., { TPDM0, TPDM1 } -> TPDA0 -> TPDA1 ?

A TPDM may not connect to the TPDA directly, for example,

TPDM0 ->FUNNEL0->FUNNEL1->TPDA0

And many TPDMs can connect to one TPDA, one input port on TPDA only has

one TPDM connected. Therefore, we use a recursive method to find the TPDM

corresponding to the input port of TPDA.


And you want to figure out the DSB size of TPDM0 when you want to enable
TPDA1 ? How do you choose between that size of TPDM0 vs TPDM1 ?

Please add a proper documentation for this function ? If TPDA0 is in the
the path, it should have been enabled before you reach TPDA1. Thus,
the dsb_esize array must have been initialised for TPDA0 and thus, you
could simply read it from the dsb_esize[] of TPDA0.
You could always look at the device_type and sub_type to detect a
TPDA{0} connected into TPDA{1}
TPDA will not connect to another TPDA in our design.

+    }
+
+    return 0;
+}
+
  /* Settings pre enabling port control register */
  static void tpda_enable_pre_port(struct tpda_drvdata *drvdata)
  {
@@ -37,6 +78,18 @@ static void tpda_enable_port(struct tpda_drvdata *drvdata, int port)
      u32 val;
        val = readl_relaxed(drvdata->base + TPDA_Pn_CR(port));
+    /*
+     * Configure aggregator port n DSB data set element size
+     * Set the bit to 0 if the size is 32
+     * Set the bit to 1 if the size is 64
+     */
+    if (drvdata->dsb_esize[port] == 32)
+        val &= ~TPDA_Pn_CR_DSBSIZE;
+    else if (drvdata->dsb_esize[port] == 64)
+        val |= TPDA_Pn_CR_DSBSIZE;
+    else
+        dev_err(drvdata->dev,
+            "DSB data size input from port[%d] is invalid\n", port);

WARN_ON_ONCE() and abort the enable opration ? Or say, "fallback to 32bit or 64bit" if one of them is a safer option ? Please don't
leave it unknown.
I will update this according to your advice in the next patch series.

      /* Enable the port */
      val |= TPDA_Pn_CR_ENA;
      writel_relaxed(val, drvdata->base + TPDA_Pn_CR(port));
@@ -57,6 +110,11 @@ static void __tpda_enable(struct tpda_drvdata *drvdata, int port)
  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_set_element_size(drvdata, csdev, inport, true);
+    if (ret)
+        return ret;
        spin_lock(&drvdata->spinlock);
      if (atomic_read(&csdev->refcnt[inport]) == 0)
diff --git a/drivers/hwtracing/coresight/coresight-tpda.h b/drivers/hwtracing/coresight/coresight-tpda.h
index 0399678..9ec5870 100644
--- a/drivers/hwtracing/coresight/coresight-tpda.h
+++ b/drivers/hwtracing/coresight/coresight-tpda.h
@@ -10,6 +10,8 @@
  #define TPDA_Pn_CR(n)        (0x004 + (n * 4))
  /* Aggregator port enable bit */
  #define TPDA_Pn_CR_ENA        BIT(0)
+/* Aggregator port DSB data set element size bit */
+#define TPDA_Pn_CR_DSBSIZE        BIT(8)
    #define TPDA_MAX_INPORTS    32
  @@ -23,6 +25,7 @@
   * @csdev:      component vitals needed by the framework.
   * @spinlock:   lock for the drvdata value.
   * @enable:     enable status of the component.
+ * @dsb_esize:   DSB element size

Please state, must be 32 or 64.
Sure, I will update in the next patch series.

   */
  struct tpda_drvdata {
      void __iomem        *base;
@@ -30,6 +33,7 @@ struct tpda_drvdata {
      struct coresight_device    *csdev;
      spinlock_t        spinlock;
      u8            atid;
+    u32            dsb_esize[TPDA_MAX_INPORTS];

Couldn't this be u8 ?

I will change it to u8 in the next patch series.


Tao



Suzuki

  };
    #endif  /* _CORESIGHT_CORESIGHT_TPDA_H */