Re: [PATCH v10 4/8] coresight: tmc: Enable panic sync handling

Hi Suzuki,

On 2024-10-17 at 17:42:21, Linu Cherian (lcherian@marvell.com) wrote:

Hi Suzuki,

On 2024-10-01 at 22:13:12, Suzuki K Poulose (suzuki.poulose@arm.com) wrote:

quoted

Hi Linu

On 16/09/2024 11:34, Linu Cherian wrote:

quoted

- Get reserved region from device tree node for metadata
- Define metadata format for TMC
- Add TMC ETR panic sync handler that syncs register snapshot
   to metadata region
- Add TMC ETF panic sync handler that syncs register snapshot
   to metadata region and internal SRAM to reserved trace buffer
   region.

The patch looks good overall. Some minor comments below.

quoted

Signed-off-by: Linu Cherian <redacted>
---
Changelog from v9:
- Add common helper function of_tmc_get_reserved_resource_by_name
   for better code reuse
- Inorder to keep the reserved buffer validity and crashdata validity
   independent, is_tmc_reserved_region_valid renamed to tmc_has_reserved_buffer
- drvdata->crash_tbuf renamed to drvdata->resrv_buf
- New fields added to crash metadata: version, ffcr, ffsr, mode
- Defined crashdata version with Major version 1, Minor version 0
- Set version while creating crashdata record
- Removed Reviewed-by tag due to the above changes
  .../hwtracing/coresight/coresight-tmc-core.c  | 14 ++++
  .../hwtracing/coresight/coresight-tmc-etf.c   | 77 ++++++++++++++++++
  .../hwtracing/coresight/coresight-tmc-etr.c   | 78 +++++++++++++++++++
  drivers/hwtracing/coresight/coresight-tmc.h   | 66 ++++++++++++++++
  4 files changed, 235 insertions(+)

diff --git a/drivers/hwtracing/coresight/coresight-tmc-core.c b/drivers/hwtracing/coresight/coresight-tmc-core.c
index 0764c21aba0f..54bf8ae2bff8 100644
--- a/drivers/hwtracing/coresight/coresight-tmc-core.c
+++ b/drivers/hwtracing/coresight/coresight-tmc-core.c

@@ -445,6 +445,20 @@ static void tmc_get_reserved_region(struct device *parent)
  	drvdata->resrv_buf.paddr = res.start;
  	drvdata->resrv_buf.size  = resource_size(&res);
+
+	if (of_tmc_get_reserved_resource_by_name(parent, "metadata", &res))
+		return;
+
+	drvdata->crash_mdata.vaddr = memremap(res.start,
+					       resource_size(&res),
+					       MEMREMAP_WC);
+	if (IS_ERR_OR_NULL(drvdata->crash_mdata.vaddr)) {
+		dev_err(parent, "Metadata memory mapping failed\n");
+		return;
+	}
+
+	drvdata->crash_mdata.paddr = res.start;
+	drvdata->crash_mdata.size  = resource_size(&res);
  }
  /* Detect and initialise the capabilities of a TMC ETR */

diff --git a/drivers/hwtracing/coresight/coresight-tmc-etf.c b/drivers/hwtracing/coresight/coresight-tmc-etf.c
index d4f641cd9de6..d77ec9307e98 100644
--- a/drivers/hwtracing/coresight/coresight-tmc-etf.c
+++ b/drivers/hwtracing/coresight/coresight-tmc-etf.c

@@ -590,6 +590,78 @@ static unsigned long tmc_update_etf_buffer(struct coresight_device *csdev,
  	return to_read;
  }
+static int tmc_panic_sync_etf(struct coresight_device *csdev)
+{
+	u32 val;
+	struct csdev_access *csa;
+	struct tmc_crash_metadata *mdata;
+	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
+
+	csa = &drvdata->csdev->access;
+	mdata = (struct tmc_crash_metadata *)drvdata->crash_mdata.vaddr;
+
+	/* Make sure we have valid reserved memory */
+	if (!tmc_has_reserved_buffer(drvdata) ||
+	    !tmc_has_crash_mdata_buffer(drvdata))
+		return 0;
+
+	tmc_crashdata_set_invalid(drvdata);
+
+	CS_UNLOCK(drvdata->base);
+
+	/* Proceed only if ETF is enabled or configured as sink */
+	val = readl(drvdata->base + TMC_CTL);
+	if (!(val & TMC_CTL_CAPT_EN))
+		goto out;
+

minor nit : Since the check below is "covered" by the same comment
above, please drop the extra line here to make it clear that "we check
for sink" by checking the "MODE == CIRCULAR_BUFFER".

Ack.

quoted

quoted

+	val = readl(drvdata->base + TMC_MODE);
+	if (val != TMC_MODE_CIRCULAR_BUFFER)
+		goto out;
+
+	val = readl(drvdata->base + TMC_FFSR);
+	/* Do manual flush and stop only if its not auto-stopped */
+	if (!(val & TMC_FFSR_FT_STOPPED)) {
+		dev_info(&csdev->dev,
+			 "%s: Triggering manual flush\n", __func__);

Please drop the ^^^ line. We don't want to do anything like that from a
panic callback.

Ack.

quoted

quoted

+		tmc_flush_and_stop(drvdata);
+	} else
+		tmc_wait_for_tmcready(drvdata);
+
+	/* Sync registers from hardware to metadata region */
+	mdata->sts = csdev_access_relaxed_read32(csa, TMC_STS);

Why are we using "csa" here and not for TMC_CTL etc ? Simply drop the "csa"
and use the raw reads like above. TMC doesn't have anyother modes
of access.

Okay.

quoted

quoted

+	mdata->mode = csdev_access_relaxed_read32(csa, TMC_MODE);
+	mdata->ffcr = csdev_access_relaxed_read32(csa, TMC_FFCR);
+	mdata->ffsr = csdev_access_relaxed_read32(csa, TMC_FFSR);
+	mdata->trace_paddr = drvdata->resrv_buf.paddr;
+
+	/* Sync Internal SRAM to reserved trace buffer region */
+	drvdata->buf = drvdata->resrv_buf.vaddr;
+	tmc_etb_dump_hw(drvdata);
+	/* Store as per RSZ register convention */
+	mdata->size = drvdata->len >> 2;
+	mdata->version = CS_CRASHDATA_VERSION;
+
+	/*
+	 * Make sure all previous writes are completed,
+	 * before we mark valid
+	 */
+	dsb(sy);

I don't think this matters much, as this would only be read by a
secondary kernel. In the worst case, you only need `dmb(ish)` to make
sure the writes are visible before valid is set to true.

Ack. Will change that.

quoted

quoted

+	mdata->valid = true;
+	/*
+	 * Below order need to maintained, since crc of metadata
+	 * is dependent on first
+	 */
+	mdata->crc32_tdata = find_crash_tracedata_crc(drvdata, mdata);
+	mdata->crc32_mdata = find_crash_metadata_crc(mdata);
+
+	tmc_disable_hw(drvdata);
+
+	dev_info(&csdev->dev, "%s: success\n", __func__);

Please no "prints" from a panic call back, unless it absolutely CRITICAL.

Ack.

quoted

quoted

+out:
+	CS_UNLOCK(drvdata->base);
+	return 0;
+}
+
  static const struct coresight_ops_sink tmc_etf_sink_ops = {
  	.enable		= tmc_enable_etf_sink,
  	.disable	= tmc_disable_etf_sink,

@@ -603,6 +675,10 @@ static const struct coresight_ops_link tmc_etf_link_ops = {
  	.disable	= tmc_disable_etf_link,
  };
+static const struct coresight_ops_panic tmc_etf_sync_ops = {
+	.sync		= tmc_panic_sync_etf,
+};
+
  const struct coresight_ops tmc_etb_cs_ops = {
  	.sink_ops	= &tmc_etf_sink_ops,
  };

@@ -610,6 +686,7 @@ const struct coresight_ops tmc_etb_cs_ops = {
  const struct coresight_ops tmc_etf_cs_ops = {
  	.sink_ops	= &tmc_etf_sink_ops,
  	.link_ops	= &tmc_etf_link_ops,
+	.panic_ops	= &tmc_etf_sync_ops,
  };
  int tmc_read_prepare_etb(struct tmc_drvdata *drvdata)

diff --git a/drivers/hwtracing/coresight/coresight-tmc-etr.c b/drivers/hwtracing/coresight/coresight-tmc-etr.c
index 8bca5b36334a..8228d7aaa361 100644
--- a/drivers/hwtracing/coresight/coresight-tmc-etr.c
+++ b/drivers/hwtracing/coresight/coresight-tmc-etr.c

@@ -1814,6 +1814,79 @@ static int tmc_disable_etr_sink(struct coresight_device *csdev)
  	return 0;
  }
+static int tmc_panic_sync_etr(struct coresight_device *csdev)
+{
+	u32 val;
+	struct csdev_access *csa;
+	struct tmc_crash_metadata *mdata;
+	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
+
+	csa = &drvdata->csdev->access;

As earlier, drop the csa.

Okay.

quoted

quoted

+	mdata = (struct tmc_crash_metadata *)drvdata->crash_mdata.vaddr;
+
+	if (!drvdata->etr_buf)
+		return 0;
+
+	/* Being in RESRV mode implies valid reserved memory as well */
+	if (drvdata->etr_buf->mode != ETR_MODE_RESRV)
+		return 0;
+
+	if (!tmc_has_reserved_buffer(drvdata) ||

Do we need to check this again ? We wouldn't be in ETR_MODE_RESRV
otherwise, also indicated by the comment.

Will drop.

quoted

quoted

+	    !tmc_has_crash_mdata_buffer(drvdata))
+		return 0;
+
+	tmc_crashdata_set_invalid(drvdata);
+
+	CS_UNLOCK(drvdata->base);
+
+	/* Proceed only if ETR is enabled */
+	val = readl(drvdata->base + TMC_CTL);
+	if (!(val & TMC_CTL_CAPT_EN))
+		goto out;
+
+	val = readl(drvdata->base + TMC_FFSR);
+	/* Do manual flush and stop only if its not auto-stopped */
+	if (!(val & TMC_FFSR_FT_STOPPED)) {
+		dev_info(&csdev->dev,
+			 "%s: Triggering manual flush\n", __func__);

Drop the info

Ack.

quoted

quoted

+		tmc_flush_and_stop(drvdata);
+	} else
+		tmc_wait_for_tmcready(drvdata);
+
+	/* Sync registers from hardware to metadata region */
+	mdata->size = csdev_access_relaxed_read32(csa, TMC_RSZ);
+	mdata->sts = csdev_access_relaxed_read32(csa, TMC_STS);
+	mdata->mode = csdev_access_relaxed_read32(csa, TMC_MODE);
+	mdata->ffcr = csdev_access_relaxed_read32(csa, TMC_FFCR);
+	mdata->ffsr = csdev_access_relaxed_read32(csa, TMC_FFSR);

Please use raw reads, don't mix csa and raw reads.

Ack.

quoted

quoted

+	mdata->rrp = tmc_read_rrp(drvdata);
+	mdata->rwp = tmc_read_rwp(drvdata);
+	mdata->dba = tmc_read_dba(drvdata);
+	mdata->trace_paddr = drvdata->resrv_buf.paddr;
+	mdata->version = CS_CRASHDATA_VERSION;
+
+	/*
+	 * Make sure all previous writes are completed,
+	 * before we mark valid
+	 */
+	dsb(sy);

Same as earlier, doesn't buy us much

Will convert to dmb.

quoted

quoted

+	mdata->valid = true;
+	/*
+	 * Below order need to maintained, since crc of metadata
+	 * is dependent on first
+	 */
+	mdata->crc32_tdata = find_crash_tracedata_crc(drvdata, mdata);
+	mdata->crc32_mdata = find_crash_metadata_crc(mdata);
+
+	tmc_disable_hw(drvdata);
+
+	dev_info(&csdev->dev, "%s: success\n", __func__);

Drop

quoted

+out:
+	CS_UNLOCK(drvdata->base);
+
+	return 0;
+}
+
  static const struct coresight_ops_sink tmc_etr_sink_ops = {
  	.enable		= tmc_enable_etr_sink,
  	.disable	= tmc_disable_etr_sink,

@@ -1822,8 +1895,13 @@ static const struct coresight_ops_sink tmc_etr_sink_ops = {
  	.free_buffer	= tmc_free_etr_buffer,
  };
+static const struct coresight_ops_panic tmc_etr_sync_ops = {
+	.sync		= tmc_panic_sync_etr,
+};
+
  const struct coresight_ops tmc_etr_cs_ops = {
  	.sink_ops	= &tmc_etr_sink_ops,
+	.panic_ops	= &tmc_etr_sync_ops,
  };
  int tmc_read_prepare_etr(struct tmc_drvdata *drvdata)

diff --git a/drivers/hwtracing/coresight/coresight-tmc.h b/drivers/hwtracing/coresight/coresight-tmc.h
index d2261eddab71..75e504e51956 100644
--- a/drivers/hwtracing/coresight/coresight-tmc.h
+++ b/drivers/hwtracing/coresight/coresight-tmc.h

@@ -12,6 +12,7 @@
  #include <linux/miscdevice.h>
  #include <linux/mutex.h>
  #include <linux/refcount.h>
+#include <linux/crc32.h>
  #define TMC_RSZ			0x004
  #define TMC_STS			0x00c

@@ -76,6 +77,9 @@
  #define TMC_AXICTL_AXCACHE_OS	(0xf << 2)
  #define TMC_AXICTL_ARCACHE_OS	(0xf << 16)
+/* TMC_FFSR - 0x300 */
+#define TMC_FFSR_FT_STOPPED	BIT(1)
+
  /* TMC_FFCR - 0x304 */
  #define TMC_FFCR_FLUSHMAN_BIT	6
  #define TMC_FFCR_EN_FMT		BIT(0)

@@ -94,6 +98,9 @@
  #define TMC_AUTH_NSID_MASK	GENMASK(1, 0)
+/* Major version 1 Minor version 0 */
+#define CS_CRASHDATA_VERSION	(1 << 16)
+
  enum tmc_config_type {
  	TMC_CONFIG_TYPE_ETB,
  	TMC_CONFIG_TYPE_ETR,

@@ -131,6 +138,25 @@ enum tmc_mem_intf_width {
  #define CORESIGHT_SOC_600_ETR_CAPS	\
  	(TMC_ETR_SAVE_RESTORE | TMC_ETR_AXI_ARCACHE)
+/* TMC metadata region for ETR and ETF configurations */
+struct tmc_crash_metadata {
+	uint32_t crc32_mdata;	/* crc of metadata */
+	uint32_t crc32_tdata;	/* crc of tracedata */
+	uint32_t version;	/* 31:16 Major version, 15:0 Minor version */
+	uint32_t valid;         /* Indicate if this ETF/ETR was enabled */
+	uint32_t size;          /* Ram Size register */

Please could you not keep this "plain bytes" ? Or rename the field to
tmc_ram_size if we want to stick to TMC RAM SIZE register. It is very
easy to confuse it with "normal" bytes.

Okay. Will change the name.

quoted

quoted

+	uint32_t sts;           /* Status register */

tmc_sts

quoted

+	uint32_t mode;		/* Mode register */

tmc_mode

This doesn't look packed. Please could you add a padding here to make sure
the fields are 64bit aligned ?

quoted

+	uint64_t ffcr;		/* Formatter and flush control register */

tmc_ffcr

quoted

+	uint64_t ffsr;		/* Formatter and flush status register */

tmc_ffsr

Guess we should prefix the rwp, rrp and dba also with tmc since they
are register snapshots ?