[PATCH v7 0/2] ufs: exynos: introduce the way to get cmd info

[PATCH v7 0/2] ufs: exynos: introduce the way to get cmd info

[Kiwoong Kim]

v6 -> v7
migrate this to 5.10.
shrink the patch set, two patches.
merge a fix for build errors

v5 -> v6
replace put_aligned with get_unaligned_be32 to set lba and sct
fix null pointer access symptom

v4 -> v5
Rebased on Stanley's patch (scsi: ufs: Fix and simplify ..
Change cmd history print order
rename config to SCSI_UFS_EXYNOS_DBG
feature functions in ufs-exynos-dbg.c by SCSI_UFS_EXYNOS_DBG

v3 -> v4
seperate respective implementations of the callbacks
change the location of compl_xfer_req related stuffs
fix null pointer access

v2 -> v3
fix build errors

v1 -> v2
change callbacks
allocate memory for ufs_s_dbg_mgr dynamically, not static way

Kiwoong Kim (2):
  ufs: introduce a callback to get info of command completion
  ufs: exynos: introduce command history

 drivers/scsi/ufs/Kconfig          |  14 +++
 drivers/scsi/ufs/Makefile         |   4 +-
 drivers/scsi/ufs/ufs-exynos-dbg.c | 200 ++++++++++++++++++++++++++++++++++++++
 drivers/scsi/ufs/ufs-exynos-if.h  |  17 ++++
 drivers/scsi/ufs/ufs-exynos.c     |  38 ++++++++
 drivers/scsi/ufs/ufs-exynos.h     |  35 +++++++
 drivers/scsi/ufs/ufshcd.c         |   1 +
 drivers/scsi/ufs/ufshcd.h         |   8 ++
 8 files changed, 316 insertions(+), 1 deletion(-)
 create mode 100644 drivers/scsi/ufs/ufs-exynos-dbg.c
 create mode 100644 drivers/scsi/ufs/ufs-exynos-if.h

-- 
2.7.4

[PATCH v7 1/2] ufs: introduce a callback to get info of command completion

[Kiwoong Kim]

Some SoC specific might need command history for
various reasons, such as stacking command contexts
in system memory to check for debugging in the future
or scaling some DVFS knobs to boost IO throughput.

What you would do with the information could be
variant per SoC vendor.

Signed-off-by: Kiwoong Kim <kwmad.kim@samsung.com>
Acked-by: Stanley Chu <stanley.chu@mediatek.com>
Reviewed-by: Avri Altman <avri.altman@wdc.com>
---
 drivers/scsi/ufs/ufshcd.c | 1 +
 drivers/scsi/ufs/ufshcd.h | 8 ++++++++
 2 files changed, 9 insertions(+)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 06e2439..64bd59c 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -4923,6 +4923,7 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 		lrbp = &hba->lrb[index];
 		lrbp->compl_time_stamp = ktime_get();
 		cmd = lrbp->cmd;
+		ufshcd_vops_compl_xfer_req(hba, index, (cmd) ? true : false);
 		if (cmd) {
 			ufshcd_add_command_trace(hba, index, "complete");
 			result = ufshcd_transfer_rsp_status(hba, lrbp);
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index 618b253..02bd405 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -307,6 +307,7 @@ struct ufs_hba_variant_ops {
 					struct ufs_pa_layer_attr *,
 					struct ufs_pa_layer_attr *);
 	void	(*setup_xfer_req)(struct ufs_hba *, int, bool);
+	void	(*compl_xfer_req)(struct ufs_hba *hba, int tag, bool is_scsi);
 	void	(*setup_task_mgmt)(struct ufs_hba *, int, u8);
 	void    (*hibern8_notify)(struct ufs_hba *, enum uic_cmd_dme,
 					enum ufs_notify_change_status);
@@ -1111,6 +1112,13 @@ static inline void ufshcd_vops_setup_xfer_req(struct ufs_hba *hba, int tag,
 		return hba->vops->setup_xfer_req(hba, tag, is_scsi_cmd);
 }
 
+static inline void ufshcd_vops_compl_xfer_req(struct ufs_hba *hba,
+					      int tag, bool is_scsi)
+{
+	if (hba->vops && hba->vops->compl_xfer_req)
+		hba->vops->compl_xfer_req(hba, tag, is_scsi);
+}
+
 static inline void ufshcd_vops_setup_task_mgmt(struct ufs_hba *hba,
 					int tag, u8 tm_function)
 {
-- 
2.7.4

[PATCH v7 2/2] ufs: exynos: introduce command history

[Kiwoong Kim]

This includes functions to record contexts of incoming commands
in a circular queue. ufshcd.c has already some function
tracer calls to get command history but ftrace would be
gone when system dies before you get the information,
such as panic cases.

This patch also implements callbacks compl_xfer_req
to store IO contexts at completion times.

When you turn on CONFIG_SCSI_UFS_EXYNOS_CMD_LOG,
the driver collects the information from incoming commands
in the circular queue.

Signed-off-by: Kiwoong Kim <kwmad.kim@samsung.com>
---
 drivers/scsi/ufs/Kconfig          |  14 +++
 drivers/scsi/ufs/Makefile         |   4 +-
 drivers/scsi/ufs/ufs-exynos-dbg.c | 200 ++++++++++++++++++++++++++++++++++++++
 drivers/scsi/ufs/ufs-exynos-if.h  |  17 ++++
 drivers/scsi/ufs/ufs-exynos.c     |  38 ++++++++
 drivers/scsi/ufs/ufs-exynos.h     |  35 +++++++
 6 files changed, 307 insertions(+), 1 deletion(-)
 create mode 100644 drivers/scsi/ufs/ufs-exynos-dbg.c
 create mode 100644 drivers/scsi/ufs/ufs-exynos-if.h

diff --git a/drivers/scsi/ufs/Kconfig b/drivers/scsi/ufs/Kconfig
index f639499..29aa4d2 100644
--- a/drivers/scsi/ufs/Kconfig
+++ b/drivers/scsi/ufs/Kconfig
@@ -174,6 +174,20 @@ config SCSI_UFS_EXYNOS
 	  Select this if you have UFS host controller on EXYNOS chipset.
 	  If unsure, say N.
 
+config SCSI_UFS_EXYNOS_DBG
+	bool "EXYNOS specific debug functions"
+	default n
+	depends on SCSI_UFS_EXYNOS
+	help
+	  This selects EXYNOS specific functions to get and even print
+	  some information to see what's happening at both command
+	  issue time completion time.
+	  The information may contain general things as well as
+	  EXYNOS specific, such as vendor specific hardware contexts.
+
+	  Select this if you want to get and print the information.
+	  If unsure, say N.
+
 config SCSI_UFS_CRYPTO
 	bool "UFS Crypto Engine Support"
 	depends on SCSI_UFSHCD && BLK_INLINE_ENCRYPTION
diff --git a/drivers/scsi/ufs/Makefile b/drivers/scsi/ufs/Makefile
index 4679af1..8d425a4 100644
--- a/drivers/scsi/ufs/Makefile
+++ b/drivers/scsi/ufs/Makefile
@@ -6,7 +6,9 @@ obj-$(CONFIG_SCSI_UFS_CDNS_PLATFORM) += cdns-pltfrm.o
 obj-$(CONFIG_SCSI_UFS_QCOM) += ufs_qcom.o
 ufs_qcom-y += ufs-qcom.o
 ufs_qcom-$(CONFIG_SCSI_UFS_CRYPTO) += ufs-qcom-ice.o
-obj-$(CONFIG_SCSI_UFS_EXYNOS) += ufs-exynos.o
+obj-$(CONFIG_SCSI_UFS_EXYNOS) += ufs-exynos-core.o
+ufs-exynos-core-y += ufs-exynos.o
+ufs-exynos-core-$(CONFIG_SCSI_UFS_EXYNOS_DBG) += ufs-exynos-dbg.o
 obj-$(CONFIG_SCSI_UFSHCD) += ufshcd-core.o
 ufshcd-core-y				+= ufshcd.o ufs-sysfs.o
 ufshcd-core-$(CONFIG_SCSI_UFS_BSG)	+= ufs_bsg.o
diff --git a/drivers/scsi/ufs/ufs-exynos-dbg.c b/drivers/scsi/ufs/ufs-exynos-dbg.c
new file mode 100644
index 0000000..89ce3c2
--- /dev/null
+++ b/drivers/scsi/ufs/ufs-exynos-dbg.c
@@ -0,0 +1,200 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * UFS Exynos debugging functions
+ *
+ * Copyright (C) 2020 Samsung Electronics Co., Ltd.
+ * Author: Kiwoong Kim <kwmad.kim@samsung.com>
+ *
+ */
+#include <linux/platform_device.h>
+#include <linux/module.h>
+#include <asm-generic/unaligned.h>
+#include "ufshcd.h"
+#include "ufs-exynos-if.h"
+
+#define MAX_CMD_LOGS    32
+
+struct cmd_data {
+	unsigned int tag;
+	u32 sct;
+	u64 lba;
+	u64 start_time;
+	u64 end_time;
+	u64 outstanding_reqs;
+	int retries;
+	u8 op;
+};
+
+struct ufs_cmd_info {
+	u32 total;
+	u32 last;
+	struct cmd_data data[MAX_CMD_LOGS];
+	struct cmd_data *pdata[MAX_CMD_LOGS];
+};
+
+/*
+ * This structure points out several contexts on debugging
+ * per one host instant.
+ * Now command history exists in here but later handle may
+ * contains some mmio base addresses including vendor specific
+ * regions to get hardware contexts.
+ */
+struct ufs_s_dbg_mgr {
+	struct ufs_exynos_handle *handle;
+	int active;
+	u64 first_time;
+	u64 time;
+
+	/* cmd log */
+	struct ufs_cmd_info cmd_info;
+	struct cmd_data cmd_log;		/* temp buffer to put */
+	spinlock_t cmd_lock;
+};
+
+#ifdef CONFIG_SCSI_UFS_EXYNOS_DBG
+static void ufs_s_print_cmd_log(struct ufs_s_dbg_mgr *mgr, struct device *dev)
+{
+	struct ufs_cmd_info *cmd_info = &mgr->cmd_info;
+	struct cmd_data *data;
+	u32 i, idx;
+	u32 last;
+	u32 max = MAX_CMD_LOGS;
+	unsigned long flags;
+	u32 total;
+
+	spin_lock_irqsave(&mgr->cmd_lock, flags);
+	total = cmd_info->total;
+	if (cmd_info->total < max)
+		max = cmd_info->total;
+	last = (cmd_info->last + MAX_CMD_LOGS - 1) % MAX_CMD_LOGS;
+	spin_unlock_irqrestore(&mgr->cmd_lock, flags);
+
+	dev_err(dev, ":---------------------------------------------------\n");
+	dev_err(dev, ":\t\tSCSI CMD(%u)\n", total - 1);
+	dev_err(dev, ":---------------------------------------------------\n");
+	dev_err(dev, ":OP, TAG, LBA, SCT, RETRIES, STIME, ETIME, REQS\n\n");
+
+	idx = (last == max - 1) ? 0 : last + 1;
+	data = &cmd_info->data[idx];
+	for (i = 0 ; i < max ; i++, data = &cmd_info->data[idx]) {
+		dev_err(dev, ": 0x%02x, %02d, 0x%08llx, 0x%04x, %d, %llu, %llu, 0x%llx",
+			data->op, data->tag, data->lba, data->sct, data->retries,
+			data->start_time, data->end_time, data->outstanding_reqs);
+		idx = (idx == max - 1) ? 0 : idx + 1;
+	}
+}
+
+static void ufs_s_put_cmd_log(struct ufs_s_dbg_mgr *mgr,
+			      struct cmd_data *cmd_data)
+{
+	struct ufs_cmd_info *cmd_info = &mgr->cmd_info;
+	unsigned long flags;
+	struct cmd_data *pdata;
+
+	spin_lock_irqsave(&mgr->cmd_lock, flags);
+	pdata = &cmd_info->data[cmd_info->last];
+	++cmd_info->total;
+	++cmd_info->last;
+	cmd_info->last = cmd_info->last % MAX_CMD_LOGS;
+	spin_unlock_irqrestore(&mgr->cmd_lock, flags);
+
+	pdata->op = cmd_data->op;
+	pdata->tag = cmd_data->tag;
+	pdata->lba = cmd_data->lba;
+	pdata->sct = cmd_data->sct;
+	pdata->retries = cmd_data->retries;
+	pdata->start_time = cmd_data->start_time;
+	pdata->end_time = 0;
+	pdata->outstanding_reqs = cmd_data->outstanding_reqs;
+	cmd_info->pdata[cmd_data->tag] = pdata;
+}
+
+/*
+ * EXTERNAL FUNCTIONS
+ *
+ * There are two classes that are to initialize data structures for debug
+ * and to define actual behavior.
+ */
+void exynos_ufs_dump_info(struct ufs_exynos_handle *handle, struct device *dev)
+{
+	struct ufs_s_dbg_mgr *mgr = (struct ufs_s_dbg_mgr *)handle->private;
+
+	if (mgr->active == 0)
+		goto out;
+
+	mgr->time = cpu_clock(raw_smp_processor_id());
+
+	ufs_s_print_cmd_log(mgr, dev);
+
+	if (mgr->first_time == 0ULL)
+		mgr->first_time = mgr->time;
+out:
+	return;
+}
+
+void exynos_ufs_cmd_log_start(struct ufs_exynos_handle *handle,
+			      struct ufs_hba *hba, int tag)
+{
+	struct ufs_s_dbg_mgr *mgr = (struct ufs_s_dbg_mgr *)handle->private;
+	struct scsi_cmnd *cmd = hba->lrb[tag].cmd;
+	int cpu = raw_smp_processor_id();
+	struct cmd_data *cmd_log = &mgr->cmd_log;	/* temp buffer to put */
+	u64 lba;
+	u32 sct;
+
+	if (mgr->active == 0)
+		return;
+
+	cmd_log->start_time = cpu_clock(cpu);
+	cmd_log->op = cmd->cmnd[0];
+	cmd_log->tag = tag;
+
+	/* This function runtime is protected by spinlock from outside */
+	cmd_log->outstanding_reqs = hba->outstanding_reqs;
+
+	/* Now assume using WRITE_10 and READ_10 */
+	lba = get_unaligned_be32(&cmd->cmnd[2]);
+	sct = get_unaligned_be32(&cmd->cmnd[7]);
+	if (cmd->cmnd[0] != UNMAP)
+		cmd_log->lba = lba;
+
+	cmd_log->sct = sct;
+	cmd_log->retries = cmd->allowed;
+
+	ufs_s_put_cmd_log(mgr, cmd_log);
+}
+
+void exynos_ufs_cmd_log_end(struct ufs_exynos_handle *handle,
+			    struct ufs_hba *hba, int tag)
+{
+	struct ufs_s_dbg_mgr *mgr = (struct ufs_s_dbg_mgr *)handle->private;
+	struct ufs_cmd_info *cmd_info = &mgr->cmd_info;
+	int cpu = raw_smp_processor_id();
+
+	if (mgr->active == 0)
+		return;
+
+	cmd_info->pdata[tag]->end_time = cpu_clock(cpu);
+}
+
+int exynos_ufs_init_dbg(struct ufs_exynos_handle *handle, struct device *dev)
+{
+	struct ufs_s_dbg_mgr *mgr;
+
+	mgr = devm_kzalloc(dev, sizeof(struct ufs_s_dbg_mgr), GFP_KERNEL);
+	if (!mgr)
+		return -ENOMEM;
+	handle->private = (void *)mgr;
+	mgr->handle = handle;
+	mgr->active = 1;
+
+	/* cmd log */
+	spin_lock_init(&mgr->cmd_lock);
+
+	return 0;
+}
+#endif
+MODULE_AUTHOR("Kiwoong Kim <kwmad.kim@samsung.com>");
+MODULE_DESCRIPTION("Exynos UFS debug information");
+MODULE_LICENSE("GPL v2");
+MODULE_VERSION("0.1");
diff --git a/drivers/scsi/ufs/ufs-exynos-if.h b/drivers/scsi/ufs/ufs-exynos-if.h
new file mode 100644
index 0000000..c746f59
--- /dev/null
+++ b/drivers/scsi/ufs/ufs-exynos-if.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * UFS Exynos debugging functions
+ *
+ * Copyright (C) 2020 Samsung Electronics Co., Ltd.
+ * Author: Kiwoong Kim <kwmad.kim@samsung.com>
+ *
+ */
+#ifndef _UFS_EXYNOS_IF_H_
+#define _UFS_EXYNOS_IF_H_
+
+/* more members would be added in the future */
+struct ufs_exynos_handle {
+	void *private;
+};
+
+#endif /* _UFS_EXYNOS_IF_H_ */
diff --git a/drivers/scsi/ufs/ufs-exynos.c b/drivers/scsi/ufs/ufs-exynos.c
index 8f1b6f6..7d383e3 100644
--- a/drivers/scsi/ufs/ufs-exynos.c
+++ b/drivers/scsi/ufs/ufs-exynos.c
@@ -700,12 +700,26 @@ static int exynos_ufs_post_pwr_mode(struct ufs_hba *hba,
 	return 0;
 }
 
+static void exynos_ufs_cmd_log(struct ufs_hba *hba, int tag, int start)
+{
+	struct exynos_ufs *ufs = ufshcd_get_variant(hba);
+	struct ufs_exynos_handle *handle = &ufs->handle;
+
+	if (start == 1)
+		exynos_ufs_cmd_log_start(handle, hba, tag);
+	else if (start == 2)
+		exynos_ufs_cmd_log_end(handle, hba, tag);
+}
+
 static void exynos_ufs_specify_nexus_t_xfer_req(struct ufs_hba *hba,
 						int tag, bool op)
 {
 	struct exynos_ufs *ufs = ufshcd_get_variant(hba);
 	u32 type;
 
+	if (op)
+		exynos_ufs_cmd_log(hba, tag, 1);
+
 	type =  hci_readl(ufs, HCI_UTRL_NEXUS_TYPE);
 
 	if (op)
@@ -714,6 +728,12 @@ static void exynos_ufs_specify_nexus_t_xfer_req(struct ufs_hba *hba,
 		hci_writel(ufs, type & ~(1 << tag), HCI_UTRL_NEXUS_TYPE);
 }
 
+static void exynos_ufs_compl_xfer_req(struct ufs_hba *hba, int tag, bool op)
+{
+	if (op)
+		exynos_ufs_cmd_log(hba, tag, 0);
+}
+
 static void exynos_ufs_specify_nexus_t_tm_req(struct ufs_hba *hba,
 						int tag, u8 func)
 {
@@ -1008,6 +1028,11 @@ static int exynos_ufs_init(struct ufs_hba *hba)
 		goto out;
 	exynos_ufs_specify_phy_time_attr(ufs);
 	exynos_ufs_config_smu(ufs);
+
+	/* init dbg */
+	ret = exynos_ufs_init_dbg(&ufs->handle, dev);
+	if (ret)
+		return ret;
 	return 0;
 
 phy_off:
@@ -1210,6 +1235,17 @@ static int exynos_ufs_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op)
 	return 0;
 }
 
+static void exynos_ufs_dbg_register_dump(struct ufs_hba *hba)
+{
+	struct exynos_ufs *ufs = ufshcd_get_variant(hba);
+
+	if (!test_and_set_bit(EXYNOS_UFS_DBG_DUMP_CXT, &ufs->dbg_flag)) {
+		exynos_ufs_dump_info(&ufs->handle, hba->dev);
+		clear_bit(EXYNOS_UFS_DBG_DUMP_CXT, &ufs->dbg_flag);
+	}
+	return;
+}
+
 static struct ufs_hba_variant_ops ufs_hba_exynos_ops = {
 	.name				= "exynos_ufs",
 	.init				= exynos_ufs_init,
@@ -1217,10 +1253,12 @@ static struct ufs_hba_variant_ops ufs_hba_exynos_ops = {
 	.link_startup_notify		= exynos_ufs_link_startup_notify,
 	.pwr_change_notify		= exynos_ufs_pwr_change_notify,
 	.setup_xfer_req			= exynos_ufs_specify_nexus_t_xfer_req,
+	.compl_xfer_req			= exynos_ufs_compl_xfer_req,
 	.setup_task_mgmt		= exynos_ufs_specify_nexus_t_tm_req,
 	.hibern8_notify			= exynos_ufs_hibern8_notify,
 	.suspend			= exynos_ufs_suspend,
 	.resume				= exynos_ufs_resume,
+	.dbg_register_dump		= exynos_ufs_dbg_register_dump,
 };
 
 static int exynos_ufs_probe(struct platform_device *pdev)
diff --git a/drivers/scsi/ufs/ufs-exynos.h b/drivers/scsi/ufs/ufs-exynos.h
index 76d6e39..eed8728 100644
--- a/drivers/scsi/ufs/ufs-exynos.h
+++ b/drivers/scsi/ufs/ufs-exynos.h
@@ -8,6 +8,7 @@
 
 #ifndef _UFS_EXYNOS_H_
 #define _UFS_EXYNOS_H_
+#include "ufs-exynos-if.h"
 
 /*
  * UNIPRO registers
@@ -212,6 +213,10 @@ struct exynos_ufs {
 #define EXYNOS_UFS_OPT_BROKEN_AUTO_CLK_CTRL	BIT(2)
 #define EXYNOS_UFS_OPT_BROKEN_RX_SEL_IDX	BIT(3)
 #define EXYNOS_UFS_OPT_USE_SW_HIBERN8_TIMER	BIT(4)
+
+	struct ufs_exynos_handle handle;
+	unsigned long dbg_flag;
+#define EXYNOS_UFS_DBG_DUMP_CXT			0
 };
 
 #define for_each_ufs_rx_lane(ufs, i) \
@@ -284,4 +289,34 @@ struct exynos_ufs_uic_attr exynos7_uic_attr = {
 	.rx_hs_g3_prep_sync_len_cap	= PREP_LEN(0xf),
 	.pa_dbg_option_suite		= 0x30103,
 };
+
+/* public function declarations */
+#ifdef CONFIG_SCSI_UFS_EXYNOS_DBG
+void exynos_ufs_cmd_log_start(struct ufs_exynos_handle *handle,
+			      struct ufs_hba *hba, int tag);
+void exynos_ufs_cmd_log_end(struct ufs_exynos_handle *handle,
+			    struct ufs_hba *hba, int tag);
+int exynos_ufs_init_dbg(struct ufs_exynos_handle *handle, struct device *dev);
+void exynos_ufs_dump_info(struct ufs_exynos_handle *handle, struct device *dev);
+#else
+void exynos_ufs_cmd_log_start(struct ufs_exynos_handle *handle,
+			      struct ufs_hba *hba, int tag)
+{
+}
+
+void exynos_ufs_cmd_log_end(struct ufs_exynos_handle *handle,
+			    struct ufs_hba *hba, int tag)
+{
+}
+
+int exynos_ufs_init_dbg(struct ufs_exynos_handle *handle, struct device *dev)
+{
+	return 0;
+}
+
+void exynos_ufs_dump_info(struct ufs_exynos_handle *handle, struct device *dev)
+{
+	return;
+}
+#endif
 #endif /* _UFS_EXYNOS_H_ */
-- 
2.7.4

RE: [PATCH v7 2/2] ufs: exynos: introduce command history

[Avri Altman]

> 
> When you turn on CONFIG_SCSI_UFS_EXYNOS_CMD_LOG,
I guess you meant CONFIG_ SCSI_UFS_EXYNOS_DBG


> +struct ufs_cmd_info {
> +       u32 total;
> +       u32 last;
> +       struct cmd_data data[MAX_CMD_LOGS];
> +       struct cmd_data *pdata[MAX_CMD_LOGS];
What is the use of pdata?
Looks like it is just a copy of data, is it?

> +
> +#ifdef CONFIG_SCSI_UFS_EXYNOS_DBG
Not needed inside exynos-dbg

> +static void ufs_s_print_cmd_log(struct ufs_s_dbg_mgr *mgr, struct device
> *dev)
> +{
> +       struct ufs_cmd_info *cmd_info = &mgr->cmd_info;
> +       struct cmd_data *data;
> +       u32 i, idx;
> +       u32 last;
> +       u32 max = MAX_CMD_LOGS;
> +       unsigned long flags;
> +       u32 total;
> +
> +       spin_lock_irqsave(&mgr->cmd_lock, flags);
> +       total = cmd_info->total;
> +       if (cmd_info->total < max)
> +               max = cmd_info->total;
> +       last = (cmd_info->last + MAX_CMD_LOGS - 1) % MAX_CMD_LOGS;
> +       spin_unlock_irqrestore(&mgr->cmd_lock, flags);
> +
> +       dev_err(dev, ":---------------------------------------------------\n");
> +       dev_err(dev, ":\t\tSCSI CMD(%u)\n", total - 1);
> +       dev_err(dev, ":---------------------------------------------------\n");
> +       dev_err(dev, ":OP, TAG, LBA, SCT, RETRIES, STIME, ETIME, REQS\n\n");
> +
> +       idx = (last == max - 1) ? 0 : last + 1;
cmd_info->last points to the current index which is the oldest data in your circular buffer.
Why not just idx = cmd_info->last ?
And then a better name for cmd_info->last would be cmd_info->pos or cmd_info->current,
Or did I get it wrong?
 
I would also not worried about max and total - 
so the first round after platform boot time it might print zeros.
Really don't worth the unnecessary compilation.

> +       data = &cmd_info->data[idx];
> +       for (i = 0 ; i < max ; i++, data = &cmd_info->data[idx]) {
> +               dev_err(dev, ": 0x%02x, %02d, 0x%08llx, 0x%04x, %d, %llu, %llu,
> 0x%llx",
> +                       data->op, data->tag, data->lba, data->sct, data->retries,
> +                       data->start_time, data->end_time, data->outstanding_reqs);
> +               idx = (idx == max - 1) ? 0 : idx + 1;
> +       }
> +}


> diff --git a/drivers/scsi/ufs/ufs-exynos-if.h b/drivers/scsi/ufs/ufs-exynos-if.h
> new file mode 100644
> index 0000000..c746f59
> --- /dev/null
> +++ b/drivers/scsi/ufs/ufs-exynos-if.h
> @@ -0,0 +1,17 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * UFS Exynos debugging functions
So this header contains a single struct with an opaque pointer.
If it is just for debugging, why  not ufs-exynos-dbg.h?


Thanks,
Avri

RE: [PATCH v7 2/2] ufs: exynos: introduce command history

[Kiwoong Kim]

> > When you turn on CONFIG_SCSI_UFS_EXYNOS_CMD_LOG,
> I guess you meant CONFIG_ SCSI_UFS_EXYNOS_DBG
Okay 

> 
> 
> > +struct ufs_cmd_info {
> > +       u32 total;
> > +       u32 last;
> > +       struct cmd_data data[MAX_CMD_LOGS];
> > +       struct cmd_data *pdata[MAX_CMD_LOGS];
> What is the use of pdata?
> Looks like it is just a copy of data, is it?
Yes

> 
> > +
> > +#ifdef CONFIG_SCSI_UFS_EXYNOS_DBG
> Not needed inside exynos-dbg
> 
> > +static void ufs_s_print_cmd_log(struct ufs_s_dbg_mgr *mgr, struct
> > +device
> > *dev)
> > +{
> > +       struct ufs_cmd_info *cmd_info = &mgr->cmd_info;
> > +       struct cmd_data *data;
> > +       u32 i, idx;
> > +       u32 last;
> > +       u32 max = MAX_CMD_LOGS;
> > +       unsigned long flags;
> > +       u32 total;
> > +
> > +       spin_lock_irqsave(&mgr->cmd_lock, flags);
> > +       total = cmd_info->total;
> > +       if (cmd_info->total < max)
> > +               max = cmd_info->total;
> > +       last = (cmd_info->last + MAX_CMD_LOGS - 1) % MAX_CMD_LOGS;
> > +       spin_unlock_irqrestore(&mgr->cmd_lock, flags);
> > +
> > +       dev_err(dev, ":--------------------------------------------------
> -\n");
> > +       dev_err(dev, ":\t\tSCSI CMD(%u)\n", total - 1);
> > +       dev_err(dev, ":--------------------------------------------------
> -\n");
> > +       dev_err(dev, ":OP, TAG, LBA, SCT, RETRIES, STIME, ETIME,
> > + REQS\n\n");
> > +
> > +       idx = (last == max - 1) ? 0 : last + 1;
> cmd_info->last points to the current index which is the oldest data in
> your circular buffer.
> Why not just idx = cmd_info->last ?
> And then a better name for cmd_info->last would be cmd_info->pos or
> cmd_info->current, Or did I get it wrong?

The member 'last' means a next available slot.
I agree this name is a little bit confusing and will change it.

> 
> I would also not worried about max and total - so the first round after
> platform boot time it might print zeros.
> Really don't worth the unnecessary compilation.
> 
> > +       data = &cmd_info->data[idx];
> > +       for (i = 0 ; i < max ; i++, data = &cmd_info->data[idx]) {
> > +               dev_err(dev, ": 0x%02x, %02d, 0x%08llx, 0x%04x, %d,
> > + %llu, %llu,
> > 0x%llx",
> > +                       data->op, data->tag, data->lba, data->sct, data-
> >retries,
> > +                       data->start_time, data->end_time, data-
> >outstanding_reqs);
> > +               idx = (idx == max - 1) ? 0 : idx + 1;
> > +       }
> > +}
> 
> 
> > diff --git a/drivers/scsi/ufs/ufs-exynos-if.h
> > b/drivers/scsi/ufs/ufs-exynos-if.h
> > new file mode 100644
> > index 0000000..c746f59
> > --- /dev/null
> > +++ b/drivers/scsi/ufs/ufs-exynos-if.h
> > @@ -0,0 +1,17 @@
> > +/* SPDX-License-Identifier: GPL-2.0-only */
> > +/*
> > + * UFS Exynos debugging functions
> So this header contains a single struct with an opaque pointer.
> If it is just for debugging, why  not ufs-exynos-dbg.h?
Okay. 

> 
> 
> Thanks,
> Avri