Re: [PATCH V22 2/3] misc: dcc: Add driver support for Data Capture and Compare unit(DCC)

From: Souradeep Chowdhury
Date: Wed Apr 19 2023 - 03:01:26 EST




On 4/18/2023 9:15 PM, Greg Kroah-Hartman wrote:
On Tue, Apr 18, 2023 at 08:52:36PM +0530, Souradeep Chowdhury wrote:
The DCC is a DMA Engine designed to capture and store data during system
crash or software triggers. The DCC operates based on user inputs via
the debugfs interface. The user gives addresses as inputs and these
addresses are stored in the dcc sram. In case of a system crash or a
manual software trigger by the user through the debugfs interface, the
dcc captures and stores the values at these addresses. This patch
contains the driver which has all the methods pertaining to the debugfs
interface, auxiliary functions to support all the four fundamental
operations of dcc namely read, write, read/modify/write and loop. The
probe method here instantiates all the resources necessary for dcc to
operate mainly the dedicated dcc sram where it stores the values. The
DCC driver can be used for debugging purposes without going for a reboot
since it can perform software triggers as well based on user inputs.

Also add the documentation for debugfs entries which explains the
functionalities of each debugfs file that has been created for dcc.

I see no documentation entries in this commit :(

My apologies, this patch was given from qcom-next tree which already has the documentation merged. Will include it from the next versions.


The following is the justification of using debugfs interface over the
other alternatives like sysfs/ioctls

i) As can be seen from the debugfs attribute descriptions, some of the
debugfs attribute files here contains multiple arguments which needs to
be accepted from the user. This goes against the design style of sysfs.

ii) The user input patterns have been made simple and convenient in this
case with the use of debugfs interface as user doesn't need to shuffle
between different files to execute one instruction as was the case on
using other alternatives.

Why do you have debugfs and also a misc device? How are they related?
Why both? Why not just one? What userspace tools are going to use
either of these interfaces and where are they published to show how this
all was tested?

DCC has two fundamental steps of usage:-

1.Configuring the register addresses on the dcc_sram which is done by user through the debugfs interface. For example:-

echo R 0x10c004 > /sys/kernel/debug/dcc/../3/config

Here we are configuring the register addresses for list 3, the 'R'
indicates a read operation, so this register value will be read
in case of a software trigger or kernel panic/watchdog bite and
dumped into the dcc_sram.

2.The dcc_sram content is exposed to the user in the form of a misc device. The user can parse the content of this dcc_sram to get the
register values. There is an opensource parser available for dcc at
the following location:-

https://git.codelinaro.org/clo/le/platform/vendor/qcom-opensource/tools/-/tree/opensource-tools.lnx.1.0.r176-rel/dcc_parser


Signed-off-by: Souradeep Chowdhury <quic_schowdhu@xxxxxxxxxxx>
Reviewed-by: Alex Elder <elder@xxxxxxxxxx>
---
drivers/misc/Kconfig | 8 +
drivers/misc/Makefile | 1 +
drivers/misc/dcc.c | 1300 +++++++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 1309 insertions(+)
create mode 100644 drivers/misc/dcc.c

diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig
index 433aa41..e2bc652 100644
--- a/drivers/misc/Kconfig
+++ b/drivers/misc/Kconfig
@@ -276,6 +276,14 @@ config QCOM_COINCELL
to maintain PMIC register and RTC state in the absence of
external power.
+config QCOM_DCC
+ tristate "Qualcomm Technologies, Inc. Data Capture and Compare(DCC) engine driver"
+ depends on ARCH_QCOM || COMPILE_TEST
+ help
+ This option enables driver for Data Capture and Compare engine. DCC
+ driver provides interface to configure DCC block and read back
+ captured data from DCC's internal SRAM.


What is the module name?

It's qcom-dcc, will update the name here.


+
config QCOM_FASTRPC
tristate "Qualcomm FastRPC"
depends on ARCH_QCOM || COMPILE_TEST
diff --git a/drivers/misc/Makefile b/drivers/misc/Makefile
index 56de439..6fa8efa 100644
--- a/drivers/misc/Makefile
+++ b/drivers/misc/Makefile
@@ -16,6 +16,7 @@ obj-$(CONFIG_TIFM_CORE) += tifm_core.o
obj-$(CONFIG_TIFM_7XX1) += tifm_7xx1.o
obj-$(CONFIG_PHANTOM) += phantom.o
obj-$(CONFIG_QCOM_COINCELL) += qcom-coincell.o
+obj-$(CONFIG_QCOM_DCC) += dcc.o

Why such a generic name? Shouldn't it be qcom-dcc?

Ack




obj-$(CONFIG_QCOM_FASTRPC) += fastrpc.o
obj-$(CONFIG_SENSORS_BH1770) += bh1770glc.o
obj-$(CONFIG_SENSORS_APDS990X) += apds990x.o
diff --git a/drivers/misc/dcc.c b/drivers/misc/dcc.c
new file mode 100644
index 0000000..7231ed9
--- /dev/null
+++ b/drivers/misc/dcc.c
@@ -0,0 +1,1300 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2015-2021, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2023, Qualcomm Innovation Center, Inc. All rights reserved.

No work happened on this code in 2022? All 22 of these entries were
only in 2021 and 2023?

Ack


+ */
+
+#include <linux/bitfield.h>
+#include <linux/bitops.h>
+#include <linux/debugfs.h>
+#include <linux/delay.h>
+#include <linux/fs.h>
+#include <linux/io.h>
+#include <linux/iopoll.h>
+#include <linux/miscdevice.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+
+#define STATUS_READY_TIMEOUT 5000 /* microseconds */
+
+/* DCC registers */
+#define DCC_HW_INFO 0x04
+#define DCC_LL_NUM_INFO 0x10
+#define DCC_STATUS(vers) ((vers) == 1 ? 0x0c : 0x1c)

Why isn't this just an inline function?

Ack. Will make this inline


+#define DCC_LL_LOCK 0x00
+#define DCC_LL_CFG 0x04
+#define DCC_LL_BASE 0x08
+#define DCC_FD_BASE 0x0c
+#define DCC_LL_TIMEOUT 0x10
+#define DCC_LL_INT_ENABLE 0x18
+#define DCC_LL_INT_STATUS 0x1c
+#define DCC_LL_SW_TRIGGER 0x2c
+#define DCC_LL_BUS_ACCESS_STATUS 0x30
+
+/* Default value used if a bit 6 in the HW_INFO register is set. */
+#define DCC_FIX_LOOP_OFFSET 16
+
+/* Mask to find version info from HW_Info register */
+#define DCC_VER_INFO_MASK BIT(9)
+
+#define MAX_DCC_OFFSET GENMASK(9, 2)
+#define MAX_DCC_LEN GENMASK(6, 0)
+#define MAX_LOOP_CNT GENMASK(7, 0)
+#define MAX_LOOP_ADDR 10
+
+#define DCC_ADDR_DESCRIPTOR 0x00
+#define DCC_ADDR_LIMIT 27
+#define DCC_WORD_SIZE sizeof(u32)
+#define DCC_ADDR_RANGE_MASK GENMASK(31, 4)
+#define DCC_LOOP_DESCRIPTOR BIT(30)
+#define DCC_RD_MOD_WR_DESCRIPTOR BIT(31)
+#define DCC_LINK_DESCRIPTOR GENMASK(31, 30)
+#define DCC_STATUS_MASK GENMASK(1, 0)
+#define DCC_LOCK_MASK BIT(0)
+#define DCC_LOOP_OFFSET_MASK BIT(6)
+#define DCC_TRIGGER_MASK BIT(9)
+
+#define DCC_WRITE_MASK BIT(15)
+#define DCC_WRITE_OFF_MASK GENMASK(7, 0)
+#define DCC_WRITE_LEN_MASK GENMASK(14, 8)
+
+#define DCC_READ_IND 0x00
+#define DCC_WRITE_IND (BIT(28))
+
+#define DCC_AHB_IND 0x00
+#define DCC_APB_IND BIT(29)
+
+#define DCC_MAX_LINK_LIST 8
+
+#define DCC_VER_MASK2 GENMASK(5, 0)
+
+#define DCC_SRAM_WORD_LENGTH 4
+
+#define DCC_RD_MOD_WR_ADDR 0xC105E
+
+enum dcc_descriptor_type {
+ DCC_READ_TYPE,
+ DCC_LOOP_TYPE,
+ DCC_READ_WRITE_TYPE,
+ DCC_WRITE_TYPE
+};
+
+struct dcc_config_entry {
+ u32 base;
+ u32 offset;
+ u32 len;
+ u32 loop_cnt;
+ u32 write_val;
+ u32 mask;
+ bool apb_bus;
+ enum dcc_descriptor_type desc_type;
+ struct list_head list;
+};

No documentation for this structure?

Ack. Will add documentation to this structure as well.


+
+/**
+ * struct dcc_drvdata - configuration information related to a dcc device
+ * @base: Base Address of the dcc device
+ * @dev: The device attached to the driver data
+ * @mutex: Lock to protect access and manipulation of dcc_drvdata
+ * @ram_base: Base address for the SRAM dedicated for the dcc device
+ * @ram_size: Total size of the SRAM dedicated for the dcc device
+ * @ram_offset: Offset to the SRAM dedicated for dcc device
+ * @mem_map_ver: Memory map version of DCC hardware
+ * @ram_cfg: Used for address limit calculation for dcc
+ * @ram_start: Starting address of DCC SRAM
+ * @sram_dev: Miscellaneous device equivalent of dcc SRAM
+ * @cfg_head: Points to the head of the linked list of addresses
+ * @dbg_dir: The dcc debugfs directory under which all the debugfs files are placed
+ * @nr_link_list: Total number of linkedlists supported by the DCC configuration
+ * @loop_shift: Loop offset bits range for the addresses
+ * @enable_bitmap: Bitmap to capture the enabled status of each linked list of addresses
+ */
+struct dcc_drvdata {
+ void __iomem *base;
+ void __iomem *ram_base;
+ struct device *dev;

Why do you need this back-pointer here?

This is getting used at multiple places to log
dev_err and also for resource allocation using
devm_kzalloc.


+ struct mutex mutex;
+ size_t ram_size;
+ size_t ram_offset;
+ int mem_map_ver;
+ unsigned int ram_cfg;
+ unsigned int ram_start;
+ struct miscdevice sram_dev;

Wait, this is the struct device, right? Or not?

miscdevice here represents the dcc_sram, an io-memory
dedicated to dcc for configuring/storing register values.
Whereas struct device represents the dcc_device which can
be used to write control signals on the bus to handle dcc
hardware operation sequence(like config_reset,sw_trigger etc.)


+ struct list_head *cfg_head;
+ struct dentry *dbg_dir;

Why is this needed and not just looked up when necessary?

This needs to be passed while creating sub-directories and also
while removing. Rather than looking up everytime,saving and
re-using this in here.


+ size_t nr_link_list;
+ u8 loop_shift;
+ unsigned long *enable_bitmap;

So this is a list of bitmaps? Why "unsigned long"? Why not u64?

Ack


+};
+
+struct dcc_cfg_attr {
+ u32 addr;
+ u32 prev_addr;
+ u32 prev_off;
+ u32 link;
+ u32 sram_offset;
+};
+
+struct dcc_cfg_loop_attr {
+ u32 loop_cnt;
+ u32 loop_len;
+ u32 loop_off;
+ bool loop_start;
+};
+
+static inline u32 dcc_list_offset(int version)
+{
+ return version == 1 ? 0x1c : version == 2 ? 0x2c : 0x34;
+}

Ah, you do have an inline function for the above mentioned macro.
Please drop the macro.

And write this inline function out to be readable, single-level ?:
comments are impossible to read, let alone double-level ones.

Write code for people first, compilers second. You gain nothing by
making this terse except to confuse people.

Ack. This inline function is different from the previous macro.

Will keep both as inline functions.


+
+static inline void dcc_list_writel(struct dcc_drvdata *drvdata,
+ u32 ll, u32 val, u32 off)
+{
+ u32 offset = dcc_list_offset(drvdata->mem_map_ver) + off;
+
+ writel(val, drvdata->base + ll * 0x80 + offset);

What is this magic 0x80 for?

This is the list offset needed for address calculation as per the dcc-hardware specification. Will declare a macro for this.


+}
+
+static inline u32 dcc_list_readl(struct dcc_drvdata *drvdata, u32 ll, u32 off)
+{
+ u32 offset = dcc_list_offset(drvdata->mem_map_ver) + off;
+
+ return readl(drvdata->base + ll * 0x80 + offset);

Again, where did 0x80 come from?

Same as above.


+}
+
+static void dcc_sram_write_auto(struct dcc_drvdata *drvdata,
+ u32 val, u32 *off)
+{
+ /* If the overflow condition is met increment the offset
+ * and return to indicate that overflow has occurred
+ */
+ if (unlikely(*off > drvdata->ram_size - 4)) {
+ *off += 4;
+ return;

You didn't indicate anything here, all you did was succeed at the call,
the caller has no way of ever knowing this failed.

Why not return an error?

As per previous discussions it was decided to perform the write speculatively. So that while writing to the dcc_sram if we exceed
the memory size, dcc will skip the write and keep incrementing
the offset. In the method "dcc_emit_config" we have the check to
finally detect if we have exceeded the sram offset

if (cfg.sram_offset + total_len > drvdata->ram_size) {
cfg.sram_offset += total_len;
goto overstep;
}


+ }
+
+ writel(val, drvdata->ram_base + *off);
+
+ *off += 4;

See, same modification as your "error" above.

How was this tested?

This increment is needed to update the offset for the next memory
position in dcc_sram.


+static int dcc_emit_config(struct dcc_drvdata *drvdata, unsigned int curr_list)
+{
+ int ret;
+ u32 total_len, pos;
+ struct dcc_config_entry *entry;
+ struct dcc_cfg_attr cfg;
+ struct dcc_cfg_loop_attr cfg_loop;
+
+ memset(&cfg, 0, sizeof(cfg));
+ memset(&cfg_loop, 0, sizeof(cfg_loop));

Why are these large structures on the stack?

cfg_loop is needed for offset calculation in case of dcc loop instructions based on the way it needs to be configured in dcc_sram
for dcc hardware to interpret it. entry, cfg is a generic structure used
across all dcc instructions. All these structures are needed for the
memory checks after we are done with configuring all the dcc instructions.


And if on the stack, why not have the compiler initialize them to 0 for
you automatically?

Ack


I stopped reviewing here...

greg k-h