Re: [PATCH v13 1/2] drivers/coresight: Add UltraSoc System Memory Buffer driver

From: hejunhao
Date: Fri Nov 18 2022 - 07:46:08 EST


Hi Suzuki ,


On 2022/11/15 19:06, Suzuki K Poulose wrote:
On 14/11/2022 09:03, Junhao He wrote:
From: Qi Liu <liuqi115@xxxxxxxxxx>

This patch adds driver for UltraSoc SMB(System Memory Buffer)
device. SMB provides a way to buffer messages from ETM, and
store these "CPU instructions trace" in system memory.

SMB is developed by UltraSoc technology, which is acquired by
Siemens, and we still use "UltraSoc" to name driver.

Signed-off-by: Qi Liu <liuqi115@xxxxxxxxxx>
Signed-off-by: Junhao He <hejunhao3@xxxxxxxxxx>
Tested-by: JunHao He <hejunhao3@xxxxxxxxxx>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
---
drivers/hwtracing/coresight/Kconfig | 11 +
drivers/hwtracing/coresight/Makefile | 1 +
drivers/hwtracing/coresight/ultrasoc-smb.c | 656 +++++++++++++++++++++
drivers/hwtracing/coresight/ultrasoc-smb.h | 120 ++++
4 files changed, 788 insertions(+)
create mode 100644 drivers/hwtracing/coresight/ultrasoc-smb.c
create mode 100644 drivers/hwtracing/coresight/ultrasoc-smb.h

diff --git a/drivers/hwtracing/coresight/Kconfig b/drivers/hwtracing/coresight/Kconfig
index 45c1eb5dfcb7..cb17c207a728 100644
--- a/drivers/hwtracing/coresight/Kconfig
+++ b/drivers/hwtracing/coresight/Kconfig
@@ -201,4 +201,15 @@ config CORESIGHT_TRBE
To compile this driver as a module, choose M here: the module will be
called coresight-trbe.
+
+config ULTRASOC_SMB
+ tristate "Ultrasoc system memory buffer drivers"
+ depends on ARM64 && CORESIGHT_LINKS_AND_SINKS

I still think it is not a good idea to leave the ACPI dependency out
of this. THe driver is unusable on a system without ACPI and the only
other reason to build this would when using COMPILE_TEST.
So, we could add something like :

depends on ACPI || COMPILE_TEST
depends on ARM64 && CORESIGHT_LINKS_AND_SINKS

OK, I will do that.
Thanks.

+ help
+ This driver provides support for the Ultrasoc system memory buffer (SMB).
+ SMB is responsible for receiving the trace data from Coresight ETM devices
+ and storing them to a system buffer.
+
+ To compile this driver as a module, choose M here: the module will be
+ called ultrasoc-smb.
endif
diff --git a/drivers/hwtracing/coresight/Makefile b/drivers/hwtracing/coresight/Makefile
index b6c4a48140ec..344dba8d6ff8 100644
--- a/drivers/hwtracing/coresight/Makefile
+++ b/drivers/hwtracing/coresight/Makefile
@@ -27,3 +27,4 @@ obj-$(CONFIG_CORESIGHT_CTI) += coresight-cti.o
obj-$(CONFIG_CORESIGHT_TRBE) += coresight-trbe.o
coresight-cti-y := coresight-cti-core.o coresight-cti-platform.o \
coresight-cti-sysfs.o
+obj-$(CONFIG_ULTRASOC_SMB) += ultrasoc-smb.o
diff --git a/drivers/hwtracing/coresight/ultrasoc-smb.c b/drivers/hwtracing/coresight/ultrasoc-smb.c
new file mode 100644
index 000000000000..1957796cbab2
--- /dev/null
+++ b/drivers/hwtracing/coresight/ultrasoc-smb.c
@@ -0,0 +1,656 @@
+// SPDX-License-Identifier: (GPL-2.0 OR MIT)
+/*
+ * Siemens System Memory Buffer driver.
+ * Copyright(c) 2022, HiSilicon Limited.
+ */
+
+#include <linux/atomic.h>
+#include <linux/acpi.h>
+#include <linux/circ_buf.h>
+#include <linux/err.h>
+#include <linux/fs.h>
+#include <linux/module.h>
+#include <linux/mod_devicetable.h>
+#include <linux/platform_device.h>
+
+#include "coresight-etm-perf.h"
+#include "coresight-priv.h"
+#include "ultrasoc-smb.h"
+
+DEFINE_CORESIGHT_DEVLIST(sink_devs, "ultra_smb");
+
+#define ULTRASOC_SMB_DSM_UUID "82ae1283-7f6a-4cbe-aa06-53e8fb24db18"
+
+static bool smb_buffer_is_empty(struct smb_drv_data *drvdata)
+{
+ u32 buf_status = readl(drvdata->base + SMB_LB_INT_STS_REG);
+
+ return !FIELD_GET(SMB_LB_INT_STS_NOT_EMPTY_MSK, buf_status);
+}
+
+static void smb_buffer_sync_status(struct smb_drv_data *drvdata)
+{
+ struct smb_data_buffer *sdb = &drvdata->sdb;
+
+ sdb->wr_offset = readl(drvdata->base + SMB_LB_WR_ADDR_REG) -
+ sdb->start_addr;
+ sdb->rd_offset = readl(drvdata->base + SMB_LB_RD_ADDR_REG) -
+ sdb->start_addr;
+ sdb->full = sdb->wr_offset == sdb->rd_offset &&
+ !smb_buffer_is_empty(drvdata);
+}
+
+static void smb_reset_buffer_status(struct smb_drv_data *drvdata)
+{
+ /* All other bits are reserved and shall be 0 */
+ writel(SMB_LB_INT_STS_RESET, drvdata->base + SMB_LB_INT_STS_REG);
+}
+
+/* Purge data remaining in hardware path in case them influence next trace */

minor nit:

/*
* Purge data remaining in hardware path to avoid corrupting the next
* session.
*/

Ok, will fix it.
Thanks.

+static void smb_purge_data(struct smb_drv_data *drvdata)
+{
+ /* All other bits are reserved and shall be 0 */
+ writel(SMB_LB_PURGE_PURGED, drvdata->base + SMB_LB_PURGE_REG);
+}
+
+static void smb_update_data_size(struct smb_drv_data *drvdata)
+{
+ struct smb_data_buffer *sdb = &drvdata->sdb;
+
+ smb_purge_data(drvdata);
+ smb_buffer_sync_status(drvdata);
+ if (sdb->full) {
+ sdb->data_size = sdb->buf_size;
+ return;
+ }
+
+ sdb->data_size = CIRC_CNT(sdb->wr_offset, sdb->rd_offset,
+ sdb->buf_size);
+}
+
+static int smb_open(struct inode *inode, struct file *file)
+{
+ struct smb_drv_data *drvdata = container_of(file->private_data,
+ struct smb_drv_data, miscdev);
+ int ret = 0;
+
+ mutex_lock(&drvdata->mutex);
+
+ if (drvdata->reading) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ if (atomic_read(drvdata->csdev->refcnt)) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ drvdata->reading = true;
+out:
+ mutex_unlock(&drvdata->mutex);
+
+ return ret;
+}
+
+static ssize_t smb_read(struct file *file, char __user *data, size_t len,
+ loff_t *ppos)
+{
+ struct smb_drv_data *drvdata = container_of(file->private_data,
+ struct smb_drv_data, miscdev);
+ struct smb_data_buffer *sdb = &drvdata->sdb;
+ struct device *dev = &drvdata->csdev->dev;
+ ssize_t to_copy = 0;
+
+ mutex_lock(&drvdata->mutex);
+
+ if (!len)
+ goto out;
+
+ /*
+ * In sysfs mode, size need to be update in the following two cases:
+ * 1) Start dumping data.
+ * 2) End dump data, make sure there is no remaining data in
+ * the hardware path. Because the remaining data cannot be purged
+ * when the buffer is full.
+ */
+ if (!sdb->data_size) {
+ smb_update_data_size(drvdata);
+ if (!sdb->data_size)
+ goto out;
+ }
+
+ to_copy = min(sdb->data_size, len);
+
+ /* Copy parts of trace data when read pointer wrap around SMB buffer */
+ if (sdb->rd_offset + to_copy > sdb->buf_size)
+ to_copy = sdb->buf_size - sdb->rd_offset;
+
+ if (copy_to_user(data, (void *)sdb->buf_base + sdb->rd_offset,
+ to_copy)) {
+ dev_dbg(dev, "Failed to copy data to user\n");
+ to_copy = -EFAULT;
+ goto out;
+ }
+
+ *ppos += to_copy;
+ sdb->data_size -= to_copy;
+ sdb->rd_offset += to_copy;
+ sdb->rd_offset %= sdb->buf_size;
+ writel(sdb->start_addr + sdb->rd_offset,
+ drvdata->base + SMB_LB_RD_ADDR_REG);
+ dev_dbg(dev, "%zu bytes copied\n", to_copy);
+out:
+ if (!sdb->data_size)
+ smb_reset_buffer_status(drvdata);
+ mutex_unlock(&drvdata->mutex);
+
+ return to_copy;
+}
+
+static int smb_release(struct inode *inode, struct file *file)
+{
+ struct smb_drv_data *drvdata = container_of(file->private_data,
+ struct smb_drv_data, miscdev);
+
+ mutex_lock(&drvdata->mutex);
+ drvdata->reading = false;
+ mutex_unlock(&drvdata->mutex);
+
+ return 0;
+}
+
+static const struct file_operations smb_fops = {
+ .owner = THIS_MODULE,
+ .open = smb_open,
+ .read = smb_read,
+ .release = smb_release,
+ .llseek = no_llseek,
+};
+
+static ssize_t buf_size_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(dev->parent);
+
+ return sysfs_emit(buf, "0x%lx\n", drvdata->sdb.buf_size);
+}
+static DEVICE_ATTR_RO(buf_size);
+
+static struct attribute *smb_sink_attrs[] = {
+ coresight_simple_reg32(read_pos, SMB_LB_RD_ADDR_REG),
+ coresight_simple_reg32(write_pos, SMB_LB_WR_ADDR_REG),
+ coresight_simple_reg32(buf_status, SMB_LB_INT_STS_REG),
+ &dev_attr_buf_size.attr,
+ NULL
+};
+
+static const struct attribute_group smb_sink_group = {
+ .attrs = smb_sink_attrs,
+ .name = "mgmt",
+};
+
+static const struct attribute_group *smb_sink_groups[] = {
+ &smb_sink_group,
+ NULL
+};
+
+static void smb_enable_hw(struct smb_drv_data *drvdata)
+{
+ writel(SMB_GLB_EN_HW_ENABLE, drvdata->base + SMB_GLB_EN_REG);

What happens to the RD_ADDR_REG and WR_ADDR_REG at enable ? Do they
reset to the base of the buf or do we need to program at enable ?
Are there any other modes of operation ? Or is it always in circular
buffer mode ?
The RD_ADDR_REG and WR_ADDR_REG register is filled in BIOS, and
WR_ADDR_REG is RO mode.
When the SMB is enabled, the values of RD_ADDR_REG and WR_ADDR_REG
remain unchanged.
UltraSoc SMB is always in circular buffer mode.
Thanks.

+}
+
+static void smb_disable_hw(struct smb_drv_data *drvdata)
+{
+ writel(0x0, drvdata->base + SMB_GLB_EN_REG);
+}
+
+static void smb_enable_sysfs(struct coresight_device *csdev)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(csdev->dev.parent);
+
+ if (drvdata->mode != CS_MODE_DISABLED)
+ return;
+
+ smb_enable_hw(drvdata);
+ drvdata->mode = CS_MODE_SYSFS;
+}
+
+static int smb_enable_perf(struct coresight_device *csdev, void *data)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(csdev->dev.parent);
+ struct perf_output_handle *handle = data;
+ struct cs_buffers *buf = etm_perf_sink_config(handle);
+ pid_t pid;
+
+ if (!buf)
+ return -EINVAL;
+
+ /* Get a handle on the pid of the target process */
+ pid = buf->pid;
+
+ /* Device is already in used by other session */
+ if (drvdata->pid != -1 && drvdata->pid != pid)
+ return -EBUSY;
+
+ if (drvdata->pid == -1) {
+ smb_enable_hw(drvdata);
+ drvdata->pid = pid;
+ drvdata->mode = CS_MODE_PERF;
+ }
+
+ return 0;
+}
+
+static int smb_enable(struct coresight_device *csdev, u32 mode, void *data)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(csdev->dev.parent);
+ int ret = 0;
+
+ mutex_lock(&drvdata->mutex);
+
+ /* Do nothing, the trace data is reading by other interface now */
+ if (drvdata->reading) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ /* Do nothing, the SMB is already enabled as other mode */
+ if (drvdata->mode != CS_MODE_DISABLED && drvdata->mode != mode) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ switch (mode) {
+ case CS_MODE_SYSFS:
+ smb_enable_sysfs(csdev);
+ break;
+ case CS_MODE_PERF:
+ ret = smb_enable_perf(csdev, data);
+ break;
+ default:
+ ret = -EINVAL;
+ }
+
+ if (ret)
+ goto out;
+
+ atomic_inc(csdev->refcnt);
+
+ dev_dbg(&csdev->dev, "Ultrasoc SMB enabled\n");
+out:
+ mutex_unlock(&drvdata->mutex);
+
+ return ret;
+}
+
+static int smb_disable(struct coresight_device *csdev)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(csdev->dev.parent);
+ int ret = 0;
+
+ mutex_lock(&drvdata->mutex);
+
+ if (drvdata->reading) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ if (atomic_dec_return(csdev->refcnt)) {
+ ret = -EBUSY;
+ goto out;
+ }
+
+ /* Complain if we (somehow) got out of sync */
+ WARN_ON_ONCE(drvdata->mode == CS_MODE_DISABLED);
+
+ smb_disable_hw(drvdata);
+ smb_purge_data(drvdata);
+
+ /* Dissociate from the target process. */
+ drvdata->pid = -1;
+ drvdata->mode = CS_MODE_DISABLED;
+
+ dev_dbg(&csdev->dev, "Ultrasoc SMB disabled\n");
+out:
+ mutex_unlock(&drvdata->mutex);
+
+ return ret;
+}
+
+static void *smb_alloc_buffer(struct coresight_device *csdev,
+ struct perf_event *event, void **pages,
+ int nr_pages, bool overwrite)
+{
+ struct cs_buffers *buf;
+ int node;
+
+ node = (event->cpu == -1) ? NUMA_NO_NODE : cpu_to_node(event->cpu);
+ buf = kzalloc_node(sizeof(struct cs_buffers), GFP_KERNEL, node);
+ if (!buf)
+ return NULL;
+
+ buf->snapshot = overwrite;
+ buf->nr_pages = nr_pages;
+ buf->data_pages = pages;
+ buf->pid = task_pid_nr(event->owner);
+
+ return buf;
+}
+
+static void smb_free_buffer(void *config)
+{
+ struct cs_buffers *buf = config;
+
+ kfree(buf);
+}
+
+static void smb_sync_perf_buffer(struct smb_drv_data *drvdata,
+ struct cs_buffers *buf,
+ unsigned long head,
+ unsigned long data_size)
+{
+ struct smb_data_buffer *sdb = &drvdata->sdb;
+ char **dst_pages = (char **)buf->data_pages;
+ unsigned long to_copy;
+ long pg_idx, pg_offset;
+
+ pg_idx = head >> PAGE_SHIFT;
+ pg_offset = head & (PAGE_SIZE - 1);
+
+ while (data_size) {
+ unsigned long pg_space = PAGE_SIZE - pg_offset;
+
+ /* Copy parts of trace data when read pointer wrap around */
+ if (sdb->rd_offset + pg_space > sdb->buf_size)
+ to_copy = sdb->buf_size - sdb->rd_offset;
+ else
+ to_copy = min(data_size, pg_space);
+
+ memcpy(dst_pages[pg_idx] + pg_offset,
+ sdb->buf_base + sdb->rd_offset, to_copy);
+
+ pg_offset += to_copy;
+ if (pg_offset >= PAGE_SIZE) {
+ pg_offset = 0;
+ pg_idx++;
+ pg_idx %= buf->nr_pages;
+ }
+ data_size -= to_copy;
+ sdb->rd_offset += to_copy;
+ sdb->rd_offset %= sdb->buf_size;
+ }
+
+ sdb->data_size = 0;

+ writel(sdb->start_addr + sdb->rd_offset,
+ drvdata->base + SMB_LB_RD_ADDR_REG);

Is this safe ? When the buffer was not full and the buffer
was not copied in full (i.e., perf_buffer < buffer_size),
the RD_ADDR_REG won't reach the WR_ADDR_REG. So can that affect
the next session ? See more below :

Yes, In this case. "sdb->rd_offset" will add the offset of discarded data
(buffer_size - perf_buffer) in the smb_update_buffer() function. If the buffer
was not full, the values of RD_ADDR_REG and WR_ADDR_REG ​​will always
be the same.
When SMB is full, the data remained in link only can be written into buffer
after updated RD_ADDR_REG, and the value of WR_ADDR_REG will be advanced.
So in this case we need synchronize the WR_ADDR_REG to RD_ADDR_REG again.
Thanks.

+
+ /*
+ * Data remained in link cannot be purged when SMB is full, so
+ * synchronize the read pointer to write pointer, to make sure
+ * these remained data won't influence next trace.
+ */
+ if (sdb->full) {
+ smb_purge_data(drvdata);
+ writel(readl(drvdata->base + SMB_LB_WR_ADDR_REG),
+ drvdata->base + SMB_LB_RD_ADDR_REG);

Or, in other words, shouldn't we do this irrespective of whether the
buffer->full or not ? We set the TRUNCATED flag appropriately when
there was data lost and thus we must not leave any trace from a previous
"schedule" in for the "next" collection. So, we must reset the RD/WR
registers to the base.

Yes, You are right, but the WR_ADDR_REG register is RO mode.
Thanks.

+ }


+ smb_reset_buffer_status(drvdata);
+}
+
+static unsigned long smb_update_buffer(struct coresight_device *csdev,
+ struct perf_output_handle *handle,
+ void *sink_config)
+{
+ struct smb_drv_data *drvdata = dev_get_drvdata(csdev->dev.parent);
+ struct smb_data_buffer *sdb = &drvdata->sdb;
+ struct cs_buffers *buf = sink_config;
+ unsigned long data_size = 0;
+ bool lost = false;
+
+ if (!buf)
+ return 0;
+
+ mutex_lock(&drvdata->mutex);
+
+ /* Don't do anything if another tracer is using this sink. */
+ if (atomic_read(csdev->refcnt) != 1)
+ goto out;
+
+ smb_disable_hw(drvdata);
+ smb_update_data_size(drvdata);
+ data_size = sdb->data_size;
+
+ /*
+ * The SMB buffer may be bigger than the space available in the
+ * perf ring buffer (handle->size). If so advance the offset so
+ * that we get the latest trace data.
+ */
+ if (data_size > handle->size) {
+ sdb->rd_offset += data_size - handle->size;
+ sdb->rd_offset %= sdb->buf_size;
+ data_size = handle->size;
+ lost = true;
+ } > +
+ smb_sync_perf_buffer(drvdata, buf, handle->head, data_size);
+ if (!buf->snapshot && lost)
+ perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED);
+

+ smb_enable_hw(drvdata);

We do not need this and could leave this disabled at this stage. The
perf session has stopped the "source" anyway.

Yes, I will do that.
Thanks.
+out:
+ mutex_unlock(&drvdata->mutex);
+
+ return data_size;
+}
+
+static const struct coresight_ops_sink smb_cs_ops = {
+ .enable = smb_enable,
+ .disable = smb_disable,
+ .alloc_buffer = smb_alloc_buffer,
+ .free_buffer = smb_free_buffer,
+ .update_buffer = smb_update_buffer,
+};
+
+static const struct coresight_ops cs_ops = {
+ .sink_ops = &smb_cs_ops,
+};
+
+static int smb_init_data_buffer(struct platform_device *pdev,
+ struct smb_data_buffer *sdb)
+{
+ struct resource *res;
+ void *base;
+
+ res = platform_get_resource(pdev, IORESOURCE_MEM, SMB_BUF_ADDR_RES);
+ if (IS_ERR(res)) {
+ dev_err(&pdev->dev, "SMB device failed to get resource\n");
+ return -EINVAL;
+ }
+
+ sdb->start_addr = FIELD_GET(SMB_BUF_ADDR_LO_MSK, res->start);
+ sdb->buf_size = resource_size(res);
+ if (sdb->buf_size == 0)
+ return -EINVAL;
+
+ /*
+ * This is a chunk of memory, use classic mapping with better
+ * performance.
+ */
+ base = devm_memremap(&pdev->dev, sdb->start_addr, sdb->buf_size,
+ MEMREMAP_WB);
+ if (IS_ERR(base))
+ return PTR_ERR(base);
+
+ sdb->buf_base = base;
+
+ return 0;
+}
+
+static void smb_init_hw(struct smb_drv_data *drvdata)
+{
+ /* First disable SMB and clear the status of SMB buffer */
+ smb_reset_buffer_status(drvdata);
+ smb_disable_hw(drvdata);
+ smb_purge_data(drvdata);
+
+ writel(SMB_LB_CFG_LO_DEFAULT, drvdata->base + SMB_LB_CFG_LO_REG);
+ writel(SMB_LB_CFG_HI_DEFAULT, drvdata->base + SMB_LB_CFG_HI_REG);
+ writel(SMB_GLB_CFG_DEFAULT, drvdata->base + SMB_GLB_CFG_REG);
+ writel(SMB_GLB_INT_CFG, drvdata->base + SMB_GLB_INT_REG);
+ writel(SMB_LB_INT_CTRL_CFG, drvdata->base + SMB_LB_INT_CTRL_REG);

Does this come with interrupt on overflow ? Do we not use this ?

When the buffer overflow, no interrupt will come.
Interrupt will upgrade SMB_LB_INT_STS_REG register status if start trace.
Thanks.
Rest looks fine to me.


Suzuki
Best regards,
HeJunhao.
_______________________________________________
CoreSight mailing list -- coresight@xxxxxxxxxxxxxxxx
To unsubscribe send an email to coresight-leave@xxxxxxxxxxxxxxxx

.