RE: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log

From: Liming Sun
Date: Thu May 11 2023 - 11:02:45 EST

Next message: Rsplwe: "[PATCH] ASoC: amd: yc: Add MECHREVO Jiaolong Series MRID6 into DMI table"
Previous message: Zi Yan: "Re: [RESEND PATCH v1 3/5] mm: Fix failure to unmap pte on highmem systems"
In reply to: Ilpo Järvinen: "Re: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log"
Next in thread: David Thompson: "RE: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Thanks for the comments.

> -----Original Message-----
> From: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>
> Sent: Wednesday, May 10, 2023 10:05 AM
> To: Liming Sun <limings@xxxxxxxxxx>
> Cc: Vadim Pasternak <vadimp@xxxxxxxxxx>; David Thompson
> <davthompson@xxxxxxxxxx>; Hans de Goede <hdegoede@xxxxxxxxxx>; Mark
> Gross <markgross@xxxxxxxxxx>; platform-driver-x86@xxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log
>
> On Wed, 10 May 2023, Liming Sun wrote:
>
> > This commit adds sysfs interface to be used to write into the
> > boot log which is 1KB HW buffer on BlueField SoC. The same log
> > buffer is also used by firmware code like ATF/UEFI, and can be
> > displayed by userspace tools or from external host via USB/PCIe.
> >
> > Signed-off-by: Liming Sun <limings@xxxxxxxxxx>
> > Reviewed-by: Vadim Pasternak <vadimp@xxxxxxxxxx>
> > ---
> > v1->v2:
> > Fixes for comments from Hans:
> > - Add more details in Documentation about the log levels;
> > - Replace 0x0a with '\n';
> > - Solve comment 'Why len + 1, this seems intended to assume some
> > separator between prefix'. The change is to remove the '+ 1'
> > here to avoid confusion. Yes, it's trying to remove the space
> > separator. Since the next block 'Ignore leading spaces' already
> > has similar logic, no need for the '+ 1" here.
> > v1: Initial version.
> > ---
> > .../testing/sysfs-platform-mellanox-bootctl | 9 ++
> > drivers/platform/mellanox/mlxbf-bootctl.c | 141 ++++++++++++++++++
> > 2 files changed, 150 insertions(+)
> >
> > diff --git a/Documentation/ABI/testing/sysfs-platform-mellanox-bootctl
> b/Documentation/ABI/testing/sysfs-platform-mellanox-bootctl
> > index 9b99a81babb1..4c5c02d8f870 100644
> > --- a/Documentation/ABI/testing/sysfs-platform-mellanox-bootctl
> > +++ b/Documentation/ABI/testing/sysfs-platform-mellanox-bootctl
> > @@ -75,3 +75,12 @@ KernelVersion: 6.4
> > Contact: "Liming Sun <limings@xxxxxxxxxx>"
> > Description:
> > The file used to access the BlueField boot fifo.
> > +
> > +What: /sys/bus/platform/devices/MLNXBF04:00/rsh_log
> > +Date: May 2023
> > +KernelVersion: 6.4
> > +Contact: "Liming Sun <limings@xxxxxxxxxx>"
> > +Description:
> > + The file used to write BlueField boot log with the format
> > + "[INFO|WARN|ERR|ASSERT ]<msg>". Log level 'INFO' is used by
> > + default if not specified.
> > diff --git a/drivers/platform/mellanox/mlxbf-bootctl.c
> b/drivers/platform/mellanox/mlxbf-bootctl.c
> > index 1bad1d278672..e88ce68acb89 100644
> > --- a/drivers/platform/mellanox/mlxbf-bootctl.c
> > +++ b/drivers/platform/mellanox/mlxbf-bootctl.c
> > @@ -45,10 +45,38 @@ static const char * const
> mlxbf_bootctl_lifecycle_states[] = {
> > [3] = "RMA",
> > };
> >
> > +/* Log header format. */
> > +#define MLXBF_RSH_LOG_TYPE_SHIFT 56
> > +#define MLXBF_RSH_LOG_LEN_SHIFT 48
> > +#define MLXBF_RSH_LOG_LEVEL_SHIFT 0
>
> Use GENMASK_ULL() instead here and FIELD_PREP() below.

Updated in v3.

>
> > +
> > +/* Log module ID and type (only MSG type in Linux driver for now). */
> > +#define MLXBF_RSH_LOG_TYPE_MSG 0x04ULL
> > +
> > +/* Log ctl/data register offset. */
> > +#define MLXBF_RSH_SCRATCH_BUF_CTL_OFF 0
> > +#define MLXBF_RSH_SCRATCH_BUF_DATA_OFF 0x10
> > +
> > +/* Log message levels. */
> > +enum {
> > + MLXBF_RSH_LOG_INFO,
> > + MLXBF_RSH_LOG_WARN,
> > + MLXBF_RSH_LOG_ERR
> > +};
> > +
> > /* Mapped pointer for RSH_BOOT_FIFO_DATA and RSH_BOOT_FIFO_COUNT
> register. */
> > static void __iomem *mlxbf_rsh_boot_data;
> > static void __iomem *mlxbf_rsh_boot_cnt;
> >
> > +/* Mapped pointer for rsh log semaphore/ctrl/data register. */
> > +static void __iomem *mlxbf_rsh_semaphore;
> > +static void __iomem *mlxbf_rsh_scratch_buf_ctl;
> > +static void __iomem *mlxbf_rsh_scratch_buf_data;
> > +
> > +/* Rsh log levels. */
> > +static const char * const mlxbf_rsh_log_level[] = {
> > + "INFO", "WARN", "ERR", "ASSERT"};
> > +
> > /* ARM SMC call which is atomic and no need for lock. */
> > static int mlxbf_bootctl_smc(unsigned int smc_op, int smc_arg)
> > {
> > @@ -266,12 +294,111 @@ static ssize_t fw_reset_store(struct device *dev,
> > return count;
> > }
> >
> > +/* Size(8-byte words) of the log buffer. */
> > +#define RSH_SCRATCH_BUF_CTL_IDX_MASK 0x7f
> > +
> > +static int mlxbf_rsh_log_sem_lock(void)
> > +{
> > + unsigned long timeout;
> > +
> > + /* Take the semaphore. */
> > + timeout = jiffies + msecs_to_jiffies(100);
> > + while (readq(mlxbf_rsh_semaphore)) {
> > + if (time_after(jiffies, timeout))
> > + return -ETIMEDOUT;
> > + }
> > +
> > + return 0;
>
> readq_poll_timeout()

Updated in v3.

>
> > +}
> > +
> > +static void mlxbf_rsh_log_sem_unlock(void)
> > +{
> > + writeq(0, mlxbf_rsh_semaphore);
> > +}
> > +
> > +static ssize_t rsh_log_store(struct device *dev,
> > + struct device_attribute *attr,
> > + const char *buf, size_t count)
> > +{
> > + int rc, idx, num, len, size = (int)count, level = MLXBF_RSH_LOG_INFO;
>
> Why casting to int, why not keep sizes as size_t??

Updated in v3 to keep type 'size_t'.

>
> > + u64 data;
> > +
> > + if (!size)
> > + return -EINVAL;
> > +
> > + if (!mlxbf_rsh_semaphore || !mlxbf_rsh_scratch_buf_ctl)
> > + return -EOPNOTSUPP;
> > +
> > + /* Ignore line break at the end. */
> > + if (buf[size - 1] == '\n')
> > + size--;
> > +
> > + /* Check the message prefix. */
> > + for (idx = 0; idx < ARRAY_SIZE(mlxbf_rsh_log_level); idx++) {
> > + len = strlen(mlxbf_rsh_log_level[idx]);
> > + if (len + 1 < size &&
> > + !strncmp(buf, mlxbf_rsh_log_level[idx], len)) {
> > + buf += len;
> > + size -= len;
> > + level = idx;
> > + break;
> > + }
> > + }
> > +
> > + /* Ignore leading spaces. */
> > + while (size > 0 && buf[0] == ' ') {
> > + size--;
> > + buf++;
> > + }
> > +
> > + /* Take the semaphore. */
> > + rc = mlxbf_rsh_log_sem_lock();
> > + if (rc)
> > + return rc;
> > +
> > + /* Calculate how many words are available. */
> > + num = (size + sizeof(u64) - 1) / sizeof(u64);
>
> DIV_ROUND_UP()

Updated in v3 to use DIV_ROUND_UP().

>
> > + idx = readq(mlxbf_rsh_scratch_buf_ctl);
> > + if (idx + num + 1 >= RSH_SCRATCH_BUF_CTL_IDX_MASK)
> > + num = RSH_SCRATCH_BUF_CTL_IDX_MASK - idx - 1;
>
> min() ?

Updated in v3 to use min().

>
> > + if (num <= 0)
> > + goto done;
> > +
> > + /* Write Header. */
> > + data = (MLXBF_RSH_LOG_TYPE_MSG <<
> MLXBF_RSH_LOG_TYPE_SHIFT) |
> > + ((u64)num << MLXBF_RSH_LOG_LEN_SHIFT) |
> > + ((u64)level << MLXBF_RSH_LOG_LEVEL_SHIFT);
> > + writeq(data, mlxbf_rsh_scratch_buf_data);
> > +
> > + /* Write message. */
> > + for (idx = 0, len = size; idx < num && len > 0; idx++) {
>
> size_t -> int conversion, why? Why you need another variable besides size
> anyway?
>
> How can len <= 0 occur when idx < num is true?

Updated in v3. Simplified the code to remove the 'len' from this loop.

>
> > + if (len <= sizeof(u64)) {
>
> Why = ?

Updated in v3. '<' or '<=' will have the same effect in this block.
But changed it to '<' to avoid confusion.

>
> > + data = 0;
> > + memcpy(&data, buf, len);
> > + len = 0;
> > + } else {
> > + memcpy(&data, buf, sizeof(u64));
> > + len -= sizeof(u64);
> > + buf += sizeof(u64);
> > + }
> > + writeq(data, mlxbf_rsh_scratch_buf_data);
> > + }
> > +
> > +done:
> > + /* Release the semaphore. */
> > + mlxbf_rsh_log_sem_unlock();
> > +
> > + /* Ignore the rest if no more space. */
> > + return count;
> > +}
> > +
> > static DEVICE_ATTR_RW(post_reset_wdog);
> > static DEVICE_ATTR_RW(reset_action);
> > static DEVICE_ATTR_RW(second_reset_action);
> > static DEVICE_ATTR_RO(lifecycle_state);
> > static DEVICE_ATTR_RO(secure_boot_fuse_state);
> > static DEVICE_ATTR_WO(fw_reset);
> > +static DEVICE_ATTR_WO(rsh_log);
> >
> > static struct attribute *mlxbf_bootctl_attrs[] = {
> > &dev_attr_post_reset_wdog.attr,
> > @@ -280,6 +407,7 @@ static struct attribute *mlxbf_bootctl_attrs[] = {
> > &dev_attr_lifecycle_state.attr,
> > &dev_attr_secure_boot_fuse_state.attr,
> > &dev_attr_fw_reset.attr,
> > + &dev_attr_rsh_log.attr,
> > NULL
> > };
> >
> > @@ -345,6 +473,7 @@ static bool mlxbf_bootctl_guid_match(const guid_t
> *guid,
> > static int mlxbf_bootctl_probe(struct platform_device *pdev)
> > {
> > struct arm_smccc_res res = { 0 };
> > + void __iomem *reg;
> > guid_t guid;
> > int ret;
> >
> > @@ -358,6 +487,18 @@ static int mlxbf_bootctl_probe(struct
> platform_device *pdev)
> > if (IS_ERR(mlxbf_rsh_boot_cnt))
> > return PTR_ERR(mlxbf_rsh_boot_cnt);
> >
> > + /* Get the resource of the bootfifo counter register. */
> > + mlxbf_rsh_semaphore = devm_platform_ioremap_resource(pdev, 2);
> > + if (IS_ERR(mlxbf_rsh_semaphore))
> > + return PTR_ERR(mlxbf_rsh_semaphore);
> > +
> > + /* Get the resource of the bootfifo counter register. */
> > + reg = devm_platform_ioremap_resource(pdev, 3);
> > + if (IS_ERR(reg))
> > + return PTR_ERR(reg);
> > + mlxbf_rsh_scratch_buf_ctl = reg +
> MLXBF_RSH_SCRATCH_BUF_CTL_OFF;
> > + mlxbf_rsh_scratch_buf_data = reg +
> MLXBF_RSH_SCRATCH_BUF_DATA_OFF;
> > +
> > /* Ensure we have the UUID we expect for this service. */
> > arm_smccc_smc(MLXBF_BOOTCTL_SIP_SVC_UID, 0, 0, 0, 0, 0, 0, 0,
> &res);
> > guid_parse(mlxbf_bootctl_svc_uuid_str, &guid);
> >
>
> --
> i.

Next message: Rsplwe: "[PATCH] ASoC: amd: yc: Add MECHREVO Jiaolong Series MRID6 into DMI table"
Previous message: Zi Yan: "Re: [RESEND PATCH v1 3/5] mm: Fix failure to unmap pte on highmem systems"
In reply to: Ilpo Järvinen: "Re: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log"
Next in thread: David Thompson: "RE: [PATCH v2] mlxbf-bootctl: Add sysfs file for BlueField boot log"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]