Re: [PATCH v7 3/3] drm/xe/sysctrl: Add generic mailbox passthrough debugfs entry

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Fri, Sep 25, 2026 at 06:55:23AM -0700, Anoop, Vijay wrote:
> From: Anoop Vijay <anoop.c.vijay@intel.com>
> 
> Add a "mailbox" debugfs entry that allows any System Controller
> mailbox command to be issued for bring-up and debug. group and
> command are supplied by the caller as the first two tokens of every
> write, followed by optional per-command response timeout and
> response-length overrides and the payload bytes:
> 
> - Usage: echo "<group> <command> [timeout=<ms>] [resplen=<N>] [byte0 byte1 ...]" \
>              > /sys/kernel/debug/dri/0/sc/mailbox
>          cat /sys/kernel/debug/dri/0/sc/mailbox
> 
> Signed-off-by: Anoop Vijay <anoop.c.vijay@intel.com>
> ---
> v7:
> - Add per-command timeout override to mailbox debugfs
> - Add per-command response-length override to mailbox debugfs
> 
> v6 (Rodrigo):
> - Reject non-zero offset writes for mailbox entry
> - Account for mailbox header size when validating input payload length
> 
> v5:
> - Add per-entry locking for debugfs accesses
> - Skip empty mailbox group/command tokens
> 
> v4:
> - Added a generic mailbox debugfs entry
> ---
>  drivers/gpu/drm/xe/xe_sysctrl_debugfs.c       | 172 ++++++++++++++++++
>  drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h |   1 +
>  drivers/gpu/drm/xe/xe_sysctrl_types.h         |   6 +
>  3 files changed, 179 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_sysctrl_debugfs.c b/drivers/gpu/drm/xe/xe_sysctrl_debugfs.c
> index 10675cd774ee..ba904cb24512 100644
> --- a/drivers/gpu/drm/xe/xe_sysctrl_debugfs.c
> +++ b/drivers/gpu/drm/xe/xe_sysctrl_debugfs.c
> @@ -299,10 +299,179 @@ static void xe_sysctrl_register_entry(struct dentry *root, struct xe_sysctrl_deb
>  	entry->command = command;
>  	entry->response_len = 0;
>  	entry->status = 0;
> +	entry->timeout_ms = 0;
>  
>  	debugfs_create_file(name, 0644, root, entry, fops);
>  }
>  
> +static ssize_t xe_sysctrl_mailbox_write(struct file *file, const char __user *ubuf,
> +					size_t len, loff_t *offp)
> +{
> +	char *kbuf __free(kfree) = NULL;
> +	u8 *input __free(kfree) = NULL;
> +	struct seq_file *m = file->private_data;
> +	struct xe_sysctrl_debugfs_entry *entry = m->private;
> +	struct xe_device *xe = sc_to_xe(entry->sc);
> +	struct xe_sysctrl_mailbox_command cmd = {};
> +	char *token, *tmp;
> +	unsigned long val;
> +	unsigned int timeout_ms = 0;
> +	size_t input_len = 0;
> +	size_t max_input;
> +	size_t out_len = 0;
> +	size_t resp_len = XE_SYSCTRL_MB_MAX_DATA_SIZE;
> +	u8 group, command;
> +	int status;
> +
> +	if (*offp)
> +		return -ESPIPE;
> +
> +	if (len == 0 || len >= PAGE_SIZE)
> +		return -EINVAL;
> +
> +	kbuf = kmalloc(len + 1, GFP_KERNEL);
> +	if (!kbuf)
> +		return -ENOMEM;
> +
> +	max_input = min_t(size_t, len, XE_SYSCTRL_MB_MAX_DATA_SIZE);
> +	input = kmalloc(max_input, GFP_KERNEL);
> +	if (!input)
> +		return -ENOMEM;
> +
> +	if (copy_from_user(kbuf, ubuf, len))
> +		return -EFAULT;
> +	kbuf[len] = '\0';
> +
> +	tmp = kbuf;
> +
> +	do {
> +		token = strsep(&tmp, " \t\n");
> +	} while (token && *token == '\0');
> +	if (!token || kstrtoul(token, 0, &val) || val > 0xFF) {
> +		xe_err(xe, "sysctrl: invalid mailbox group id\n");
> +		return -EINVAL;
> +	}
> +	group = (u8)val;
> +
> +	do {
> +		token = strsep(&tmp, " \t\n");
> +	} while (token && *token == '\0');
> +	if (!token || kstrtoul(token, 0, &val) || val > SYSCTRL_HDR_COMMAND_MAX) {
> +		xe_err(xe, "sysctrl: invalid mailbox command id (max 0x%x)\n",
> +		       SYSCTRL_HDR_COMMAND_MAX);
> +		return -EINVAL;
> +	}
> +	command = (u8)val;
> +
> +	do {
> +		token = strsep(&tmp, " \t\n");
> +	} while (token && *token == '\0');
> +	if (token && strstarts(token, "timeout=")) {
> +		if (kstrtouint(token + strlen("timeout="), 0, &timeout_ms) ||
> +		    timeout_ms > XE_SYSCTRL_MB_MAX_TIMEOUT_MS) {
> +			xe_err(xe, "sysctrl: invalid mailbox timeout (max %u ms)\n",
> +			       XE_SYSCTRL_MB_MAX_TIMEOUT_MS);
> +			return -EINVAL;
> +		}
> +		token = strsep(&tmp, " \t\n");

Sashiko noticed that in this and below case you might fail on double space.
I wonder if we have a cleaner way of doing these parses...

> +	}
> +
> +	if (token && strstarts(token, "resplen=")) {
> +		unsigned int resplen;
> +
> +		if (kstrtouint(token + strlen("resplen="), 0, &resplen) ||
> +		    resplen > XE_SYSCTRL_MB_MAX_DATA_SIZE) {
> +			xe_err(xe, "sysctrl: invalid mailbox resplen (max %zu bytes)\n",
> +			       (size_t)XE_SYSCTRL_MB_MAX_DATA_SIZE);
> +			return -EINVAL;
> +		}
> +		resp_len = resplen;
> +		token = strsep(&tmp, " \t\n");
> +	}
> +
> +	while (token) {
> +		if (*token == '\0') {
> +			token = strsep(&tmp, " \t\n");
> +			continue;
> +		}
> +
> +		if (input_len >= max_input) {
> +			xe_err(xe, "sysctrl: mailbox payload too large (max %zu bytes)\n",
> +			       max_input);
> +			return -EINVAL;
> +		}
> +
> +		if (kstrtoul(token, 0, &val) || val > 0xFF) {
> +			xe_err(xe, "sysctrl: invalid mailbox payload byte '%s'\n", token);
> +			return -EINVAL;
> +		}
> +
> +		input[input_len++] = (u8)val;
> +		token = strsep(&tmp, " \t\n");
> +	}
> +
> +	scoped_guard(mutex, &entry->lock) {
> +		entry->group = group;
> +		entry->command = command;
> +		entry->timeout_ms = timeout_ms;
> +
> +		xe_sysctrl_create_command(&cmd, group, command, input_len ? input : NULL, input_len,
> +					  entry->response_buf, resp_len);
> +		cmd.timeout_ms = timeout_ms;
> +
> +		guard(xe_pm_runtime)(xe);
> +		status = xe_sysctrl_send_command(entry->sc, &cmd, &out_len);
> +		entry->status = status;
> +		entry->response_len = status ? 0 : out_len;
> +	}
> +
> +	return status ? status : len;
> +}
> +
> +static int xe_sysctrl_mailbox_show(struct seq_file *m, void *data)
> +{
> +	struct xe_sysctrl_debugfs_entry *entry = m->private;
> +	size_t i;
> +
> +	guard(mutex)(&entry->lock);
> +
> +	seq_printf(m, "Command: group=0x%02x cmd=0x%02x\n", entry->group, entry->command);
> +	seq_printf(m, "Timeout: %u ms%s\n", entry->timeout_ms ?: XE_SYSCTRL_MB_DEFAULT_TIMEOUT_MS,
> +		   entry->timeout_ms ? "" : " (default)");
> +	seq_printf(m, "Status: %d (%s)\n", entry->status, entry->status ? "FAILED" : "SUCCESS");
> +	seq_printf(m, "Response: %zu bytes\n", entry->response_len);
> +
> +	if (entry->response_len) {
> +		seq_puts(m, "Response data:\n");
> +		for (i = 0; i < entry->response_len; i++) {
> +			if (i && (i % 16) == 0)
> +				seq_putc(m, '\n');
> +			seq_printf(m, "%02x ", entry->response_buf[i]);
> +		}
> +		seq_putc(m, '\n');
> +	}
> +
> +	seq_puts(m, "\nUsage:\n");
> +	seq_puts(m, "  echo \"<group> <command> [timeout=<ms>] [resplen=<N>] [byte0 byte1 ...]\" > mailbox\n");
> +	seq_puts(m, "  cat mailbox\n");
> +
> +	return 0;
> +}
> +
> +static int xe_sysctrl_mailbox_open(struct inode *inode, struct file *file)
> +{
> +	return single_open(file, xe_sysctrl_mailbox_show, inode->i_private);
> +}
> +
> +static const struct file_operations xe_sysctrl_mailbox_fops = {
> +	.owner = THIS_MODULE,
> +	.open = xe_sysctrl_mailbox_open,
> +	.read = seq_read,
> +	.write = xe_sysctrl_mailbox_write,
> +	.llseek = seq_lseek,
> +	.release = single_release,
> +};
> +
>  /**
>   * xe_sysctrl_debugfs_register - Register debugfs entries for System Controller
>   * @sc: xe_sysctrl instance
> @@ -325,4 +494,7 @@ void xe_sysctrl_debugfs_register(struct xe_sysctrl *sc, struct dentry *parent)
>  	xe_sysctrl_register_entry(root, &sc->debugfs.ras_error_inject, sc, "ras_error_inject",
>  				  XE_SYSCTRL_GROUP_DIAG, XE_SYSCTRL_CMD_DIAG_RAS_ERR_INJECT,
>  				  &xe_sysctrl_ras_error_inject_fops);
> +
> +	xe_sysctrl_register_entry(root, &sc->debugfs.mailbox, sc, "mailbox", 0, 0,
> +				  &xe_sysctrl_mailbox_fops);

Should we put this behind the CONFIG_DRM_XE_DEBUG ?

>  }
> diff --git a/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h b/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> index 2554848e7484..a0628eb47bff 100644
> --- a/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> +++ b/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> @@ -155,5 +155,6 @@ struct xe_sysctrl_mailbox_command {
>  	(XE_SYSCTRL_MB_MAX_MESSAGE_SIZE - sizeof(u32))
>  
>  #define XE_SYSCTRL_MB_DEFAULT_TIMEOUT_MS	500
> +#define XE_SYSCTRL_MB_MAX_TIMEOUT_MS		60000
>  
>  #endif
> diff --git a/drivers/gpu/drm/xe/xe_sysctrl_types.h b/drivers/gpu/drm/xe/xe_sysctrl_types.h
> index 8ea6e1f29ddd..f12455f2886d 100644
> --- a/drivers/gpu/drm/xe/xe_sysctrl_types.h
> +++ b/drivers/gpu/drm/xe/xe_sysctrl_types.h
> @@ -39,6 +39,9 @@ struct xe_sysctrl_debugfs_entry {
>  
>  	/** @status: Last command result */
>  	int status;
> +
> +	/** @timeout_ms: Last mailbox response timeout used (0 for default) */
> +	unsigned int timeout_ms;
>  };
>  
>  /**
> @@ -71,6 +74,9 @@ struct xe_sysctrl {
>  
>  		/** @debugfs.ras_error_inject: RAS error injection test entry */
>  		struct xe_sysctrl_debugfs_entry ras_error_inject;
> +
> +		/** @debugfs.mailbox: Generic, user-parameterized mailbox entry */
> +		struct xe_sysctrl_debugfs_entry mailbox;
>  	} debugfs;
>  };
>  
> -- 
> 2.43.0
>