On Thu, Oct 18, 2012 at 02:06:01PM +0300, dragos.tatu...@intel.com wrote:
> From: Adrian Hunter <adrian.hun...@intel.com>
>
> Other modules that may wish to write to persistent storage
> are supported by adding a notifier and write function.
>
> The notifier has 3 events: PSTORE_BEGIN, PSTORE_DUMP and
> PSTORE_END. External writers use the PSTORE_DUMP event
> whereas the PSTORE_BEGIN and PSTORE_END can be used by
> platform code to ensure the back end is powered up.
>
> Signed-off-by: Adrian Hunter <adrian.hun...@intel.com>
> ---
> fs/pstore/platform.c | 88
> ++++++++++++++++++++++++++++++++++++++++++++++++
> include/linux/pstore.h | 43 +++++++++++++++++++----
> 2 files changed, 124 insertions(+), 7 deletions(-)
>
> diff --git a/fs/pstore/platform.c b/fs/pstore/platform.c
> index 9d65723..63ff377 100644
> --- a/fs/pstore/platform.c
> +++ b/fs/pstore/platform.c
[...]
> +/* pstore_write must only be called from PSTORE_DUMP notifier callbacks */
> +int pstore_write(enum pstore_type_id type, const char *buf, size_t size)
> +{
> + size_t len;
> + int err = 0, err2;
I guess there is no need for err2 here, you can place it in the while
loop, in the if block.
> +
> + if (!psinfo)
> + return -ENODEV;
> +
> + /*
> + * No locking is needed because pstore_write is called only from
> + * PSTORE_DUMP notifier callbacks.
> + */
> +
> + if (type != psinfo->ext_type) {
> + err = pstore_ext_flush();
> + psinfo->ext_type = type;
> + psinfo->ext_part = 1;
> + }
> +
> + while (size) {
> + len = min(size, psinfo->bufsize - psinfo->ext_len);
> + memcpy(psinfo->buf + psinfo->ext_len, buf, len);
> + psinfo->ext_len += len;
> + buf += len;
> + size -= len;
> + if (psinfo->ext_len == psinfo->bufsize) {
> + err2 = pstore_ext_flush();
> + if (err2 && !err)
> + err = err2;
> + }
> + }
> +
> + return err;
> +}
> +EXPORT_SYMBOL_GPL(pstore_write);
> +
> module_param(backend, charp, 0444);
> MODULE_PARM_DESC(backend, "Pstore backend to use");
> diff --git a/include/linux/pstore.h b/include/linux/pstore.h
> index 55ab23f..1bf7f4b 100644
> --- a/include/linux/pstore.h
> +++ b/include/linux/pstore.h
> @@ -40,17 +40,27 @@ enum pstore_type_id {
>
> struct module;
>
> +/* Notifier events */
> +#define PSTORE_BEGIN 1
> +#define PSTORE_DUMP 2
> +#define PSTORE_END 3
> +
> #define PSTORE_NO_HEADINGS BIT(0)
> #define PSTORE_MAX_KMSG_BYTES BIT(1)
>
> struct pstore_info {
> - struct module *owner;
> - char *name;
> - unsigned int flags;
> - spinlock_t buf_lock; /* serialize access to 'buf' */
> - char *buf;
> - size_t bufsize;
> - struct mutex read_mutex; /* serialize open/read/close */
That's why I don't like "pretty" formatting for structs. :) The change is
unrelated, so please don't do this.
> + struct module *owner;
> + char *name;
> + unsigned int flags;
> + spinlock_t buf_lock; /* serialize access to 'buf' */
> + char *buf;
> + size_t bufsize;
> + struct mutex read_mutex; /* serialize open/read/close */
> + u64 ext_id;
> + size_t ext_len;
> + unsigned int ext_part;
> + enum pstore_type_id ext_type;
> + enum kmsg_dump_reason ext_reason;
> int (*open)(struct pstore_info *psi);
> int (*close)(struct pstore_info *psi);
> ssize_t (*read)(u64 *id, enum pstore_type_id *type,
> @@ -70,12 +80,31 @@ struct pstore_info {
>
> #ifdef CONFIG_PSTORE
> extern int pstore_register(struct pstore_info *);
> +extern int pstore_notifier_register(struct notifier_block *n);
> +extern int pstore_notifier_unregister(struct notifier_block *n);
> +/* pstore_write must only be called from PSTORE_DUMP notifier callbacks */
> +extern int pstore_write(enum pstore_type_id type, const char *buf, size_t
> size);
> #else
> static inline int
> pstore_register(struct pstore_info *psi)
> {
> return -ENODEV;
> }
> +static inline int
> +pstore_notifier_register(struct notifier_block *n)
> +{
> + return 0;
> +}
> +static inline int
> +pstore_notifier_unregister(struct notifier_block *n)
> +{
> + return 0;
> +}
> +static inline int
> +pstore_write(enum pstore_type_id type, const char *buf, size_t size)
> +{
> + return 0;
> +}
> #endif
>
> #endif /*_LINUX_PSTORE_H*/
> --
> 1.7.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
