wimax/i2400m: clarify and fix i2400m->{ready,updown}

The i2400m driver uses two different bits to distinguish how much the
driver is up. i2400m->ready is used to denote that the infrastructure
to communicate with the device is up and running. i2400m->updown is
used to indicate if 'ready' and the device is up and running, ready to
take control and data traffic.

However, all this was pretty dirty and not clear, with many open spots
where race conditions were present.

This commit cleans up the situation by:

 - documenting the usage of both bits

 - setting them only in specific, well controlled places
   (i2400m_dev_start, i2400m_dev_stop)

 - ensuring the i2400m workqueue can't get in the middle of the
   setting by flushing it when i2400m->ready is set to zero. This
   allows the report hook not having to check again for the bit to be
   set [rx.c:i2400m_report_hook_work()].

 - using i2400m->updown to determine if the device is up and running
   instead of the wimax state in i2400m_dev_reset_handle().

 - not loosing missed messages sent by the hardware before
   i2400m->ready is set. In rx.c, whatever the device sends can be
   sent to user space over the message pipes as soon as the wimax
   device is registered, so don't wait for i2400m->ready to be set.

Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>

1 files changed, 22 insertions, 1 deletions

diff --git a/drivers/net/wimax/i2400m/i2400m.h b/drivers/net/wimax/i2400m/i2400m.h
index 2b9c400810b..fbc156db5bf 100644
--- a/drivers/net/wimax/i2400m/i2400m.h
+++ b/drivers/net/wimax/i2400m/i2400m.h
@@ -307,6 +307,27 @@ struct i2400m_barker_db;
  *     force this to be the first field so that we can get from
  *     netdev_priv() the right pointer.
  *
+ * @updown: the device is up and ready for transmitting control and
+ *     data packets. This implies @ready (communication infrastructure
+ *     with the device is ready) and the device's firmware has been
+ *     loaded and the device initialized.
+ *
+ *     Write to it only inside a i2400m->init_mutex protected area
+ *     followed with a wmb(); rmb() before accesing (unless locked
+ *     inside i2400m->init_mutex). Read access can be loose like that
+ *     [just using rmb()] because the paths that use this also do
+ *     other error checks later on.
+ *
+ * @ready: Communication infrastructure with the device is ready, data
+ *     frames can start to be passed around (this is lighter than
+ *     using the WiMAX state for certain hot paths).
+ *
+ *     Write to it only inside a i2400m->init_mutex protected area
+ *     followed with a wmb(); rmb() before accesing (unless locked
+ *     inside i2400m->init_mutex). Read access can be loose like that
+ *     [just using rmb()] because the paths that use this also do
+ *     other error checks later on.
+ *
  * @rx_reorder: 1 if RX reordering is enabled; this can only be
  *     set at probe time.
  *
@@ -458,7 +479,7 @@ struct i2400m {
 	unsigned updown:1;		/* Network device is up or down */
 	unsigned boot_mode:1;		/* is the device in boot mode? */
 	unsigned sboot:1;		/* signed or unsigned fw boot */
-	unsigned ready:1;		/* all probing steps done */
+	unsigned ready:1;		/* Device comm infrastructure ready */
 	unsigned rx_reorder:1;		/* RX reorder is enabled */
 	u8 trace_msg_from_user;		/* echo rx msgs to 'trace' pipe */
 					/* typed u8 so /sys/kernel/debug/u8 can tweak */