Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-next-2.6

1 files changed, 20 insertions, 0 deletions

diff --git a/drivers/net/wireless/iwlwifi/iwl-io.h b/drivers/net/wireless/iwlwifi/iwl-io.h
index 0a078b082833..d0a358c9d96b 100644
--- a/drivers/net/wireless/iwlwifi/iwl-io.h
+++ b/drivers/net/wireless/iwlwifi/iwl-io.h
@@ -200,6 +200,26 @@ static inline int _iwl_grab_nic_access(struct iwl_priv *priv)
 
 	/* this bit wakes up the NIC */
 	_iwl_set_bit(priv, CSR_GP_CNTRL, CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
+
+	/*
+	 * These bits say the device is running, and should keep running for
+	 * at least a short while (at least as long as MAC_ACCESS_REQ stays 1),
+	 * but they do not indicate that embedded SRAM is restored yet;
+	 * 3945 and 4965 have volatile SRAM, and must save/restore contents
+	 * to/from host DRAM when sleeping/waking for power-saving.
+	 * Each direction takes approximately 1/4 millisecond; with this
+	 * overhead, it's a good idea to grab and hold MAC_ACCESS_REQUEST if a
+	 * series of register accesses are expected (e.g. reading Event Log),
+	 * to keep device from sleeping.
+	 *
+	 * CSR_UCODE_DRV_GP1 register bit MAC_SLEEP == 0 indicates that
+	 * SRAM is okay/restored.  We don't check that here because this call
+	 * is just for hardware register access; but GP1 MAC_SLEEP check is a
+	 * good idea before accessing 3945/4965 SRAM (e.g. reading Event Log).
+	 *
+	 * 5000 series and later (including 1000 series) have non-volatile SRAM,
+	 * and do not save/restore SRAM when power cycling.
+	 */
 	ret = _iwl_poll_bit(priv, CSR_GP_CNTRL,
 			   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
 			   (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |