iwlwifi: don't access the HW when it is not available

When we kill the radio with the RF kill button we could access
the HW after having stopped the APM which would result in the
warning below.

The flow goes like this:
* RF kill
	iwlwifi notifies the stack which stops the driver
	fw sends CARD_STATE_NOTIFICATION
* iwl_trans_pcie_stop_device stops the APM
* the tasklet runs and calls to iwl_rx_handle
* iwl_rx_handle calls iwl_rx_queue_restock
* iwl_rx_queue_restock tries to access the HW...

[255908.543823] ------------[ cut here ]------------
[255908.543843] WARNING: at drivers/net/wireless/iwlwifi/iwl-io.c:150 iwl_grab_nic_access+0x79/0xb0 [iwlwifi]()
[255908.543849] Hardware name: Latitude E6410
[255908.543852] Timeout waiting for hardware access (CSR_GP_CNTRL 0x000003d8)
[255908.543856] Modules linked in: iwlmvm iwlwifi mac80211 [...]
[255908.543935] Pid: 0, comm: swapper Tainted: G        W   3.1.0 #1
[255908.543939] Call Trace:
[255908.543950]  [<c1046e42>] warn_slowpath_common+0x72/0xa0
[255908.543980]  [<c1046f13>] warn_slowpath_fmt+0x33/0x40
[255908.543992]  [<fa4bb3b9>] iwl_grab_nic_access+0x79/0xb0 [iwlwifi]
[255908.544004]  [<fa4bb9eb>] iwl_write_direct32+0x2b/0xa0 [iwlwifi]
[255908.544018]  [<fa4c0ff9>] iwl_rx_queue_update_write_ptr+0x89/0x1d0 [iwlwifi]
[255908.544054]  [<fa4c1250>] iwlagn_rx_queue_restock+0x110/0x140 [iwlwifi]
[255908.544067]  [<fa4c234d>] iwl_irq_tasklet+0x82d/0xf40 [iwlwifi]
[255908.544096]  [<c104e11e>] tasklet_action+0xbe/0x100
[255908.544102]  [<c104d91e>] __do_softirq+0xae/0x1f0
[255908.544227] ---[ end trace d150f49345d85009 ]---

Prevent this.

Signed-off-by: Emmanuel Grumbach <[email protected]>
Reviewed-by: Johannes Berg <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>

Author: egrumbach

drivers/net/wireless/iwlwifi/pcie/rx.c

@@ -203,6 +203,17 @@ static void iwl_rx_queue_restock(struct iwl_trans *trans)
 	struct iwl_rx_mem_buffer *rxb;
 	unsigned long flags;
 
+	/*
+	 * If the device isn't enabled - not need to try to add buffers...
+	 * This can happen when we stop the device and still have an interrupt
+	 * pending. We stop the APM before we sync the interrupts / tasklets
+	 * because we have to (see comment there). On the other hand, since
+	 * the APM is stopped, we cannot access the HW (in particular not prph).
+	 * So don't try to restock if the APM has been already stopped.
+	 */
+	if (!test_bit(STATUS_DEVICE_ENABLED, &trans_pcie->status))
+		return;
+
 	spin_lock_irqsave(&rxq->lock, flags);
 	while ((iwl_rx_queue_space(rxq) > 0) && (rxq->free_count)) {
 		/* The overwritten rxb must be a used one */