iwlwifi: pcie: fix RF-Kill vs. firmware load race
authorEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Sun, 31 Jan 2016 13:02:30 +0000 (15:02 +0200)
committerEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Mon, 15 Feb 2016 11:38:25 +0000 (13:38 +0200)
When we load the firmware, we hold trans_pcie->mutex to
avoid nested flows. We also rely on the ISR to wake up the
thread when the DMA has finished copying a chunk. During
this flow, we enable the RF-Kill interrupt.

The problem is that the RF-Kill interrupt handler can take
the mutex and bring the device down. This means that if
we load the firmware while the RF-Kill switch is enabled
(which will happen when we load the INIT firmware to read
the device's capabilities and register to mac80211), we
may get an RF-Kill interrupt immediately and the ISR will
be waiting for the mutex held by the thread that is
currently loading the firmware. At this stage, the ISR
won't be able to service the DMA's interrupt needed to
wake up the thread that load the firmware. We are in a
deadlock situation which ends when the thread that loads
the firmware fails on timeout and releases the mutex.

To fix this, take the mutex later in the flow, disable
the interrupts and synchronize_irq() to give a chance to
the RF-Kill interrupt to run and complete.
After that, mask all the interrupts besides the DMA
interrupt and proceed with firmware load. Make sure to
check that there was no RF-Kill interrupt when the
interrupts were disabled.

This fixes https://bugzilla.kernel.org/show_bug.cgi?id=111361

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
drivers/net/wireless/intel/iwlwifi/pcie/internal.h
drivers/net/wireless/intel/iwlwifi/pcie/rx.c
drivers/net/wireless/intel/iwlwifi/pcie/trans.c

index cc3888e2700daf1cfe07ebb4e003ca53bcf6d934..73c95594eabe2c530d503fa51aedce05f0ffb8a9 100644 (file)
@@ -490,6 +490,15 @@ static inline void iwl_enable_interrupts(struct iwl_trans *trans)
        iwl_write32(trans, CSR_INT_MASK, trans_pcie->inta_mask);
 }
 
+static inline void iwl_enable_fw_load_int(struct iwl_trans *trans)
+{
+       struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+
+       IWL_DEBUG_ISR(trans, "Enabling FW load interrupt\n");
+       trans_pcie->inta_mask = CSR_INT_BIT_FH_TX;
+       iwl_write32(trans, CSR_INT_MASK, trans_pcie->inta_mask);
+}
+
 static inline void iwl_enable_rfkill_int(struct iwl_trans *trans)
 {
        struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
index ccafbd8cf4b3b224649b3fbc87da69207884743c..152cf9ad9566587320a4a8d5dd34140b3d6b24b9 100644 (file)
@@ -1438,9 +1438,11 @@ irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
                         inta & ~trans_pcie->inta_mask);
        }
 
-       /* Re-enable all interrupts */
-       /* only Re-enable if disabled by irq */
-       if (test_bit(STATUS_INT_ENABLED, &trans->status))
+       /* we are loading the firmware, enable FH_TX interrupt only */
+       if (handled & CSR_INT_BIT_FH_TX)
+               iwl_enable_fw_load_int(trans);
+       /* only Re-enable all interrupt if disabled by irq */
+       else if (test_bit(STATUS_INT_ENABLED, &trans->status))
                iwl_enable_interrupts(trans);
        /* Re-enable RF_KILL if it occurred */
        else if (handled & CSR_INT_BIT_RF_KILL)
index d60a467a983c6f220384b701953eb42d2cc4c1ae..55030729c7d9002f28a70ee6913ba02e8a0a71d4 100644 (file)
@@ -1021,82 +1021,6 @@ static int iwl_pcie_load_given_ucode_8000(struct iwl_trans *trans,
                                               &first_ucode_section);
 }
 
-static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
-                                  const struct fw_img *fw, bool run_in_rfkill)
-{
-       struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
-       bool hw_rfkill;
-       int ret;
-
-       mutex_lock(&trans_pcie->mutex);
-
-       /* Someone called stop_device, don't try to start_fw */
-       if (trans_pcie->is_down) {
-               IWL_WARN(trans,
-                        "Can't start_fw since the HW hasn't been started\n");
-               ret = EIO;
-               goto out;
-       }
-
-       /* This may fail if AMT took ownership of the device */
-       if (iwl_pcie_prepare_card_hw(trans)) {
-               IWL_WARN(trans, "Exit HW not ready\n");
-               ret = -EIO;
-               goto out;
-       }
-
-       iwl_enable_rfkill_int(trans);
-
-       /* If platform's RF_KILL switch is NOT set to KILL */
-       hw_rfkill = iwl_is_rfkill_set(trans);
-       if (hw_rfkill)
-               set_bit(STATUS_RFKILL, &trans->status);
-       else
-               clear_bit(STATUS_RFKILL, &trans->status);
-       iwl_trans_pcie_rf_kill(trans, hw_rfkill);
-       if (hw_rfkill && !run_in_rfkill) {
-               ret = -ERFKILL;
-               goto out;
-       }
-
-       iwl_write32(trans, CSR_INT, 0xFFFFFFFF);
-
-       ret = iwl_pcie_nic_init(trans);
-       if (ret) {
-               IWL_ERR(trans, "Unable to init nic\n");
-               goto out;
-       }
-
-       /* make sure rfkill handshake bits are cleared */
-       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
-       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR,
-                   CSR_UCODE_DRV_GP1_BIT_CMD_BLOCKED);
-
-       /* clear (again), then enable host interrupts */
-       iwl_write32(trans, CSR_INT, 0xFFFFFFFF);
-       iwl_enable_interrupts(trans);
-
-       /* really make sure rfkill handshake bits are cleared */
-       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
-       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
-
-       /* Load the given image to the HW */
-       if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
-               ret = iwl_pcie_load_given_ucode_8000(trans, fw);
-       else
-               ret = iwl_pcie_load_given_ucode(trans, fw);
-
-out:
-       mutex_unlock(&trans_pcie->mutex);
-       return ret;
-}
-
-static void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr)
-{
-       iwl_pcie_reset_ict(trans);
-       iwl_pcie_tx_start(trans, scd_addr);
-}
-
 static void _iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
 {
        struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
@@ -1127,7 +1051,8 @@ static void _iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
         * already dead.
         */
        if (test_and_clear_bit(STATUS_DEVICE_ENABLED, &trans->status)) {
-               IWL_DEBUG_INFO(trans, "DEVICE_ENABLED bit was set and is now cleared\n");
+               IWL_DEBUG_INFO(trans,
+                              "DEVICE_ENABLED bit was set and is now cleared\n");
                iwl_pcie_tx_stop(trans);
                iwl_pcie_rx_stop(trans);
 
@@ -1161,7 +1086,6 @@ static void _iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
        iwl_disable_interrupts(trans);
        spin_unlock(&trans_pcie->irq_lock);
 
-
        /* clear all status bits */
        clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
        clear_bit(STATUS_INT_ENABLED, &trans->status);
@@ -1194,10 +1118,116 @@ static void _iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
        if (hw_rfkill != was_hw_rfkill)
                iwl_trans_pcie_rf_kill(trans, hw_rfkill);
 
-       /* re-take ownership to prevent other users from stealing the deivce */
+       /* re-take ownership to prevent other users from stealing the device */
        iwl_pcie_prepare_card_hw(trans);
 }
 
+static int iwl_trans_pcie_start_fw(struct iwl_trans *trans,
+                                  const struct fw_img *fw, bool run_in_rfkill)
+{
+       struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
+       bool hw_rfkill;
+       int ret;
+
+       /* This may fail if AMT took ownership of the device */
+       if (iwl_pcie_prepare_card_hw(trans)) {
+               IWL_WARN(trans, "Exit HW not ready\n");
+               ret = -EIO;
+               goto out;
+       }
+
+       iwl_enable_rfkill_int(trans);
+
+       iwl_write32(trans, CSR_INT, 0xFFFFFFFF);
+
+       /*
+        * We enabled the RF-Kill interrupt and the handler may very
+        * well be running. Disable the interrupts to make sure no other
+        * interrupt can be fired.
+        */
+       iwl_disable_interrupts(trans);
+
+       /* Make sure it finished running */
+       synchronize_irq(trans_pcie->pci_dev->irq);
+
+       mutex_lock(&trans_pcie->mutex);
+
+       /* If platform's RF_KILL switch is NOT set to KILL */
+       hw_rfkill = iwl_is_rfkill_set(trans);
+       if (hw_rfkill)
+               set_bit(STATUS_RFKILL, &trans->status);
+       else
+               clear_bit(STATUS_RFKILL, &trans->status);
+       iwl_trans_pcie_rf_kill(trans, hw_rfkill);
+       if (hw_rfkill && !run_in_rfkill) {
+               ret = -ERFKILL;
+               goto out;
+       }
+
+       /* Someone called stop_device, don't try to start_fw */
+       if (trans_pcie->is_down) {
+               IWL_WARN(trans,
+                        "Can't start_fw since the HW hasn't been started\n");
+               ret = EIO;
+               goto out;
+       }
+
+       /* make sure rfkill handshake bits are cleared */
+       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
+       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR,
+                   CSR_UCODE_DRV_GP1_BIT_CMD_BLOCKED);
+
+       /* clear (again), then enable host interrupts */
+       iwl_write32(trans, CSR_INT, 0xFFFFFFFF);
+
+       ret = iwl_pcie_nic_init(trans);
+       if (ret) {
+               IWL_ERR(trans, "Unable to init nic\n");
+               goto out;
+       }
+
+       /*
+        * Now, we load the firmware and don't want to be interrupted, even
+        * by the RF-Kill interrupt (hence mask all the interrupt besides the
+        * FH_TX interrupt which is needed to load the firmware). If the
+        * RF-Kill switch is toggled, we will find out after having loaded
+        * the firmware and return the proper value to the caller.
+        */
+       iwl_enable_fw_load_int(trans);
+
+       /* really make sure rfkill handshake bits are cleared */
+       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
+       iwl_write32(trans, CSR_UCODE_DRV_GP1_CLR, CSR_UCODE_SW_BIT_RFKILL);
+
+       /* Load the given image to the HW */
+       if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
+               ret = iwl_pcie_load_given_ucode_8000(trans, fw);
+       else
+               ret = iwl_pcie_load_given_ucode(trans, fw);
+       iwl_enable_interrupts(trans);
+
+       /* re-check RF-Kill state since we may have missed the interrupt */
+       hw_rfkill = iwl_is_rfkill_set(trans);
+       if (hw_rfkill)
+               set_bit(STATUS_RFKILL, &trans->status);
+       else
+               clear_bit(STATUS_RFKILL, &trans->status);
+
+       iwl_trans_pcie_rf_kill(trans, hw_rfkill);
+       if (hw_rfkill && !run_in_rfkill)
+               ret = -ERFKILL;
+
+out:
+       mutex_unlock(&trans_pcie->mutex);
+       return ret;
+}
+
+static void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr)
+{
+       iwl_pcie_reset_ict(trans);
+       iwl_pcie_tx_start(trans, scd_addr);
+}
+
 static void iwl_trans_pcie_stop_device(struct iwl_trans *trans, bool low_power)
 {
        struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);