sfc: Allow resets to be upgraded; use atomic ops for safety

Currently an attempt to schedule any reset is ignored if a reset
is already pending.  This ignores the relative scopes - if the
requested reset is greater in scope then the scheduled reset should
be upgraded accordingly.

There are also some race conditions which could lead to a reset
request being lost.  Deal with them by using atomic operations on a
bitmask.  This also makes tests on reset_pending easier to get right.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

4 files changed, 22 insertions, 25 deletions

diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c
index c914729f9554..9b4cfdb09516 100644
--- a/drivers/net/sfc/efx.c
+++ b/drivers/net/sfc/efx.c
@@ -229,8 +229,7 @@ static int efx_process_channel(struct efx_channel *channel, int budget)
         struct efx_nic *efx = channel->efx;
         int spent;
 
-        if (unlikely(efx->reset_pending != RESET_TYPE_NONE ||
-                     !channel->enabled))
+        if (unlikely(efx->reset_pending || !channel->enabled))
                 return 0;
 
         spent = efx_nic_process_eventq(channel, budget);
@@ -1461,7 +1460,7 @@ static void efx_start_all(struct efx_nic *efx)
          * reset_pending [modified from an atomic context], we instead guarantee
          * that efx_mcdi_mode_poll() isn't reverted erroneously */
         efx_mcdi_mode_event(efx);
-        if (efx->reset_pending != RESET_TYPE_NONE)
+        if (efx->reset_pending)
                 efx_mcdi_mode_poll(efx);
 
         /* Start the hardware monitor if there is one. Otherwise (we're link
@@ -2118,8 +2117,10 @@ int efx_reset(struct efx_nic *efx, enum reset_type method)
                 goto out;
         }
 
-        /* Allow resets to be rescheduled. */
-        efx->reset_pending = RESET_TYPE_NONE;
+        /* Clear flags for the scopes we covered.  We assume the NIC and
+         * driver are now quiescent so that there is no race here.
+         */
+        efx->reset_pending &= -(1 << (method + 1));
 
         /* Reinitialise bus-mastering, which may have been turned off before
          * the reset was scheduled. This is still appropriate, even in the
@@ -2154,12 +2155,13 @@ out:
 static void efx_reset_work(struct work_struct *data)
 {
         struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
+        unsigned long pending = ACCESS_ONCE(efx->reset_pending);
 
-        if (efx->reset_pending == RESET_TYPE_NONE)
+        if (!pending)
                 return;
 
         /* If we're not RUNNING then don't reset. Leave the reset_pending
-         * flag set so that efx_pci_probe_main will be retried */
+         * flags set so that efx_pci_probe_main will be retried */
         if (efx->state != STATE_RUNNING) {
                 netif_info(efx, drv, efx->net_dev,
                            "scheduled reset quenched. NIC not RUNNING\n");
@@ -2167,7 +2169,7 @@ static void efx_reset_work(struct work_struct *data)
         }
 
         rtnl_lock();
-        (void)efx_reset(efx, efx->reset_pending);
+        (void)efx_reset(efx, fls(pending) - 1);
         rtnl_unlock();
 }
 
@@ -2175,12 +2177,6 @@ void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
 {
         enum reset_type method;
 
-        if (efx->reset_pending != RESET_TYPE_NONE) {
-                netif_info(efx, drv, efx->net_dev,
-                           "quenching already scheduled reset\n");
-                return;
-        }
-
         switch (type) {
         case RESET_TYPE_INVISIBLE:
         case RESET_TYPE_ALL:
@@ -2208,7 +2204,7 @@ void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
                 netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
                           RESET_TYPE(method));
 
-        efx->reset_pending = method;
+        set_bit(method, &efx->reset_pending);
 
         /* efx_process_channel() will no longer read events once a
          * reset is scheduled. So switch back to poll'd MCDI completions. */
@@ -2288,7 +2284,6 @@ static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
         efx->pci_dev = pci_dev;
         efx->msg_enable = debug;
         efx->state = STATE_INIT;
-        efx->reset_pending = RESET_TYPE_NONE;
         strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
 
         efx->net_dev = net_dev;
@@ -2510,7 +2505,7 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
                 cancel_work_sync(&efx->reset_work);
 
                 if (rc == 0) {
-                        if (efx->reset_pending != RESET_TYPE_NONE) {
+                        if (efx->reset_pending) {
                                 /* If there was a scheduled reset during
                                  * probe, the NIC is probably hosed anyway */
                                 efx_pci_remove_main(efx);
@@ -2521,11 +2516,12 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
                 }
 
                 /* Retry if a recoverably reset event has been scheduled */
-                if ((efx->reset_pending != RESET_TYPE_INVISIBLE) &&
-                    (efx->reset_pending != RESET_TYPE_ALL))
+                if (efx->reset_pending &
+                    ~(1 << RESET_TYPE_INVISIBLE | 1 << RESET_TYPE_ALL) ||
+                    !efx->reset_pending)
                         goto fail3;
 
-                efx->reset_pending = RESET_TYPE_NONE;
+                efx->reset_pending = 0;
         }
 
         if (rc) {
@@ -2609,7 +2605,7 @@ static int efx_pm_poweroff(struct device *dev)
 
         efx->type->fini(efx);
 
-        efx->reset_pending = RESET_TYPE_NONE;
+        efx->reset_pending = 0;
 
         pci_save_state(pci_dev);
         return pci_set_power_state(pci_dev, PCI_D3hot);
diff --git a/drivers/net/sfc/enum.h b/drivers/net/sfc/enum.h
index 384cfe3b1be1..d725a8fbe1a6 100644
--- a/drivers/net/sfc/enum.h
+++ b/drivers/net/sfc/enum.h
@@ -134,6 +134,8 @@ enum efx_loopback_mode {
  * other valuesspecify reasons, which efx_schedule_reset() will choose
  * a method for.
  *
+ * Reset methods are numbered in order of increasing scope.
+ *
  * @RESET_TYPE_INVISIBLE: don't reset the PHYs or interrupts
  * @RESET_TYPE_ALL: reset everything but PCI core blocks
  * @RESET_TYPE_WORLD: reset everything, save & restore PCI config
@@ -147,7 +149,6 @@ enum efx_loopback_mode {
  * @RESET_TYPE_MC_FAILURE: MC reboot/assertion
  */
 enum reset_type {
-        RESET_TYPE_NONE = -1,
         RESET_TYPE_INVISIBLE = 0,
         RESET_TYPE_ALL = 1,
         RESET_TYPE_WORLD = 2,
diff --git a/drivers/net/sfc/falcon.c b/drivers/net/sfc/falcon.c
index 60176e873d62..a4c7830ec9b0 100644
--- a/drivers/net/sfc/falcon.c
+++ b/drivers/net/sfc/falcon.c
@@ -536,7 +536,7 @@ void falcon_reconfigure_mac_wrapper(struct efx_nic *efx)
         efx_oword_t reg;
         int link_speed, isolate;
 
-        isolate = (efx->reset_pending != RESET_TYPE_NONE);
+        isolate = !!ACCESS_ONCE(efx->reset_pending);
 
         switch (link_state->speed) {
         case 10000: link_speed = 3; break;
diff --git a/drivers/net/sfc/net_driver.h b/drivers/net/sfc/net_driver.h
index e8d5f03a89fe..c422eb2ce60a 100644
--- a/drivers/net/sfc/net_driver.h
+++ b/drivers/net/sfc/net_driver.h
@@ -645,7 +645,7 @@ struct efx_filter_state;
  * @irq_rx_moderation: IRQ moderation time for RX event queues
  * @msg_enable: Log message enable flags
  * @state: Device state flag. Serialised by the rtnl_lock.
- * @reset_pending: Pending reset method (normally RESET_TYPE_NONE)
+ * @reset_pending: Bitmask for pending resets
  * @tx_queue: TX DMA queues
  * @rx_queue: RX DMA queues
  * @channel: Channels
@@ -728,7 +728,7 @@ struct efx_nic {
         u32 msg_enable;
 
         enum nic_state state;
-        enum reset_type reset_pending;
+        unsigned long reset_pending;
 
         struct efx_channel *channel[EFX_MAX_CHANNELS];
         char channel_name[EFX_MAX_CHANNELS][IFNAMSIZ + 6];