8 files changed, 500 insertions, 140 deletions

diff --git a/drivers/block/rsxx/Makefile b/drivers/block/rsxx/Makefile
index f35cd0b71f7b..b1c53c0aa450 100644
--- a/drivers/block/rsxx/Makefile
+++ b/drivers/block/rsxx/Makefile
@@ -1,2 +1,2 @@
 obj-$(CONFIG_BLK_DEV_RSXX) += rsxx.o
-rsxx-y := config.o core.o cregs.o dev.o dma.o
+rsxx-objs := config.o core.o cregs.o dev.o dma.o
diff --git a/drivers/block/rsxx/config.c b/drivers/block/rsxx/config.c
index a295e7e9ee41..10cd530d3e10 100644
--- a/drivers/block/rsxx/config.c
+++ b/drivers/block/rsxx/config.c
@@ -29,15 +29,13 @@
 #include "rsxx_priv.h"
 #include "rsxx_cfg.h"
 
-static void initialize_config(void *config)
+static void initialize_config(struct rsxx_card_cfg *cfg)
 {
-        struct rsxx_card_cfg *cfg = config;
-
         cfg->hdr.version = RSXX_CFG_VERSION;
 
         cfg->data.block_size        = RSXX_HW_BLK_SIZE;
         cfg->data.stripe_size       = RSXX_HW_BLK_SIZE;
-        cfg->data.vendor_id         = RSXX_VENDOR_ID_TMS_IBM;
+        cfg->data.vendor_id         = RSXX_VENDOR_ID_IBM;
         cfg->data.cache_order       = (-1);
         cfg->data.intr_coal.mode    = RSXX_INTR_COAL_DISABLED;
         cfg->data.intr_coal.count   = 0;
@@ -181,7 +179,7 @@ int rsxx_load_config(struct rsxx_cardinfo *card)
         } else {
                 dev_info(CARD_TO_DEV(card),
                         "Initializing card configuration.\n");
-                initialize_config(card);
+                initialize_config(&card->config);
                 st = rsxx_save_config(card);
                 if (st)
                         return st;
diff --git a/drivers/block/rsxx/core.c b/drivers/block/rsxx/core.c
index e5162487686a..5af21f2db29c 100644
--- a/drivers/block/rsxx/core.c
+++ b/drivers/block/rsxx/core.c
@@ -30,6 +30,7 @@
 #include <linux/reboot.h>
 #include <linux/slab.h>
 #include <linux/bitops.h>
+#include <linux/delay.h>
 
 #include <linux/genhd.h>
 #include <linux/idr.h>
@@ -39,8 +40,8 @@
 
 #define NO_LEGACY 0
 
-MODULE_DESCRIPTION("IBM RamSan PCIe Flash SSD Device Driver");
-MODULE_AUTHOR("IBM <support@ramsan.com>");
+MODULE_DESCRIPTION("IBM FlashSystem 70/80 PCIe SSD Device Driver");
+MODULE_AUTHOR("Joshua Morris/Philip Kelleher, IBM");
 MODULE_LICENSE("GPL");
 MODULE_VERSION(DRIVER_VERSION);
 
@@ -52,6 +53,13 @@ static DEFINE_IDA(rsxx_disk_ida);
 static DEFINE_SPINLOCK(rsxx_ida_lock);
 
 /*----------------- Interrupt Control & Handling -------------------*/
+
+static void rsxx_mask_interrupts(struct rsxx_cardinfo *card)
+{
+        card->isr_mask = 0;
+        card->ier_mask = 0;
+}
+
 static void __enable_intr(unsigned int *mask, unsigned int intr)
 {
         *mask |= intr;
@@ -71,7 +79,8 @@ static void __disable_intr(unsigned int *mask, unsigned int intr)
  */
 void rsxx_enable_ier(struct rsxx_cardinfo *card, unsigned int intr)
 {
-        if (unlikely(card->halt))
+        if (unlikely(card->halt) ||
+            unlikely(card->eeh_state))
                 return;
 
         __enable_intr(&card->ier_mask, intr);
@@ -80,6 +89,9 @@ void rsxx_enable_ier(struct rsxx_cardinfo *card, unsigned int intr)
 
 void rsxx_disable_ier(struct rsxx_cardinfo *card, unsigned int intr)
 {
+        if (unlikely(card->eeh_state))
+                return;
+
         __disable_intr(&card->ier_mask, intr);
         iowrite32(card->ier_mask, card->regmap + IER);
 }
@@ -87,7 +99,8 @@ void rsxx_disable_ier(struct rsxx_cardinfo *card, unsigned int intr)
 void rsxx_enable_ier_and_isr(struct rsxx_cardinfo *card,
                                  unsigned int intr)
 {
-        if (unlikely(card->halt))
+        if (unlikely(card->halt) ||
+            unlikely(card->eeh_state))
                 return;
 
         __enable_intr(&card->isr_mask, intr);
@@ -97,6 +110,9 @@ void rsxx_enable_ier_and_isr(struct rsxx_cardinfo *card,
 void rsxx_disable_ier_and_isr(struct rsxx_cardinfo *card,
                                   unsigned int intr)
 {
+        if (unlikely(card->eeh_state))
+                return;
+
         __disable_intr(&card->isr_mask, intr);
         __disable_intr(&card->ier_mask, intr);
         iowrite32(card->ier_mask, card->regmap + IER);
@@ -115,6 +131,9 @@ static irqreturn_t rsxx_isr(int irq, void *pdata)
         do {
                 reread_isr = 0;
 
+                if (unlikely(card->eeh_state))
+                        break;
+
                 isr = ioread32(card->regmap + ISR);
                 if (isr == 0xffffffff) {
                         /*
@@ -161,9 +180,9 @@ static irqreturn_t rsxx_isr(int irq, void *pdata)
 }
 
 /*----------------- Card Event Handler -------------------*/
-static char *rsxx_card_state_to_str(unsigned int state)
+static const char * const rsxx_card_state_to_str(unsigned int state)
 {
-        static char *state_strings[] = {
+        static const char * const state_strings[] = {
                 "Unknown", "Shutdown", "Starting", "Formatting",
                 "Uninitialized", "Good", "Shutting Down",
                 "Fault", "Read Only Fault", "dStroying"
@@ -304,6 +323,192 @@ static int card_shutdown(struct rsxx_cardinfo *card)
         return 0;
 }
 
+static int rsxx_eeh_frozen(struct pci_dev *dev)
+{
+        struct rsxx_cardinfo *card = pci_get_drvdata(dev);
+        int i;
+        int st;
+
+        dev_warn(&dev->dev, "IBM FlashSystem PCI: preparing for slot reset.\n");
+
+        card->eeh_state = 1;
+        rsxx_mask_interrupts(card);
+
+        /*
+         * We need to guarantee that the write for eeh_state and masking
+         * interrupts does not become reordered. This will prevent a possible
+         * race condition with the EEH code.
+         */
+        wmb();
+
+        pci_disable_device(dev);
+
+        st = rsxx_eeh_save_issued_dmas(card);
+        if (st)
+                return st;
+
+        rsxx_eeh_save_issued_creg(card);
+
+        for (i = 0; i < card->n_targets; i++) {
+                if (card->ctrl[i].status.buf)
+                        pci_free_consistent(card->dev, STATUS_BUFFER_SIZE8,
+                                            card->ctrl[i].status.buf,
+                                            card->ctrl[i].status.dma_addr);
+                if (card->ctrl[i].cmd.buf)
+                        pci_free_consistent(card->dev, COMMAND_BUFFER_SIZE8,
+                                            card->ctrl[i].cmd.buf,
+                                            card->ctrl[i].cmd.dma_addr);
+        }
+
+        return 0;
+}
+
+static void rsxx_eeh_failure(struct pci_dev *dev)
+{
+        struct rsxx_cardinfo *card = pci_get_drvdata(dev);
+        int i;
+
+        dev_err(&dev->dev, "IBM FlashSystem PCI: disabling failed card.\n");
+
+        card->eeh_state = 1;
+
+        for (i = 0; i < card->n_targets; i++)
+                del_timer_sync(&card->ctrl[i].activity_timer);
+
+        rsxx_eeh_cancel_dmas(card);
+}
+
+static int rsxx_eeh_fifo_flush_poll(struct rsxx_cardinfo *card)
+{
+        unsigned int status;
+        int iter = 0;
+
+        /* We need to wait for the hardware to reset */
+        while (iter++ < 10) {
+                status = ioread32(card->regmap + PCI_RECONFIG);
+
+                if (status & RSXX_FLUSH_BUSY) {
+                        ssleep(1);
+                        continue;
+                }
+
+                if (status & RSXX_FLUSH_TIMEOUT)
+                        dev_warn(CARD_TO_DEV(card), "HW: flash controller timeout\n");
+                return 0;
+        }
+
+        /* Hardware failed resetting itself. */
+        return -1;
+}
+
+static pci_ers_result_t rsxx_error_detected(struct pci_dev *dev,
+                                            enum pci_channel_state error)
+{
+        int st;
+
+        if (dev->revision < RSXX_EEH_SUPPORT)
+                return PCI_ERS_RESULT_NONE;
+
+        if (error == pci_channel_io_perm_failure) {
+                rsxx_eeh_failure(dev);
+                return PCI_ERS_RESULT_DISCONNECT;
+        }
+
+        st = rsxx_eeh_frozen(dev);
+        if (st) {
+                dev_err(&dev->dev, "Slot reset setup failed\n");
+                rsxx_eeh_failure(dev);
+                return PCI_ERS_RESULT_DISCONNECT;
+        }
+
+        return PCI_ERS_RESULT_NEED_RESET;
+}
+
+static pci_ers_result_t rsxx_slot_reset(struct pci_dev *dev)
+{
+        struct rsxx_cardinfo *card = pci_get_drvdata(dev);
+        unsigned long flags;
+        int i;
+        int st;
+
+        dev_warn(&dev->dev,
+                "IBM FlashSystem PCI: recovering from slot reset.\n");
+
+        st = pci_enable_device(dev);
+        if (st)
+                goto failed_hw_setup;
+
+        pci_set_master(dev);
+
+        st = rsxx_eeh_fifo_flush_poll(card);
+        if (st)
+                goto failed_hw_setup;
+
+        rsxx_dma_queue_reset(card);
+
+        for (i = 0; i < card->n_targets; i++) {
+                st = rsxx_hw_buffers_init(dev, &card->ctrl[i]);
+                if (st)
+                        goto failed_hw_buffers_init;
+        }
+
+        if (card->config_valid)
+                rsxx_dma_configure(card);
+
+        /* Clears the ISR register from spurious interrupts */
+        st = ioread32(card->regmap + ISR);
+
+        card->eeh_state = 0;
+
+        st = rsxx_eeh_remap_dmas(card);
+        if (st)
+                goto failed_remap_dmas;
+
+        spin_lock_irqsave(&card->irq_lock, flags);
+        if (card->n_targets & RSXX_MAX_TARGETS)
+                rsxx_enable_ier_and_isr(card, CR_INTR_ALL_G);
+        else
+                rsxx_enable_ier_and_isr(card, CR_INTR_ALL_C);
+        spin_unlock_irqrestore(&card->irq_lock, flags);
+
+        rsxx_kick_creg_queue(card);
+
+        for (i = 0; i < card->n_targets; i++) {
+                spin_lock(&card->ctrl[i].queue_lock);
+                if (list_empty(&card->ctrl[i].queue)) {
+                        spin_unlock(&card->ctrl[i].queue_lock);
+                        continue;
+                }
+                spin_unlock(&card->ctrl[i].queue_lock);
+
+                queue_work(card->ctrl[i].issue_wq,
+                                &card->ctrl[i].issue_dma_work);
+        }
+
+        dev_info(&dev->dev, "IBM FlashSystem PCI: recovery complete.\n");
+
+        return PCI_ERS_RESULT_RECOVERED;
+
+failed_hw_buffers_init:
+failed_remap_dmas:
+        for (i = 0; i < card->n_targets; i++) {
+                if (card->ctrl[i].status.buf)
+                        pci_free_consistent(card->dev,
+                                        STATUS_BUFFER_SIZE8,
+                                        card->ctrl[i].status.buf,
+                                        card->ctrl[i].status.dma_addr);
+                if (card->ctrl[i].cmd.buf)
+                        pci_free_consistent(card->dev,
+                                        COMMAND_BUFFER_SIZE8,
+                                        card->ctrl[i].cmd.buf,
+                                        card->ctrl[i].cmd.dma_addr);
+        }
+failed_hw_setup:
+        rsxx_eeh_failure(dev);
+        return PCI_ERS_RESULT_DISCONNECT;
+
+}
+
 /*----------------- Driver Initialization & Setup -------------------*/
 /* Returns:   0 if the driver is compatible with the device
              -1 if the driver is NOT compatible with the device */
@@ -383,6 +588,7 @@ static int rsxx_pci_probe(struct pci_dev *dev,
 
         spin_lock_init(&card->irq_lock);
         card->halt = 0;
+        card->eeh_state = 0;
 
         spin_lock_irq(&card->irq_lock);
         rsxx_disable_ier_and_isr(card, CR_INTR_ALL);
@@ -538,9 +744,6 @@ static void rsxx_pci_remove(struct pci_dev *dev)
         rsxx_disable_ier_and_isr(card, CR_INTR_EVENT);
         spin_unlock_irqrestore(&card->irq_lock, flags);
 
-        /* Prevent work_structs from re-queuing themselves. */
-        card->halt = 1;
-
         cancel_work_sync(&card->event_work);
 
         rsxx_destroy_dev(card);
@@ -549,6 +752,10 @@ static void rsxx_pci_remove(struct pci_dev *dev)
         spin_lock_irqsave(&card->irq_lock, flags);
         rsxx_disable_ier_and_isr(card, CR_INTR_ALL);
         spin_unlock_irqrestore(&card->irq_lock, flags);
+
+        /* Prevent work_structs from re-queuing themselves. */
+        card->halt = 1;
+
         free_irq(dev->irq, card);
 
         if (!force_legacy)
@@ -592,11 +799,14 @@ static void rsxx_pci_shutdown(struct pci_dev *dev)
         card_shutdown(card);
 }
 
+static const struct pci_error_handlers rsxx_err_handler = {
+        .error_detected = rsxx_error_detected,
+        .slot_reset     = rsxx_slot_reset,
+};
+
 static DEFINE_PCI_DEVICE_TABLE(rsxx_pci_ids) = {
-        {PCI_DEVICE(PCI_VENDOR_ID_TMS_IBM, PCI_DEVICE_ID_RS70_FLASH)},
-        {PCI_DEVICE(PCI_VENDOR_ID_TMS_IBM, PCI_DEVICE_ID_RS70D_FLASH)},
-        {PCI_DEVICE(PCI_VENDOR_ID_TMS_IBM, PCI_DEVICE_ID_RS80_FLASH)},
-        {PCI_DEVICE(PCI_VENDOR_ID_TMS_IBM, PCI_DEVICE_ID_RS81_FLASH)},
+        {PCI_DEVICE(PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_FS70_FLASH)},
+        {PCI_DEVICE(PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_FS80_FLASH)},
         {0,},
 };
 
@@ -609,6 +819,7 @@ static struct pci_driver rsxx_pci_driver = {
         .remove         = rsxx_pci_remove,
         .suspend        = rsxx_pci_suspend,
         .shutdown       = rsxx_pci_shutdown,
+        .err_handler    = &rsxx_err_handler,
 };
 
 static int __init rsxx_core_init(void)
diff --git a/drivers/block/rsxx/cregs.c b/drivers/block/rsxx/cregs.c
index 80bbe639fccd..4b5c020a0a65 100644
--- a/drivers/block/rsxx/cregs.c
+++ b/drivers/block/rsxx/cregs.c
@@ -58,7 +58,7 @@ static struct kmem_cache *creg_cmd_pool;
 #error Unknown endianess!!! Aborting...
 #endif
 
-static void copy_to_creg_data(struct rsxx_cardinfo *card,
+static int copy_to_creg_data(struct rsxx_cardinfo *card,
                               int cnt8,
                               void *buf,
                               unsigned int stream)
@@ -66,6 +66,9 @@ static void copy_to_creg_data(struct rsxx_cardinfo *card,
         int i = 0;
         u32 *data = buf;
 
+        if (unlikely(card->eeh_state))
+                return -EIO;
+
         for (i = 0; cnt8 > 0; i++, cnt8 -= 4) {
                 /*
                  * Firmware implementation makes it necessary to byte swap on
@@ -76,10 +79,12 @@ static void copy_to_creg_data(struct rsxx_cardinfo *card,
                 else
                         iowrite32(data[i], card->regmap + CREG_DATA(i));
         }
+
+        return 0;
 }
 
 
-static void copy_from_creg_data(struct rsxx_cardinfo *card,
+static int copy_from_creg_data(struct rsxx_cardinfo *card,
                                 int cnt8,
                                 void *buf,
                                 unsigned int stream)
@@ -87,6 +92,9 @@ static void copy_from_creg_data(struct rsxx_cardinfo *card,
         int i = 0;
         u32 *data = buf;
 
+        if (unlikely(card->eeh_state))
+                return -EIO;
+
         for (i = 0; cnt8 > 0; i++, cnt8 -= 4) {
                 /*
                  * Firmware implementation makes it necessary to byte swap on
@@ -97,41 +105,31 @@ static void copy_from_creg_data(struct rsxx_cardinfo *card,
                 else
                         data[i] = ioread32(card->regmap + CREG_DATA(i));
         }
-}
-
-static struct creg_cmd *pop_active_cmd(struct rsxx_cardinfo *card)
-{
-        struct creg_cmd *cmd;
 
-        /*
-         * Spin lock is needed because this can be called in atomic/interrupt
-         * context.
-         */
-        spin_lock_bh(&card->creg_ctrl.lock);
-        cmd = card->creg_ctrl.active_cmd;
-        card->creg_ctrl.active_cmd = NULL;
-        spin_unlock_bh(&card->creg_ctrl.lock);
-
-        return cmd;
+        return 0;
 }
 
 static void creg_issue_cmd(struct rsxx_cardinfo *card, struct creg_cmd *cmd)
 {
+        int st;
+
+        if (unlikely(card->eeh_state))
+                return;
+
         iowrite32(cmd->addr, card->regmap + CREG_ADD);
         iowrite32(cmd->cnt8, card->regmap + CREG_CNT);
 
         if (cmd->op == CREG_OP_WRITE) {
-                if (cmd->buf)
-                        copy_to_creg_data(card, cmd->cnt8,
-                                          cmd->buf, cmd->stream);
+                if (cmd->buf) {
+                        st = copy_to_creg_data(card, cmd->cnt8,
+                                               cmd->buf, cmd->stream);
+                        if (st)
+                                return;
+                }
         }
 
-        /*
-         * Data copy must complete before initiating the command. This is
-         * needed for weakly ordered processors (i.e. PowerPC), so that all
-         * neccessary registers are written before we kick the hardware.
-         */
-        wmb();
+        if (unlikely(card->eeh_state))
+                return;
 
         /* Setting the valid bit will kick off the command. */
         iowrite32(cmd->op, card->regmap + CREG_CMD);
@@ -196,11 +194,11 @@ static int creg_queue_cmd(struct rsxx_cardinfo *card,
         cmd->cb_private = cb_private;
         cmd->status     = 0;
 
-        spin_lock(&card->creg_ctrl.lock);
+        spin_lock_bh(&card->creg_ctrl.lock);
         list_add_tail(&cmd->list, &card->creg_ctrl.queue);
         card->creg_ctrl.q_depth++;
         creg_kick_queue(card);
-        spin_unlock(&card->creg_ctrl.lock);
+        spin_unlock_bh(&card->creg_ctrl.lock);
 
         return 0;
 }
@@ -210,7 +208,11 @@ static void creg_cmd_timed_out(unsigned long data)
         struct rsxx_cardinfo *card = (struct rsxx_cardinfo *) data;
         struct creg_cmd *cmd;
 
-        cmd = pop_active_cmd(card);
+        spin_lock(&card->creg_ctrl.lock);
+        cmd = card->creg_ctrl.active_cmd;
+        card->creg_ctrl.active_cmd = NULL;
+        spin_unlock(&card->creg_ctrl.lock);
+
         if (cmd == NULL) {
                 card->creg_ctrl.creg_stats.creg_timeout++;
                 dev_warn(CARD_TO_DEV(card),
@@ -247,7 +249,11 @@ static void creg_cmd_done(struct work_struct *work)
         if (del_timer_sync(&card->creg_ctrl.cmd_timer) == 0)
                 card->creg_ctrl.creg_stats.failed_cancel_timer++;
 
-        cmd = pop_active_cmd(card);
+        spin_lock_bh(&card->creg_ctrl.lock);
+        cmd = card->creg_ctrl.active_cmd;
+        card->creg_ctrl.active_cmd = NULL;
+        spin_unlock_bh(&card->creg_ctrl.lock);
+
         if (cmd == NULL) {
                 dev_err(CARD_TO_DEV(card),
                         "Spurious creg interrupt!\n");
@@ -287,7 +293,7 @@ static void creg_cmd_done(struct work_struct *work)
                         goto creg_done;
                 }
 
-                copy_from_creg_data(card, cnt8, cmd->buf, cmd->stream);
+                st = copy_from_creg_data(card, cnt8, cmd->buf, cmd->stream);
         }
 
 creg_done:
@@ -296,10 +302,10 @@ creg_done:
 
         kmem_cache_free(creg_cmd_pool, cmd);
 
-        spin_lock(&card->creg_ctrl.lock);
+        spin_lock_bh(&card->creg_ctrl.lock);
         card->creg_ctrl.active = 0;
         creg_kick_queue(card);
-        spin_unlock(&card->creg_ctrl.lock);
+        spin_unlock_bh(&card->creg_ctrl.lock);
 }
 
 static void creg_reset(struct rsxx_cardinfo *card)
@@ -324,7 +330,7 @@ static void creg_reset(struct rsxx_cardinfo *card)
                 "Resetting creg interface for recovery\n");
 
         /* Cancel outstanding commands */
-        spin_lock(&card->creg_ctrl.lock);
+        spin_lock_bh(&card->creg_ctrl.lock);
         list_for_each_entry_safe(cmd, tmp, &card->creg_ctrl.queue, list) {
                 list_del(&cmd->list);
                 card->creg_ctrl.q_depth--;
@@ -345,7 +351,7 @@ static void creg_reset(struct rsxx_cardinfo *card)
 
                 card->creg_ctrl.active = 0;
         }
-        spin_unlock(&card->creg_ctrl.lock);
+        spin_unlock_bh(&card->creg_ctrl.lock);
 
         card->creg_ctrl.reset = 0;
         spin_lock_irqsave(&card->irq_lock, flags);
@@ -399,12 +405,12 @@ static int __issue_creg_rw(struct rsxx_cardinfo *card,
                 return st;
 
         /*
-         * This timeout is neccessary for unresponsive hardware. The additional
+         * This timeout is necessary for unresponsive hardware. The additional
          * 20 seconds to used to guarantee that each cregs requests has time to
          * complete.
          */
-        timeout = msecs_to_jiffies((CREG_TIMEOUT_MSEC *
-                                card->creg_ctrl.q_depth) + 20000);
+        timeout = msecs_to_jiffies(CREG_TIMEOUT_MSEC *
+                                   card->creg_ctrl.q_depth + 20000);
 
         /*
          * The creg interface is guaranteed to complete. It has a timeout
@@ -690,6 +696,32 @@ int rsxx_reg_access(struct rsxx_cardinfo *card,
         return 0;
 }
 
+void rsxx_eeh_save_issued_creg(struct rsxx_cardinfo *card)
+{
+        struct creg_cmd *cmd = NULL;
+
+        cmd = card->creg_ctrl.active_cmd;
+        card->creg_ctrl.active_cmd = NULL;
+
+        if (cmd) {
+                del_timer_sync(&card->creg_ctrl.cmd_timer);
+
+                spin_lock_bh(&card->creg_ctrl.lock);
+                list_add(&cmd->list, &card->creg_ctrl.queue);
+                card->creg_ctrl.q_depth++;
+                card->creg_ctrl.active = 0;
+                spin_unlock_bh(&card->creg_ctrl.lock);
+        }
+}
+
+void rsxx_kick_creg_queue(struct rsxx_cardinfo *card)
+{
+        spin_lock_bh(&card->creg_ctrl.lock);
+        if (!list_empty(&card->creg_ctrl.queue))
+                creg_kick_queue(card);
+        spin_unlock_bh(&card->creg_ctrl.lock);
+}
+
 /*------------ Initialization & Setup --------------*/
 int rsxx_creg_setup(struct rsxx_cardinfo *card)
 {
@@ -712,7 +744,7 @@ void rsxx_creg_destroy(struct rsxx_cardinfo *card)
         int cnt = 0;
 
         /* Cancel outstanding commands */
-        spin_lock(&card->creg_ctrl.lock);
+        spin_lock_bh(&card->creg_ctrl.lock);
         list_for_each_entry_safe(cmd, tmp, &card->creg_ctrl.queue, list) {
                 list_del(&cmd->list);
                 if (cmd->cb)
@@ -737,7 +769,7 @@ void rsxx_creg_destroy(struct rsxx_cardinfo *card)
                         "Canceled active creg command\n");
                 kmem_cache_free(creg_cmd_pool, cmd);
         }
-        spin_unlock(&card->creg_ctrl.lock);
+        spin_unlock_bh(&card->creg_ctrl.lock);
 
         cancel_work_sync(&card->creg_ctrl.done_work);
 }
diff --git a/drivers/block/rsxx/dma.c b/drivers/block/rsxx/dma.c
index 63176e67662f..0607513cfb41 100644
--- a/drivers/block/rsxx/dma.c
+++ b/drivers/block/rsxx/dma.c
@@ -28,7 +28,7 @@
 struct rsxx_dma {
         struct list_head         list;
         u8                       cmd;
-        unsigned int             laddr;     /* Logical address on the ramsan */
+        unsigned int             laddr;     /* Logical address */
         struct {
                 u32              off;
                 u32              cnt;
@@ -81,9 +81,6 @@ enum rsxx_hw_status {
         HW_STATUS_FAULT         = 0x08,
 };
 
-#define STATUS_BUFFER_SIZE8     4096
-#define COMMAND_BUFFER_SIZE8    4096
-
 static struct kmem_cache *rsxx_dma_pool;
 
 struct dma_tracker {
@@ -122,7 +119,7 @@ static unsigned int rsxx_get_dma_tgt(struct rsxx_cardinfo *card, u64 addr8)
         return tgt;
 }
 
-static void rsxx_dma_queue_reset(struct rsxx_cardinfo *card)
+void rsxx_dma_queue_reset(struct rsxx_cardinfo *card)
 {
         /* Reset all DMA Command/Status Queues */
         iowrite32(DMA_QUEUE_RESET, card->regmap + RESET);
@@ -210,7 +207,8 @@ static void dma_intr_coal_auto_tune(struct rsxx_cardinfo *card)
         u32 q_depth = 0;
         u32 intr_coal;
 
-        if (card->config.data.intr_coal.mode != RSXX_INTR_COAL_AUTO_TUNE)
+        if (card->config.data.intr_coal.mode != RSXX_INTR_COAL_AUTO_TUNE ||
+            unlikely(card->eeh_state))
                 return;
 
         for (i = 0; i < card->n_targets; i++)
@@ -223,31 +221,26 @@ static void dma_intr_coal_auto_tune(struct rsxx_cardinfo *card)
 }
 
 /*----------------- RSXX DMA Handling -------------------*/
-static void rsxx_complete_dma(struct rsxx_cardinfo *card,
+static void rsxx_complete_dma(struct rsxx_dma_ctrl *ctrl,
                                   struct rsxx_dma *dma,
                                   unsigned int status)
 {
         if (status & DMA_SW_ERR)
-                printk_ratelimited(KERN_ERR
-                                   "SW Error in DMA(cmd x%02x, laddr x%08x)\n",
-                                   dma->cmd, dma->laddr);
+                ctrl->stats.dma_sw_err++;
         if (status & DMA_HW_FAULT)
-                printk_ratelimited(KERN_ERR
-                                   "HW Fault in DMA(cmd x%02x, laddr x%08x)\n",
-                                   dma->cmd, dma->laddr);
+                ctrl->stats.dma_hw_fault++;
         if (status & DMA_CANCELLED)
-                printk_ratelimited(KERN_ERR
-                                   "DMA Cancelled(cmd x%02x, laddr x%08x)\n",
-                                   dma->cmd, dma->laddr);
+                ctrl->stats.dma_cancelled++;
 
         if (dma->dma_addr)
-                pci_unmap_page(card->dev, dma->dma_addr, get_dma_size(dma),
+                pci_unmap_page(ctrl->card->dev, dma->dma_addr,
+                               get_dma_size(dma),
                                dma->cmd == HW_CMD_BLK_WRITE ?
                                            PCI_DMA_TODEVICE :
                                            PCI_DMA_FROMDEVICE);
 
         if (dma->cb)
-                dma->cb(card, dma->cb_data, status ? 1 : 0);
+                dma->cb(ctrl->card, dma->cb_data, status ? 1 : 0);
 
         kmem_cache_free(rsxx_dma_pool, dma);
 }
@@ -330,14 +323,15 @@ static void rsxx_handle_dma_error(struct rsxx_dma_ctrl *ctrl,
         if (requeue_cmd)
                 rsxx_requeue_dma(ctrl, dma);
         else
-                rsxx_complete_dma(ctrl->card, dma, status);
+                rsxx_complete_dma(ctrl, dma, status);
 }
 
 static void dma_engine_stalled(unsigned long data)
 {
         struct rsxx_dma_ctrl *ctrl = (struct rsxx_dma_ctrl *)data;
 
-        if (atomic_read(&ctrl->stats.hw_q_depth) == 0)
+        if (atomic_read(&ctrl->stats.hw_q_depth) == 0 ||
+            unlikely(ctrl->card->eeh_state))
                 return;
 
         if (ctrl->cmd.idx != ioread32(ctrl->regmap + SW_CMD_IDX)) {
@@ -369,7 +363,8 @@ static void rsxx_issue_dmas(struct work_struct *work)
         ctrl = container_of(work, struct rsxx_dma_ctrl, issue_dma_work);
         hw_cmd_buf = ctrl->cmd.buf;
 
-        if (unlikely(ctrl->card->halt))
+        if (unlikely(ctrl->card->halt) ||
+            unlikely(ctrl->card->eeh_state))
                 return;
 
         while (1) {
@@ -397,7 +392,7 @@ static void rsxx_issue_dmas(struct work_struct *work)
                  */
                 if (unlikely(ctrl->card->dma_fault)) {
                         push_tracker(ctrl->trackers, tag);
-                        rsxx_complete_dma(ctrl->card, dma, DMA_CANCELLED);
+                        rsxx_complete_dma(ctrl, dma, DMA_CANCELLED);
                         continue;
                 }
 
@@ -432,19 +427,15 @@ static void rsxx_issue_dmas(struct work_struct *work)
 
         /* Let HW know we've queued commands. */
         if (cmds_pending) {
-                /*
-                 * We must guarantee that the CPU writes to 'ctrl->cmd.buf'
-                 * (which is in PCI-consistent system-memory) from the loop
-                 * above make it into the coherency domain before the
-                 * following PIO "trigger" updating the cmd.idx.  A WMB is
-                 * sufficient. We need not explicitly CPU cache-flush since
-                 * the memory is a PCI-consistent (ie; coherent) mapping.
-                 */
-                wmb();
-
                 atomic_add(cmds_pending, &ctrl->stats.hw_q_depth);
                 mod_timer(&ctrl->activity_timer,
                           jiffies + DMA_ACTIVITY_TIMEOUT);
+
+                if (unlikely(ctrl->card->eeh_state)) {
+                        del_timer_sync(&ctrl->activity_timer);
+                        return;
+                }
+
                 iowrite32(ctrl->cmd.idx, ctrl->regmap + SW_CMD_IDX);
         }
 }
@@ -463,7 +454,8 @@ static void rsxx_dma_done(struct work_struct *work)
         hw_st_buf = ctrl->status.buf;
 
         if (unlikely(ctrl->card->halt) ||
-            unlikely(ctrl->card->dma_fault))
+            unlikely(ctrl->card->dma_fault) ||
+            unlikely(ctrl->card->eeh_state))
                 return;
 
         count = le16_to_cpu(hw_st_buf[ctrl->status.idx].count);
@@ -508,7 +500,7 @@ static void rsxx_dma_done(struct work_struct *work)
                 if (status)
                         rsxx_handle_dma_error(ctrl, dma, status);
                 else
-                        rsxx_complete_dma(ctrl->card, dma, 0);
+                        rsxx_complete_dma(ctrl, dma, 0);
 
                 push_tracker(ctrl->trackers, tag);
 
@@ -727,20 +719,54 @@ bvec_err:
 
 
 /*----------------- DMA Engine Initialization & Setup -------------------*/
+int rsxx_hw_buffers_init(struct pci_dev *dev, struct rsxx_dma_ctrl *ctrl)
+{
+        ctrl->status.buf = pci_alloc_consistent(dev, STATUS_BUFFER_SIZE8,
+                                &ctrl->status.dma_addr);
+        ctrl->cmd.buf = pci_alloc_consistent(dev, COMMAND_BUFFER_SIZE8,
+                                &ctrl->cmd.dma_addr);
+        if (ctrl->status.buf == NULL || ctrl->cmd.buf == NULL)
+                return -ENOMEM;
+
+        memset(ctrl->status.buf, 0xac, STATUS_BUFFER_SIZE8);
+        iowrite32(lower_32_bits(ctrl->status.dma_addr),
+                ctrl->regmap + SB_ADD_LO);
+        iowrite32(upper_32_bits(ctrl->status.dma_addr),
+                ctrl->regmap + SB_ADD_HI);
+
+        memset(ctrl->cmd.buf, 0x83, COMMAND_BUFFER_SIZE8);
+        iowrite32(lower_32_bits(ctrl->cmd.dma_addr), ctrl->regmap + CB_ADD_LO);
+        iowrite32(upper_32_bits(ctrl->cmd.dma_addr), ctrl->regmap + CB_ADD_HI);
+
+        ctrl->status.idx = ioread32(ctrl->regmap + HW_STATUS_CNT);
+        if (ctrl->status.idx > RSXX_MAX_OUTSTANDING_CMDS) {
+                dev_crit(&dev->dev, "Failed reading status cnt x%x\n",
+                        ctrl->status.idx);
+                return -EINVAL;
+        }
+        iowrite32(ctrl->status.idx, ctrl->regmap + HW_STATUS_CNT);
+        iowrite32(ctrl->status.idx, ctrl->regmap + SW_STATUS_CNT);
+
+        ctrl->cmd.idx = ioread32(ctrl->regmap + HW_CMD_IDX);
+        if (ctrl->cmd.idx > RSXX_MAX_OUTSTANDING_CMDS) {
+                dev_crit(&dev->dev, "Failed reading cmd cnt x%x\n",
+                        ctrl->status.idx);
+                return -EINVAL;
+        }
+        iowrite32(ctrl->cmd.idx, ctrl->regmap + HW_CMD_IDX);
+        iowrite32(ctrl->cmd.idx, ctrl->regmap + SW_CMD_IDX);
+
+        return 0;
+}
+
 static int rsxx_dma_ctrl_init(struct pci_dev *dev,
                                   struct rsxx_dma_ctrl *ctrl)
 {
         int i;
+        int st;
 
         memset(&ctrl->stats, 0, sizeof(ctrl->stats));
 
-        ctrl->status.buf = pci_alloc_consistent(dev, STATUS_BUFFER_SIZE8,
-                                                &ctrl->status.dma_addr);
-        ctrl->cmd.buf = pci_alloc_consistent(dev, COMMAND_BUFFER_SIZE8,
-                                             &ctrl->cmd.dma_addr);
-        if (ctrl->status.buf == NULL || ctrl->cmd.buf == NULL)
-                return -ENOMEM;
-
         ctrl->trackers = vmalloc(DMA_TRACKER_LIST_SIZE8);
         if (!ctrl->trackers)
                 return -ENOMEM;
@@ -770,35 +796,9 @@ static int rsxx_dma_ctrl_init(struct pci_dev *dev,
         INIT_WORK(&ctrl->issue_dma_work, rsxx_issue_dmas);
         INIT_WORK(&ctrl->dma_done_work, rsxx_dma_done);
 
-        memset(ctrl->status.buf, 0xac, STATUS_BUFFER_SIZE8);
-        iowrite32(lower_32_bits(ctrl->status.dma_addr),
-                  ctrl->regmap + SB_ADD_LO);
-        iowrite32(upper_32_bits(ctrl->status.dma_addr),
-                  ctrl->regmap + SB_ADD_HI);
-
-        memset(ctrl->cmd.buf, 0x83, COMMAND_BUFFER_SIZE8);
-        iowrite32(lower_32_bits(ctrl->cmd.dma_addr), ctrl->regmap + CB_ADD_LO);
-        iowrite32(upper_32_bits(ctrl->cmd.dma_addr), ctrl->regmap + CB_ADD_HI);
-
-        ctrl->status.idx = ioread32(ctrl->regmap + HW_STATUS_CNT);
-        if (ctrl->status.idx > RSXX_MAX_OUTSTANDING_CMDS) {
-                dev_crit(&dev->dev, "Failed reading status cnt x%x\n",
-                         ctrl->status.idx);
-                return -EINVAL;
-        }
-        iowrite32(ctrl->status.idx, ctrl->regmap + HW_STATUS_CNT);
-        iowrite32(ctrl->status.idx, ctrl->regmap + SW_STATUS_CNT);
-
-        ctrl->cmd.idx = ioread32(ctrl->regmap + HW_CMD_IDX);
-        if (ctrl->cmd.idx > RSXX_MAX_OUTSTANDING_CMDS) {
-                dev_crit(&dev->dev, "Failed reading cmd cnt x%x\n",
-                         ctrl->status.idx);
-                return -EINVAL;
-        }
-        iowrite32(ctrl->cmd.idx, ctrl->regmap + HW_CMD_IDX);
-        iowrite32(ctrl->cmd.idx, ctrl->regmap + SW_CMD_IDX);
-
-        wmb();
+        st = rsxx_hw_buffers_init(dev, ctrl);
+        if (st)
+                return st;
 
         return 0;
 }
@@ -834,7 +834,7 @@ static int rsxx_dma_stripe_setup(struct rsxx_cardinfo *card,
         return 0;
 }
 
-static int rsxx_dma_configure(struct rsxx_cardinfo *card)
+int rsxx_dma_configure(struct rsxx_cardinfo *card)
 {
         u32 intr_coal;
 
@@ -980,6 +980,103 @@ void rsxx_dma_destroy(struct rsxx_cardinfo *card)
         }
 }
 
+int rsxx_eeh_save_issued_dmas(struct rsxx_cardinfo *card)
+{
+        int i;
+        int j;
+        int cnt;
+        struct rsxx_dma *dma;
+        struct list_head *issued_dmas;
+
+        issued_dmas = kzalloc(sizeof(*issued_dmas) * card->n_targets,
+                              GFP_KERNEL);
+        if (!issued_dmas)
+                return -ENOMEM;
+
+        for (i = 0; i < card->n_targets; i++) {
+                INIT_LIST_HEAD(&issued_dmas[i]);
+                cnt = 0;
+                for (j = 0; j < RSXX_MAX_OUTSTANDING_CMDS; j++) {
+                        dma = get_tracker_dma(card->ctrl[i].trackers, j);
+                        if (dma == NULL)
+                                continue;
+
+                        if (dma->cmd == HW_CMD_BLK_WRITE)
+                                card->ctrl[i].stats.writes_issued--;
+                        else if (dma->cmd == HW_CMD_BLK_DISCARD)
+                                card->ctrl[i].stats.discards_issued--;
+                        else
+                                card->ctrl[i].stats.reads_issued--;
+
+                        list_add_tail(&dma->list, &issued_dmas[i]);
+                        push_tracker(card->ctrl[i].trackers, j);
+                        cnt++;
+                }
+
+                spin_lock(&card->ctrl[i].queue_lock);
+                list_splice(&issued_dmas[i], &card->ctrl[i].queue);
+
+                atomic_sub(cnt, &card->ctrl[i].stats.hw_q_depth);
+                card->ctrl[i].stats.sw_q_depth += cnt;
+                card->ctrl[i].e_cnt = 0;
+
+                list_for_each_entry(dma, &card->ctrl[i].queue, list) {
+                        if (dma->dma_addr)
+                                pci_unmap_page(card->dev, dma->dma_addr,
+                                               get_dma_size(dma),
+                                               dma->cmd == HW_CMD_BLK_WRITE ?
+                                               PCI_DMA_TODEVICE :
+                                               PCI_DMA_FROMDEVICE);
+                }
+                spin_unlock(&card->ctrl[i].queue_lock);
+        }
+
+        kfree(issued_dmas);
+
+        return 0;
+}
+
+void rsxx_eeh_cancel_dmas(struct rsxx_cardinfo *card)
+{
+        struct rsxx_dma *dma;
+        struct rsxx_dma *tmp;
+        int i;
+
+        for (i = 0; i < card->n_targets; i++) {
+                spin_lock(&card->ctrl[i].queue_lock);
+                list_for_each_entry_safe(dma, tmp, &card->ctrl[i].queue, list) {
+                        list_del(&dma->list);
+
+                        rsxx_complete_dma(&card->ctrl[i], dma, DMA_CANCELLED);
+                }
+                spin_unlock(&card->ctrl[i].queue_lock);
+        }
+}
+
+int rsxx_eeh_remap_dmas(struct rsxx_cardinfo *card)
+{
+        struct rsxx_dma *dma;
+        int i;
+
+        for (i = 0; i < card->n_targets; i++) {
+                spin_lock(&card->ctrl[i].queue_lock);
+                list_for_each_entry(dma, &card->ctrl[i].queue, list) {
+                        dma->dma_addr = pci_map_page(card->dev, dma->page,
+                                        dma->pg_off, get_dma_size(dma),
+                                        dma->cmd == HW_CMD_BLK_WRITE ?
+                                        PCI_DMA_TODEVICE :
+                                        PCI_DMA_FROMDEVICE);
+                        if (!dma->dma_addr) {
+                                spin_unlock(&card->ctrl[i].queue_lock);
+                                kmem_cache_free(rsxx_dma_pool, dma);
+                                return -ENOMEM;
+                        }
+                }
+                spin_unlock(&card->ctrl[i].queue_lock);
+        }
+
+        return 0;
+}
 
 int rsxx_dma_init(void)
 {
diff --git a/drivers/block/rsxx/rsxx.h b/drivers/block/rsxx/rsxx.h
index 2e50b65902b7..24ba3642bd89 100644
--- a/drivers/block/rsxx/rsxx.h
+++ b/drivers/block/rsxx/rsxx.h
@@ -27,15 +27,17 @@
 
 /*----------------- IOCTL Definitions -------------------*/
 
+#define RSXX_MAX_DATA 8
+
 struct rsxx_reg_access {
         __u32 addr;
         __u32 cnt;
         __u32 stat;
         __u32 stream;
-        __u32 data[8];
+        __u32 data[RSXX_MAX_DATA];
 };
 
-#define RSXX_MAX_REG_CNT        (8 * (sizeof(__u32)))
+#define RSXX_MAX_REG_CNT        (RSXX_MAX_DATA * (sizeof(__u32)))
 
 #define RSXX_IOC_MAGIC 'r'
 
diff --git a/drivers/block/rsxx/rsxx_cfg.h b/drivers/block/rsxx/rsxx_cfg.h
index c025fe5fdb70..f384c943846d 100644
--- a/drivers/block/rsxx/rsxx_cfg.h
+++ b/drivers/block/rsxx/rsxx_cfg.h
@@ -58,7 +58,7 @@ struct rsxx_card_cfg {
 };
 
 /* Vendor ID Values */
-#define RSXX_VENDOR_ID_TMS_IBM          0
+#define RSXX_VENDOR_ID_IBM              0
 #define RSXX_VENDOR_ID_DSI              1
 #define RSXX_VENDOR_COUNT               2
 
diff --git a/drivers/block/rsxx/rsxx_priv.h b/drivers/block/rsxx/rsxx_priv.h
index a1ac907d8f4c..382e8bf5c03b 100644
--- a/drivers/block/rsxx/rsxx_priv.h
+++ b/drivers/block/rsxx/rsxx_priv.h
@@ -45,16 +45,13 @@
 
 struct proc_cmd;
 
-#define PCI_VENDOR_ID_TMS_IBM           0x15B6
-#define PCI_DEVICE_ID_RS70_FLASH        0x0019
-#define PCI_DEVICE_ID_RS70D_FLASH       0x001A
-#define PCI_DEVICE_ID_RS80_FLASH        0x001C
-#define PCI_DEVICE_ID_RS81_FLASH        0x001E
+#define PCI_DEVICE_ID_FS70_FLASH        0x04A9
+#define PCI_DEVICE_ID_FS80_FLASH        0x04AA
 
 #define RS70_PCI_REV_SUPPORTED  4
 
 #define DRIVER_NAME "rsxx"
-#define DRIVER_VERSION "3.7"
+#define DRIVER_VERSION "4.0"
 
 /* Block size is 4096 */
 #define RSXX_HW_BLK_SHIFT               12
@@ -67,6 +64,9 @@ struct proc_cmd;
 #define RSXX_MAX_OUTSTANDING_CMDS       255
 #define RSXX_CS_IDX_MASK                0xff
 
+#define STATUS_BUFFER_SIZE8     4096
+#define COMMAND_BUFFER_SIZE8    4096
+
 #define RSXX_MAX_TARGETS        8
 
 struct dma_tracker_list;
@@ -91,6 +91,9 @@ struct rsxx_dma_stats {
         u32 discards_failed;
         u32 done_rescheduled;
         u32 issue_rescheduled;
+        u32 dma_sw_err;
+        u32 dma_hw_fault;
+        u32 dma_cancelled;
         u32 sw_q_depth;         /* Number of DMAs on the SW queue. */
         atomic_t hw_q_depth;    /* Number of DMAs queued to HW. */
 };
@@ -116,6 +119,7 @@ struct rsxx_dma_ctrl {
 struct rsxx_cardinfo {
         struct pci_dev          *dev;
         unsigned int            halt;
+        unsigned int            eeh_state;
 
         void                    __iomem *regmap;
         spinlock_t              irq_lock;
@@ -224,6 +228,7 @@ enum rsxx_pci_regmap {
         PERF_RD512_HI   = 0xac,
         PERF_WR512_LO   = 0xb0,
         PERF_WR512_HI   = 0xb4,
+        PCI_RECONFIG    = 0xb8,
 };
 
 enum rsxx_intr {
@@ -237,6 +242,8 @@ enum rsxx_intr {
         CR_INTR_DMA5    = 0x00000080,
         CR_INTR_DMA6    = 0x00000100,
         CR_INTR_DMA7    = 0x00000200,
+        CR_INTR_ALL_C   = 0x0000003f,
+        CR_INTR_ALL_G   = 0x000003ff,
         CR_INTR_DMA_ALL = 0x000003f5,
         CR_INTR_ALL     = 0xffffffff,
 };
@@ -253,8 +260,14 @@ enum rsxx_pci_reset {
         DMA_QUEUE_RESET         = 0x00000001,
 };
 
+enum rsxx_hw_fifo_flush {
+        RSXX_FLUSH_BUSY         = 0x00000002,
+        RSXX_FLUSH_TIMEOUT      = 0x00000004,
+};
+
 enum rsxx_pci_revision {
         RSXX_DISCARD_SUPPORT = 2,
+        RSXX_EEH_SUPPORT     = 3,
 };
 
 enum rsxx_creg_cmd {
@@ -360,11 +373,17 @@ int rsxx_dma_setup(struct rsxx_cardinfo *card);
 void rsxx_dma_destroy(struct rsxx_cardinfo *card);
 int rsxx_dma_init(void);
 void rsxx_dma_cleanup(void);
+void rsxx_dma_queue_reset(struct rsxx_cardinfo *card);
+int rsxx_dma_configure(struct rsxx_cardinfo *card);
 int rsxx_dma_queue_bio(struct rsxx_cardinfo *card,
                            struct bio *bio,
                            atomic_t *n_dmas,
                            rsxx_dma_cb cb,
                            void *cb_data);
+int rsxx_hw_buffers_init(struct pci_dev *dev, struct rsxx_dma_ctrl *ctrl);
+int rsxx_eeh_save_issued_dmas(struct rsxx_cardinfo *card);
+void rsxx_eeh_cancel_dmas(struct rsxx_cardinfo *card);
+int rsxx_eeh_remap_dmas(struct rsxx_cardinfo *card);
 
 /***** cregs.c *****/
 int rsxx_creg_write(struct rsxx_cardinfo *card, u32 addr,
@@ -389,10 +408,11 @@ int rsxx_creg_setup(struct rsxx_cardinfo *card);
 void rsxx_creg_destroy(struct rsxx_cardinfo *card);
 int rsxx_creg_init(void);
 void rsxx_creg_cleanup(void);
-
 int rsxx_reg_access(struct rsxx_cardinfo *card,
                         struct rsxx_reg_access __user *ucmd,
                         int read);
+void rsxx_eeh_save_issued_creg(struct rsxx_cardinfo *card);
+void rsxx_kick_creg_queue(struct rsxx_cardinfo *card);