Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / drivers / pci / msi.c
index 132af77..9855c4c 100644 (file)
 #include <asm/io.h>
 #include <asm/smp.h>
 
+#include "pci.h"
 #include "msi.h"
 
-static spinlock_t msi_lock = SPIN_LOCK_UNLOCKED;
+#define MSI_TARGET_CPU         first_cpu(cpu_online_map)
+
+static DEFINE_SPINLOCK(msi_lock);
 static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
 static kmem_cache_t* msi_cachep;
 
 static int pci_msi_enable = 1;
-static int last_alloc_vector = 0;
-static int nr_released_vectors = 0;
+static int last_alloc_vector;
+static int nr_released_vectors;
 static int nr_reserved_vectors = NR_HP_RESERVED_VECTORS;
-static int nr_msix_devices = 0;
+static int nr_msix_devices;
 
 #ifndef CONFIG_X86_IO_APIC
 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
@@ -64,15 +67,13 @@ static void msi_set_mask_bit(unsigned int vector, int flag)
        case PCI_CAP_ID_MSI:
        {
                int             pos;
-               unsigned int    mask_bits;
+               u32             mask_bits;
 
-               pos = entry->mask_base;
-               entry->dev->bus->ops->read(entry->dev->bus, entry->dev->devfn,
-                               pos, 4, &mask_bits);
+               pos = (long)entry->mask_base;
+               pci_read_config_dword(entry->dev, pos, &mask_bits);
                mask_bits &= ~(1);
                mask_bits |= flag;
-               entry->dev->bus->ops->write(entry->dev->bus, entry->dev->devfn,
-                               pos, 4, mask_bits);
+               pci_write_config_dword(entry->dev, pos, mask_bits);
                break;
        }
        case PCI_CAP_ID_MSIX:
@@ -92,6 +93,8 @@ static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
 {
        struct msi_desc *entry;
        struct msg_address address;
+       unsigned int irq = vector;
+       unsigned int dest_cpu = first_cpu(cpu_mask);
 
        entry = (struct msi_desc *)msi_desc[vector];
        if (!entry || !entry->dev)
@@ -100,21 +103,20 @@ static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
        switch (entry->msi_attrib.type) {
        case PCI_CAP_ID_MSI:
        {
-               int pos;
+               int pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI);
 
-               if (!(pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI)))
+               if (!pos)
                        return;
 
-               entry->dev->bus->ops->read(entry->dev->bus, entry->dev->devfn,
-                       msi_lower_address_reg(pos), 4,
+               pci_read_config_dword(entry->dev, msi_lower_address_reg(pos),
                        &address.lo_address.value);
                address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
-               address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) <<
-                       MSI_TARGET_CPU_SHIFT);
-               entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask);
-               entry->dev->bus->ops->write(entry->dev->bus, entry->dev->devfn,
-                       msi_lower_address_reg(pos), 4,
+               address.lo_address.value |= (cpu_physical_id(dest_cpu) <<
+                                                                       MSI_TARGET_CPU_SHIFT);
+               entry->msi_attrib.current_cpu = cpu_physical_id(dest_cpu);
+               pci_write_config_dword(entry->dev, msi_lower_address_reg(pos),
                        address.lo_address.value);
+               set_native_irq_info(irq, cpu_mask);
                break;
        }
        case PCI_CAP_ID_MSIX:
@@ -124,26 +126,19 @@ static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
 
                address.lo_address.value = readl(entry->mask_base + offset);
                address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
-               address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) <<
-                       MSI_TARGET_CPU_SHIFT);
-               entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask);
+               address.lo_address.value |= (cpu_physical_id(dest_cpu) <<
+                                                                       MSI_TARGET_CPU_SHIFT);
+               entry->msi_attrib.current_cpu = cpu_physical_id(dest_cpu);
                writel(address.lo_address.value, entry->mask_base + offset);
+               set_native_irq_info(irq, cpu_mask);
                break;
        }
        default:
                break;
        }
 }
-
-#ifdef CONFIG_IRQBALANCE
-static inline void move_msi(int vector)
-{
-       if (!cpus_empty(pending_irq_balance_cpumask[vector])) {
-               set_msi_affinity(vector, pending_irq_balance_cpumask[vector]);
-               cpus_clear(pending_irq_balance_cpumask[vector]);
-       }
-}
-#endif /* CONFIG_IRQBALANCE */
+#else
+#define set_msi_affinity NULL
 #endif /* CONFIG_SMP */
 
 static void mask_MSI_irq(unsigned int vector)
@@ -158,41 +153,55 @@ static void unmask_MSI_irq(unsigned int vector)
 
 static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector)
 {
+       struct msi_desc *entry;
+       unsigned long flags;
+
+       spin_lock_irqsave(&msi_lock, flags);
+       entry = msi_desc[vector];
+       if (!entry || !entry->dev) {
+               spin_unlock_irqrestore(&msi_lock, flags);
+               return 0;
+       }
+       entry->msi_attrib.state = 1;    /* Mark it active */
+       spin_unlock_irqrestore(&msi_lock, flags);
+
+       return 0;       /* never anything pending */
+}
+
+static unsigned int startup_msi_irq_w_maskbit(unsigned int vector)
+{
+       startup_msi_irq_wo_maskbit(vector);
+       unmask_MSI_irq(vector);
        return 0;       /* never anything pending */
 }
 
-static void pci_disable_msi(unsigned int vector);
 static void shutdown_msi_irq(unsigned int vector)
 {
-       pci_disable_msi(vector);
+       struct msi_desc *entry;
+       unsigned long flags;
+
+       spin_lock_irqsave(&msi_lock, flags);
+       entry = msi_desc[vector];
+       if (entry && entry->dev)
+               entry->msi_attrib.state = 0;    /* Mark it not active */
+       spin_unlock_irqrestore(&msi_lock, flags);
 }
 
-#define shutdown_msi_irq_wo_maskbit    shutdown_msi_irq
-static void enable_msi_irq_wo_maskbit(unsigned int vector) {}
-static void disable_msi_irq_wo_maskbit(unsigned int vector) {}
-static void ack_msi_irq_wo_maskbit(unsigned int vector) {}
 static void end_msi_irq_wo_maskbit(unsigned int vector)
 {
-       move_msi(vector);
+       move_native_irq(vector);
        ack_APIC_irq();
 }
 
-static unsigned int startup_msi_irq_w_maskbit(unsigned int vector)
+static void end_msi_irq_w_maskbit(unsigned int vector)
 {
+       move_native_irq(vector);
        unmask_MSI_irq(vector);
-       return 0;       /* never anything pending */
+       ack_APIC_irq();
 }
 
-#define shutdown_msi_irq_w_maskbit     shutdown_msi_irq
-#define enable_msi_irq_w_maskbit       unmask_MSI_irq
-#define disable_msi_irq_w_maskbit      mask_MSI_irq
-#define ack_msi_irq_w_maskbit          mask_MSI_irq
-
-static void end_msi_irq_w_maskbit(unsigned int vector)
+static void do_nothing(unsigned int vector)
 {
-       move_msi(vector);
-       unmask_MSI_irq(vector);
-       ack_APIC_irq();
 }
 
 /*
@@ -200,14 +209,14 @@ static void end_msi_irq_w_maskbit(unsigned int vector)
  * which implement the MSI-X Capability Structure.
  */
 static struct hw_interrupt_type msix_irq_type = {
-       .typename       = "PCI MSI-X",
+       .typename       = "PCI-MSI-X",
        .startup        = startup_msi_irq_w_maskbit,
-       .shutdown       = shutdown_msi_irq_w_maskbit,
-       .enable         = enable_msi_irq_w_maskbit,
-       .disable        = disable_msi_irq_w_maskbit,
-       .ack            = ack_msi_irq_w_maskbit,
+       .shutdown       = shutdown_msi_irq,
+       .enable         = unmask_MSI_irq,
+       .disable        = mask_MSI_irq,
+       .ack            = mask_MSI_irq,
        .end            = end_msi_irq_w_maskbit,
-       .set_affinity   = set_msi_irq_affinity
+       .set_affinity   = set_msi_affinity
 };
 
 /*
@@ -216,14 +225,14 @@ static struct hw_interrupt_type msix_irq_type = {
  * Mask-and-Pending Bits.
  */
 static struct hw_interrupt_type msi_irq_w_maskbit_type = {
-       .typename       = "PCI MSI",
+       .typename       = "PCI-MSI",
        .startup        = startup_msi_irq_w_maskbit,
-       .shutdown       = shutdown_msi_irq_w_maskbit,
-       .enable         = enable_msi_irq_w_maskbit,
-       .disable        = disable_msi_irq_w_maskbit,
-       .ack            = ack_msi_irq_w_maskbit,
+       .shutdown       = shutdown_msi_irq,
+       .enable         = unmask_MSI_irq,
+       .disable        = mask_MSI_irq,
+       .ack            = mask_MSI_irq,
        .end            = end_msi_irq_w_maskbit,
-       .set_affinity   = set_msi_irq_affinity
+       .set_affinity   = set_msi_affinity
 };
 
 /*
@@ -232,14 +241,14 @@ static struct hw_interrupt_type msi_irq_w_maskbit_type = {
  * Mask-and-Pending Bits.
  */
 static struct hw_interrupt_type msi_irq_wo_maskbit_type = {
-       .typename       = "PCI MSI",
+       .typename       = "PCI-MSI",
        .startup        = startup_msi_irq_wo_maskbit,
-       .shutdown       = shutdown_msi_irq_wo_maskbit,
-       .enable         = enable_msi_irq_wo_maskbit,
-       .disable        = disable_msi_irq_wo_maskbit,
-       .ack            = ack_msi_irq_wo_maskbit,
+       .shutdown       = shutdown_msi_irq,
+       .enable         = do_nothing,
+       .disable        = do_nothing,
+       .ack            = do_nothing,
        .end            = end_msi_irq_wo_maskbit,
-       .set_affinity   = set_msi_irq_affinity
+       .set_affinity   = set_msi_affinity
 };
 
 static void msi_data_init(struct msg_data *msi_data,
@@ -255,16 +264,18 @@ static void msi_data_init(struct msg_data *msi_data,
 static void msi_address_init(struct msg_address *msi_address)
 {
        unsigned int    dest_id;
+       unsigned long   dest_phys_id = cpu_physical_id(MSI_TARGET_CPU);
 
        memset(msi_address, 0, sizeof(struct msg_address));
        msi_address->hi_address = (u32)0;
        dest_id = (MSI_ADDRESS_HEADER << MSI_ADDRESS_HEADER_SHIFT);
-       msi_address->lo_address.u.dest_mode = MSI_DEST_MODE;
+       msi_address->lo_address.u.dest_mode = MSI_PHYSICAL_MODE;
        msi_address->lo_address.u.redirection_hint = MSI_REDIRECTION_HINT_MODE;
        msi_address->lo_address.u.dest_id = dest_id;
-       msi_address->lo_address.value |= (MSI_TARGET_CPU << MSI_TARGET_CPU_SHIFT);
+       msi_address->lo_address.value |= (dest_phys_id << MSI_TARGET_CPU_SHIFT);
 }
 
+static int msi_free_vector(struct pci_dev* dev, int vector, int reassign);
 static int assign_msi_vector(void)
 {
        static int new_vector_avail = 1;
@@ -278,6 +289,8 @@ static int assign_msi_vector(void)
        spin_lock_irqsave(&msi_lock, flags);
 
        if (!new_vector_avail) {
+               int free_vector = 0;
+
                /*
                 * vector_irq[] = -1 indicates that this specific vector is:
                 * - assigned for MSI (since MSI have no associated IRQ) or
@@ -294,13 +307,34 @@ static int assign_msi_vector(void)
                for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
                        if (vector_irq[vector] != 0)
                                continue;
-                       vector_irq[vector] = -1;
-                       nr_released_vectors--;
+                       free_vector = vector;
+                       if (!msi_desc[vector])
+                               break;
+                       else
+                               continue;
+               }
+               if (!free_vector) {
                        spin_unlock_irqrestore(&msi_lock, flags);
-                       return vector;
+                       return -EBUSY;
                }
+               vector_irq[free_vector] = -1;
+               nr_released_vectors--;
                spin_unlock_irqrestore(&msi_lock, flags);
-               return -EBUSY;
+               if (msi_desc[free_vector] != NULL) {
+                       struct pci_dev *dev;
+                       int tail;
+
+                       /* free all linked vectors before re-assign */
+                       do {
+                               spin_lock_irqsave(&msi_lock, flags);
+                               dev = msi_desc[free_vector]->dev;
+                               tail = msi_desc[free_vector]->link.tail;
+                               spin_unlock_irqrestore(&msi_lock, flags);
+                               msi_free_vector(dev, tail, 1);
+                       } while (free_vector != tail);
+               }
+
+               return free_vector;
        }
        vector = assign_irq_vector(AUTO_ASSIGN);
        last_alloc_vector = vector;
@@ -313,9 +347,9 @@ static int assign_msi_vector(void)
 
 static int get_new_vector(void)
 {
-       int vector;
+       int vector = assign_msi_vector();
 
-       if ((vector = assign_msi_vector()) > 0)
+       if (vector > 0)
                set_intr_gate(vector, interrupt[vector]);
 
        return vector;
@@ -328,12 +362,28 @@ static int msi_init(void)
        if (!status)
                return status;
 
-       if ((status = msi_cache_init()) < 0) {
+       if (pci_msi_quirk) {
+               pci_msi_enable = 0;
+               printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n");
+               status = -EINVAL;
+               return status;
+       }
+
+       status = msi_cache_init();
+       if (status < 0) {
+               pci_msi_enable = 0;
+               printk(KERN_WARNING "PCI: MSI cache init failed\n");
+               return status;
+       }
+       last_alloc_vector = assign_irq_vector(AUTO_ASSIGN);
+       if (last_alloc_vector < 0) {
                pci_msi_enable = 0;
-               printk(KERN_INFO "WARNING: MSI INIT FAILURE\n");
+               printk(KERN_WARNING "PCI: No interrupt vectors available for MSI\n");
+               status = -EBUSY;
                return status;
        }
-       printk(KERN_INFO "MSI INIT SUCCESS\n");
+       vector_irq[last_alloc_vector] = 0;
+       nr_released_vectors++;
 
        return status;
 }
@@ -347,7 +397,7 @@ static struct msi_desc* alloc_msi_entry(void)
 {
        struct msi_desc *entry;
 
-       entry = (struct msi_desc*) kmem_cache_alloc(msi_cachep, SLAB_KERNEL);
+       entry = kmem_cache_alloc(msi_cachep, SLAB_KERNEL);
        if (!entry)
                return NULL;
 
@@ -369,7 +419,9 @@ static void attach_msi_entry(struct msi_desc *entry, int vector)
 
 static void irq_handler_init(int cap_id, int pos, int mask)
 {
-       spin_lock(&irq_desc[pos].lock);
+       unsigned long flags;
+
+       spin_lock_irqsave(&irq_desc[pos].lock, flags);
        if (cap_id == PCI_CAP_ID_MSIX)
                irq_desc[pos].handler = &msix_irq_type;
        else {
@@ -378,60 +430,48 @@ static void irq_handler_init(int cap_id, int pos, int mask)
                else
                        irq_desc[pos].handler = &msi_irq_w_maskbit_type;
        }
-       spin_unlock(&irq_desc[pos].lock);
+       spin_unlock_irqrestore(&irq_desc[pos].lock, flags);
 }
 
 static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
 {
-       u32 control;
+       u16 control;
 
-       dev->bus->ops->read(dev->bus, dev->devfn,
-               msi_control_reg(pos), 2, &control);
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
        if (type == PCI_CAP_ID_MSI) {
                /* Set enabled bits to single MSI & enable MSI_enable bit */
                msi_enable(control, 1);
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_control_reg(pos), 2, control);
+               pci_write_config_word(dev, msi_control_reg(pos), control);
        } else {
                msix_enable(control);
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_control_reg(pos), 2, control);
+               pci_write_config_word(dev, msi_control_reg(pos), control);
        }
        if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
                /* PCI Express Endpoint device detected */
-               u32 cmd;
-               dev->bus->ops->read(dev->bus, dev->devfn, PCI_COMMAND, 2, &cmd);
-               cmd |= PCI_COMMAND_INTX_DISABLE;
-               dev->bus->ops->write(dev->bus, dev->devfn, PCI_COMMAND, 2, cmd);
+               pci_intx(dev, 0);  /* disable intx */
        }
 }
 
-static void disable_msi_mode(struct pci_dev *dev, int pos, int type)
+void disable_msi_mode(struct pci_dev *dev, int pos, int type)
 {
-       u32 control;
+       u16 control;
 
-       dev->bus->ops->read(dev->bus, dev->devfn,
-               msi_control_reg(pos), 2, &control);
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
        if (type == PCI_CAP_ID_MSI) {
                /* Set enabled bits to single MSI & enable MSI_enable bit */
                msi_disable(control);
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_control_reg(pos), 2, control);
+               pci_write_config_word(dev, msi_control_reg(pos), control);
        } else {
                msix_disable(control);
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_control_reg(pos), 2, control);
+               pci_write_config_word(dev, msi_control_reg(pos), control);
        }
        if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
                /* PCI Express Endpoint device detected */
-               u32 cmd;
-               dev->bus->ops->read(dev->bus, dev->devfn, PCI_COMMAND, 2, &cmd);
-               cmd &= ~PCI_COMMAND_INTX_DISABLE;
-               dev->bus->ops->write(dev->bus, dev->devfn, PCI_COMMAND, 2, cmd);
+               pci_intx(dev, 1);  /* enable intx */
        }
 }
 
-static int msi_lookup_vector(struct pci_dev *dev)
+static int msi_lookup_vector(struct pci_dev *dev, int type)
 {
        int vector;
        unsigned long flags;
@@ -439,11 +479,11 @@ static int msi_lookup_vector(struct pci_dev *dev)
        spin_lock_irqsave(&msi_lock, flags);
        for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
                if (!msi_desc[vector] || msi_desc[vector]->dev != dev ||
-                       msi_desc[vector]->msi_attrib.entry_nr ||
+                       msi_desc[vector]->msi_attrib.type != type ||
                        msi_desc[vector]->msi_attrib.default_vector != dev->irq)
-                       continue;       /* not entry 0, skip */
+                       continue;
                spin_unlock_irqrestore(&msi_lock, flags);
-               /* This pre-assigned entry-0 MSI vector for this device
+               /* This pre-assigned MSI vector for this device
                   already exits. Override dev->irq with this vector */
                dev->irq = vector;
                return 0;
@@ -458,18 +498,212 @@ void pci_scan_msi_device(struct pci_dev *dev)
        if (!dev)
                return;
 
-       if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0) {
-               nr_reserved_vectors++;
+       if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0)
                nr_msix_devices++;
-       else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0)
+       else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0)
                nr_reserved_vectors++;
 }
 
+#ifdef CONFIG_PM
+int pci_save_msi_state(struct pci_dev *dev)
+{
+       int pos, i = 0;
+       u16 control;
+       struct pci_cap_saved_state *save_state;
+       u32 *cap;
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       if (pos <= 0 || dev->no_msi)
+               return 0;
+
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (!(control & PCI_MSI_FLAGS_ENABLE))
+               return 0;
+
+       save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
+               GFP_KERNEL);
+       if (!save_state) {
+               printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
+               return -ENOMEM;
+       }
+       cap = &save_state->data[0];
+
+       pci_read_config_dword(dev, pos, &cap[i++]);
+       control = cap[0] >> 16;
+       pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
+       if (control & PCI_MSI_FLAGS_64BIT) {
+               pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
+               pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
+       } else
+               pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
+       if (control & PCI_MSI_FLAGS_MASKBIT)
+               pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
+       disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
+       save_state->cap_nr = PCI_CAP_ID_MSI;
+       pci_add_saved_cap(dev, save_state);
+       return 0;
+}
+
+void pci_restore_msi_state(struct pci_dev *dev)
+{
+       int i = 0, pos;
+       u16 control;
+       struct pci_cap_saved_state *save_state;
+       u32 *cap;
+
+       save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       if (!save_state || pos <= 0)
+               return;
+       cap = &save_state->data[0];
+
+       control = cap[i++] >> 16;
+       pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
+       if (control & PCI_MSI_FLAGS_64BIT) {
+               pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
+               pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
+       } else
+               pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
+       if (control & PCI_MSI_FLAGS_MASKBIT)
+               pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
+       pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
+       enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
+       pci_remove_saved_cap(save_state);
+       kfree(save_state);
+}
+
+int pci_save_msix_state(struct pci_dev *dev)
+{
+       int pos;
+       u16 control;
+       struct pci_cap_saved_state *save_state;
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (pos <= 0 || dev->no_msi)
+               return 0;
+
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (!(control & PCI_MSIX_FLAGS_ENABLE))
+               return 0;
+       save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
+               GFP_KERNEL);
+       if (!save_state) {
+               printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
+               return -ENOMEM;
+       }
+       *((u16 *)&save_state->data[0]) = control;
+
+       disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
+       save_state->cap_nr = PCI_CAP_ID_MSIX;
+       pci_add_saved_cap(dev, save_state);
+       return 0;
+}
+
+void pci_restore_msix_state(struct pci_dev *dev)
+{
+       u16 save;
+       int pos;
+       int vector, head, tail = 0;
+       void __iomem *base;
+       int j;
+       struct msg_address address;
+       struct msg_data data;
+       struct msi_desc *entry;
+       int temp;
+       struct pci_cap_saved_state *save_state;
+
+       save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
+       if (!save_state)
+               return;
+       save = *((u16 *)&save_state->data[0]);
+       pci_remove_saved_cap(save_state);
+       kfree(save_state);
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (pos <= 0)
+               return;
+
+       /* route the table */
+       temp = dev->irq;
+       if (msi_lookup_vector(dev, PCI_CAP_ID_MSIX))
+               return;
+       vector = head = dev->irq;
+       while (head != tail) {
+               entry = msi_desc[vector];
+               base = entry->mask_base;
+               j = entry->msi_attrib.entry_nr;
+
+               msi_address_init(&address);
+               msi_data_init(&data, vector);
+
+               address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
+               address.lo_address.value |= entry->msi_attrib.current_cpu <<
+                                       MSI_TARGET_CPU_SHIFT;
+
+               writel(address.lo_address.value,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
+                       PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
+               writel(address.hi_address,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
+                       PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
+               writel(*(u32*)&data,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
+                       PCI_MSIX_ENTRY_DATA_OFFSET);
+
+               tail = msi_desc[vector]->link.tail;
+               vector = tail;
+       }
+       dev->irq = temp;
+
+       pci_write_config_word(dev, msi_control_reg(pos), save);
+       enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
+}
+#endif
+
+static void msi_register_init(struct pci_dev *dev, struct msi_desc *entry)
+{
+       struct msg_address address;
+       struct msg_data data;
+       int pos, vector = dev->irq;
+       u16 control;
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       /* Configure MSI capability structure */
+       msi_address_init(&address);
+       msi_data_init(&data, vector);
+       entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
+                               MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
+       pci_write_config_dword(dev, msi_lower_address_reg(pos),
+                       address.lo_address.value);
+       if (is_64bit_address(control)) {
+               pci_write_config_dword(dev,
+                       msi_upper_address_reg(pos), address.hi_address);
+               pci_write_config_word(dev,
+                       msi_data_reg(pos, 1), *((u32*)&data));
+       } else
+               pci_write_config_word(dev,
+                       msi_data_reg(pos, 0), *((u32*)&data));
+       if (entry->msi_attrib.maskbit) {
+               unsigned int maskbits, temp;
+               /* All MSIs are unmasked by default, Mask them all */
+               pci_read_config_dword(dev,
+                       msi_mask_bits_reg(pos, is_64bit_address(control)),
+                       &maskbits);
+               temp = (1 << multi_msi_capable(control));
+               temp = ((temp - 1) & ~temp);
+               maskbits |= temp;
+               pci_write_config_dword(dev,
+                       msi_mask_bits_reg(pos, is_64bit_address(control)),
+                       maskbits);
+       }
+}
+
 /**
  * msi_capability_init - configure device's MSI capability structure
  * @dev: pointer to the pci_dev data structure of MSI device function
  *
- * Setup the MSI capability structure of device funtion with a single
+ * Setup the MSI capability structure of device function with a single
  * MSI vector, regardless of device function is capable of handling
  * multiple messages. A return of zero indicates the successful setup
  * of an entry zero with the new MSI vector or non-zero for otherwise.
@@ -477,73 +711,39 @@ void pci_scan_msi_device(struct pci_dev *dev)
 static int msi_capability_init(struct pci_dev *dev)
 {
        struct msi_desc *entry;
-       struct msg_address address;
-       struct msg_data data;
        int pos, vector;
-       u32 control;
+       u16 control;
 
        pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
-       if (!pos)
-               return -EINVAL;
-
-       dev->bus->ops->read(dev->bus, dev->devfn, msi_control_reg(pos),
-               2, &control);
-       if (control & PCI_MSI_FLAGS_ENABLE)
-               return 0;
-
-       if (!msi_lookup_vector(dev)) {
-               /* Lookup Sucess */
-               enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
-               return 0;
-       }
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
        /* MSI Entry Initialization */
-       if (!(entry = alloc_msi_entry()))
+       entry = alloc_msi_entry();
+       if (!entry)
                return -ENOMEM;
 
-       if ((vector = get_msi_vector(dev)) < 0) {
+       vector = get_msi_vector(dev);
+       if (vector < 0) {
                kmem_cache_free(msi_cachep, entry);
                return -EBUSY;
        }
+       entry->link.head = vector;
+       entry->link.tail = vector;
        entry->msi_attrib.type = PCI_CAP_ID_MSI;
+       entry->msi_attrib.state = 0;                    /* Mark it not active */
        entry->msi_attrib.entry_nr = 0;
        entry->msi_attrib.maskbit = is_mask_bit_support(control);
-       entry->msi_attrib.default_vector = dev->irq;
-       dev->irq = vector;      /* save default pre-assigned ioapic vector */
+       entry->msi_attrib.default_vector = dev->irq;    /* Save IOAPIC IRQ */
+       dev->irq = vector;
        entry->dev = dev;
        if (is_mask_bit_support(control)) {
-               entry->mask_base = msi_mask_bits_reg(pos,
+               entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
                                is_64bit_address(control));
        }
        /* Replace with MSI handler */
        irq_handler_init(PCI_CAP_ID_MSI, vector, entry->msi_attrib.maskbit);
        /* Configure MSI capability structure */
-       msi_address_init(&address);
-       msi_data_init(&data, vector);
-       entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
-                               MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
-       dev->bus->ops->write(dev->bus, dev->devfn, msi_lower_address_reg(pos),
-                               4, address.lo_address.value);
-       if (is_64bit_address(control)) {
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_upper_address_reg(pos), 4, address.hi_address);
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_data_reg(pos, 1), 2, *((u32*)&data));
-       } else
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_data_reg(pos, 0), 2, *((u32*)&data));
-       if (entry->msi_attrib.maskbit) {
-               unsigned int maskbits, temp;
-               /* All MSIs are unmasked by default, Mask them all */
-               dev->bus->ops->read(dev->bus, dev->devfn,
-                       msi_mask_bits_reg(pos, is_64bit_address(control)), 4,
-                       &maskbits);
-               temp = (1 << multi_msi_capable(control));
-               temp = ((temp - 1) & ~temp);
-               maskbits |= temp;
-               dev->bus->ops->write(dev->bus, dev->devfn,
-                       msi_mask_bits_reg(pos, is_64bit_address(control)), 4,
-                       maskbits);
-       }
+       msi_register_init(dev, entry);
+
        attach_msi_entry(entry, vector);
        /* Set MSI enabled bits  */
        enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
@@ -554,241 +754,230 @@ static int msi_capability_init(struct pci_dev *dev)
 /**
  * msix_capability_init - configure device's MSI-X capability
  * @dev: pointer to the pci_dev data structure of MSI-X device function
+ * @entries: pointer to an array of struct msix_entry entries
+ * @nvec: number of @entries
  *
- * Setup the MSI-X capability structure of device funtion with a
- * single MSI-X vector. A return of zero indicates the successful setup
- * of an entry zero with the new MSI-X vector or non-zero for otherwise.
- * To request for additional MSI-X vectors, the device drivers are
- * required to utilize the following supported APIs:
- * 1) msi_alloc_vectors(...) for requesting one or more MSI-X vectors
- * 2) msi_free_vectors(...) for releasing one or more MSI-X vectors
- *    back to PCI subsystem before calling free_irq(...)
+ * Setup the MSI-X capability structure of device function with a
+ * single MSI-X vector. A return of zero indicates the successful setup of
+ * requested MSI-X entries with allocated vectors or non-zero for otherwise.
  **/
-static int msix_capability_init(struct pci_dev *dev)
+static int msix_capability_init(struct pci_dev *dev,
+                               struct msix_entry *entries, int nvec)
 {
-       struct msi_desc *entry;
+       struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
        struct msg_address address;
        struct msg_data data;
-       int vector = 0, pos, dev_msi_cap;
-       u32 phys_addr, table_offset;
-       u32 control;
+       int vector, pos, i, j, nr_entries, temp = 0;
+       unsigned long phys_addr;
+       u32 table_offset;
+       u16 control;
        u8 bir;
-       void *base;
+       void __iomem *base;
 
        pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
-       if (!pos)
-               return -EINVAL;
-
        /* Request & Map MSI-X table region */
-       dev->bus->ops->read(dev->bus, dev->devfn, msi_control_reg(pos), 2,
-               &control);
-       if (control & PCI_MSIX_FLAGS_ENABLE)
-               return 0;
-
-       if (!msi_lookup_vector(dev)) {
-               /* Lookup Sucess */
-               enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
-               return 0;
-       }
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       nr_entries = multi_msix_capable(control);
 
-       dev_msi_cap = multi_msix_capable(control);
-       dev->bus->ops->read(dev->bus, dev->devfn,
-               msix_table_offset_reg(pos), 4, &table_offset);
+       pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
        bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
-       phys_addr = pci_resource_start (dev, bir);
-       phys_addr += (u32)(table_offset & ~PCI_MSIX_FLAGS_BIRMASK);
-       if (!request_mem_region(phys_addr,
-               dev_msi_cap * PCI_MSIX_ENTRY_SIZE,
-               "MSI-X iomap Failure"))
-               return -ENOMEM;
-       base = ioremap_nocache(phys_addr, dev_msi_cap * PCI_MSIX_ENTRY_SIZE);
+       table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
+       phys_addr = pci_resource_start (dev, bir) + table_offset;
+       base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
        if (base == NULL)
-               goto free_region;
-       /* MSI Entry Initialization */
-       entry = alloc_msi_entry();
-       if (!entry)
-               goto free_iomap;
-       if ((vector = get_msi_vector(dev)) < 0)
-               goto free_entry;
+               return -ENOMEM;
 
-       entry->msi_attrib.type = PCI_CAP_ID_MSIX;
-       entry->msi_attrib.entry_nr = 0;
-       entry->msi_attrib.maskbit = 1;
-       entry->msi_attrib.default_vector = dev->irq;
-       dev->irq = vector;      /* save default pre-assigned ioapic vector */
-       entry->dev = dev;
-       entry->mask_base = (unsigned long)base;
-       /* Replace with MSI handler */
-       irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
-       /* Configure MSI-X capability structure */
-       msi_address_init(&address);
-       msi_data_init(&data, vector);
-       entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
-                               MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
-       writel(address.lo_address.value, base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
-       writel(address.hi_address, base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
-       writel(*(u32*)&data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
-       /* Initialize all entries from 1 up to 0 */
-       for (pos = 1; pos < dev_msi_cap; pos++) {
-               writel(0, base + pos * PCI_MSIX_ENTRY_SIZE +
+       /* MSI-X Table Initialization */
+       for (i = 0; i < nvec; i++) {
+               entry = alloc_msi_entry();
+               if (!entry)
+                       break;
+               vector = get_msi_vector(dev);
+               if (vector < 0) {
+                       kmem_cache_free(msi_cachep, entry);
+                       break;
+               }
+
+               j = entries[i].entry;
+               entries[i].vector = vector;
+               entry->msi_attrib.type = PCI_CAP_ID_MSIX;
+               entry->msi_attrib.state = 0;            /* Mark it not active */
+               entry->msi_attrib.entry_nr = j;
+               entry->msi_attrib.maskbit = 1;
+               entry->msi_attrib.default_vector = dev->irq;
+               entry->dev = dev;
+               entry->mask_base = base;
+               if (!head) {
+                       entry->link.head = vector;
+                       entry->link.tail = vector;
+                       head = entry;
+               } else {
+                       entry->link.head = temp;
+                       entry->link.tail = tail->link.tail;
+                       tail->link.tail = vector;
+                       head->link.head = vector;
+               }
+               temp = vector;
+               tail = entry;
+               /* Replace with MSI-X handler */
+               irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
+               /* Configure MSI-X capability structure */
+               msi_address_init(&address);
+               msi_data_init(&data, vector);
+               entry->msi_attrib.current_cpu =
+                       ((address.lo_address.u.dest_id >>
+                       MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
+               writel(address.lo_address.value,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
                        PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
-               writel(0, base + pos * PCI_MSIX_ENTRY_SIZE +
+               writel(address.hi_address,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
                        PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
-               writel(0, base + pos * PCI_MSIX_ENTRY_SIZE +
+               writel(*(u32*)&data,
+                       base + j * PCI_MSIX_ENTRY_SIZE +
                        PCI_MSIX_ENTRY_DATA_OFFSET);
+               attach_msi_entry(entry, vector);
        }
-       attach_msi_entry(entry, vector);
-       /* Set MSI enabled bits  */
+       if (i != nvec) {
+               i--;
+               for (; i >= 0; i--) {
+                       vector = (entries + i)->vector;
+                       msi_free_vector(dev, vector, 0);
+                       (entries + i)->vector = 0;
+               }
+               return -EBUSY;
+       }
+       /* Set MSI-X enabled bits */
        enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
 
        return 0;
-
-free_entry:
-       kmem_cache_free(msi_cachep, entry);
-free_iomap:
-       iounmap(base);
-free_region:
-       release_mem_region(phys_addr, dev_msi_cap * PCI_MSIX_ENTRY_SIZE);
-
-       return ((vector < 0) ? -EBUSY : -ENOMEM);
 }
 
 /**
- * pci_enable_msi - configure device's MSI(X) capability structure
- * @dev: pointer to the pci_dev data structure of MSI(X) device function
+ * pci_enable_msi - configure device's MSI capability structure
+ * @dev: pointer to the pci_dev data structure of MSI device function
  *
- * Setup the MSI/MSI-X capability structure of device function with
- * a single MSI(X) vector upon its software driver call to request for
- * MSI(X) mode enabled on its hardware device function. A return of zero
- * indicates the successful setup of an entry zero with the new MSI(X)
+ * Setup the MSI capability structure of device function with
+ * a single MSI vector upon its software driver call to request for
+ * MSI mode enabled on its hardware device function. A return of zero
+ * indicates the successful setup of an entry zero with the new MSI
  * vector or non-zero for otherwise.
  **/
 int pci_enable_msi(struct pci_dev* dev)
 {
-       int status = -EINVAL;
+       int pos, temp, status = -EINVAL;
+       u16 control;
 
        if (!pci_msi_enable || !dev)
                return status;
 
-       if (msi_init() < 0)
-               return -ENOMEM;
+       if (dev->no_msi)
+               return status;
 
-       if ((status = msix_capability_init(dev)) == -EINVAL)
-               status = msi_capability_init(dev);
-       if (!status)
-               nr_reserved_vectors--;
+       if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
+               return -EINVAL;
 
-       return status;
-}
+       temp = dev->irq;
 
-static int msi_free_vector(struct pci_dev* dev, int vector);
-static void pci_disable_msi(unsigned int vector)
-{
-       int head, tail, type, default_vector;
-       struct msi_desc *entry;
-       struct pci_dev *dev;
-       unsigned long flags;
+       status = msi_init();
+       if (status < 0)
+               return status;
 
-       spin_lock_irqsave(&msi_lock, flags);
-       entry = msi_desc[vector];
-       if (!entry || !entry->dev) {
-               spin_unlock_irqrestore(&msi_lock, flags);
-               return;
-       }
-       dev = entry->dev;
-       type = entry->msi_attrib.type;
-       head = entry->link.head;
-       tail = entry->link.tail;
-       default_vector = entry->msi_attrib.default_vector;
-       spin_unlock_irqrestore(&msi_lock, flags);
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       if (!pos)
+               return -EINVAL;
 
-       disable_msi_mode(dev, pci_find_capability(dev, type), type);
-       /* Restore dev->irq to its default pin-assertion vector */
-       dev->irq = default_vector;
-       if (type == PCI_CAP_ID_MSIX && head != tail) {
-               /* Bad driver, which do not call msi_free_vectors before exit.
-                  We must do a cleanup here */
-               while (1) {
-                       spin_lock_irqsave(&msi_lock, flags);
-                       entry = msi_desc[vector];
-                       head = entry->link.head;
-                       tail = entry->link.tail;
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (control & PCI_MSI_FLAGS_ENABLE)
+               return 0;                       /* Already in MSI mode */
+
+       if (!msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
+               /* Lookup Sucess */
+               unsigned long flags;
+
+               spin_lock_irqsave(&msi_lock, flags);
+               if (!vector_irq[dev->irq]) {
+                       msi_desc[dev->irq]->msi_attrib.state = 0;
+                       vector_irq[dev->irq] = -1;
+                       nr_released_vectors--;
                        spin_unlock_irqrestore(&msi_lock, flags);
-                       if (tail == head)
-                               break;
-                       if (msi_free_vector(dev, entry->link.tail))
-                               break;
+                       msi_register_init(dev, msi_desc[dev->irq]);
+                       enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
+                       return 0;
                }
+               spin_unlock_irqrestore(&msi_lock, flags);
+               dev->irq = temp;
        }
+       /* Check whether driver already requested for MSI-X vectors */
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
+                       printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
+                              "Device already has MSI-X vectors assigned\n",
+                              pci_name(dev));
+                       dev->irq = temp;
+                       return -EINVAL;
+       }
+       status = msi_capability_init(dev);
+       if (!status) {
+               if (!pos)
+                       nr_reserved_vectors--;  /* Only MSI capable */
+               else if (nr_msix_devices > 0)
+                       nr_msix_devices--;      /* Both MSI and MSI-X capable,
+                                                  but choose enabling MSI */
+       }
+
+       return status;
 }
 
-static int msi_alloc_vector(struct pci_dev* dev, int head)
+void pci_disable_msi(struct pci_dev* dev)
 {
        struct msi_desc *entry;
-       struct msg_address address;
-       struct msg_data data;
-       int i, offset, pos, dev_msi_cap, vector;
-       u32 low_address, control;
-       unsigned long base = 0L;
+       int pos, default_vector;
+       u16 control;
        unsigned long flags;
 
+       if (!pci_msi_enable)
+               return;
+       if (!dev)
+               return;
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       if (!pos)
+               return;
+
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (!(control & PCI_MSI_FLAGS_ENABLE))
+               return;
+
        spin_lock_irqsave(&msi_lock, flags);
        entry = msi_desc[dev->irq];
-       if (!entry) {
+       if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
                spin_unlock_irqrestore(&msi_lock, flags);
-               return -EINVAL;
+               return;
        }
-       base = entry->mask_base;
-       spin_unlock_irqrestore(&msi_lock, flags);
-
-       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
-       dev->bus->ops->read(dev->bus, dev->devfn, msi_control_reg(pos),
-               2, &control);
-       dev_msi_cap = multi_msix_capable(control);
-       for (i = 1; i < dev_msi_cap; i++) {
-               if (!(low_address = readl(base + i * PCI_MSIX_ENTRY_SIZE)))
-                        break;
-       }
-       if (i >= dev_msi_cap)
-               return -EINVAL;
-
-       /* MSI Entry Initialization */
-       if (!(entry = alloc_msi_entry()))
-               return -ENOMEM;
-
-       if ((vector = get_new_vector()) < 0) {
-               kmem_cache_free(msi_cachep, entry);
-               return vector;
+       if (entry->msi_attrib.state) {
+               spin_unlock_irqrestore(&msi_lock, flags);
+               printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
+                      "free_irq() on MSI vector %d\n",
+                      pci_name(dev), dev->irq);
+               BUG_ON(entry->msi_attrib.state > 0);
+       } else {
+               vector_irq[dev->irq] = 0; /* free it */
+               nr_released_vectors++;
+               default_vector = entry->msi_attrib.default_vector;
+               spin_unlock_irqrestore(&msi_lock, flags);
+               /* Restore dev->irq to its default pin-assertion vector */
+               dev->irq = default_vector;
+               disable_msi_mode(dev, pci_find_capability(dev, PCI_CAP_ID_MSI),
+                                       PCI_CAP_ID_MSI);
        }
-       entry->msi_attrib.type = PCI_CAP_ID_MSIX;
-       entry->msi_attrib.entry_nr = i;
-       entry->msi_attrib.maskbit = 1;
-       entry->dev = dev;
-       entry->link.head = head;
-       entry->mask_base = base;
-       irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
-       /* Configure MSI-X capability structure */
-       msi_address_init(&address);
-       msi_data_init(&data, vector);
-       entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
-                               MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
-       offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
-       writel(address.lo_address.value, base + offset +
-               PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
-       writel(address.hi_address, base + offset +
-               PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
-       writel(*(u32*)&data, base + offset + PCI_MSIX_ENTRY_DATA_OFFSET);
-       writel(1, base + offset + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
-       attach_msi_entry(entry, vector);
-
-       return vector;
 }
 
-static int msi_free_vector(struct pci_dev* dev, int vector)
+static int msi_free_vector(struct pci_dev* dev, int vector, int reassign)
 {
        struct msi_desc *entry;
-       int entry_nr, type;
-       unsigned long base = 0L;
+       int head, entry_nr, type;
+       void __iomem *base;
        unsigned long flags;
 
        spin_lock_irqsave(&msi_lock, flags);
@@ -799,66 +988,184 @@ static int msi_free_vector(struct pci_dev* dev, int vector)
        }
        type = entry->msi_attrib.type;
        entry_nr = entry->msi_attrib.entry_nr;
+       head = entry->link.head;
        base = entry->mask_base;
-       if (entry->link.tail != entry->link.head) {
-               msi_desc[entry->link.head]->link.tail = entry->link.tail;
-               if (entry->link.tail)
-                       msi_desc[entry->link.tail]->link.head = entry->link.head;
-       }
+       msi_desc[entry->link.head]->link.tail = entry->link.tail;
+       msi_desc[entry->link.tail]->link.head = entry->link.head;
        entry->dev = NULL;
-       vector_irq[vector] = 0;
-       nr_released_vectors++;
+       if (!reassign) {
+               vector_irq[vector] = 0;
+               nr_released_vectors++;
+       }
        msi_desc[vector] = NULL;
        spin_unlock_irqrestore(&msi_lock, flags);
 
        kmem_cache_free(msi_cachep, entry);
+
        if (type == PCI_CAP_ID_MSIX) {
-               int offset;
+               if (!reassign)
+                       writel(1, base +
+                               entry_nr * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
+
+               if (head == vector) {
+                       /*
+                        * Detect last MSI-X vector to be released.
+                        * Release the MSI-X memory-mapped table.
+                        */
+#if 0
+                       int pos, nr_entries;
+                       unsigned long phys_addr;
+                       u32 table_offset;
+                       u16 control;
+                       u8 bir;
+
+                       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+                       pci_read_config_word(dev, msi_control_reg(pos),
+                               &control);
+                       nr_entries = multi_msix_capable(control);
+                       pci_read_config_dword(dev, msix_table_offset_reg(pos),
+                               &table_offset);
+                       bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
+                       table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
+                       phys_addr = pci_resource_start(dev, bir) + table_offset;
+/*
+ * FIXME!  and what did you want to do with phys_addr?
+ */
+#endif
+                       iounmap(base);
+               }
+       }
+
+       return 0;
+}
+
+static int reroute_msix_table(int head, struct msix_entry *entries, int *nvec)
+{
+       int vector = head, tail = 0;
+       int i, j = 0, nr_entries = 0;
+       void __iomem *base;
+       unsigned long flags;
 
-               offset = entry_nr * PCI_MSIX_ENTRY_SIZE;
-               writel(1, base + offset + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
-               writel(0, base + offset + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
+       spin_lock_irqsave(&msi_lock, flags);
+       while (head != tail) {
+               nr_entries++;
+               tail = msi_desc[vector]->link.tail;
+               if (entries[0].entry == msi_desc[vector]->msi_attrib.entry_nr)
+                       j = vector;
+               vector = tail;
+       }
+       if (*nvec > nr_entries) {
+               spin_unlock_irqrestore(&msi_lock, flags);
+               *nvec = nr_entries;
+               return -EINVAL;
+       }
+       vector = ((j > 0) ? j : head);
+       for (i = 0; i < *nvec; i++) {
+               j = msi_desc[vector]->msi_attrib.entry_nr;
+               msi_desc[vector]->msi_attrib.state = 0; /* Mark it not active */
+               vector_irq[vector] = -1;                /* Mark it busy */
+               nr_released_vectors--;
+               entries[i].vector = vector;
+               if (j != (entries + i)->entry) {
+                       base = msi_desc[vector]->mask_base;
+                       msi_desc[vector]->msi_attrib.entry_nr =
+                               (entries + i)->entry;
+                       writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET), base +
+                               (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
+                       writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET), base +
+                               (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
+                       writel( (readl(base + j * PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_DATA_OFFSET) & 0xff00) | vector,
+                               base + (entries+i)->entry*PCI_MSIX_ENTRY_SIZE +
+                               PCI_MSIX_ENTRY_DATA_OFFSET);
+               }
+               vector = msi_desc[vector]->link.tail;
        }
+       spin_unlock_irqrestore(&msi_lock, flags);
 
        return 0;
 }
 
 /**
- * msi_alloc_vectors - allocate additional MSI-X vectors
+ * pci_enable_msix - configure device's MSI-X capability structure
  * @dev: pointer to the pci_dev data structure of MSI-X device function
- * @vector: pointer to an array of new allocated MSI-X vectors
+ * @entries: pointer to an array of MSI-X entries
  * @nvec: number of MSI-X vectors requested for allocation by device driver
  *
- * Allocate additional MSI-X vectors requested by device driver. A
- * return of zero indicates the successful setup of MSI-X capability
- * structure with new allocated MSI-X vectors or non-zero for otherwise.
+ * Setup the MSI-X capability structure of device function with the number
+ * of requested vectors upon its software driver call to request for
+ * MSI-X mode enabled on its hardware device function. A return of zero
+ * indicates the successful configuration of MSI-X capability structure
+ * with new allocated MSI-X vectors. A return of < 0 indicates a failure.
+ * Or a return of > 0 indicates that driver request is exceeding the number
+ * of vectors available. Driver should use the returned value to re-send
+ * its request.
  **/
-int msi_alloc_vectors(struct pci_dev* dev, int *vector, int nvec)
+int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
 {
-       struct msi_desc *entry;
-       int i, head, pos, vec, free_vectors, alloc_vectors;
-       int *vectors = (int *)vector;
-       u32 control;
+       int status, pos, nr_entries, free_vectors;
+       int i, j, temp;
+       u16 control;
        unsigned long flags;
 
-       if (!pci_msi_enable || !dev)
+       if (!pci_msi_enable || !dev || !entries)
                return -EINVAL;
 
-       if (!(pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)))
+       status = msi_init();
+       if (status < 0)
+               return status;
+
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (!pos)
                return -EINVAL;
 
-       dev->bus->ops->read(dev->bus, dev->devfn, msi_control_reg(pos),                         2, &control);
-       if (nvec > multi_msix_capable(control))
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (control & PCI_MSIX_FLAGS_ENABLE)
+               return -EINVAL;                 /* Already in MSI-X mode */
+
+       nr_entries = multi_msix_capable(control);
+       if (nvec > nr_entries)
                return -EINVAL;
 
-       spin_lock_irqsave(&msi_lock, flags);
-       entry = msi_desc[dev->irq];
-       if (!entry || entry->dev != dev ||              /* legal call */
-          entry->msi_attrib.type != PCI_CAP_ID_MSIX || /* must be MSI-X */
-          entry->link.head != entry->link.tail) {      /* already multi */
-               spin_unlock_irqrestore(&msi_lock, flags);
+       /* Check for any invalid entries */
+       for (i = 0; i < nvec; i++) {
+               if (entries[i].entry >= nr_entries)
+                       return -EINVAL;         /* invalid entry */
+               for (j = i + 1; j < nvec; j++) {
+                       if (entries[i].entry == entries[j].entry)
+                               return -EINVAL; /* duplicate entry */
+               }
+       }
+       temp = dev->irq;
+       if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
+               /* Lookup Sucess */
+               nr_entries = nvec;
+               /* Reroute MSI-X table */
+               if (reroute_msix_table(dev->irq, entries, &nr_entries)) {
+                       /* #requested > #previous-assigned */
+                       dev->irq = temp;
+                       return nr_entries;
+               }
+               dev->irq = temp;
+               enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
+               return 0;
+       }
+       /* Check whether driver already requested for MSI vector */
+       if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
+               !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
+               printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
+                      "Device already has an MSI vector assigned\n",
+                      pci_name(dev));
+               dev->irq = temp;
                return -EINVAL;
        }
+
+       spin_lock_irqsave(&msi_lock, flags);
        /*
         * msi_lock is provided to ensure that enough vectors resources are
         * available before granting.
@@ -874,140 +1181,159 @@ int msi_alloc_vectors(struct pci_dev* dev, int *vector, int nvec)
                free_vectors /= nr_msix_devices;
        spin_unlock_irqrestore(&msi_lock, flags);
 
-       if (nvec > free_vectors)
-               return -EBUSY;
-
-       alloc_vectors = 0;
-       head = dev->irq;
-       for (i = 0; i < nvec; i++) {
-               if ((vec = msi_alloc_vector(dev, head)) < 0)
-                       break;
-               *(vectors + i) = vec;
-               head = vec;
-               alloc_vectors++;
-       }
-       if (alloc_vectors != nvec) {
-               for (i = 0; i < alloc_vectors; i++) {
-                       vec = *(vectors + i);
-                       msi_free_vector(dev, vec);
-               }
-               spin_lock_irqsave(&msi_lock, flags);
-               msi_desc[dev->irq]->link.tail = msi_desc[dev->irq]->link.head;
-               spin_unlock_irqrestore(&msi_lock, flags);
-               return -EBUSY;
+       if (nvec > free_vectors) {
+               if (free_vectors > 0)
+                       return free_vectors;
+               else
+                       return -EBUSY;
        }
-       if (nr_msix_devices > 0)
+
+       status = msix_capability_init(dev, entries, nvec);
+       if (!status && nr_msix_devices > 0)
                nr_msix_devices--;
 
-       return 0;
+       return status;
 }
 
-/**
- * msi_free_vectors - reclaim MSI-X vectors to unused state
- * @dev: pointer to the pci_dev data structure of MSI-X device function
- * @vector: pointer to an array of released MSI-X vectors
- * @nvec: number of MSI-X vectors requested for release by device driver
- *
- * Reclaim MSI-X vectors released by device driver to unused state,
- * which may be used later on. A return of zero indicates the
- * success or non-zero for otherwise. Device driver should call this
- * before calling function free_irq.
- **/
-int msi_free_vectors(struct pci_dev* dev, int *vector, int nvec)
+void pci_disable_msix(struct pci_dev* dev)
 {
-       struct msi_desc *entry;
-       int i;
-       unsigned long flags;
+       int pos, temp;
+       u16 control;
 
        if (!pci_msi_enable)
-               return -EINVAL;
+               return;
+       if (!dev)
+               return;
 
-       spin_lock_irqsave(&msi_lock, flags);
-       entry = msi_desc[dev->irq];
-       if (!entry || entry->dev != dev ||
-               entry->msi_attrib.type != PCI_CAP_ID_MSIX ||
-               entry->link.head == entry->link.tail) { /* Nothing to free */
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (!pos)
+               return;
+
+       pci_read_config_word(dev, msi_control_reg(pos), &control);
+       if (!(control & PCI_MSIX_FLAGS_ENABLE))
+               return;
+
+       temp = dev->irq;
+       if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
+               int state, vector, head, tail = 0, warning = 0;
+               unsigned long flags;
+
+               vector = head = dev->irq;
+               spin_lock_irqsave(&msi_lock, flags);
+               while (head != tail) {
+                       state = msi_desc[vector]->msi_attrib.state;
+                       if (state)
+                               warning = 1;
+                       else {
+                               vector_irq[vector] = 0; /* free it */
+                               nr_released_vectors++;
+                       }
+                       tail = msi_desc[vector]->link.tail;
+                       vector = tail;
+               }
                spin_unlock_irqrestore(&msi_lock, flags);
-               return -EINVAL;
-       }
-       spin_unlock_irqrestore(&msi_lock, flags);
+               if (warning) {
+                       dev->irq = temp;
+                       printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
+                              "free_irq() on all MSI-X vectors\n",
+                              pci_name(dev));
+                       BUG_ON(warning > 0);
+               } else {
+                       dev->irq = temp;
+                       disable_msi_mode(dev,
+                               pci_find_capability(dev, PCI_CAP_ID_MSIX),
+                               PCI_CAP_ID_MSIX);
 
-       for (i = 0; i < nvec; i++) {
-               if (*(vector + i) == dev->irq)
-                       continue;/* Don't free entry 0 if mistaken by driver */
-               msi_free_vector(dev, *(vector + i));
+               }
        }
-
-       return 0;
 }
 
 /**
  * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state
  * @dev: pointer to the pci_dev data structure of MSI(X) device function
  *
- * Being called during hotplug remove, from which the device funciton
+ * Being called during hotplug remove, from which the device function
  * is hot-removed. All previous assigned MSI/MSI-X vectors, if
  * allocated for this device function, are reclaimed to unused state,
  * which may be used later on.
  **/
 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
 {
-       struct msi_desc *entry;
-       int type, temp;
+       int state, pos, temp;
        unsigned long flags;
 
        if (!pci_msi_enable || !dev)
                return;
 
-       if (!pci_find_capability(dev, PCI_CAP_ID_MSI)) {
-               if (!pci_find_capability(dev, PCI_CAP_ID_MSIX))
-                       return;
-       }
-       temp = dev->irq;
-       if (msi_lookup_vector(dev))
-               return;
-
-       spin_lock_irqsave(&msi_lock, flags);
-       entry = msi_desc[dev->irq];
-       if (!entry || entry->dev != dev) {
+       temp = dev->irq;                /* Save IOAPIC IRQ */
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
+       if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
+               spin_lock_irqsave(&msi_lock, flags);
+               state = msi_desc[dev->irq]->msi_attrib.state;
                spin_unlock_irqrestore(&msi_lock, flags);
-               return;
+               if (state) {
+                       printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
+                              "called without free_irq() on MSI vector %d\n",
+                              pci_name(dev), dev->irq);
+                       BUG_ON(state > 0);
+               } else /* Release MSI vector assigned to this device */
+                       msi_free_vector(dev, dev->irq, 0);
+               dev->irq = temp;                /* Restore IOAPIC IRQ */
        }
-       type = entry->msi_attrib.type;
-       spin_unlock_irqrestore(&msi_lock, flags);
+       pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
+       if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
+               int vector, head, tail = 0, warning = 0;
+               void __iomem *base = NULL;
 
-       msi_free_vector(dev, dev->irq);
-       if (type == PCI_CAP_ID_MSIX) {
-               int i, pos, dev_msi_cap;
-               u32 phys_addr, table_offset;
-               u32 control;
-               u8 bir;
-
-               pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
-               dev->bus->ops->read(dev->bus, dev->devfn, msi_control_reg(pos),                         2, &control);
-               dev_msi_cap = multi_msix_capable(control);
-               dev->bus->ops->read(dev->bus, dev->devfn,
-                       msix_table_offset_reg(pos), 4, &table_offset);
-               bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
-               phys_addr = pci_resource_start (dev, bir);
-               phys_addr += (u32)(table_offset & ~PCI_MSIX_FLAGS_BIRMASK);
-               for (i = FIRST_DEVICE_VECTOR; i < NR_IRQS; i++) {
+               vector = head = dev->irq;
+               while (head != tail) {
                        spin_lock_irqsave(&msi_lock, flags);
-                       if (!msi_desc[i] || msi_desc[i]->dev != dev) {
-                               spin_unlock_irqrestore(&msi_lock, flags);
-                               continue;
-                       }
+                       state = msi_desc[vector]->msi_attrib.state;
+                       tail = msi_desc[vector]->link.tail;
+                       base = msi_desc[vector]->mask_base;
                        spin_unlock_irqrestore(&msi_lock, flags);
-                       msi_free_vector(dev, i);
+                       if (state)
+                               warning = 1;
+                       else if (vector != head) /* Release MSI-X vector */
+                               msi_free_vector(dev, vector, 0);
+                       vector = tail;
+               }
+               msi_free_vector(dev, vector, 0);
+               if (warning) {
+                       /* Force to release the MSI-X memory-mapped table */
+#if 0
+                       unsigned long phys_addr;
+                       u32 table_offset;
+                       u16 control;
+                       u8 bir;
+
+                       pci_read_config_word(dev, msi_control_reg(pos),
+                               &control);
+                       pci_read_config_dword(dev, msix_table_offset_reg(pos),
+                               &table_offset);
+                       bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
+                       table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
+                       phys_addr = pci_resource_start(dev, bir) + table_offset;
+/*
+ * FIXME! and what did you want to do with phys_addr?
+ */
+#endif
+                       iounmap(base);
+                       printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
+                              "called without free_irq() on all MSI-X vectors\n",
+                              pci_name(dev));
+                       BUG_ON(warning > 0);
                }
-               writel(1, entry->mask_base + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
-               iounmap((void*)entry->mask_base);
-               release_mem_region(phys_addr, dev_msi_cap * PCI_MSIX_ENTRY_SIZE);
+               dev->irq = temp;                /* Restore IOAPIC IRQ */
        }
-       dev->irq = temp;
-       nr_reserved_vectors++;
+}
+
+void pci_no_msi(void)
+{
+       pci_msi_enable = 0;
 }
 
 EXPORT_SYMBOL(pci_enable_msi);
-EXPORT_SYMBOL(msi_alloc_vectors);
-EXPORT_SYMBOL(msi_free_vectors);
+EXPORT_SYMBOL(pci_disable_msi);
+EXPORT_SYMBOL(pci_enable_msix);
+EXPORT_SYMBOL(pci_disable_msix);