bitkeeper revision 1.820 (40600453bCkH6oPCubNeqIe3OBUXGQ)

io.h:
  new file
Many files:
  Further IO virtualisation patches.

6 files changed, 114 insertions, 39 deletions

diff --git a/xenolinux-2.4.25-sparse/arch/xeno/config.in b/xenolinux-2.4.25-sparse/arch/xeno/config.in
index 39e821556a..209bbe6d51 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/config.in
+++ b/xenolinux-2.4.25-sparse/arch/xeno/config.in
@@ -107,6 +107,10 @@ bool 'Networking support' CONFIG_NET
 bool 'PCI support' CONFIG_PCI
 if [ "$CONFIG_PCI" = "y" ]; then
    tristate '    3c590/3c900 series (592/595/597) "Vortex/Boomerang" support' CONFIG_VORTEX
+   tristate 'Intel(R) PRO/1000 Gigabit Ethernet support' CONFIG_E1000
+   if [ "$CONFIG_E1000" != "n" ]; then
+      bool '  Use Rx Polling (NAPI)' CONFIG_E1000_NAPI
+   fi
 fi
 source drivers/pci/Config.in
 
diff --git a/xenolinux-2.4.25-sparse/arch/xeno/drivers/network/network.c b/xenolinux-2.4.25-sparse/arch/xeno/drivers/network/network.c
index 0a1bce2bfa..c5d25442e2 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/drivers/network/network.c
+++ b/xenolinux-2.4.25-sparse/arch/xeno/drivers/network/network.c
@@ -58,8 +58,8 @@ struct net_private
      * {tx,rx}_skbs store outstanding skbuffs. The first entry in each
      * array is an index into a chain of free entries.
      */
-    struct sk_buff *tx_skbs[TX_RING_SIZE+1];
-    struct sk_buff *rx_skbs[RX_RING_SIZE+1];
+    struct sk_buff *tx_skbs[XENNET_TX_RING_SIZE+1];
+    struct sk_buff *rx_skbs[XENNET_RX_RING_SIZE+1];
 };
 
 /* Access macros for acquiring freeing slots in {tx,rx}_skbs[]. */
@@ -143,9 +143,9 @@ static int network_open(struct net_device *dev)
     memset(np->net_idx, 0, sizeof(*np->net_idx));
 
     /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
-    for ( i = 0; i <= TX_RING_SIZE; i++ )
+    for ( i = 0; i <= XENNET_TX_RING_SIZE; i++ )
         np->tx_skbs[i] = (void *)(i+1);
-    for ( i = 0; i <= RX_RING_SIZE; i++ )
+    for ( i = 0; i <= XENNET_RX_RING_SIZE; i++ )
         np->rx_skbs[i] = (void *)(i+1);
 
     wmb();
@@ -196,7 +196,8 @@ static void network_tx_buf_gc(struct net_device *dev)
     }
     while ( prod != np->net_idx->tx_resp_prod );
 
-    if ( np->tx_full && ((np->net_idx->tx_req_prod - prod) < TX_RING_SIZE) )
+    if ( np->tx_full && 
+         ((np->net_idx->tx_req_prod - prod) < XENNET_TX_RING_SIZE) )
     {
         np->tx_full = 0;
         if ( np->state == STATE_ACTIVE )
@@ -223,7 +224,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
     netop_t netop;
     NET_RING_IDX i = np->net_idx->rx_req_prod;
 
-    if ( unlikely((i - np->rx_resp_cons) == RX_RING_SIZE) || 
+    if ( unlikely((i - np->rx_resp_cons) == XENNET_RX_RING_SIZE) || 
          unlikely(np->state != STATE_ACTIVE) )
         return;
 
@@ -246,7 +247,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
 
         np->rx_bufs_to_notify++;
     }
-    while ( (++i - np->rx_resp_cons) != RX_RING_SIZE );
+    while ( (++i - np->rx_resp_cons) != XENNET_RX_RING_SIZE );
 
     /*
      * We may have allocated buffers which have entries outstanding in the page
@@ -258,7 +259,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
     np->net_idx->rx_event    = np->rx_resp_cons + 1;
         
     /* Batch Xen notifications. */
-    if ( np->rx_bufs_to_notify > (RX_RING_SIZE/4) )
+    if ( np->rx_bufs_to_notify > (XENNET_RX_RING_SIZE/4) )
     {
         netop.cmd = NETOP_PUSH_BUFFERS;
         netop.vif = np->idx;
@@ -313,7 +314,7 @@ static int network_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
     network_tx_buf_gc(dev);
 
-    if ( (i - np->tx_resp_cons) == (TX_RING_SIZE - 1) )
+    if ( (i - np->tx_resp_cons) == (XENNET_TX_RING_SIZE - 1) )
     {
         np->tx_full = 1;
         netif_stop_queue(dev);
diff --git a/xenolinux-2.4.25-sparse/arch/xeno/drivers/vnetif/vnetif.c b/xenolinux-2.4.25-sparse/arch/xeno/drivers/vnetif/vnetif.c
index 465dd18233..91f3c5c17e 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/drivers/vnetif/vnetif.c
+++ b/xenolinux-2.4.25-sparse/arch/xeno/drivers/vnetif/vnetif.c
@@ -58,8 +58,8 @@ struct net_private
      * {tx,rx}_skbs store outstanding skbuffs. The first entry in each
      * array is an index into a chain of free entries.
      */
-    struct sk_buff *tx_skbs[TX_RING_SIZE+1];
-    struct sk_buff *rx_skbs[RX_RING_SIZE+1];
+    struct sk_buff *tx_skbs[XENNET_TX_RING_SIZE+1];
+    struct sk_buff *rx_skbs[XENNET_RX_RING_SIZE+1];
 };
 
 /* Access macros for acquiring freeing slots in {tx,rx}_skbs[]. */
@@ -143,9 +143,9 @@ static int network_open(struct net_device *dev)
     memset(np->net_idx, 0, sizeof(*np->net_idx));
 
     /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
-    for ( i = 0; i <= TX_RING_SIZE; i++ )
+    for ( i = 0; i <= XENNET_TX_RING_SIZE; i++ )
         np->tx_skbs[i] = (void *)(i+1);
-    for ( i = 0; i <= RX_RING_SIZE; i++ )
+    for ( i = 0; i <= XENNET_RX_RING_SIZE; i++ )
         np->rx_skbs[i] = (void *)(i+1);
 
     wmb();
@@ -196,7 +196,8 @@ static void network_tx_buf_gc(struct net_device *dev)
     }
     while ( prod != np->net_idx->tx_resp_prod );
 
-    if ( np->tx_full && ((np->net_idx->tx_req_prod - prod) < TX_RING_SIZE) )
+    if ( np->tx_full && 
+         ((np->net_idx->tx_req_prod - prod) < XENNET_TX_RING_SIZE) )
     {
         np->tx_full = 0;
         if ( np->state == STATE_ACTIVE )
@@ -223,7 +224,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
     netop_t netop;
     NET_RING_IDX i = np->net_idx->rx_req_prod;
 
-    if ( unlikely((i - np->rx_resp_cons) == RX_RING_SIZE) || 
+    if ( unlikely((i - np->rx_resp_cons) == XENNET_RX_RING_SIZE) || 
          unlikely(np->state != STATE_ACTIVE) )
         return;
 
@@ -246,7 +247,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
 
         np->rx_bufs_to_notify++;
     }
-    while ( (++i - np->rx_resp_cons) != RX_RING_SIZE );
+    while ( (++i - np->rx_resp_cons) != XENNET_RX_RING_SIZE );
 
     /*
      * We may have allocated buffers which have entries outstanding in the page
@@ -258,7 +259,7 @@ static void network_alloc_rx_buffers(struct net_device *dev)
     np->net_idx->rx_event    = np->rx_resp_cons + 1;
         
     /* Batch Xen notifications. */
-    if ( np->rx_bufs_to_notify > (RX_RING_SIZE/4) )
+    if ( np->rx_bufs_to_notify > (XENNET_RX_RING_SIZE/4) )
     {
         netop.cmd = NETOP_PUSH_BUFFERS;
         netop.vif = np->idx;
@@ -313,7 +314,7 @@ static int network_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
     network_tx_buf_gc(dev);
 
-    if ( (i - np->tx_resp_cons) == (TX_RING_SIZE - 1) )
+    if ( (i - np->tx_resp_cons) == (XENNET_TX_RING_SIZE - 1) )
     {
         np->tx_full = 1;
         netif_stop_queue(dev);
diff --git a/xenolinux-2.4.25-sparse/arch/xeno/kernel/hypervisor.c b/xenolinux-2.4.25-sparse/arch/xeno/kernel/hypervisor.c
index 64a3590362..7c6aca05c5 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/kernel/hypervisor.c
+++ b/xenolinux-2.4.25-sparse/arch/xeno/kernel/hypervisor.c
@@ -7,8 +7,9 @@
  */
 
 #include <linux/config.h>
-#include <asm/atomic.h>
 #include <linux/irq.h>
+#include <linux/kernel_stat.h>
+#include <asm/atomic.h>
 #include <asm/hypervisor.h>
 #include <asm/system.h>
 #include <asm/ptrace.h>
@@ -18,6 +19,40 @@ int nr_multicall_ents = 0;
 
 static unsigned long event_mask = 0;
 
+asmlinkage unsigned int do_physirq(int irq, struct pt_regs *regs)
+{
+    int cpu = smp_processor_id();
+    unsigned long irqs;
+    shared_info_t *shared = HYPERVISOR_shared_info;
+
+    /* do this manually */
+    kstat.irqs[cpu][irq]++;
+    ack_hypervisor_event(irq);
+
+    barrier();
+    irqs  = xchg(&shared->physirq_pend, 0);
+
+    __asm__ __volatile__ (
+        "   push %1                            ;"
+        "   sub  $4,%%esp                      ;"
+        "   jmp  3f                            ;"
+        "1: btrl %%eax,%0                      ;" /* clear bit     */
+        "   mov  %%eax,(%%esp)                 ;"
+        "   call do_IRQ                        ;" /* do_IRQ(event) */
+        "3: bsfl %0,%%eax                      ;" /* %eax == bit # */
+        "   jnz  1b                            ;"
+        "   add  $8,%%esp                      ;"
+        /* we use %ebx because it is callee-saved */
+        : : "b" (irqs), "r" (regs)
+        /* clobbered by callback function calls */
+        : "eax", "ecx", "edx", "memory" ); 
+
+    /* do this manually */
+    end_hypervisor_event(irq);
+
+    return 0;
+}
+
 void do_hypervisor_callback(struct pt_regs *regs)
 {
     unsigned long events, flags;
@@ -32,6 +67,12 @@ void do_hypervisor_callback(struct pt_regs *regs)
         events  = xchg(&shared->events, 0);
         events &= event_mask;
 
+        if ( (events & EVENT_PHYSIRQ) != 0 )
+        {
+            do_physirq(_EVENT_PHYSIRQ, regs);
+            events &= ~EVENT_PHYSIRQ;
+        }
+
         __asm__ __volatile__ (
             "   push %1                            ;"
             "   sub  $4,%%esp                      ;"
diff --git a/xenolinux-2.4.25-sparse/arch/xeno/kernel/i386_ksyms.c b/xenolinux-2.4.25-sparse/arch/xeno/kernel/i386_ksyms.c
index e62ca85c04..6744999039 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/kernel/i386_ksyms.c
+++ b/xenolinux-2.4.25-sparse/arch/xeno/kernel/i386_ksyms.c
@@ -9,7 +9,7 @@
 #include <linux/interrupt.h>
 #include <linux/smp_lock.h>
 #include <linux/pm.h>
-//XXX ??? #include <linux/pci.h>
+#include <linux/pci.h>
 #include <linux/apm_bios.h>
 #include <linux/kernel.h>
 #include <linux/string.h>
@@ -68,6 +68,8 @@ EXPORT_SYMBOL(pm_power_off);
 EXPORT_SYMBOL(apm_info);
 //EXPORT_SYMBOL(gdt);
 EXPORT_SYMBOL(empty_zero_page);
+EXPORT_SYMBOL(phys_to_machine_mapping);
+
 
 #ifdef CONFIG_DEBUG_IOVIRT
 EXPORT_SYMBOL(__io_virt_debug);
@@ -101,6 +103,16 @@ EXPORT_SYMBOL(__generic_copy_from_user);
 EXPORT_SYMBOL(__generic_copy_to_user);
 EXPORT_SYMBOL(strnlen_user);
 
+
+EXPORT_SYMBOL(pci_alloc_consistent);
+EXPORT_SYMBOL(pci_free_consistent);
+
+#ifdef CONFIG_PCI
+EXPORT_SYMBOL(pcibios_penalize_isa_irq);
+EXPORT_SYMBOL(pci_mem_start);
+#endif
+
+
 #ifdef CONFIG_X86_USE_3DNOW
 EXPORT_SYMBOL(_mmx_memcpy);
 EXPORT_SYMBOL(mmx_clear_page);
diff --git a/xenolinux-2.4.25-sparse/arch/xeno/kernel/physirq.c b/xenolinux-2.4.25-sparse/arch/xeno/kernel/physirq.c
index 3a6083b24e..1f7a8e4fee 100644
--- a/xenolinux-2.4.25-sparse/arch/xeno/kernel/physirq.c
+++ b/xenolinux-2.4.25-sparse/arch/xeno/kernel/physirq.c
@@ -38,7 +38,9 @@ static unsigned int startup_physirq_event(unsigned int irq)
     printk("startup_physirq_event %d\n", irq);
 
     /*
-     * install a interrupt handler for physirq event when called thefirst tim
+     * install a interrupt handler for physirq event when called first time
+     * we actually are never executing the handler as _EVENT_PHYSIRQ is 
+     * handled specially in hypervisor.c But we need to enable the event etc.
      */
     if ( !setup_event_handler )
     {
@@ -66,23 +68,51 @@ static unsigned int startup_physirq_event(unsigned int irq)
     }
     return 0;
 }
+/*
+ * This is a dummy interrupt handler.
+ * It should never be called. events for physical interrupts are handled
+ * differently in hypervisor.c
+ */
+static void physirq_interrupt(int irq, void *unused, struct pt_regs *ptregs)
+{
+    printk("XXX This should never be called!");
+}
+
 
+/*
+ * IRQ is not needed anymore.
+ */
 static void shutdown_physirq_event(unsigned int irq)
 {
+    physdev_op_t op;
+    int err;
 
-    /* call xen to free IRQ */
+    printk("shutdown_phys_irq called.");
 
+    /*
+     * tell hypervisor
+     */
+    op.cmd = PHYSDEVOP_FREE_IRQ;
+    op.u.free_irq.irq   = irq;
+    if ( (err = HYPERVISOR_physdev_op(&op)) != 0 )
+    {
+        printk(KERN_ALERT "could not free IRQ %d\n", irq);
+        return;
+    }
+    return;
 }
 
 
 static void enable_physirq_event(unsigned int irq)
 {
-    /* XXX just enable all interrupts for now */
+    /* XXX just enable all phys interrupts for now */
+    enable_irq(HYPEREVENT_IRQ(_EVENT_PHYSIRQ));
 }
 
 static void disable_physirq_event(unsigned int irq)
 {
-    /* XXX just disable all interrupts for now */
+    /* XXX just disable all phys interrupts for now */
+    disable_irq(HYPEREVENT_IRQ(_EVENT_PHYSIRQ));
 }
 
 static void ack_physirq_event(unsigned int irq)
@@ -100,6 +130,7 @@ static void end_physirq_event(unsigned int irq)
 {
     int err;
     physdev_op_t op;
+
     /* call hypervisor */
     op.cmd = PHYSDEVOP_FINISHED_IRQ;
     op.u.finished_irq.irq   = irq;
@@ -123,21 +154,6 @@ static struct hw_interrupt_type physirq_irq_type = {
 };
 
 
-/*
- * this interrupt handler demuxes the virt phys event and the virt phys 
- * bitmask and calls the interrupt handlers for virtualised physical interrupts
- */
-static void physirq_interrupt(int irq, void *unused, struct pt_regs *ptregs)
-{
-#if 0
-    unsigned long flags;
-    int virq;
-    local_irq_save(flags);
-    do_IRQ(virq);
-    local_irq_restore(flags);
-#endif
-}
-
 
 void __init physirq_init(void)
 {