bitkeeper revision 1.913 (40ab4ea8vu84ako0LR_rQN0fDuznVw)

Fix the mmu_update hypercall interface.

11 files changed, 62 insertions, 80 deletions

diff --git a/extras/mini-os/h/hypervisor.h b/extras/mini-os/h/hypervisor.h
index c0f9275945..b1cfd60dc6 100644
--- a/extras/mini-os/h/hypervisor.h
+++ b/extras/mini-os/h/hypervisor.h
@@ -50,13 +50,15 @@ static __inline__ int HYPERVISOR_set_trap_table(trap_info_t *table)
     return ret;
 }
 
-static __inline__ int HYPERVISOR_mmu_update(mmu_update_t *req, int count)
+static __inline__ int HYPERVISOR_mmu_update(mmu_update_t *req, 
+                                            int count, 
+                                            int *success_count)
 {
     int ret;
     __asm__ __volatile__ (
         TRAP_INSTR
         : "=a" (ret) : "0" (__HYPERVISOR_mmu_update), 
-        "b" (req), "c" (count) : "memory" );
+        "b" (req), "c" (count), "d" (success_count)  : "memory" );
 
     return ret;
 }
diff --git a/tools/xc/lib/xc_private.c b/tools/xc/lib/xc_private.c
index 430dc6ec11..015354fde5 100644
--- a/tools/xc/lib/xc_private.c
+++ b/tools/xc/lib/xc_private.c
@@ -323,7 +323,8 @@ static int flush_mmu_updates(int xc_handle, mmu_t *mmu)
 
     hypercall.op     = __HYPERVISOR_mmu_update;
     hypercall.arg[0] = (unsigned long)mmu->updates;
-    hypercall.arg[1] = (unsigned long)&(mmu->idx);
+    hypercall.arg[1] = (unsigned long)mmu->idx;
+    hypercall.arg[2] = 0;
 
     if ( mlock(mmu->updates, sizeof(mmu->updates)) != 0 )
     {
diff --git a/xen/common/memory.c b/xen/common/memory.c
index e5d4db2e81..5b03588b2c 100644
--- a/xen/common/memory.c
+++ b/xen/common/memory.c
@@ -988,9 +988,8 @@ static int do_extended_command(unsigned long ptr, unsigned long val)
 }
 
 
-int do_mmu_update(mmu_update_t *ureqs, int * p_count)
+int do_mmu_update(mmu_update_t *ureqs, int count, int *success_count)
 {
-    int count;
     mmu_update_t req;
     unsigned long va = 0, deferred_ops, pfn, prev_pfn = 0;
     struct pfn_info *page;
@@ -999,11 +998,6 @@ int do_mmu_update(mmu_update_t *ureqs, int * p_count)
     unsigned long prev_spfn = 0;
     l1_pgentry_t *prev_spl1e = 0;
 
-    if ( unlikely( get_user(count, p_count) ) )
-    {
-	return -EFAULT;
-    }
-
     perfc_incrc(calls_to_mmu_update); 
     perfc_addc(num_page_updates, count);
 
@@ -1160,8 +1154,8 @@ int do_mmu_update(mmu_update_t *ureqs, int * p_count)
         percpu_info[cpu].gps = percpu_info[cpu].pts = NULL;
     }
 
-    if ( unlikely(rc) )
-	put_user( count, p_count );
+    if ( unlikely(success_count != NULL) )
+	put_user(count, success_count);
 
     return rc;
 }
diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h
index c132ad9662..ecb73627a3 100644
--- a/xen/include/xen/mm.h
+++ b/xen/include/xen/mm.h
@@ -314,7 +314,7 @@ int check_descriptor(unsigned long a, unsigned long b);
 #define machine_to_phys_mapping ((unsigned long *)RDWR_MPT_VIRT_START)
 
 /* Part of the domain API. */
-int do_mmu_update(mmu_update_t *updates, int *count);
+int do_mmu_update(mmu_update_t *updates, int count, int *success_count);
 
 #define DEFAULT_GDT_ENTRIES     ((LAST_RESERVED_GDT_ENTRY*8)+7)
 #define DEFAULT_GDT_ADDRESS     ((unsigned long)gdt_table)
diff --git a/xenolinux-2.4.26-sparse/arch/xen/drivers/dom0/core.c b/xenolinux-2.4.26-sparse/arch/xen/drivers/dom0/core.c
index 2fc577061e..5bb4f4f4b9 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/drivers/dom0/core.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/drivers/dom0/core.c
@@ -89,18 +89,18 @@ static int privcmd_ioctl(struct inode *inode, struct file *file,
 	    if ( copy_from_user(&msg, p, n*sizeof(privcmd_mmap_entry_t)) )
 		return -EFAULT;
 	    
-	    for (j=0;j<n;j++)
+	    for ( j = 0; j < n; j++ )
 	    {
 		struct vm_area_struct *vma = 
 		    find_vma( current->mm, msg[j].va );
 
-		if (!vma)
+		if ( !vma )
 		    return -EINVAL;
 
-		if (msg[j].va > PAGE_OFFSET)
+		if ( msg[j].va > PAGE_OFFSET )
 		    return -EINVAL;
 
-		if (msg[j].va + (msg[j].npages<<PAGE_SHIFT) > vma->vm_end)
+		if ( (msg[j].va + (msg[j].npages<<PAGE_SHIFT)) > vma->vm_end )
 		    return -EINVAL;
 
 		if ( (rc = direct_remap_area_pages(vma->vm_mm, 
@@ -108,7 +108,7 @@ static int privcmd_ioctl(struct inode *inode, struct file *file,
 					    msg[j].mfn<<PAGE_SHIFT, 
 					    msg[j].npages<<PAGE_SHIFT, 
 					    vma->vm_page_prot,
-					    mmapcmd.dom)) <0)
+					    mmapcmd.dom)) < 0 )
 		    return rc;
 	    }
 	}
@@ -131,21 +131,15 @@ static int privcmd_ioctl(struct inode *inode, struct file *file,
 
 	vma = find_vma( current->mm, m.addr );
 
-	if (!vma)
+	if ( !vma )
 	{ ret = -EINVAL; goto batch_err; }
 
-	if (m.addr > PAGE_OFFSET)
+	if ( m.addr > PAGE_OFFSET )
 	{ ret = -EFAULT; goto batch_err; }
 
-	if (m.addr + (m.num<<PAGE_SHIFT) > vma->vm_end)
+	if ( (m.addr + (m.num<<PAGE_SHIFT)) > vma->vm_end )
 	{ ret = -EFAULT; goto batch_err; }
 
-	// everything fits inside the vma
-
-//printk("direct_r_a_p sx=%ld address=%lx macaddr=%lx dom=%lld\n",size,address,machine_addr,domid);
-//    memset( u, 0, sizeof(mmu_update_t)*MAX_DIRECTMAP_MMU_QUEUE );// XXX
-
-
 	if ( m.dom != 0 )
 	{
 	    u[0].val  = (unsigned long)(m.dom<<16) & ~0xFFFFUL;
@@ -165,35 +159,28 @@ static int privcmd_ioctl(struct inode *inode, struct file *file,
 
 	p = m.arr;
 	addr = m.addr;
-//printk("BATCH: arr=%p addr=%lx num=%d u=%p,w=%p\n",p,addr,m.num,u,w);
-	for (i=0; i<m.num; i++, addr+=PAGE_SIZE, p++)
+	for ( i = 0; i < m.num; i++, addr += PAGE_SIZE, p++ )
 	{
-	    unsigned int count;
 	    if ( get_user(mfn, p) ) return -EFAULT;
 
 	    v->val = (mfn << PAGE_SHIFT) | pgprot_val(vma->vm_page_prot) |
 		_PAGE_IO;
 
-	    __direct_remap_area_pages( vma->vm_mm,
-				       addr, 
-				       PAGE_SIZE, 
-				       v);
-	    v++;
-	    count = v-u;
-//printk("Q i=%d mfn=%x co=%d v=%p : %lx %lx\n",i,mfn,count,v, w->val,w->ptr);
+	    __direct_remap_area_pages(vma->vm_mm,
+                                      addr, 
+                                      PAGE_SIZE, 
+                                      v);
 
-	    if ( HYPERVISOR_mmu_update(u, &count) < 0 )
-	    {
-		//printk("Fail %d->%d mfn=%lx\n",v-u,count, w->val);
+	    if ( unlikely(HYPERVISOR_mmu_update(u, v - u + 1, NULL) < 0) )
 		put_user( 0xe0000000 | mfn, p );
-	    }
-	    v=w;
+
+	    v = w;
 	}
 	ret = 0;
 	break;
 
     batch_err:
-	printk("batch_err ret=%d vma=%p addr=%lx num=%d arr=%lx %lx-%lx\n", 
+	printk("batch_err ret=%d vma=%p addr=%lx num=%d arr=%p %lx-%lx\n", 
 	       ret, vma, m.addr, m.num, m.arr, vma->vm_start, vma->vm_end);
 	break;
     }
diff --git a/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/backend/main.c b/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/backend/main.c
index 5b563f41d9..44ca0d0554 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/backend/main.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/backend/main.c
@@ -231,6 +231,7 @@ static void net_rx_action(unsigned long unused)
         mcl[1].op = __HYPERVISOR_mmu_update;
         mcl[1].args[0] = (unsigned long)mmu;
         mcl[1].args[1] = 4;
+        mcl[1].args[2] = 0;
 
         mmu += 4;
         mcl += 2;
diff --git a/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/frontend/main.c b/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/frontend/main.c
index 43e1ba6ed7..0d27bdaeb2 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/frontend/main.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/drivers/netif/frontend/main.c
@@ -384,6 +384,7 @@ static int netif_poll(struct net_device *dev, int *pbudget)
         mcl->op = __HYPERVISOR_mmu_update;
         mcl->args[0] = (unsigned long)rx_mmu;
         mcl->args[1] = mmu - rx_mmu;
+        mcl->args[2] = 0;
         mcl++;
         (void)HYPERVISOR_multicall(rx_mcl, mcl - rx_mcl);
     }
diff --git a/xenolinux-2.4.26-sparse/arch/xen/kernel/traps.c b/xenolinux-2.4.26-sparse/arch/xen/kernel/traps.c
index 78dbb9ef23..8e3523fd3c 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/kernel/traps.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/kernel/traps.c
@@ -317,12 +317,11 @@ asmlinkage void do_general_protection(struct pt_regs * regs, long error_code)
 		__asm__ __volatile__ ( "sldt %0" : "=r" (ldt) );
 		if ( ldt == 0 )
 		{
-		    int count = 1;
 		    mmu_update_t u;
 		    u.ptr  = MMU_EXTENDED_COMMAND;
 		    u.ptr |= (unsigned long)&default_ldt[0];
 		    u.val  = MMUEXT_SET_LDT | (5 << MMUEXT_CMD_SHIFT);
-		    if ( unlikely(HYPERVISOR_mmu_update(&u, &count) < 0) )
+		    if ( unlikely(HYPERVISOR_mmu_update(&u, 1, NULL) < 0) )
 		    {
 			show_trace(NULL);
 			panic("Failed to install default LDT");
@@ -644,7 +643,7 @@ void __init trap_init(void)
  * don't set them to safe values on entry to the kernel). At *any* point Xen 
  * may be entered due to a hardware interrupt --- on exit from Xen an invalid 
  * FS/GS will cause our failsafe_callback to be executed. This could occur, 
- * for example, while the mmmu_update_queue is in an inconsistent state. This
+ * for example, while the mmu_update_queue is in an inconsistent state. This
  * is disastrous because the normal page-fault handler touches the update
  * queue!
  * 
diff --git a/xenolinux-2.4.26-sparse/arch/xen/mm/hypervisor.c b/xenolinux-2.4.26-sparse/arch/xen/mm/hypervisor.c
index 9c4997e3ec..1b73feee35 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/mm/hypervisor.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/mm/hypervisor.c
@@ -40,13 +40,12 @@ static void DEBUG_allow_pt_reads(void)
     int i;
     for ( i = idx-1; i >= 0; i-- )
     {
-	int count = 1;
         pte = update_debug_queue[i].ptep;
         if ( pte == NULL ) continue;
         update_debug_queue[i].ptep = NULL;
         update.ptr = virt_to_machine(pte);
         update.val = update_debug_queue[i].pteval;
-        HYPERVISOR_mmu_update(&update, &count);
+        HYPERVISOR_mmu_update(&update, 1, NULL);
     }
 }
 static void DEBUG_disallow_pt_read(unsigned long va)
@@ -55,7 +54,6 @@ static void DEBUG_disallow_pt_read(unsigned long va)
     pmd_t *pmd;
     pgd_t *pgd;
     unsigned long pteval;
-    int count = 1;
     /*
      * We may fault because of an already outstanding update.
      * That's okay -- it'll get fixed up in the fault handler.
@@ -67,7 +65,7 @@ static void DEBUG_disallow_pt_read(unsigned long va)
     update.ptr = virt_to_machine(pte);
     pteval = *(unsigned long *)pte;
     update.val = pteval & ~_PAGE_PRESENT;
-    HYPERVISOR_mmu_update(&update, &count);
+    HYPERVISOR_mmu_update(&update, 1, NULL);
     update_debug_queue[idx].ptep = pte;
     update_debug_queue[idx].pteval = pteval;
 }
@@ -103,9 +101,10 @@ void MULTICALL_flush_page_update_queue(void)
 #endif
         idx = 0;
         wmb(); /* Make sure index is cleared first to avoid double updates. */
-        queue_multicall2(__HYPERVISOR_mmu_update, 
+        queue_multicall3(__HYPERVISOR_mmu_update, 
                          (unsigned long)update_queue, 
-                         &_idx);
+                         (unsigned long)_idx, 
+                         (unsigned long)NULL);
     }
     spin_unlock_irqrestore(&update_lock, flags);
 }
@@ -121,7 +120,7 @@ static inline void __flush_page_update_queue(void)
 #endif
     idx = 0;
     wmb(); /* Make sure index is cleared first to avoid double updates. */
-    if ( unlikely(HYPERVISOR_mmu_update(update_queue, &_idx) < 0) )
+    if ( unlikely(HYPERVISOR_mmu_update(update_queue, _idx, NULL) < 0) )
         panic("Failed to execute MMU updates");
 }
 
diff --git a/xenolinux-2.4.26-sparse/arch/xen/mm/ioremap.c b/xenolinux-2.4.26-sparse/arch/xen/mm/ioremap.c
index 09a677cc6f..4fd28897a4 100644
--- a/xenolinux-2.4.26-sparse/arch/xen/mm/ioremap.c
+++ b/xenolinux-2.4.26-sparse/arch/xen/mm/ioremap.c
@@ -42,18 +42,11 @@ static inline void direct_remap_area_pte(pte_t *pte,
         BUG();
 
     do {
-#if 0 // XXX
-        if (!pte_none(*pte)) {
-            printk("direct_remap_area_pte: page already exists\n");
-            BUG();
-        }
-#endif
         (*v)->ptr = virt_to_machine(pte);
         (*v)++;
         address += PAGE_SIZE;
         pte++;
     } while (address && (address < end));
-    return ;
 }
 
 static inline int direct_remap_area_pmd(struct mm_struct *mm,
@@ -71,7 +64,7 @@ static inline int direct_remap_area_pmd(struct mm_struct *mm,
     if (address >= end)
         BUG();
     do {
-        pte_t * pte = pte_alloc(mm, pmd, address);
+        pte_t *pte = pte_alloc(mm, pmd, address);
         if (!pte)
             return -ENOMEM;
         direct_remap_area_pte(pte, address, end - address, v);
@@ -117,7 +110,7 @@ int direct_remap_area_pages(struct mm_struct *mm,
                             pgprot_t prot,
                             domid_t  domid)
 {
-    int i, count;
+    int i;
     unsigned long start_address;
 #define MAX_DIRECTMAP_MMU_QUEUE 130
     mmu_update_t u[MAX_DIRECTMAP_MMU_QUEUE], *w, *v;
@@ -141,39 +134,42 @@ int direct_remap_area_pages(struct mm_struct *mm,
 
     start_address = address;
 
-    for(i=0; i<size; 
-	i+=PAGE_SIZE, machine_addr+=PAGE_SIZE, address+=PAGE_SIZE, v++)
+    for( i = 0; i < size; i += PAGE_SIZE )
     {
-	if( (v-u) == MAX_DIRECTMAP_MMU_QUEUE )
+	if ( (v - u) == MAX_DIRECTMAP_MMU_QUEUE )
 	{
-	    /* get the ptep's filled in */
+	    /* Fill in the PTE pointers. */
 	    __direct_remap_area_pages( mm,
 				       start_address, 
 				       address-start_address, 
 				       w);
 	    
-	    count = v-u;
-	    if ( HYPERVISOR_mmu_update(u, &count) < 0 )
+	    if ( HYPERVISOR_mmu_update(u, v - u, NULL) < 0 )
 		return -EFAULT;	    
-	    v=w;
+	    v = w;
 	    start_address = address;
 	}
 
-	/* fill in the machine addresses */
+	/*
+         * Fill in the machine address: PTE ptr is done later by
+         * __direct_remap_area_pages(). 
+         */
         v->val = (machine_addr & PAGE_MASK) | pgprot_val(prot) | _PAGE_IO;
+
+        machine_addr += PAGE_SIZE;
+        address += PAGE_SIZE; 
+        v++;
     }
 
-    if(v!=w)
+    if ( v != w )
     {
 	/* get the ptep's filled in */
-	__direct_remap_area_pages( mm,
-				   start_address, 
-				   address-start_address, 
-				   w);	 
-	count = v-u;
-	if ( HYPERVISOR_mmu_update(u, &count) < 0 )
+	__direct_remap_area_pages(mm,
+                                  start_address, 
+                                  address-start_address, 
+                                  w);	 
+	if ( unlikely(HYPERVISOR_mmu_update(u, v - u, NULL) < 0) )
 	    return -EFAULT;	    
-
     }
     
     return 0;
diff --git a/xenolinux-2.4.26-sparse/include/asm-xen/hypervisor.h b/xenolinux-2.4.26-sparse/include/asm-xen/hypervisor.h
index dddfc25cc3..a2f9653b6f 100644
--- a/xenolinux-2.4.26-sparse/include/asm-xen/hypervisor.h
+++ b/xenolinux-2.4.26-sparse/include/asm-xen/hypervisor.h
@@ -160,13 +160,15 @@ static inline int HYPERVISOR_set_trap_table(trap_info_t *table)
     return ret;
 }
 
-static inline int HYPERVISOR_mmu_update(mmu_update_t *req, int *count)
+static inline int HYPERVISOR_mmu_update(mmu_update_t *req, 
+                                        int count, 
+                                        int *success_count)
 {
     int ret;
     __asm__ __volatile__ (
         TRAP_INSTR
         : "=a" (ret) : "0" (__HYPERVISOR_mmu_update), 
-        "b" (req), "c" (count) : "memory" );
+        "b" (req), "c" (count), "d" (success_count) : "memory" );
 
     return ret;
 }