[HVM] Move shadow initialisation into domain-creation hypercall.

Allocate HVM guest memory in the libxc builder function rather
than in xend. Clean up fall out from these changes.

Todo: Fix ia64. Move PV builder to same model (it should allocate the
memory rather than xend doing so -- then it can avoid using
xc_get_pfn_list()).

Signed-off-by: Keir Fraser <keir@xensource.com>

8 files changed, 64 insertions, 164 deletions

diff --git a/tools/ioemu/vl.c b/tools/ioemu/vl.c
index fafb325835..c6798e5c92 100644
--- a/tools/ioemu/vl.c
+++ b/tools/ioemu/vl.c
@@ -6420,15 +6420,14 @@ int main(int argc, char **argv)
     }
 
 #if defined(__i386__) || defined(__x86_64__)
-    if (xc_get_pfn_list(xc_handle, domid, page_array, nr_pages) != nr_pages) {
+    for ( i = 0; i < tmp_nr_pages; i++)
+        page_array[i] = i;
+    if (xc_domain_translate_gpfn_list(xc_handle, domid, tmp_nr_pages,
+                                      page_array, page_array)) {
         fprintf(logfile, "xc_get_pfn_list returned error %d\n", errno);
         exit(-1);
     }
 
-    if (ram_size > HVM_BELOW_4G_RAM_END)
-        for (i = 0; i < nr_pages - (HVM_BELOW_4G_RAM_END >> PAGE_SHIFT); i++)
-            page_array[tmp_nr_pages - 1 - i] = page_array[nr_pages - 1 - i];
-
     phys_ram_base = xc_map_foreign_batch(xc_handle, domid,
                                          PROT_READ|PROT_WRITE, page_array,
                                          tmp_nr_pages);
diff --git a/tools/libxc/xc_hvm_build.c b/tools/libxc/xc_hvm_build.c
index f2bd062952..52e935d3e0 100644
--- a/tools/libxc/xc_hvm_build.c
+++ b/tools/libxc/xc_hvm_build.c
@@ -196,7 +196,6 @@ static int set_hvm_info(int xc_handle, uint32_t dom,
 static int setup_guest(int xc_handle,
                        uint32_t dom, int memsize,
                        char *image, unsigned long image_size,
-                       unsigned long nr_pages,
                        vcpu_guest_context_t *ctxt,
                        unsigned long shared_info_frame,
                        unsigned int vcpus,
@@ -207,18 +206,13 @@ static int setup_guest(int xc_handle,
                        unsigned long *store_mfn)
 {
     xen_pfn_t *page_array = NULL;
-    unsigned long count, i;
-    unsigned long long ptr;
-    xc_mmu_t *mmu = NULL;
-
+    unsigned long i, nr_pages = (unsigned long)memsize << (20 - PAGE_SHIFT);
+    unsigned long shared_page_nr;
     shared_info_t *shared_info;
     void *e820_page;
-
     struct domain_setup_info dsi;
     uint64_t v_end;
 
-    unsigned long shared_page_nr;
-
     memset(&dsi, 0, sizeof(struct domain_setup_info));
 
     if ( (parseelfimage(image, image_size, &dsi)) != 0 )
@@ -230,7 +224,6 @@ static int setup_guest(int xc_handle,
         goto error_out;
     }
 
-    /* memsize is in megabytes */
     v_end = (unsigned long long)memsize << 20;
 
     IPRINTF("VIRTUAL MEMORY ARRANGEMENT:\n"
@@ -255,53 +248,27 @@ static int setup_guest(int xc_handle,
         goto error_out;
     }
 
-    if ( xc_get_pfn_list(xc_handle, dom, page_array, nr_pages) != nr_pages )
+    for ( i = 0; i < nr_pages; i++ )
+        page_array[i] = i;
+    for ( i = HVM_BELOW_4G_RAM_END >> PAGE_SHIFT; i < nr_pages; i++ )
+        page_array[i] += HVM_BELOW_4G_MMIO_LENGTH >> PAGE_SHIFT;
+
+    if ( xc_domain_memory_populate_physmap(xc_handle, dom, nr_pages,
+                                           0, 0, page_array) )
     {
-        PERROR("Could not get the page frame list.\n");
+        PERROR("Could not allocate memory for HVM guest.\n");
         goto error_out;
     }
 
-    /* HVM domains must be put into shadow mode at the start of day. */
-    /* XXX *After* xc_get_pfn_list()!! */
-    if ( xc_shadow_control(xc_handle, dom, XEN_DOMCTL_SHADOW_OP_ENABLE,
-                           NULL, 0, NULL, 
-                           XEN_DOMCTL_SHADOW_ENABLE_REFCOUNT  |
-                           XEN_DOMCTL_SHADOW_ENABLE_TRANSLATE |
-                           XEN_DOMCTL_SHADOW_ENABLE_EXTERNAL, 
-                           NULL) )
+    if ( xc_domain_translate_gpfn_list(xc_handle, dom, nr_pages,
+                                       page_array, page_array) )
     {
-        PERROR("Could not enable shadow paging for domain.\n");
+        PERROR("Could not translate addresses of HVM guest.\n");
         goto error_out;
-    }        
+    }
 
     loadelfimage(image, xc_handle, dom, page_array, &dsi);
 
-    if ( (mmu = xc_init_mmu_updates(xc_handle, dom)) == NULL )
-        goto error_out;
-
-    /* Write the machine->phys table entries. */
-    for ( count = 0; count < nr_pages; count++ )
-    {
-        unsigned long gpfn_count_skip;
-
-        ptr = (unsigned long long)page_array[count] << PAGE_SHIFT;
-
-        gpfn_count_skip = 0;
-
-        /*
-         * physical address space from HVM_BELOW_4G_RAM_END to 4G is reserved
-         * for PCI devices MMIO. So if HVM has more than HVM_BELOW_4G_RAM_END
-         * RAM, memory beyond HVM_BELOW_4G_RAM_END will go to 4G above.
-         */
-        if ( count >= (HVM_BELOW_4G_RAM_END >> PAGE_SHIFT) )
-            gpfn_count_skip = HVM_BELOW_4G_MMIO_LENGTH >> PAGE_SHIFT;
-
-        if ( xc_add_mmu_update(xc_handle, mmu,
-                               ptr | MMU_MACHPHYS_UPDATE,
-                               count + gpfn_count_skip) )
-            goto error_out;
-    }
-
     if ( set_hvm_info(xc_handle, dom, page_array, vcpus, acpi) )
     {
         ERROR("Couldn't set hvm info for HVM guest.\n");
@@ -352,22 +319,13 @@ static int setup_guest(int xc_handle,
     if ( xc_clear_domain_page(xc_handle, dom, *store_mfn) )
         goto error_out;
 
-    /* Send the page update requests down to the hypervisor. */
-    if ( xc_finish_mmu_updates(xc_handle, mmu) )
-        goto error_out;
-
-    free(mmu);
     free(page_array);
 
-    /*
-     * Initial register values:
-     */
     ctxt->user_regs.eip = dsi.v_kernentry;
 
     return 0;
 
  error_out:
-    free(mmu);
     free(page_array);
     return -1;
 }
@@ -387,8 +345,6 @@ static int xc_hvm_build_internal(int xc_handle,
     struct xen_domctl launch_domctl, domctl;
     int rc, i;
     vcpu_guest_context_t st_ctxt, *ctxt = &st_ctxt;
-    unsigned long nr_pages;
-    xen_capabilities_info_t xen_caps;
 
     if ( (image == NULL) || (image_size == 0) )
     {
@@ -396,25 +352,6 @@ static int xc_hvm_build_internal(int xc_handle,
         goto error_out;
     }
 
-    if ( (rc = xc_version(xc_handle, XENVER_capabilities, &xen_caps)) != 0 )
-    {
-        PERROR("Failed to get xen version info");
-        goto error_out;
-    }
-
-    if ( !strstr(xen_caps, "hvm") )
-    {
-        PERROR("CPU doesn't support HVM extensions or "
-               "the extensions are not enabled");
-        goto error_out;
-    }
-
-    if ( (nr_pages = xc_get_tot_pages(xc_handle, domid)) < 0 )
-    {
-        PERROR("Could not find total pages for domain");
-        goto error_out;
-    }
-
     if ( lock_pages(&st_ctxt, sizeof(st_ctxt) ) )
     {
         PERROR("%s: ctxt mlock failed", __func__);
@@ -430,24 +367,10 @@ static int xc_hvm_build_internal(int xc_handle,
         goto error_out;
     }
 
-#if 0
-    /* HVM domains must be put into shadow mode at the start of day */
-    if ( xc_shadow_control(xc_handle, domid, XEN_DOMCTL_SHADOW_OP_ENABLE,
-                           NULL, 0, NULL, 
-                           XEN_DOMCTL_SHADOW_ENABLE_REFCOUNT  |
-                           XEN_DOMCTL_SHADOW_ENABLE_TRANSLATE |
-                           XEN_DOMCTL_SHADOW_ENABLE_EXTERNAL, 
-                           NULL) )
-    {
-        PERROR("Could not enable shadow paging for domain.\n");
-        goto error_out;
-    }        
-#endif
-
     memset(ctxt, 0, sizeof(*ctxt));
-
     ctxt->flags = VGCF_HVM_GUEST;
-    if ( setup_guest(xc_handle, domid, memsize, image, image_size, nr_pages,
+
+    if ( setup_guest(xc_handle, domid, memsize, image, image_size,
                      ctxt, domctl.u.getdomaininfo.shared_info_frame,
                      vcpus, pae, acpi, apic, store_evtchn, store_mfn) < 0)
     {
diff --git a/tools/python/xen/xend/XendDomainInfo.py b/tools/python/xen/xend/XendDomainInfo.py
index ae163a3a60..5d9a9a98be 100644
--- a/tools/python/xen/xend/XendDomainInfo.py
+++ b/tools/python/xen/xend/XendDomainInfo.py
@@ -1295,9 +1295,11 @@ class XendDomainInfo:
             shadow_cur = xc.shadow_mem_control(self.domid, shadow / 1024)
             self.info['shadow_memory'] = shadow_cur
 
-            # initial memory reservation
-            xc.domain_memory_increase_reservation(self.domid, reservation, 0,
-                                                  0)
+            # Initial memory reservation
+            if not (self._infoIsSet('image') and
+                    sxp.name(self.info['image']) == "hvm"):
+                xc.domain_memory_increase_reservation(
+                    self.domid, reservation, 0, 0)
 
             self._createChannels()
 
diff --git a/tools/python/xen/xend/image.py b/tools/python/xen/xend/image.py
index 0d07fbe928..cd380e02e1 100644
--- a/tools/python/xen/xend/image.py
+++ b/tools/python/xen/xend/image.py
@@ -478,22 +478,12 @@ class X86_HVM_ImageHandler(HVMImageHandler):
 
     def getRequiredAvailableMemory(self, mem_kb):
         # Add 8 MiB overhead for QEMU's video RAM.
-        return self.getRequiredInitialReservation(mem_kb) + 8192
+        return mem_kb + 8192
 
     def getRequiredInitialReservation(self, mem_kb):
-        page_kb = 4
-        # This was derived emperically:
-        #   2.4 MB overhead per 1024 MB RAM
-        #   + 4 to avoid low-memory condition
-        extra_mb = (2.4/1024) * (mem_kb/1024.0) + 4;
-        extra_pages = int( math.ceil( extra_mb*1024 / page_kb ))
-        return mem_kb + extra_pages * page_kb
+        return mem_kb
 
     def getRequiredShadowMemory(self, shadow_mem_kb, maxmem_kb):
-        # The given value is the configured value -- we need to include the
-        # overhead due to getRequiredInitialReservation.
-        maxmem_kb = self.getRequiredInitialReservation(maxmem_kb)
-
         # 256 pages (1MB) per vcpu,
         # plus 1 page per MiB of RAM for the P2M map,
         # plus 1 page per MiB of RAM to shadow the resident processes.  
diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index d8a8f111bc..4b8a124270 100644
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -155,19 +155,12 @@ int arch_domain_create(struct domain *d)
 {
     l1_pgentry_t gdt_l1e;
     int vcpuid, pdpt_order;
-    int i;
-
-    if ( is_hvm_domain(d) && !hvm_enabled )
-    {
-        gdprintk(XENLOG_WARNING, "Attempt to create a HVM guest "
-                 "on a non-VT/AMDV platform.\n");
-        return -EINVAL;
-    }
+    int i, rc = -ENOMEM;
 
     pdpt_order = get_order_from_bytes(PDPT_L1_ENTRIES * sizeof(l1_pgentry_t));
     d->arch.mm_perdomain_pt = alloc_xenheap_pages(pdpt_order);
     if ( d->arch.mm_perdomain_pt == NULL )
-        goto fail_nomem;
+        goto fail;
     memset(d->arch.mm_perdomain_pt, 0, PAGE_SIZE << pdpt_order);
 
     /*
@@ -192,7 +185,7 @@ int arch_domain_create(struct domain *d)
     d->arch.mm_perdomain_l3 = alloc_xenheap_page();
     if ( (d->arch.mm_perdomain_l2 == NULL) ||
          (d->arch.mm_perdomain_l3 == NULL) )
-        goto fail_nomem;
+        goto fail;
 
     memset(d->arch.mm_perdomain_l2, 0, PAGE_SIZE);
     for ( i = 0; i < (1 << pdpt_order); i++ )
@@ -219,26 +212,41 @@ int arch_domain_create(struct domain *d)
         d->arch.ioport_caps = 
             rangeset_new(d, "I/O Ports", RANGESETF_prettyprint_hex);
         if ( d->arch.ioport_caps == NULL )
-            goto fail_nomem;
+            goto fail;
 
         if ( (d->shared_info = alloc_xenheap_page()) == NULL )
-            goto fail_nomem;
+            goto fail;
 
         memset(d->shared_info, 0, PAGE_SIZE);
         share_xen_page_with_guest(
             virt_to_page(d->shared_info), d, XENSHARE_writable);
     }
 
+    if ( is_hvm_domain(d) )
+    {
+        if ( !hvm_enabled )
+        {
+            gdprintk(XENLOG_WARNING, "Attempt to create a HVM guest "
+                     "on a non-VT/AMDV platform.\n");
+            rc = -EINVAL;
+            goto fail;
+        }
+
+        rc = shadow_enable(d, SHM2_refcounts|SHM2_translate|SHM2_external);
+        if ( rc != 0 )
+            goto fail;
+    }
+
     return 0;
 
- fail_nomem:
+ fail:
     free_xenheap_page(d->shared_info);
 #ifdef __x86_64__
     free_xenheap_page(d->arch.mm_perdomain_l2);
     free_xenheap_page(d->arch.mm_perdomain_l3);
 #endif
     free_xenheap_pages(d->arch.mm_perdomain_pt, pdpt_order);
-    return -ENOMEM;
+    return rc;
 }
 
 void arch_domain_destroy(struct domain *d)
diff --git a/xen/arch/x86/domctl.c b/xen/arch/x86/domctl.c
index f900650729..362365dd2f 100644
--- a/xen/arch/x86/domctl.c
+++ b/xen/arch/x86/domctl.c
@@ -224,45 +224,18 @@ long arch_do_domctl(
 
             spin_lock(&d->page_alloc_lock);
 
-            if ( is_hvm_domain(d) && shadow_mode_translate(d) )
+            list_ent = d->page_list.next;
+            for ( i = 0; (i < max_pfns) && (list_ent != &d->page_list); i++ )
             {
-                /* HVM domain: scan P2M to get guaranteed physmap order. */
-                for ( i = 0, gmfn = 0;
-                      (i < max_pfns) && (i < d->tot_pages); 
-                      i++, gmfn++ )
+                mfn = page_to_mfn(list_entry(
+                    list_ent, struct page_info, list));
+                if ( copy_to_guest_offset(domctl->u.getmemlist.buffer,
+                                          i, &mfn, 1) )
                 {
-                    if ( unlikely(i == (HVM_BELOW_4G_MMIO_START>>PAGE_SHIFT)) )
-                    {
-                        /* skip MMIO range */
-                        gmfn += HVM_BELOW_4G_MMIO_LENGTH >> PAGE_SHIFT;
-                    }
-                    mfn = gmfn_to_mfn(d, gmfn);
-                    if ( copy_to_guest_offset(domctl->u.getmemlist.buffer,
-                                              i, &mfn, 1) )
-                    {
-                        ret = -EFAULT;
-                        break;
-                    }
-                }
-            }
-            else 
-            {        
-                /* Other guests: return in order of ownership list. */
-                list_ent = d->page_list.next;
-                for ( i = 0;
-                      (i < max_pfns) && (list_ent != &d->page_list);
-                      i++ )
-                {
-                    mfn = page_to_mfn(list_entry(
-                        list_ent, struct page_info, list));
-                    if ( copy_to_guest_offset(domctl->u.getmemlist.buffer,
-                                              i, &mfn, 1) )
-                    {
-                        ret = -EFAULT;
-                        break;
-                    }
-                    list_ent = mfn_to_page(mfn)->list.next;
+                    ret = -EFAULT;
+                    break;
                 }
+                list_ent = mfn_to_page(mfn)->list.next;
             }
             
             spin_unlock(&d->page_alloc_lock);
diff --git a/xen/arch/x86/mm/shadow/common.c b/xen/arch/x86/mm/shadow/common.c
index 2e48164ca7..aa217975ed 100644
--- a/xen/arch/x86/mm/shadow/common.c
+++ b/xen/arch/x86/mm/shadow/common.c
@@ -2461,7 +2461,7 @@ static void sh_new_mode(struct domain *d, u32 new_mode)
         sh_update_paging_modes(v);
 }
 
-static int shadow_enable(struct domain *d, u32 mode)
+int shadow_enable(struct domain *d, u32 mode)
 /* Turn on "permanent" shadow features: external, translate, refcount.
  * Can only be called once on a domain, and these features cannot be
  * disabled. 
@@ -3092,6 +3092,8 @@ int shadow_domctl(struct domain *d,
         if ( shadow_mode_log_dirty(d) )
             if ( (rc = shadow_log_dirty_disable(d)) != 0 ) 
                 return rc;
+        if ( is_hvm_domain(d) )
+            return -EINVAL;
         if ( d->arch.shadow.mode & SHM2_enable )
             if ( (rc = shadow_test_disable(d)) != 0 ) 
                 return rc;
diff --git a/xen/include/asm-x86/shadow.h b/xen/include/asm-x86/shadow.h
index 1cf2e5041d..292ed43916 100644
--- a/xen/include/asm-x86/shadow.h
+++ b/xen/include/asm-x86/shadow.h
@@ -313,6 +313,9 @@ static inline int shadow_guest_paging_levels(struct vcpu *v)
 /**************************************************************************/
 /* Entry points into the shadow code */
 
+/* Enable arbitrary shadow mode. */
+int shadow_enable(struct domain *d, u32 mode);
+
 /* Turning on shadow test mode */
 int shadow_test_enable(struct domain *d);