[Qemu-devel] [PATCH v4 0/1] Add support for Xen access to vmport

[Qemu-devel] [PATCH v4 0/1] Add support for Xen access to vmport

[Don Slutz]

Changes RFC-v2x to v4:
  Stefano Stabellini
    Please try to get rid of the #ifdefs.
      Moved 2 #ifdefs into hw/xen/xen_common.h

Changes v2 to RFC-v2x:
  Paul Durrant
    Use a 2nd shared page.
      Added HVM_PARAM_VMPORT_IOREQ_PFN usage.

Changes v1 to v2:
   More info in commit message.

  Stefano Stabellini
    the registers being passes explicitely by Xen rather than
    "hiding" them into other ioreq fields.
       Added vmware_ioreq_t
  Paolo Bonzini & Alexander Graf
    Fixup env access
      Added cpu_by_ioreq_id.
      Set current_cpu in regs_to_cpu(), clear in regs_from_cpu().
      Drop all changes to vmport.c

Note: to use this with Xen either a version of:

[Qemu-devel] [PATCH] -machine vmport=off: Allow disabling of VMWare ioport emulation

or

>From f70663d9fb86914144ba340b6186cb1e67ac6eec Mon Sep 17 00:00:00 2001
From: Don Slutz <dslutz@verizon.com>
Date: Fri, 26 Sep 2014 08:11:39 -0400
Subject: [PATCH 1/2] hack: force enable vmport

Signed-off-by: Don Slutz <dslutz@verizon.com>
---
 hw/i386/pc_piix.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/hw/i386/pc_piix.c b/hw/i386/pc_piix.c
index 103d756..b76dfbc 100644
--- a/hw/i386/pc_piix.c
+++ b/hw/i386/pc_piix.c
@@ -234,7 +234,7 @@ static void pc_init1(MachineState *machine,
     pc_vga_init(isa_bus, pci_enabled ? pci_bus : NULL);
 
     /* init basic PC hardware */
-    pc_basic_device_init(isa_bus, gsi, &rtc_state, &floppy, xen_enabled(),
+    pc_basic_device_init(isa_bus, gsi, &rtc_state, &floppy, false,
         0x4);
 
     pc_nic_init(isa_bus, pci_bus);
-- 
1.8.4

needs to be done to QEMU.

And the Xen RFC patch:

[RFC][PATCH v2 1/1] Add IOREQ_TYPE_VMWARE_PORT

needs to be done to Xen.

Don Slutz (1):
  xen-hvm.c: Add support for Xen access to vmport

 include/hw/xen/xen_common.h |  15 ++++++
 xen-hvm.c                   | 121 ++++++++++++++++++++++++++++++++++++++++++--
 2 files changed, 131 insertions(+), 5 deletions(-)

-- 
1.8.4

[Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Don Slutz]

This adds synchronisation of the 6 vcpu registers (only 32bits of
them) that vmport.c needs between Xen and QEMU.

This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
fetch and put these 6 vcpu registers used by the code in vmport.c
and vmmouse.c

The registers are passed in the new shared page provided by
HVM_PARAM_VMPORT_IOREQ_PFN.

Add new array to XenIOState that allows selection of current_cpu by
ioreq_id.

Now pass XenIOState to handle_ioreq().

Add new routines regs_to_cpu(), regs_from_cpu(), and
handle_vmport_ioreq().

Signed-off-by: Don Slutz <dslutz@verizon.com>
---
 include/hw/xen/xen_common.h |  15 ++++++
 xen-hvm.c                   | 121 ++++++++++++++++++++++++++++++++++++++++++--
 2 files changed, 131 insertions(+), 5 deletions(-)

diff --git a/include/hw/xen/xen_common.h b/include/hw/xen/xen_common.h
index 07731b9..9542756 100644
--- a/include/hw/xen/xen_common.h
+++ b/include/hw/xen/xen_common.h
@@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
 /* shutdown/destroy current domain because of an error */
 void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
 
+#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
+static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
+                                           unsigned long *vmport_ioreq_pfn)
+{
+    return xc_get_hvm_param(xc, dom, HVM_PARAM_VMPORT_IOREQ_PFN,
+                            vmport_ioreq_pfn);
+}
+#else
+static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
+                                           unsigned long *vmport_ioreq_pfn)
+{
+    return -ENOSYS;
+}
+#endif
+
 #endif /* QEMU_HW_XEN_COMMON_H */
diff --git a/xen-hvm.c b/xen-hvm.c
index 05e522c..17b3cbd 100644
--- a/xen-hvm.c
+++ b/xen-hvm.c
@@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
 static bool xen_in_migration;
 
 /* Compatibility with older version */
+
+/* This allows QEMU to build on a system that has Xen 4.5 or earlier
+ * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
+ * needs to be included before this block and hw/xen/xen_common.h needs to
+ * be included before xen/hvm/ioreq.h
+ */
+#ifndef IOREQ_TYPE_VMWARE_PORT
+#define IOREQ_TYPE_VMWARE_PORT  3
+struct vmware_ioreq {
+    uint32_t esi;
+    uint32_t edi;
+    uint32_t ebx;
+    uint32_t ecx;
+    uint32_t edx;
+};
+typedef struct vmware_ioreq vmware_ioreq_t;
+
+struct shared_vmport_iopage {
+    struct vmware_ioreq vcpu_vmport_ioreq[1];
+};
+typedef struct shared_vmport_iopage shared_vmport_iopage_t;
+#endif
+
 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
 {
@@ -79,8 +102,10 @@ typedef struct XenPhysmap {
 
 typedef struct XenIOState {
     shared_iopage_t *shared_page;
+    shared_vmport_iopage_t *shared_vmport_page;
     buffered_iopage_t *buffered_io_page;
     QEMUTimer *buffered_io_timer;
+    CPUState **cpu_by_ioreq_id;
     /* the evtchn port for polling the notification, */
     evtchn_port_t *ioreq_local_port;
     /* evtchn local port for buffered io */
@@ -101,6 +126,8 @@ typedef struct XenIOState {
     Notifier wakeup;
 } XenIOState;
 
+static void handle_ioreq(XenIOState *state, ioreq_t *req);
+
 /* Xen specific function for piix pci */
 
 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
@@ -610,6 +637,20 @@ static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
     return req;
 }
 
+/* get the vmport ioreq packets from share mem */
+static vmware_ioreq_t *cpu_get_vmport_ioreq_from_shared_memory(
+    XenIOState *state, int vcpu)
+{
+    vmware_ioreq_t *vmport_req;
+
+    assert(state->shared_vmport_page);
+    vmport_req = &state->shared_vmport_page->vcpu_vmport_ioreq[vcpu];
+
+    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
+
+    return vmport_req;
+}
+
 /* use poll to get the port notification */
 /* ioreq_vec--out,the */
 /* retval--the number of ioreq packet */
@@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
     }
 }
 
-static void handle_ioreq(ioreq_t *req)
+static void regs_to_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
+                        ioreq_t *req)
+{
+    X86CPU *cpu;
+    CPUX86State *env;
+
+    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
+    cpu = X86_CPU(current_cpu);
+    env = &cpu->env;
+    env->regs[R_EAX] = req->data;
+    env->regs[R_EBX] = vmport_req->ebx;
+    env->regs[R_ECX] = vmport_req->ecx;
+    env->regs[R_EDX] = vmport_req->edx;
+    env->regs[R_ESI] = vmport_req->esi;
+    env->regs[R_EDI] = vmport_req->edi;
+}
+
+static void regs_from_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
+                          ioreq_t *req)
+{
+    X86CPU *cpu = X86_CPU(current_cpu);
+    CPUX86State *env = &cpu->env;
+
+    assert(sizeof(*vmport_req) <= sizeof(*req));
+
+    vmport_req->ebx = env->regs[R_EBX];
+    vmport_req->ecx = env->regs[R_ECX];
+    vmport_req->edx = env->regs[R_EDX];
+    vmport_req->esi = env->regs[R_ESI];
+    vmport_req->edi = env->regs[R_EDI];
+    current_cpu = NULL;
+}
+
+static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
+{
+    vmware_ioreq_t *vmport_req =
+        cpu_get_vmport_ioreq_from_shared_memory(state, state->send_vcpu);
+
+    regs_to_cpu(state, vmport_req, req);
+    cpu_ioreq_pio(req);
+    regs_from_cpu(state, vmport_req, req);
+}
+
+static void handle_ioreq(XenIOState *state, ioreq_t *req)
 {
     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
             (req->size < sizeof (target_ulong))) {
@@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
         case IOREQ_TYPE_COPY:
             cpu_ioreq_move(req);
             break;
+        case IOREQ_TYPE_VMWARE_PORT:
+            handle_vmport_ioreq(state, req);
+            break;
         case IOREQ_TYPE_TIMEOFFSET:
             break;
         case IOREQ_TYPE_INVALIDATE:
@@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState *state)
             req.data |= ((uint64_t)buf_req->data) << 32;
         }
 
-        handle_ioreq(&req);
+        handle_ioreq(state, &req);
 
         xen_mb();
         state->buffered_io_page->read_pointer += qw ? 2 : 1;
@@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
 
     handle_buffered_iopage(state);
     if (req) {
-        handle_ioreq(req);
+        handle_ioreq(state, req);
 
         if (req->state != STATE_IOREQ_INPROCESS) {
             fprintf(stderr, "Badness in I/O request ... not in service?!: "
                     "%x, ptr: %x, port: %"PRIx64", "
-                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
+                    "data: %"PRIx64", count: %" FMT_ioreq_size
+                    ", size: %" FMT_ioreq_size
+                    ", type: %"FMT_ioreq_size"\n",
                     req->state, req->data_is_ptr, req->addr,
-                    req->data, req->count, req->size);
+                    req->data, req->count, req->size, req->type);
             destroy_hvm_domain(false);
             return;
         }
@@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState *state)
                                                  state);
 
     if (evtchn_fd != -1) {
+        CPUState *cpu_state;
+
+        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
+        CPU_FOREACH(cpu_state) {
+            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
+                    __func__, cpu_state->cpu_index, cpu_state);
+            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
+        }
         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
     }
 }
@@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
                  errno, xen_xc);
     }
 
+    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
+        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
+        state->shared_vmport_page =
+            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
+                                 PROT_READ|PROT_WRITE, ioreq_pfn);
+        if (state->shared_vmport_page == NULL) {
+            hw_error("map shared vmport IO page returned error %d handle="
+                     XC_INTERFACE_FMT, errno, xen_xc);
+        }
+    }
+
     xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
     DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
     state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
@@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
         hw_error("map buffered IO page returned error %d", errno);
     }
 
+    /* Note: cpus is empty at this point in init */
+    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
+
     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
 
     /* FIXME: how about if we overflow the page here? */
-- 
1.8.4

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Paul Durrant]

> -----Original Message-----
> From: Don Slutz [mailto:dslutz@verizon.com]
> Sent: 17 October 2014 13:14
> To: qemu-devel@nongnu.org; Paul Durrant
> Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
> Anthony Liguori; Don Slutz; Marcel Apfelbaum; Markus Armbruster; Michael
> S. Tsirkin; Stefano Stabellini
> Subject: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport
> 
> This adds synchronisation of the 6 vcpu registers (only 32bits of
> them) that vmport.c needs between Xen and QEMU.
> 
> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
> fetch and put these 6 vcpu registers used by the code in vmport.c
> and vmmouse.c
> 
> The registers are passed in the new shared page provided by
> HVM_PARAM_VMPORT_IOREQ_PFN.
> 
> Add new array to XenIOState that allows selection of current_cpu by
> ioreq_id.
> 
> Now pass XenIOState to handle_ioreq().
> 
> Add new routines regs_to_cpu(), regs_from_cpu(), and
> handle_vmport_ioreq().
> 
> Signed-off-by: Don Slutz <dslutz@verizon.com>
> ---
>  include/hw/xen/xen_common.h |  15 ++++++
>  xen-hvm.c                   | 121
> ++++++++++++++++++++++++++++++++++++++++++--
>  2 files changed, 131 insertions(+), 5 deletions(-)
> 
> diff --git a/include/hw/xen/xen_common.h
> b/include/hw/xen/xen_common.h
> index 07731b9..9542756 100644
> --- a/include/hw/xen/xen_common.h
> +++ b/include/hw/xen/xen_common.h
> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
>  /* shutdown/destroy current domain because of an error */
>  void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
> 
> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> +                                           unsigned long *vmport_ioreq_pfn)
> +{
> +    return xc_get_hvm_param(xc, dom,
> HVM_PARAM_VMPORT_IOREQ_PFN,
> +                            vmport_ioreq_pfn);
> +}
> +#else
> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> +                                           unsigned long *vmport_ioreq_pfn)
> +{
> +    return -ENOSYS;
> +}
> +#endif
> +
>  #endif /* QEMU_HW_XEN_COMMON_H */
> diff --git a/xen-hvm.c b/xen-hvm.c
> index 05e522c..17b3cbd 100644
> --- a/xen-hvm.c
> +++ b/xen-hvm.c
> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
>  static bool xen_in_migration;
> 
>  /* Compatibility with older version */
> +
> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
> + * installed.  This here (not in hw/xen/xen_common.h) because
> xen/hvm/ioreq.h
> + * needs to be included before this block and hw/xen/xen_common.h
> needs to
> + * be included before xen/hvm/ioreq.h
> + */
> +#ifndef IOREQ_TYPE_VMWARE_PORT
> +#define IOREQ_TYPE_VMWARE_PORT  3
> +struct vmware_ioreq {
> +    uint32_t esi;
> +    uint32_t edi;
> +    uint32_t ebx;
> +    uint32_t ecx;
> +    uint32_t edx;
> +};
> +typedef struct vmware_ioreq vmware_ioreq_t;

This struct is not really a request any more. Maybe vmware_regs_t?

> +
> +struct shared_vmport_iopage {
> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
> +};
> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
> +#endif
> +
>  #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
>  static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
>  {
> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
> 
>  typedef struct XenIOState {
>      shared_iopage_t *shared_page;
> +    shared_vmport_iopage_t *shared_vmport_page;
>      buffered_iopage_t *buffered_io_page;
>      QEMUTimer *buffered_io_timer;
> +    CPUState **cpu_by_ioreq_id;

This name implies the array is indexed by an id carries in the ioreq. That doesn't seem to be the case; it's just indexed by guest vcpu id isn't it?

>      /* the evtchn port for polling the notification, */
>      evtchn_port_t *ioreq_local_port;
>      /* evtchn local port for buffered io */
> @@ -101,6 +126,8 @@ typedef struct XenIOState {
>      Notifier wakeup;
>  } XenIOState;
> 
> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
> +
>  /* Xen specific function for piix pci */
> 
>  int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
> @@ -610,6 +637,20 @@ static ioreq_t
> *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
>      return req;
>  }
> 
> +/* get the vmport ioreq packets from share mem */
> +static vmware_ioreq_t *cpu_get_vmport_ioreq_from_shared_memory(

Is this worth its own function? I can only see one use of it in handle_vmport_ioreq below...

> +    XenIOState *state, int vcpu)
> +{
> +    vmware_ioreq_t *vmport_req;
> +
> +    assert(state->shared_vmport_page);
> +    vmport_req = &state->shared_vmport_page-
> >vcpu_vmport_ioreq[vcpu];
> +
> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */

I don't think that comment is applicable here, and I don't think you need the barrier anyway.

> +
> +    return vmport_req;
> +}
> +
>  /* use poll to get the port notification */
>  /* ioreq_vec--out,the */
>  /* retval--the number of ioreq packet */
> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
>      }
>  }
> 
> -static void handle_ioreq(ioreq_t *req)
> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
> +                        ioreq_t *req)
> +{
> +    X86CPU *cpu;
> +    CPUX86State *env;
> +
> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
> +    cpu = X86_CPU(current_cpu);
> +    env = &cpu->env;
> +    env->regs[R_EAX] = req->data;
> +    env->regs[R_EBX] = vmport_req->ebx;
> +    env->regs[R_ECX] = vmport_req->ecx;
> +    env->regs[R_EDX] = vmport_req->edx;
> +    env->regs[R_ESI] = vmport_req->esi;
> +    env->regs[R_EDI] = vmport_req->edi;
> +}
> +
> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t
> *vmport_req,
> +                          ioreq_t *req)
> +{
> +    X86CPU *cpu = X86_CPU(current_cpu);
> +    CPUX86State *env = &cpu->env;
> +
> +    assert(sizeof(*vmport_req) <= sizeof(*req));

Why?

> +
> +    vmport_req->ebx = env->regs[R_EBX];
> +    vmport_req->ecx = env->regs[R_ECX];
> +    vmport_req->edx = env->regs[R_EDX];
> +    vmport_req->esi = env->regs[R_ESI];
> +    vmport_req->edi = env->regs[R_EDI];
> +    current_cpu = NULL;
> +}
> +
> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
> +{
> +    vmware_ioreq_t *vmport_req =
> +        cpu_get_vmport_ioreq_from_shared_memory(state, state-
> >send_vcpu);
> +
> +    regs_to_cpu(state, vmport_req, req);
> +    cpu_ioreq_pio(req);
> +    regs_from_cpu(state, vmport_req, req);

state appears to be unused in this function, and you could avoid passing it to both of them by setting current_cpu here couldn't you?

  Paul

> +}
> +
> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
>  {
>      if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
>              (req->size < sizeof (target_ulong))) {
> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
>          case IOREQ_TYPE_COPY:
>              cpu_ioreq_move(req);
>              break;
> +        case IOREQ_TYPE_VMWARE_PORT:
> +            handle_vmport_ioreq(state, req);
> +            break;
>          case IOREQ_TYPE_TIMEOFFSET:
>              break;
>          case IOREQ_TYPE_INVALIDATE:
> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState
> *state)
>              req.data |= ((uint64_t)buf_req->data) << 32;
>          }
> 
> -        handle_ioreq(&req);
> +        handle_ioreq(state, &req);
> 
>          xen_mb();
>          state->buffered_io_page->read_pointer += qw ? 2 : 1;
> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
> 
>      handle_buffered_iopage(state);
>      if (req) {
> -        handle_ioreq(req);
> +        handle_ioreq(state, req);
> 
>          if (req->state != STATE_IOREQ_INPROCESS) {
>              fprintf(stderr, "Badness in I/O request ... not in service?!: "
>                      "%x, ptr: %x, port: %"PRIx64", "
> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %"
> FMT_ioreq_size "\n",
> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
> +                    ", size: %" FMT_ioreq_size
> +                    ", type: %"FMT_ioreq_size"\n",
>                      req->state, req->data_is_ptr, req->addr,
> -                    req->data, req->count, req->size);
> +                    req->data, req->count, req->size, req->type);
>              destroy_hvm_domain(false);
>              return;
>          }
> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState
> *state)
>                                                   state);
> 
>      if (evtchn_fd != -1) {
> +        CPUState *cpu_state;
> +
> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
> +        CPU_FOREACH(cpu_state) {
> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
> +                    __func__, cpu_state->cpu_index, cpu_state);
> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
> +        }
>          qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
>      }
>  }
> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t
> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>                   errno, xen_xc);
>      }
> 
> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
> +        state->shared_vmport_page =
> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
> +        if (state->shared_vmport_page == NULL) {
> +            hw_error("map shared vmport IO page returned error %d handle="
> +                     XC_INTERFACE_FMT, errno, xen_xc);
> +        }
> +    }
> +
>      xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN,
> &ioreq_pfn);
>      DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
>      state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid,
> XC_PAGE_SIZE,
> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t
> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>          hw_error("map buffered IO page returned error %d", errno);
>      }
> 
> +    /* Note: cpus is empty at this point in init */
> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
> +
>      state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
> 
>      /* FIXME: how about if we overflow the page here? */
> --
> 1.8.4

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Don Slutz]

On 10/17/14 08:34, Paul Durrant wrote:
>> -----Original Message-----
>> From: Don Slutz [mailto:dslutz@verizon.com]
>> Sent: 17 October 2014 13:14
>> To: qemu-devel@nongnu.org; Paul Durrant
>> Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
>> Anthony Liguori; Don Slutz; Marcel Apfelbaum; Markus Armbruster; Michael
>> S. Tsirkin; Stefano Stabellini
>> Subject: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport
>>
>> This adds synchronisation of the 6 vcpu registers (only 32bits of
>> them) that vmport.c needs between Xen and QEMU.
>>
>> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
>> fetch and put these 6 vcpu registers used by the code in vmport.c
>> and vmmouse.c
>>
>> The registers are passed in the new shared page provided by
>> HVM_PARAM_VMPORT_IOREQ_PFN.
>>
>> Add new array to XenIOState that allows selection of current_cpu by
>> ioreq_id.
>>
>> Now pass XenIOState to handle_ioreq().
>>
>> Add new routines regs_to_cpu(), regs_from_cpu(), and
>> handle_vmport_ioreq().
>>
>> Signed-off-by: Don Slutz <dslutz@verizon.com>
>> ---
>>   include/hw/xen/xen_common.h |  15 ++++++
>>   xen-hvm.c                   | 121
>> ++++++++++++++++++++++++++++++++++++++++++--
>>   2 files changed, 131 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/hw/xen/xen_common.h
>> b/include/hw/xen/xen_common.h
>> index 07731b9..9542756 100644
>> --- a/include/hw/xen/xen_common.h
>> +++ b/include/hw/xen/xen_common.h
>> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
>>   /* shutdown/destroy current domain because of an error */
>>   void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>>
>> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
>> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
>> +                                           unsigned long *vmport_ioreq_pfn)
>> +{
>> +    return xc_get_hvm_param(xc, dom,
>> HVM_PARAM_VMPORT_IOREQ_PFN,
>> +                            vmport_ioreq_pfn);
>> +}
>> +#else
>> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
>> +                                           unsigned long *vmport_ioreq_pfn)
>> +{
>> +    return -ENOSYS;
>> +}
>> +#endif
>> +
>>   #endif /* QEMU_HW_XEN_COMMON_H */
>> diff --git a/xen-hvm.c b/xen-hvm.c
>> index 05e522c..17b3cbd 100644
>> --- a/xen-hvm.c
>> +++ b/xen-hvm.c
>> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
>>   static bool xen_in_migration;
>>
>>   /* Compatibility with older version */
>> +
>> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
>> + * installed.  This here (not in hw/xen/xen_common.h) because
>> xen/hvm/ioreq.h
>> + * needs to be included before this block and hw/xen/xen_common.h
>> needs to
>> + * be included before xen/hvm/ioreq.h
>> + */
>> +#ifndef IOREQ_TYPE_VMWARE_PORT
>> +#define IOREQ_TYPE_VMWARE_PORT  3
>> +struct vmware_ioreq {
>> +    uint32_t esi;
>> +    uint32_t edi;
>> +    uint32_t ebx;
>> +    uint32_t ecx;
>> +    uint32_t edx;
>> +};
>> +typedef struct vmware_ioreq vmware_ioreq_t;
> This struct is not really a request any more. Maybe vmware_regs_t?

Sure.

>> +
>> +struct shared_vmport_iopage {
>> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
>> +};
>> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
>> +#endif
>> +
>>   #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
>>   static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
>>   {
>> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
>>
>>   typedef struct XenIOState {
>>       shared_iopage_t *shared_page;
>> +    shared_vmport_iopage_t *shared_vmport_page;
>>       buffered_iopage_t *buffered_io_page;
>>       QEMUTimer *buffered_io_timer;
>> +    CPUState **cpu_by_ioreq_id;
> This name implies the array is indexed by an id carries in the ioreq. That doesn't seem to be the case; it's just indexed by guest vcpu id isn't it?

Yes.  How about cpu_by_vcpu_id ?

>>       /* the evtchn port for polling the notification, */
>>       evtchn_port_t *ioreq_local_port;
>>       /* evtchn local port for buffered io */
>> @@ -101,6 +126,8 @@ typedef struct XenIOState {
>>       Notifier wakeup;
>>   } XenIOState;
>>
>> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
>> +
>>   /* Xen specific function for piix pci */
>>
>>   int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
>> @@ -610,6 +637,20 @@ static ioreq_t
>> *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
>>       return req;
>>   }
>>
>> +/* get the vmport ioreq packets from share mem */
>> +static vmware_ioreq_t *cpu_get_vmport_ioreq_from_shared_memory(
> Is this worth its own function? I can only see one use of it in handle_vmport_ioreq below...

Will move it in-line.  I was just following
cpu_get_ioreq_from_shared_memory which also has 1 use.

>> +    XenIOState *state, int vcpu)
>> +{
>> +    vmware_ioreq_t *vmport_req;
>> +
>> +    assert(state->shared_vmport_page);
>> +    vmport_req = &state->shared_vmport_page-
>>> vcpu_vmport_ioreq[vcpu];
>> +
>> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
> I don't think that comment is applicable here, and I don't think you need the barrier anyway.

I was not sure either way, so went with the safe way.  Will drop.

>> +
>> +    return vmport_req;
>> +}
>> +
>>   /* use poll to get the port notification */
>>   /* ioreq_vec--out,the */
>>   /* retval--the number of ioreq packet */
>> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
>>       }
>>   }
>>
>> -static void handle_ioreq(ioreq_t *req)
>> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
>> +                        ioreq_t *req)
>> +{
>> +    X86CPU *cpu;
>> +    CPUX86State *env;
>> +
>> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
>> +    cpu = X86_CPU(current_cpu);
>> +    env = &cpu->env;
>> +    env->regs[R_EAX] = req->data;
>> +    env->regs[R_EBX] = vmport_req->ebx;
>> +    env->regs[R_ECX] = vmport_req->ecx;
>> +    env->regs[R_EDX] = vmport_req->edx;
>> +    env->regs[R_ESI] = vmport_req->esi;
>> +    env->regs[R_EDI] = vmport_req->edi;
>> +}
>> +
>> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t
>> *vmport_req,
>> +                          ioreq_t *req)
>> +{
>> +    X86CPU *cpu = X86_CPU(current_cpu);
>> +    CPUX86State *env = &cpu->env;
>> +
>> +    assert(sizeof(*vmport_req) <= sizeof(*req));
> Why?

Since there is 1 page of ioreq_t indexed by vcpu id this checks that
the 1 page of vmport data has at least the same number of vcpu
slots.  This prevents a bad memory read.  Looks to not be the right place
for this, will move.

>> +
>> +    vmport_req->ebx = env->regs[R_EBX];
>> +    vmport_req->ecx = env->regs[R_ECX];
>> +    vmport_req->edx = env->regs[R_EDX];
>> +    vmport_req->esi = env->regs[R_ESI];
>> +    vmport_req->edi = env->regs[R_EDI];
>> +    current_cpu = NULL;
>> +}
>> +
>> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
>> +{
>> +    vmware_ioreq_t *vmport_req =
>> +        cpu_get_vmport_ioreq_from_shared_memory(state, state-
>>> send_vcpu);
>> +
>> +    regs_to_cpu(state, vmport_req, req);
>> +    cpu_ioreq_pio(req);
>> +    regs_from_cpu(state, vmport_req, req);
> state appears to be unused in this function, and you could avoid passing it to both of them by setting current_cpu here couldn't you?

Yes, will move state code here.

    -Don Slutz

>    Paul
>
>> +}
>> +
>> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
>>   {
>>       if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
>>               (req->size < sizeof (target_ulong))) {
>> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
>>           case IOREQ_TYPE_COPY:
>>               cpu_ioreq_move(req);
>>               break;
>> +        case IOREQ_TYPE_VMWARE_PORT:
>> +            handle_vmport_ioreq(state, req);
>> +            break;
>>           case IOREQ_TYPE_TIMEOFFSET:
>>               break;
>>           case IOREQ_TYPE_INVALIDATE:
>> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState
>> *state)
>>               req.data |= ((uint64_t)buf_req->data) << 32;
>>           }
>>
>> -        handle_ioreq(&req);
>> +        handle_ioreq(state, &req);
>>
>>           xen_mb();
>>           state->buffered_io_page->read_pointer += qw ? 2 : 1;
>> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
>>
>>       handle_buffered_iopage(state);
>>       if (req) {
>> -        handle_ioreq(req);
>> +        handle_ioreq(state, req);
>>
>>           if (req->state != STATE_IOREQ_INPROCESS) {
>>               fprintf(stderr, "Badness in I/O request ... not in service?!: "
>>                       "%x, ptr: %x, port: %"PRIx64", "
>> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %"
>> FMT_ioreq_size "\n",
>> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
>> +                    ", size: %" FMT_ioreq_size
>> +                    ", type: %"FMT_ioreq_size"\n",
>>                       req->state, req->data_is_ptr, req->addr,
>> -                    req->data, req->count, req->size);
>> +                    req->data, req->count, req->size, req->type);
>>               destroy_hvm_domain(false);
>>               return;
>>           }
>> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState
>> *state)
>>                                                    state);
>>
>>       if (evtchn_fd != -1) {
>> +        CPUState *cpu_state;
>> +
>> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
>> +        CPU_FOREACH(cpu_state) {
>> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
>> +                    __func__, cpu_state->cpu_index, cpu_state);
>> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
>> +        }
>>           qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
>>       }
>>   }
>> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t
>> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>>                    errno, xen_xc);
>>       }
>>
>> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
>> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
>> +        state->shared_vmport_page =
>> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
>> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
>> +        if (state->shared_vmport_page == NULL) {
>> +            hw_error("map shared vmport IO page returned error %d handle="
>> +                     XC_INTERFACE_FMT, errno, xen_xc);
>> +        }
>> +    }
>> +
>>       xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN,
>> &ioreq_pfn);
>>       DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
>>       state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid,
>> XC_PAGE_SIZE,
>> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t
>> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>>           hw_error("map buffered IO page returned error %d", errno);
>>       }
>>
>> +    /* Note: cpus is empty at this point in init */
>> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
>> +
>>       state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
>>
>>       /* FIXME: how about if we overflow the page here? */
>> --
>> 1.8.4

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Paul Durrant]

> -----Original Message-----
> From: Don Slutz [mailto:dslutz@verizon.com]
> Sent: 17 October 2014 14:31
> To: Paul Durrant; Don Slutz; qemu-devel@nongnu.org
> Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
> Anthony Liguori; Marcel Apfelbaum; Markus Armbruster; Michael S. Tsirkin;
> Stefano Stabellini
> Subject: Re: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to
> vmport
> 
> On 10/17/14 08:34, Paul Durrant wrote:
> >> -----Original Message-----
> >> From: Don Slutz [mailto:dslutz@verizon.com]
> >> Sent: 17 October 2014 13:14
> >> To: qemu-devel@nongnu.org; Paul Durrant
> >> Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
> >> Anthony Liguori; Don Slutz; Marcel Apfelbaum; Markus Armbruster;
> Michael
> >> S. Tsirkin; Stefano Stabellini
> >> Subject: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport
> >>
> >> This adds synchronisation of the 6 vcpu registers (only 32bits of
> >> them) that vmport.c needs between Xen and QEMU.
> >>
> >> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
> >> fetch and put these 6 vcpu registers used by the code in vmport.c
> >> and vmmouse.c
> >>
> >> The registers are passed in the new shared page provided by
> >> HVM_PARAM_VMPORT_IOREQ_PFN.
> >>
> >> Add new array to XenIOState that allows selection of current_cpu by
> >> ioreq_id.
> >>
> >> Now pass XenIOState to handle_ioreq().
> >>
> >> Add new routines regs_to_cpu(), regs_from_cpu(), and
> >> handle_vmport_ioreq().
> >>
> >> Signed-off-by: Don Slutz <dslutz@verizon.com>
> >> ---
> >>   include/hw/xen/xen_common.h |  15 ++++++
> >>   xen-hvm.c                   | 121
> >> ++++++++++++++++++++++++++++++++++++++++++--
> >>   2 files changed, 131 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/include/hw/xen/xen_common.h
> >> b/include/hw/xen/xen_common.h
> >> index 07731b9..9542756 100644
> >> --- a/include/hw/xen/xen_common.h
> >> +++ b/include/hw/xen/xen_common.h
> >> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
> >>   /* shutdown/destroy current domain because of an error */
> >>   void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1,
> 2);
> >>
> >> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
> >> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> >> +                                           unsigned long *vmport_ioreq_pfn)
> >> +{
> >> +    return xc_get_hvm_param(xc, dom,
> >> HVM_PARAM_VMPORT_IOREQ_PFN,
> >> +                            vmport_ioreq_pfn);
> >> +}
> >> +#else
> >> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> >> +                                           unsigned long *vmport_ioreq_pfn)
> >> +{
> >> +    return -ENOSYS;
> >> +}
> >> +#endif
> >> +
> >>   #endif /* QEMU_HW_XEN_COMMON_H */
> >> diff --git a/xen-hvm.c b/xen-hvm.c
> >> index 05e522c..17b3cbd 100644
> >> --- a/xen-hvm.c
> >> +++ b/xen-hvm.c
> >> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
> >>   static bool xen_in_migration;
> >>
> >>   /* Compatibility with older version */
> >> +
> >> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
> >> + * installed.  This here (not in hw/xen/xen_common.h) because
> >> xen/hvm/ioreq.h
> >> + * needs to be included before this block and hw/xen/xen_common.h
> >> needs to
> >> + * be included before xen/hvm/ioreq.h
> >> + */
> >> +#ifndef IOREQ_TYPE_VMWARE_PORT
> >> +#define IOREQ_TYPE_VMWARE_PORT  3
> >> +struct vmware_ioreq {
> >> +    uint32_t esi;
> >> +    uint32_t edi;
> >> +    uint32_t ebx;
> >> +    uint32_t ecx;
> >> +    uint32_t edx;
> >> +};
> >> +typedef struct vmware_ioreq vmware_ioreq_t;
> > This struct is not really a request any more. Maybe vmware_regs_t?
> 
> Sure.
> 
> >> +
> >> +struct shared_vmport_iopage {
> >> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
> >> +};
> >> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
> >> +#endif
> >> +
> >>   #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
> >>   static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page,
> int i)
> >>   {
> >> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
> >>
> >>   typedef struct XenIOState {
> >>       shared_iopage_t *shared_page;
> >> +    shared_vmport_iopage_t *shared_vmport_page;
> >>       buffered_iopage_t *buffered_io_page;
> >>       QEMUTimer *buffered_io_timer;
> >> +    CPUState **cpu_by_ioreq_id;
> > This name implies the array is indexed by an id carries in the ioreq. That
> doesn't seem to be the case; it's just indexed by guest vcpu id isn't it?
> 
> Yes.  How about cpu_by_vcpu_id ?

Ok.

> 
> >>       /* the evtchn port for polling the notification, */
> >>       evtchn_port_t *ioreq_local_port;
> >>       /* evtchn local port for buffered io */
> >> @@ -101,6 +126,8 @@ typedef struct XenIOState {
> >>       Notifier wakeup;
> >>   } XenIOState;
> >>
> >> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
> >> +
> >>   /* Xen specific function for piix pci */
> >>
> >>   int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
> >> @@ -610,6 +637,20 @@ static ioreq_t
> >> *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
> >>       return req;
> >>   }
> >>
> >> +/* get the vmport ioreq packets from share mem */
> >> +static vmware_ioreq_t
> *cpu_get_vmport_ioreq_from_shared_memory(
> > Is this worth its own function? I can only see one use of it in
> handle_vmport_ioreq below...
> 
> Will move it in-line.  I was just following
> cpu_get_ioreq_from_shared_memory which also has 1 use.
> 
> >> +    XenIOState *state, int vcpu)
> >> +{
> >> +    vmware_ioreq_t *vmport_req;
> >> +
> >> +    assert(state->shared_vmport_page);
> >> +    vmport_req = &state->shared_vmport_page-
> >>> vcpu_vmport_ioreq[vcpu];
> >> +
> >> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
> > I don't think that comment is applicable here, and I don't think you need
> the barrier anyway.
> 
> I was not sure either way, so went with the safe way.  Will drop.
> 
> >> +
> >> +    return vmport_req;
> >> +}
> >> +
> >>   /* use poll to get the port notification */
> >>   /* ioreq_vec--out,the */
> >>   /* retval--the number of ioreq packet */
> >> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
> >>       }
> >>   }
> >>
> >> -static void handle_ioreq(ioreq_t *req)
> >> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t
> *vmport_req,
> >> +                        ioreq_t *req)
> >> +{
> >> +    X86CPU *cpu;
> >> +    CPUX86State *env;
> >> +
> >> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
> >> +    cpu = X86_CPU(current_cpu);
> >> +    env = &cpu->env;
> >> +    env->regs[R_EAX] = req->data;
> >> +    env->regs[R_EBX] = vmport_req->ebx;
> >> +    env->regs[R_ECX] = vmport_req->ecx;
> >> +    env->regs[R_EDX] = vmport_req->edx;
> >> +    env->regs[R_ESI] = vmport_req->esi;
> >> +    env->regs[R_EDI] = vmport_req->edi;
> >> +}
> >> +
> >> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t
> >> *vmport_req,
> >> +                          ioreq_t *req)
> >> +{
> >> +    X86CPU *cpu = X86_CPU(current_cpu);
> >> +    CPUX86State *env = &cpu->env;
> >> +
> >> +    assert(sizeof(*vmport_req) <= sizeof(*req));
> > Why?
> 
> Since there is 1 page of ioreq_t indexed by vcpu id this checks that
> the 1 page of vmport data has at least the same number of vcpu
> slots.

Does it? It looks like it checks that the size of the vmware_ioreq_t structure is less than or equal to the size of the ioreq_t structure to me. I see nothing checking number of array elements.

  Paul

>  This prevents a bad memory read.  Looks to not be the right place
> for this, will move.
> 
> >> +
> >> +    vmport_req->ebx = env->regs[R_EBX];
> >> +    vmport_req->ecx = env->regs[R_ECX];
> >> +    vmport_req->edx = env->regs[R_EDX];
> >> +    vmport_req->esi = env->regs[R_ESI];
> >> +    vmport_req->edi = env->regs[R_EDI];
> >> +    current_cpu = NULL;
> >> +}
> >> +
> >> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
> >> +{
> >> +    vmware_ioreq_t *vmport_req =
> >> +        cpu_get_vmport_ioreq_from_shared_memory(state, state-
> >>> send_vcpu);
> >> +
> >> +    regs_to_cpu(state, vmport_req, req);
> >> +    cpu_ioreq_pio(req);
> >> +    regs_from_cpu(state, vmport_req, req);
> > state appears to be unused in this function, and you could avoid passing it
> to both of them by setting current_cpu here couldn't you?
> 
> Yes, will move state code here.
> 
>     -Don Slutz
> 
> >    Paul
> >
> >> +}
> >> +
> >> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
> >>   {
> >>       if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
> >>               (req->size < sizeof (target_ulong))) {
> >> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
> >>           case IOREQ_TYPE_COPY:
> >>               cpu_ioreq_move(req);
> >>               break;
> >> +        case IOREQ_TYPE_VMWARE_PORT:
> >> +            handle_vmport_ioreq(state, req);
> >> +            break;
> >>           case IOREQ_TYPE_TIMEOFFSET:
> >>               break;
> >>           case IOREQ_TYPE_INVALIDATE:
> >> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState
> >> *state)
> >>               req.data |= ((uint64_t)buf_req->data) << 32;
> >>           }
> >>
> >> -        handle_ioreq(&req);
> >> +        handle_ioreq(state, &req);
> >>
> >>           xen_mb();
> >>           state->buffered_io_page->read_pointer += qw ? 2 : 1;
> >> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
> >>
> >>       handle_buffered_iopage(state);
> >>       if (req) {
> >> -        handle_ioreq(req);
> >> +        handle_ioreq(state, req);
> >>
> >>           if (req->state != STATE_IOREQ_INPROCESS) {
> >>               fprintf(stderr, "Badness in I/O request ... not in service?!: "
> >>                       "%x, ptr: %x, port: %"PRIx64", "
> >> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %"
> >> FMT_ioreq_size "\n",
> >> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
> >> +                    ", size: %" FMT_ioreq_size
> >> +                    ", type: %"FMT_ioreq_size"\n",
> >>                       req->state, req->data_is_ptr, req->addr,
> >> -                    req->data, req->count, req->size);
> >> +                    req->data, req->count, req->size, req->type);
> >>               destroy_hvm_domain(false);
> >>               return;
> >>           }
> >> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState
> >> *state)
> >>                                                    state);
> >>
> >>       if (evtchn_fd != -1) {
> >> +        CPUState *cpu_state;
> >> +
> >> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
> >> +        CPU_FOREACH(cpu_state) {
> >> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
> >> +                    __func__, cpu_state->cpu_index, cpu_state);
> >> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
> >> +        }
> >>           qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
> >>       }
> >>   }
> >> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t
> >> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
> >>                    errno, xen_xc);
> >>       }
> >>
> >> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
> >> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
> >> +        state->shared_vmport_page =
> >> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
> >> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
> >> +        if (state->shared_vmport_page == NULL) {
> >> +            hw_error("map shared vmport IO page returned error %d
> handle="
> >> +                     XC_INTERFACE_FMT, errno, xen_xc);
> >> +        }
> >> +    }
> >> +
> >>       xc_get_hvm_param(xen_xc, xen_domid,
> HVM_PARAM_BUFIOREQ_PFN,
> >> &ioreq_pfn);
> >>       DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
> >>       state->buffered_io_page = xc_map_foreign_range(xen_xc,
> xen_domid,
> >> XC_PAGE_SIZE,
> >> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t
> >> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
> >>           hw_error("map buffered IO page returned error %d", errno);
> >>       }
> >>
> >> +    /* Note: cpus is empty at this point in init */
> >> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
> >> +
> >>       state->ioreq_local_port = g_malloc0(max_cpus * sizeof
> (evtchn_port_t));
> >>
> >>       /* FIXME: how about if we overflow the page here? */
> >> --
> >> 1.8.4

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Paul Durrant]

> -----Original Message-----
> From: qemu-devel-bounces+paul.durrant=citrix.com@nongnu.org
> [mailto:qemu-devel-bounces+paul.durrant=citrix.com@nongnu.org] On
> Behalf Of Paul Durrant
> Sent: 17 October 2014 14:36
> To: Don Slutz; qemu-devel@nongnu.org
> Cc: xen-devel@lists.xensource.com; Marcel Apfelbaum; Michael S. Tsirkin;
> Markus Armbruster; Alexander Graf; Stefano Stabellini; Anthony Liguori;
> Andreas Färber
> Subject: Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen
> access to vmport
> 
> > -----Original Message-----
> > From: Don Slutz [mailto:dslutz@verizon.com]
> > Sent: 17 October 2014 14:31
> > To: Paul Durrant; Don Slutz; qemu-devel@nongnu.org
> > Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
> > Anthony Liguori; Marcel Apfelbaum; Markus Armbruster; Michael S. Tsirkin;
> > Stefano Stabellini
> > Subject: Re: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to
> > vmport
> >
> > On 10/17/14 08:34, Paul Durrant wrote:
> > >> -----Original Message-----
> > >> From: Don Slutz [mailto:dslutz@verizon.com]
> > >> Sent: 17 October 2014 13:14
> > >> To: qemu-devel@nongnu.org; Paul Durrant
> > >> Cc: xen-devel@lists.xensource.com; Alexander Graf; Andreas Färber;
> > >> Anthony Liguori; Don Slutz; Marcel Apfelbaum; Markus Armbruster;
> > Michael
> > >> S. Tsirkin; Stefano Stabellini
> > >> Subject: [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to
> vmport
> > >>
> > >> This adds synchronisation of the 6 vcpu registers (only 32bits of
> > >> them) that vmport.c needs between Xen and QEMU.
> > >>
> > >> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
> > >> fetch and put these 6 vcpu registers used by the code in vmport.c
> > >> and vmmouse.c
> > >>
> > >> The registers are passed in the new shared page provided by
> > >> HVM_PARAM_VMPORT_IOREQ_PFN.
> > >>
> > >> Add new array to XenIOState that allows selection of current_cpu by
> > >> ioreq_id.
> > >>
> > >> Now pass XenIOState to handle_ioreq().
> > >>
> > >> Add new routines regs_to_cpu(), regs_from_cpu(), and
> > >> handle_vmport_ioreq().
> > >>
> > >> Signed-off-by: Don Slutz <dslutz@verizon.com>
> > >> ---
> > >>   include/hw/xen/xen_common.h |  15 ++++++
> > >>   xen-hvm.c                   | 121
> > >> ++++++++++++++++++++++++++++++++++++++++++--
> > >>   2 files changed, 131 insertions(+), 5 deletions(-)
> > >>
> > >> diff --git a/include/hw/xen/xen_common.h
> > >> b/include/hw/xen/xen_common.h
> > >> index 07731b9..9542756 100644
> > >> --- a/include/hw/xen/xen_common.h
> > >> +++ b/include/hw/xen/xen_common.h
> > >> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
> > >>   /* shutdown/destroy current domain because of an error */
> > >>   void xen_shutdown_fatal_error(const char *fmt, ...)
> GCC_FMT_ATTR(1,
> > 2);
> > >>
> > >> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
> > >> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> > >> +                                           unsigned long *vmport_ioreq_pfn)
> > >> +{
> > >> +    return xc_get_hvm_param(xc, dom,
> > >> HVM_PARAM_VMPORT_IOREQ_PFN,
> > >> +                            vmport_ioreq_pfn);
> > >> +}
> > >> +#else
> > >> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> > >> +                                           unsigned long *vmport_ioreq_pfn)
> > >> +{
> > >> +    return -ENOSYS;
> > >> +}
> > >> +#endif
> > >> +
> > >>   #endif /* QEMU_HW_XEN_COMMON_H */
> > >> diff --git a/xen-hvm.c b/xen-hvm.c
> > >> index 05e522c..17b3cbd 100644
> > >> --- a/xen-hvm.c
> > >> +++ b/xen-hvm.c
> > >> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
> > >>   static bool xen_in_migration;
> > >>
> > >>   /* Compatibility with older version */
> > >> +
> > >> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
> > >> + * installed.  This here (not in hw/xen/xen_common.h) because
> > >> xen/hvm/ioreq.h
> > >> + * needs to be included before this block and hw/xen/xen_common.h
> > >> needs to
> > >> + * be included before xen/hvm/ioreq.h
> > >> + */
> > >> +#ifndef IOREQ_TYPE_VMWARE_PORT
> > >> +#define IOREQ_TYPE_VMWARE_PORT  3
> > >> +struct vmware_ioreq {
> > >> +    uint32_t esi;
> > >> +    uint32_t edi;
> > >> +    uint32_t ebx;
> > >> +    uint32_t ecx;
> > >> +    uint32_t edx;
> > >> +};
> > >> +typedef struct vmware_ioreq vmware_ioreq_t;
> > > This struct is not really a request any more. Maybe vmware_regs_t?
> >
> > Sure.
> >
> > >> +
> > >> +struct shared_vmport_iopage {
> > >> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
> > >> +};
> > >> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
> > >> +#endif
> > >> +
> > >>   #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
> > >>   static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page,
> > int i)
> > >>   {
> > >> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
> > >>
> > >>   typedef struct XenIOState {
> > >>       shared_iopage_t *shared_page;
> > >> +    shared_vmport_iopage_t *shared_vmport_page;
> > >>       buffered_iopage_t *buffered_io_page;
> > >>       QEMUTimer *buffered_io_timer;
> > >> +    CPUState **cpu_by_ioreq_id;
> > > This name implies the array is indexed by an id carries in the ioreq. That
> > doesn't seem to be the case; it's just indexed by guest vcpu id isn't it?
> >
> > Yes.  How about cpu_by_vcpu_id ?
> 
> Ok.
> 
> >
> > >>       /* the evtchn port for polling the notification, */
> > >>       evtchn_port_t *ioreq_local_port;
> > >>       /* evtchn local port for buffered io */
> > >> @@ -101,6 +126,8 @@ typedef struct XenIOState {
> > >>       Notifier wakeup;
> > >>   } XenIOState;
> > >>
> > >> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
> > >> +
> > >>   /* Xen specific function for piix pci */
> > >>
> > >>   int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
> > >> @@ -610,6 +637,20 @@ static ioreq_t
> > >> *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
> > >>       return req;
> > >>   }
> > >>
> > >> +/* get the vmport ioreq packets from share mem */
> > >> +static vmware_ioreq_t
> > *cpu_get_vmport_ioreq_from_shared_memory(
> > > Is this worth its own function? I can only see one use of it in
> > handle_vmport_ioreq below...
> >
> > Will move it in-line.  I was just following
> > cpu_get_ioreq_from_shared_memory which also has 1 use.
> >
> > >> +    XenIOState *state, int vcpu)
> > >> +{
> > >> +    vmware_ioreq_t *vmport_req;
> > >> +
> > >> +    assert(state->shared_vmport_page);
> > >> +    vmport_req = &state->shared_vmport_page-
> > >>> vcpu_vmport_ioreq[vcpu];
> > >> +
> > >> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
> > > I don't think that comment is applicable here, and I don't think you need
> > the barrier anyway.
> >
> > I was not sure either way, so went with the safe way.  Will drop.
> >
> > >> +
> > >> +    return vmport_req;
> > >> +}
> > >> +
> > >>   /* use poll to get the port notification */
> > >>   /* ioreq_vec--out,the */
> > >>   /* retval--the number of ioreq packet */
> > >> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
> > >>       }
> > >>   }
> > >>
> > >> -static void handle_ioreq(ioreq_t *req)
> > >> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t
> > *vmport_req,
> > >> +                        ioreq_t *req)
> > >> +{
> > >> +    X86CPU *cpu;
> > >> +    CPUX86State *env;
> > >> +
> > >> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
> > >> +    cpu = X86_CPU(current_cpu);
> > >> +    env = &cpu->env;
> > >> +    env->regs[R_EAX] = req->data;
> > >> +    env->regs[R_EBX] = vmport_req->ebx;
> > >> +    env->regs[R_ECX] = vmport_req->ecx;
> > >> +    env->regs[R_EDX] = vmport_req->edx;
> > >> +    env->regs[R_ESI] = vmport_req->esi;
> > >> +    env->regs[R_EDI] = vmport_req->edi;
> > >> +}
> > >> +
> > >> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t
> > >> *vmport_req,
> > >> +                          ioreq_t *req)
> > >> +{
> > >> +    X86CPU *cpu = X86_CPU(current_cpu);
> > >> +    CPUX86State *env = &cpu->env;
> > >> +
> > >> +    assert(sizeof(*vmport_req) <= sizeof(*req));
> > > Why?
> >
> > Since there is 1 page of ioreq_t indexed by vcpu id this checks that
> > the 1 page of vmport data has at least the same number of vcpu
> > slots.
> 
> Does it? It looks like it checks that the size of the vmware_ioreq_t structure is
> less than or equal to the size of the ioreq_t structure to me. I see nothing
> checking number of array elements.
> 

Oh, I now realize you mean the same theoretical rather than actual limit, in which case this can be a build time check anyway.

  Paul

>   Paul
> 
> >  This prevents a bad memory read.  Looks to not be the right place
> > for this, will move.
> >
> > >> +
> > >> +    vmport_req->ebx = env->regs[R_EBX];
> > >> +    vmport_req->ecx = env->regs[R_ECX];
> > >> +    vmport_req->edx = env->regs[R_EDX];
> > >> +    vmport_req->esi = env->regs[R_ESI];
> > >> +    vmport_req->edi = env->regs[R_EDI];
> > >> +    current_cpu = NULL;
> > >> +}
> > >> +
> > >> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
> > >> +{
> > >> +    vmware_ioreq_t *vmport_req =
> > >> +        cpu_get_vmport_ioreq_from_shared_memory(state, state-
> > >>> send_vcpu);
> > >> +
> > >> +    regs_to_cpu(state, vmport_req, req);
> > >> +    cpu_ioreq_pio(req);
> > >> +    regs_from_cpu(state, vmport_req, req);
> > > state appears to be unused in this function, and you could avoid passing it
> > to both of them by setting current_cpu here couldn't you?
> >
> > Yes, will move state code here.
> >
> >     -Don Slutz
> >
> > >    Paul
> > >
> > >> +}
> > >> +
> > >> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
> > >>   {
> > >>       if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
> > >>               (req->size < sizeof (target_ulong))) {
> > >> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
> > >>           case IOREQ_TYPE_COPY:
> > >>               cpu_ioreq_move(req);
> > >>               break;
> > >> +        case IOREQ_TYPE_VMWARE_PORT:
> > >> +            handle_vmport_ioreq(state, req);
> > >> +            break;
> > >>           case IOREQ_TYPE_TIMEOFFSET:
> > >>               break;
> > >>           case IOREQ_TYPE_INVALIDATE:
> > >> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState
> > >> *state)
> > >>               req.data |= ((uint64_t)buf_req->data) << 32;
> > >>           }
> > >>
> > >> -        handle_ioreq(&req);
> > >> +        handle_ioreq(state, &req);
> > >>
> > >>           xen_mb();
> > >>           state->buffered_io_page->read_pointer += qw ? 2 : 1;
> > >> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
> > >>
> > >>       handle_buffered_iopage(state);
> > >>       if (req) {
> > >> -        handle_ioreq(req);
> > >> +        handle_ioreq(state, req);
> > >>
> > >>           if (req->state != STATE_IOREQ_INPROCESS) {
> > >>               fprintf(stderr, "Badness in I/O request ... not in service?!: "
> > >>                       "%x, ptr: %x, port: %"PRIx64", "
> > >> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %"
> > >> FMT_ioreq_size "\n",
> > >> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
> > >> +                    ", size: %" FMT_ioreq_size
> > >> +                    ", type: %"FMT_ioreq_size"\n",
> > >>                       req->state, req->data_is_ptr, req->addr,
> > >> -                    req->data, req->count, req->size);
> > >> +                    req->data, req->count, req->size, req->type);
> > >>               destroy_hvm_domain(false);
> > >>               return;
> > >>           }
> > >> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState
> > >> *state)
> > >>                                                    state);
> > >>
> > >>       if (evtchn_fd != -1) {
> > >> +        CPUState *cpu_state;
> > >> +
> > >> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
> > >> +        CPU_FOREACH(cpu_state) {
> > >> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
> > >> +                    __func__, cpu_state->cpu_index, cpu_state);
> > >> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
> > >> +        }
> > >>           qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL,
> state);
> > >>       }
> > >>   }
> > >> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t
> > >> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
> > >>                    errno, xen_xc);
> > >>       }
> > >>
> > >> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
> > >> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
> > >> +        state->shared_vmport_page =
> > >> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
> > >> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
> > >> +        if (state->shared_vmport_page == NULL) {
> > >> +            hw_error("map shared vmport IO page returned error %d
> > handle="
> > >> +                     XC_INTERFACE_FMT, errno, xen_xc);
> > >> +        }
> > >> +    }
> > >> +
> > >>       xc_get_hvm_param(xen_xc, xen_domid,
> > HVM_PARAM_BUFIOREQ_PFN,
> > >> &ioreq_pfn);
> > >>       DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
> > >>       state->buffered_io_page = xc_map_foreign_range(xen_xc,
> > xen_domid,
> > >> XC_PAGE_SIZE,
> > >> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t
> > >> *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
> > >>           hw_error("map buffered IO page returned error %d", errno);
> > >>       }
> > >>
> > >> +    /* Note: cpus is empty at this point in init */
> > >> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState
> *));
> > >> +
> > >>       state->ioreq_local_port = g_malloc0(max_cpus * sizeof
> > (evtchn_port_t));
> > >>
> > >>       /* FIXME: how about if we overflow the page here? */
> > >> --
> > >> 1.8.4
> 

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Stefano Stabellini]

On Fri, 17 Oct 2014, Don Slutz wrote:
> This adds synchronisation of the 6 vcpu registers (only 32bits of
> them) that vmport.c needs between Xen and QEMU.
> 
> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
> fetch and put these 6 vcpu registers used by the code in vmport.c
> and vmmouse.c
> 
> The registers are passed in the new shared page provided by
> HVM_PARAM_VMPORT_IOREQ_PFN.
> 
> Add new array to XenIOState that allows selection of current_cpu by
> ioreq_id.
> 
> Now pass XenIOState to handle_ioreq().
> 
> Add new routines regs_to_cpu(), regs_from_cpu(), and
> handle_vmport_ioreq().
> 
> Signed-off-by: Don Slutz <dslutz@verizon.com>
> ---
>  include/hw/xen/xen_common.h |  15 ++++++
>  xen-hvm.c                   | 121 ++++++++++++++++++++++++++++++++++++++++++--
>  2 files changed, 131 insertions(+), 5 deletions(-)
> 
> diff --git a/include/hw/xen/xen_common.h b/include/hw/xen/xen_common.h
> index 07731b9..9542756 100644
> --- a/include/hw/xen/xen_common.h
> +++ b/include/hw/xen/xen_common.h
> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
>  /* shutdown/destroy current domain because of an error */
>  void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>  
> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> +                                           unsigned long *vmport_ioreq_pfn)
> +{
> +    return xc_get_hvm_param(xc, dom, HVM_PARAM_VMPORT_IOREQ_PFN,
> +                            vmport_ioreq_pfn);
> +}
> +#else
> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
> +                                           unsigned long *vmport_ioreq_pfn)
> +{
> +    return -ENOSYS;
> +}
> +#endif
> +
>  #endif /* QEMU_HW_XEN_COMMON_H */
> diff --git a/xen-hvm.c b/xen-hvm.c
> index 05e522c..17b3cbd 100644
> --- a/xen-hvm.c
> +++ b/xen-hvm.c
> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
>  static bool xen_in_migration;
>  
>  /* Compatibility with older version */
> +
> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
> + * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
> + * needs to be included before this block and hw/xen/xen_common.h needs to
> + * be included before xen/hvm/ioreq.h
> + */
> +#ifndef IOREQ_TYPE_VMWARE_PORT
> +#define IOREQ_TYPE_VMWARE_PORT  3
> +struct vmware_ioreq {
> +    uint32_t esi;
> +    uint32_t edi;
> +    uint32_t ebx;
> +    uint32_t ecx;
> +    uint32_t edx;
> +};
> +typedef struct vmware_ioreq vmware_ioreq_t;
> +
> +struct shared_vmport_iopage {
> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
> +};
> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
> +#endif
> +
>  #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
>  static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
>  {
> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
>  
>  typedef struct XenIOState {
>      shared_iopage_t *shared_page;
> +    shared_vmport_iopage_t *shared_vmport_page;
>      buffered_iopage_t *buffered_io_page;
>      QEMUTimer *buffered_io_timer;
> +    CPUState **cpu_by_ioreq_id;
>      /* the evtchn port for polling the notification, */
>      evtchn_port_t *ioreq_local_port;
>      /* evtchn local port for buffered io */
> @@ -101,6 +126,8 @@ typedef struct XenIOState {
>      Notifier wakeup;
>  } XenIOState;
>  
> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
> +
>  /* Xen specific function for piix pci */
>  
>  int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
> @@ -610,6 +637,20 @@ static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
>      return req;
>  }
>  
> +/* get the vmport ioreq packets from share mem */
> +static vmware_ioreq_t *cpu_get_vmport_ioreq_from_shared_memory(
> +    XenIOState *state, int vcpu)
> +{
> +    vmware_ioreq_t *vmport_req;
> +
> +    assert(state->shared_vmport_page);
> +    vmport_req = &state->shared_vmport_page->vcpu_vmport_ioreq[vcpu];
> +
> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
> +
> +    return vmport_req;
> +}
> +
>  /* use poll to get the port notification */
>  /* ioreq_vec--out,the */
>  /* retval--the number of ioreq packet */
> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
>      }
>  }
>  
> -static void handle_ioreq(ioreq_t *req)
> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
> +                        ioreq_t *req)
> +{
> +    X86CPU *cpu;
> +    CPUX86State *env;
> +
> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
> +    cpu = X86_CPU(current_cpu);
> +    env = &cpu->env;
> +    env->regs[R_EAX] = req->data;
> +    env->regs[R_EBX] = vmport_req->ebx;
> +    env->regs[R_ECX] = vmport_req->ecx;
> +    env->regs[R_EDX] = vmport_req->edx;
> +    env->regs[R_ESI] = vmport_req->esi;
> +    env->regs[R_EDI] = vmport_req->edi;
> +}
> +
> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
> +                          ioreq_t *req)
> +{
> +    X86CPU *cpu = X86_CPU(current_cpu);
> +    CPUX86State *env = &cpu->env;
> +
> +    assert(sizeof(*vmport_req) <= sizeof(*req));
> +
> +    vmport_req->ebx = env->regs[R_EBX];
> +    vmport_req->ecx = env->regs[R_ECX];
> +    vmport_req->edx = env->regs[R_EDX];
> +    vmport_req->esi = env->regs[R_ESI];
> +    vmport_req->edi = env->regs[R_EDI];
> +    current_cpu = NULL;
> +}
> +
> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
> +{
> +    vmware_ioreq_t *vmport_req =
> +        cpu_get_vmport_ioreq_from_shared_memory(state, state->send_vcpu);
> +
> +    regs_to_cpu(state, vmport_req, req);
> +    cpu_ioreq_pio(req);
> +    regs_from_cpu(state, vmport_req, req);
> +}
> +
> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
>  {
>      if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
>              (req->size < sizeof (target_ulong))) {
> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
>          case IOREQ_TYPE_COPY:
>              cpu_ioreq_move(req);
>              break;
> +        case IOREQ_TYPE_VMWARE_PORT:
> +            handle_vmport_ioreq(state, req);
> +            break;
>          case IOREQ_TYPE_TIMEOFFSET:
>              break;
>          case IOREQ_TYPE_INVALIDATE:
> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState *state)
>              req.data |= ((uint64_t)buf_req->data) << 32;
>          }
>  
> -        handle_ioreq(&req);
> +        handle_ioreq(state, &req);
>  
>          xen_mb();
>          state->buffered_io_page->read_pointer += qw ? 2 : 1;
> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
>  
>      handle_buffered_iopage(state);
>      if (req) {
> -        handle_ioreq(req);
> +        handle_ioreq(state, req);
>  
>          if (req->state != STATE_IOREQ_INPROCESS) {
>              fprintf(stderr, "Badness in I/O request ... not in service?!: "
>                      "%x, ptr: %x, port: %"PRIx64", "
> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
> +                    ", size: %" FMT_ioreq_size
> +                    ", type: %"FMT_ioreq_size"\n",
>                      req->state, req->data_is_ptr, req->addr,
> -                    req->data, req->count, req->size);
> +                    req->data, req->count, req->size, req->type);
>              destroy_hvm_domain(false);
>              return;
>          }
> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState *state)
>                                                   state);
>  
>      if (evtchn_fd != -1) {
> +        CPUState *cpu_state;
> +
> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
> +        CPU_FOREACH(cpu_state) {
> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
> +                    __func__, cpu_state->cpu_index, cpu_state);
> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
> +        }
>          qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
>      }
>  }
> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>                   errno, xen_xc);
>      }
>  
> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {

I think you should error out if it fails with error != -ENOSYS.


> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
> +        state->shared_vmport_page =
> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
> +        if (state->shared_vmport_page == NULL) {
> +            hw_error("map shared vmport IO page returned error %d handle="
> +                     XC_INTERFACE_FMT, errno, xen_xc);
> +        }
> +    }
> +
>      xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
>      DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
>      state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>          hw_error("map buffered IO page returned error %d", errno);
>      }
>  
> +    /* Note: cpus is empty at this point in init */
> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
> +
>      state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
>  
>      /* FIXME: how about if we overflow the page here? */
> -- 
> 1.8.4
> 

Re: [Qemu-devel] [PATCH v4 1/1] xen-hvm.c: Add support for Xen access to vmport

[Don Slutz]

On 10/17/14 13:41, Stefano Stabellini wrote:
> On Fri, 17 Oct 2014, Don Slutz wrote:
>> This adds synchronisation of the 6 vcpu registers (only 32bits of
>> them) that vmport.c needs between Xen and QEMU.
>>
>> This is to avoid a 2nd and 3rd exchange between QEMU and Xen to
>> fetch and put these 6 vcpu registers used by the code in vmport.c
>> and vmmouse.c
>>
>> The registers are passed in the new shared page provided by
>> HVM_PARAM_VMPORT_IOREQ_PFN.
>>
>> Add new array to XenIOState that allows selection of current_cpu by
>> ioreq_id.
>>
>> Now pass XenIOState to handle_ioreq().
>>
>> Add new routines regs_to_cpu(), regs_from_cpu(), and
>> handle_vmport_ioreq().
>>
>> Signed-off-by: Don Slutz <dslutz@verizon.com>
>> ---
>>   include/hw/xen/xen_common.h |  15 ++++++
>>   xen-hvm.c                   | 121 ++++++++++++++++++++++++++++++++++++++++++--
>>   2 files changed, 131 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/hw/xen/xen_common.h b/include/hw/xen/xen_common.h
>> index 07731b9..9542756 100644
>> --- a/include/hw/xen/xen_common.h
>> +++ b/include/hw/xen/xen_common.h
>> @@ -164,4 +164,19 @@ void destroy_hvm_domain(bool reboot);
>>   /* shutdown/destroy current domain because of an error */
>>   void xen_shutdown_fatal_error(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>>   
>> +#ifdef HVM_PARAM_VMPORT_IOREQ_PFN
>> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
>> +                                           unsigned long *vmport_ioreq_pfn)
>> +{
>> +    return xc_get_hvm_param(xc, dom, HVM_PARAM_VMPORT_IOREQ_PFN,
>> +                            vmport_ioreq_pfn);
>> +}
>> +#else
>> +static inline int xen_get_vmport_ioreq_pfn(XenXC xc, domid_t dom,
>> +                                           unsigned long *vmport_ioreq_pfn)
>> +{
>> +    return -ENOSYS;
>> +}
>> +#endif
>> +
>>   #endif /* QEMU_HW_XEN_COMMON_H */
>> diff --git a/xen-hvm.c b/xen-hvm.c
>> index 05e522c..17b3cbd 100644
>> --- a/xen-hvm.c
>> +++ b/xen-hvm.c
>> @@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
>>   static bool xen_in_migration;
>>   
>>   /* Compatibility with older version */
>> +
>> +/* This allows QEMU to build on a system that has Xen 4.5 or earlier
>> + * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
>> + * needs to be included before this block and hw/xen/xen_common.h needs to
>> + * be included before xen/hvm/ioreq.h
>> + */
>> +#ifndef IOREQ_TYPE_VMWARE_PORT
>> +#define IOREQ_TYPE_VMWARE_PORT  3
>> +struct vmware_ioreq {
>> +    uint32_t esi;
>> +    uint32_t edi;
>> +    uint32_t ebx;
>> +    uint32_t ecx;
>> +    uint32_t edx;
>> +};
>> +typedef struct vmware_ioreq vmware_ioreq_t;
>> +
>> +struct shared_vmport_iopage {
>> +    struct vmware_ioreq vcpu_vmport_ioreq[1];
>> +};
>> +typedef struct shared_vmport_iopage shared_vmport_iopage_t;
>> +#endif
>> +
>>   #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
>>   static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
>>   {
>> @@ -79,8 +102,10 @@ typedef struct XenPhysmap {
>>   
>>   typedef struct XenIOState {
>>       shared_iopage_t *shared_page;
>> +    shared_vmport_iopage_t *shared_vmport_page;
>>       buffered_iopage_t *buffered_io_page;
>>       QEMUTimer *buffered_io_timer;
>> +    CPUState **cpu_by_ioreq_id;
>>       /* the evtchn port for polling the notification, */
>>       evtchn_port_t *ioreq_local_port;
>>       /* evtchn local port for buffered io */
>> @@ -101,6 +126,8 @@ typedef struct XenIOState {
>>       Notifier wakeup;
>>   } XenIOState;
>>   
>> +static void handle_ioreq(XenIOState *state, ioreq_t *req);
>> +
>>   /* Xen specific function for piix pci */
>>   
>>   int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
>> @@ -610,6 +637,20 @@ static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
>>       return req;
>>   }
>>   
>> +/* get the vmport ioreq packets from share mem */
>> +static vmware_ioreq_t *cpu_get_vmport_ioreq_from_shared_memory(
>> +    XenIOState *state, int vcpu)
>> +{
>> +    vmware_ioreq_t *vmport_req;
>> +
>> +    assert(state->shared_vmport_page);
>> +    vmport_req = &state->shared_vmport_page->vcpu_vmport_ioreq[vcpu];
>> +
>> +    xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
>> +
>> +    return vmport_req;
>> +}
>> +
>>   /* use poll to get the port notification */
>>   /* ioreq_vec--out,the */
>>   /* retval--the number of ioreq packet */
>> @@ -773,7 +814,50 @@ static void cpu_ioreq_move(ioreq_t *req)
>>       }
>>   }
>>   
>> -static void handle_ioreq(ioreq_t *req)
>> +static void regs_to_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
>> +                        ioreq_t *req)
>> +{
>> +    X86CPU *cpu;
>> +    CPUX86State *env;
>> +
>> +    current_cpu = state->cpu_by_ioreq_id[state->send_vcpu];
>> +    cpu = X86_CPU(current_cpu);
>> +    env = &cpu->env;
>> +    env->regs[R_EAX] = req->data;
>> +    env->regs[R_EBX] = vmport_req->ebx;
>> +    env->regs[R_ECX] = vmport_req->ecx;
>> +    env->regs[R_EDX] = vmport_req->edx;
>> +    env->regs[R_ESI] = vmport_req->esi;
>> +    env->regs[R_EDI] = vmport_req->edi;
>> +}
>> +
>> +static void regs_from_cpu(XenIOState *state, vmware_ioreq_t *vmport_req,
>> +                          ioreq_t *req)
>> +{
>> +    X86CPU *cpu = X86_CPU(current_cpu);
>> +    CPUX86State *env = &cpu->env;
>> +
>> +    assert(sizeof(*vmport_req) <= sizeof(*req));
>> +
>> +    vmport_req->ebx = env->regs[R_EBX];
>> +    vmport_req->ecx = env->regs[R_ECX];
>> +    vmport_req->edx = env->regs[R_EDX];
>> +    vmport_req->esi = env->regs[R_ESI];
>> +    vmport_req->edi = env->regs[R_EDI];
>> +    current_cpu = NULL;
>> +}
>> +
>> +static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
>> +{
>> +    vmware_ioreq_t *vmport_req =
>> +        cpu_get_vmport_ioreq_from_shared_memory(state, state->send_vcpu);
>> +
>> +    regs_to_cpu(state, vmport_req, req);
>> +    cpu_ioreq_pio(req);
>> +    regs_from_cpu(state, vmport_req, req);
>> +}
>> +
>> +static void handle_ioreq(XenIOState *state, ioreq_t *req)
>>   {
>>       if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
>>               (req->size < sizeof (target_ulong))) {
>> @@ -787,6 +871,9 @@ static void handle_ioreq(ioreq_t *req)
>>           case IOREQ_TYPE_COPY:
>>               cpu_ioreq_move(req);
>>               break;
>> +        case IOREQ_TYPE_VMWARE_PORT:
>> +            handle_vmport_ioreq(state, req);
>> +            break;
>>           case IOREQ_TYPE_TIMEOFFSET:
>>               break;
>>           case IOREQ_TYPE_INVALIDATE:
>> @@ -828,7 +915,7 @@ static int handle_buffered_iopage(XenIOState *state)
>>               req.data |= ((uint64_t)buf_req->data) << 32;
>>           }
>>   
>> -        handle_ioreq(&req);
>> +        handle_ioreq(state, &req);
>>   
>>           xen_mb();
>>           state->buffered_io_page->read_pointer += qw ? 2 : 1;
>> @@ -857,14 +944,16 @@ static void cpu_handle_ioreq(void *opaque)
>>   
>>       handle_buffered_iopage(state);
>>       if (req) {
>> -        handle_ioreq(req);
>> +        handle_ioreq(state, req);
>>   
>>           if (req->state != STATE_IOREQ_INPROCESS) {
>>               fprintf(stderr, "Badness in I/O request ... not in service?!: "
>>                       "%x, ptr: %x, port: %"PRIx64", "
>> -                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
>> +                    "data: %"PRIx64", count: %" FMT_ioreq_size
>> +                    ", size: %" FMT_ioreq_size
>> +                    ", type: %"FMT_ioreq_size"\n",
>>                       req->state, req->data_is_ptr, req->addr,
>> -                    req->data, req->count, req->size);
>> +                    req->data, req->count, req->size, req->type);
>>               destroy_hvm_domain(false);
>>               return;
>>           }
>> @@ -904,6 +993,14 @@ static void xen_main_loop_prepare(XenIOState *state)
>>                                                    state);
>>   
>>       if (evtchn_fd != -1) {
>> +        CPUState *cpu_state;
>> +
>> +        DPRINTF("%s: Init cpu_by_ioreq_id\n", __func__);
>> +        CPU_FOREACH(cpu_state) {
>> +            DPRINTF("%s: cpu_by_ioreq_id[%d]=%p\n",
>> +                    __func__, cpu_state->cpu_index, cpu_state);
>> +            state->cpu_by_ioreq_id[cpu_state->cpu_index] = cpu_state;
>> +        }
>>           qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
>>       }
>>   }
>> @@ -1020,6 +1117,17 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>>                    errno, xen_xc);
>>       }
>>   
>> +    if (!xen_get_vmport_ioreq_pfn(xen_xc, xen_domid, &ioreq_pfn)) {
> I think you should error out if it fails with error != -ENOSYS.
>

Will do.

     -Don Slutz

>> +        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
>> +        state->shared_vmport_page =
>> +            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
>> +                                 PROT_READ|PROT_WRITE, ioreq_pfn);
>> +        if (state->shared_vmport_page == NULL) {
>> +            hw_error("map shared vmport IO page returned error %d handle="
>> +                     XC_INTERFACE_FMT, errno, xen_xc);
>> +        }
>> +    }
>> +
>>       xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
>>       DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
>>       state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
>> @@ -1028,6 +1136,9 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
>>           hw_error("map buffered IO page returned error %d", errno);
>>       }
>>   
>> +    /* Note: cpus is empty at this point in init */
>> +    state->cpu_by_ioreq_id = g_malloc0(max_cpus * sizeof(CPUState *));
>> +
>>       state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
>>   
>>       /* FIXME: how about if we overflow the page here? */
>> -- 
>> 1.8.4
>>