On Thu, 2010-07-29 at 18:54 +0100, Gianni Tedesco wrote:
> Hi,
>
> The interface for PCI hotplug is flexible enough to shoot ones-self in
> the foot. It is possible to try to insert a PCI device in to a slot
> already occupied by a qemu emulated device (NIC, PIIX, ISA-bridge, etc.)
> In this case qemu (wisely) refuses to do the hotplug. Since there is no
> way for a toolstack to query qemu's pci device layout there is no way to
> check for this ahead of time. In this case the toolstack must wait for
> device-model state to change to pci-inserted which never happens.
>
> I propose that when qemu decides not to hot-plug a device that it raise
> the "pci-inserted" status anyway. The tools must then check the
> "parameter" key in xenbus for a non-error string. In other words:
>
> send_command("pci-ins")
> wait_for_device_model("pci-inserted")
> if parameter[0-2] == "0x" {
> /* success */
> }else{
> /* fail */
> }
FYI: The corresponding patch I am proposing is actually this (should
apply cleanly albeit with offsets):
diff -r 414e906dd623 tools/libxl/libxl_pci.c
--- a/tools/libxl/libxl_pci.c Thu Jul 29 17:58:38 2010 +0100
+++ b/tools/libxl/libxl_pci.c Thu Jul 29 18:59:32 2010 +0100
@@ -404,6 +479,12 @@ static int do_pci_add(libxl_ctx *ctx, ui
XL_LOG(ctx, XL_LOG_ERROR, "Device Model didn't respond in time");
path = libxl_sprintf(ctx, "/local/domain/0/device-model/%d/parameter",
domid);
vdevfn = libxl_xs_read(ctx, XBT_NULL, path);
+ if ( strncmp(vdevfn, "0x", 2) ) {
+ XL_LOG(ctx, XL_LOG_ERROR, "qemu refused to add device: %s",
vdevfn);
+ path = libxl_sprintf(ctx, "/local/domain/0/device-model/%d/state",
domid);
+ xs_write(ctx->xsh, XBT_NULL, path, "running", strlen("running"));
+ return ERROR_FAIL;
+ }
sscanf(vdevfn + 2, "%x", &pcidev->vdevfn);
path = libxl_sprintf(ctx, "/local/domain/0/device-model/%d/state",
domid);
xs_write(ctx->xsh, XBT_NULL, path, state, strlen(state));
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home