diff --git a/sys/arch/amd64/amd64/genassym.cf b/sys/arch/amd64/amd64/genassym.cf
index b5c7330a229..facd9216f75 100644
# Add convenient defines for locore.S address manipulation,
# these are the different locations of start_info structure
# and members.
--- a/sys/arch/amd64/amd64/genassym.cf
+++ b/sys/arch/amd64/amd64/genassym.cf
@@ -373,6 +373,7 @@ define BST_TYPE offsetof(struct bus_space_tag, bst_type)
define VM_GUEST_XENPV VM_GUEST_XENPV
define VM_GUEST_XENPVH VM_GUEST_XENPVH
+define VM_GUEST_GENPVH VM_GUEST_GENPVH
ifdef XEN
define CPU_INFO_VCPU offsetof(struct cpu_info, ci_vcpu)
@@ -381,6 +382,12 @@ define SIR_XENIPL_VM SIR_XENIPL_VM
define SIR_XENIPL_SCHED SIR_XENIPL_SCHED
define SIR_XENIPL_HIGH SIR_XENIPL_HIGH
define EVTCHN_UPCALL_MASK offsetof(struct vcpu_info, evtchn_upcall_mask)
+define HVM_START_INFO_SIZE sizeof(struct hvm_start_info)
+define START_INFO_VERSION offsetof(struct hvm_start_info, version)
+define MMAP_PADDR offsetof(struct hvm_start_info, memmap_paddr)
+define MMAP_ENTRIES offsetof(struct hvm_start_info, memmap_entries)
+define MMAP_ENTRY_SIZE sizeof(struct hvm_memmap_table_entry)
+define CMDLINE_PADDR offsetof(struct hvm_start_info, cmdline_paddr)
ifdef XENPV
define XEN_PT_BASE offsetof(struct start_info, pt_base)
define XEN_NR_PT_FRAMES offsetof(struct start_info, nr_pt_frames)
diff --git a/sys/arch/amd64/amd64/locore.S b/sys/arch/amd64/amd64/locore.S
index 089c7388e18..97519bcb6a1 100644
# locore.S expects start_info being passed by the calling hypervisor in
# %ebx to be located at the end of the symbol table. Qemu and Firecracker
# don't follow this rule which is not part of the official Xen ABI
# https://xenbits.xen.org/docs/unstable/misc/pvh.html
# This patch copies the start_info structure and sub-structures where
# they are expected.
# This work has been done in common with Gregory Cavelier <gcavelier@cleia.org>
--- a/sys/arch/amd64/amd64/locore.S
+++ b/sys/arch/amd64/amd64/locore.S
@@ -278,11 +278,12 @@
#ifdef XEN
#define __ASSEMBLY__
+#include <xen/include/public/arch-x86/cpuid.h>
#include <xen/include/public/elfnote.h>
#include <xen/include/public/xen.h>
#define ELFNOTE(name, type, desctype, descdata...) \
-.pushsection .note.name ; \
+.pushsection .note.name, "a", @note ; \
.align 4 ; \
.long 2f - 1f /* namesz */ ; \
.long 4f - 3f /* descsz */ ; \
@@ -306,7 +307,7 @@
ELFNOTE(Xen, XEN_ELFNOTE_ENTRY, .quad, start)
#else
ELFNOTE(Xen, XEN_ELFNOTE_PADDR_OFFSET, .quad, 0)
- ELFNOTE(Xen, XEN_ELFNOTE_PHYS32_ENTRY, .long, RELOC(start_xen32))
+ ELFNOTE(Xen, XEN_ELFNOTE_PHYS32_ENTRY, .long, RELOC(start_genpvh))
#endif /* XENPV */
ELFNOTE(Xen, XEN_ELFNOTE_HYPERCALL_PAGE, .quad, hypercall_page)
ELFNOTE(Xen, XEN_ELFNOTE_HV_START_LOW, .quad, HYPERVISOR_VIRT_START)
@@ -973,7 +974,6 @@ longmode_hi:
movl %eax,_C_LABEL(cpuid_level)
movl $VM_GUEST_XENPV, _C_LABEL(vm_guest)
-
/*
* Initialize cpu_info_primary.ci_self := &cpu_info_primary,
* and initialize some MSRs with
@@ -1035,7 +1035,7 @@ END(start)
# if !defined(XENPV)
/* entry point for Xen PVH */
.code32
-ENTRY(start_xen32)
+ENTRY(start_genpvh)
/* Xen doesn't start us with a valid gdt */
movl $RELOC(gdtdesc32), %eax
lgdt (%eax)
@@ -1051,13 +1051,98 @@ ENTRY(start_xen32)
movl $RELOC(tmpstk),%esp
/* clear BSS */
- xorl %eax,%eax
+ xorl %eax,%eax
movl $RELOC(__bss_start),%edi
movl $RELOC(_end),%ecx
subl %edi,%ecx
rep
stosb
+ /*
+ * Here, we have 2 cases :
+ *
+ * 1) We have been started by Xen
+ * 2) We have been started by another VMM (Qemu, Firecracker, ...)
+ *
+ * The main difference is that, when we are started by Xen,
+ * %ebx (addr of the hvm_start_info structure) is pointing to a
+ * location that will be mapped correctly later.
+ *
+ * In the second case, we have to copy this structure (and all
+ * the information contained in it) to a location that will be
+ * mapped later : __kernel_end
+ *
+ * To distinguish between the 2 cases, we'll use the 'cpuid' instruction
+ */
+
+ push %ebx
+ xorl %eax, %eax
+ cpuid
+ cmpl $0x1, %eax /* Check if we can call CPUID with eax=1 */
+ jb .start_genpvh
+ xorl %eax, %eax
+ inc %eax
+ cpuid
+ shr $31, %ecx
+ testb $1, %cl /* Check if bit 31 of ECX (hypervisor) is set */
+ jz .start_genpvh
+ xorl %eax, %eax
+ inc %eax
+ shl $30, %eax
+ cpuid /* Calling cpuid with eax=0x40000000 */
+ cmp $XEN_CPUID_SIGNATURE_EBX, %ebx /* "VneX" */
+ je .start_xen32
+
+ /* We have been started by a VMM that is *not* Xen */
+
+.start_genpvh:
+
+ /* First, copy the hvm_start_info structure to __kernel_end */
+ pop %ebx
+ movl %ebx, %esi
+ movl $RELOC(__kernel_end), %edi
+ movl $HVM_START_INFO_SIZE, %ecx
+ shrl $2, %ecx
+ rep movsl
+
+ /* Copy cmdline_paddr after hvm_start_info */
+ movl CMDLINE_PADDR(%ebx), %esi
+ movl $RELOC(__kernel_end), %ecx
+ movl %edi, CMDLINE_PADDR(%ecx) /* Set new cmdline_paddr in hvm_start_info */
+ .cmdline_copy:
+ movb (%esi), %al
+ movsb
+ cmp $0, %al
+ jne .cmdline_copy
+
+ /* Copy memmap_paddr after cmdline (only if hvm_start_info->version != 0) */
+ xorl %eax, %eax
+ cmpl START_INFO_VERSION(%ebx), %eax
+ je .reload_ebx
+ movl MMAP_PADDR(%ebx), %esi
+ movl $RELOC(__kernel_end), %ecx
+ movl %edi, MMAP_PADDR(%ecx) /* Set new memmap_paddr in hvm_start_info */
+ movl MMAP_ENTRIES(%ebx), %eax /* Get memmap_entries */
+ movl $MMAP_ENTRY_SIZE, %ebx
+ mull %ebx /* eax * ebx => edx:eax */
+ movl %eax, %ecx
+ shrl $2, %ecx
+ rep movsl
+
+.reload_ebx:
+ movl $RELOC(__kernel_end), %ebx
+
+ /* announce ourself */
+ movl $VM_GUEST_GENPVH, RELOC(vm_guest)
+
+ jmp .save_hvm_start_paddr
+
+.start_xen32:
+ pop %ebx
+ movl $VM_GUEST_XENPVH, RELOC(vm_guest)
+
+.save_hvm_start_paddr:
+
/*
* save addr of the hvm_start_info structure. This is also the end
* of the symbol table
@@ -1069,6 +1154,9 @@ ENTRY(start_xen32)
movl %eax,(%ebp)
movl $KERNBASE_HI,4(%ebp)
/* get a page for HYPERVISOR_shared_info */
+ /* this is only needed if we are running on Xen */
+ cmpl $VM_GUEST_XENPVH, RELOC(vm_guest)
+ jne .add_hvm_start_info_page
addl $PAGE_SIZE, %ebx
addl $PGOFSET,%ebx
andl $~PGOFSET,%ebx
@@ -1076,6 +1164,7 @@ ENTRY(start_xen32)
movl %ebx,(%ebp)
movl $0,4(%ebp)
/* XXX assume hvm_start_info+dependant structure fits in a single page */
+.add_hvm_start_info_page:
addl $PAGE_SIZE, %ebx
addl $PGOFSET,%ebx
andl $~PGOFSET,%ebx
@@ -1084,10 +1173,8 @@ ENTRY(start_xen32)
movl %ebx,(%ebp)
movl $KERNBASE_HI,4(%ebp)
- /* announce ourself */
- movl $VM_GUEST_XENPVH, RELOC(vm_guest)
jmp .Lbiosbasemem_finished
-END(start_xen32)
+END(start_genpvh)
.code64
# endif /* !XENPV */
/* space for the hypercall call page */
@@ -1099,6 +1186,7 @@ ENTRY(hypercall_page) /* Returns -1, on HYPERVISOR_xen_version() */
retq
.align HYPERCALL_PAGE_OFFSET, 0x90
END(hypercall_page)
+
#endif /* XEN */
/*
diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index 375d7a41e7f..f1c8f5c66a6 100644
# The symbol table is not loaded when booting from qemu/Firecracker.
# The ELF headers are not available and we cannot compute a valid value for esym.
# We prefer having a "table missing" than "table invalid" warning by passing
# a 0 sized symsize.
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1525,8 +1525,10 @@ init_x86_64_ksyms(void)
} else {
uintptr_t endp = (uintptr_t)(void *)&end;
- ksyms_addsyms_elf(*(long *)endp,
- ((long *)endp) + 1, esym);
+ if (vm_guest == VM_GUEST_GENPVH)
+ ksyms_addsyms_elf(0, ((long *)endp) + 1, esym);
+ else
+ ksyms_addsyms_elf(*(long *)endp, ((long *)endp) + 1, esym);
}
#endif
}
@@ -1710,7 +1712,7 @@ init_x86_64(paddr_t first_avail)
#endif
#ifdef XEN
- if (vm_guest == VM_GUEST_XENPVH)
+ if (vm_guest == VM_GUEST_XENPVH || vm_guest == VM_GUEST_GENPVH)
xen_parse_cmdline(XEN_PARSE_BOOTFLAGS, NULL);
#endif
init_pte();
diff --git a/sys/arch/x86/acpi/acpi_machdep.c b/sys/arch/x86/acpi/acpi_machdep.c
index 9dbfe49ca5f..b75767fe615 100644
# When booting on GENPVH, ACPI address informations is in start_info
--- a/sys/arch/x86/acpi/acpi_machdep.c
+++ b/sys/arch/x86/acpi/acpi_machdep.c
@@ -158,7 +158,7 @@ out:
}
#else
#ifdef XEN
- if (vm_guest == VM_GUEST_XENPVH) {
+ if (vm_guest == VM_GUEST_XENPVH || vm_guest == VM_GUEST_GENPVH) {
PhysicalAddress = hvm_start_info->rsdp_paddr;
if (PhysicalAddress)
return PhysicalAddress;
diff --git a/sys/arch/x86/include/cpu.h b/sys/arch/x86/include/cpu.h
index 0dae029202a..c82da831157 100644
# Add VM_GUEST_GENPVH as a vm_guest
--- a/sys/arch/x86/include/cpu.h
+++ b/sys/arch/x86/include/cpu.h
@@ -516,6 +516,7 @@ typedef enum vm_guest {
VM_GUEST_VMWARE,
VM_GUEST_KVM,
VM_GUEST_VIRTUALBOX,
+ VM_GUEST_GENPVH,
VM_LAST
} vm_guest_t;
extern vm_guest_t vm_guest;
diff --git a/sys/arch/x86/x86/consinit.c b/sys/arch/x86/x86/consinit.c
index 7a65f1bfa09..f72f872ab54 100644
# GENPVH doesn't need nor use Xen console, but it got its console
# information through the start_info structure which holds the
# parameters passed to the kernel, this patch enables the use of both
# consinfo and xcp.xcp_console.
--- a/sys/arch/x86/x86/consinit.c
+++ b/sys/arch/x86/x86/consinit.c
@@ -171,6 +171,7 @@ consinit(void)
#if (NCOM > 0)
int rv;
#endif
+ char console_devname[16] = "";
#ifdef XENPVHVM
if (vm_guest == VM_GUEST_XENPVH) {
@@ -178,6 +179,13 @@ consinit(void)
return;
/* fallback to native console selection, usefull for dom0 PVH */
}
+ if (vm_guest == VM_GUEST_GENPVH) {
+ union xen_cmdline_parseinfo xcp;
+ /* get console= parameter from generic PVH VMM */
+ xen_parse_cmdline(XEN_PARSE_CONSOLE, &xcp);
+ strncpy(console_devname, xcp.xcp_console,
+ sizeof(console_devname));
+ }
#endif
if (initted)
return;
@@ -188,7 +196,10 @@ consinit(void)
if (!consinfo)
#endif
consinfo = &default_consinfo;
-
+ /* console= parameter was not passed via a generic PVH VMM */
+ if (!console_devname[0])
+ strncpy(console_devname, consinfo->devname,
+ sizeof(console_devname));
#if (NGENFB > 0)
#if defined(XENPVHVM) && defined(DOM0OPS)
if (vm_guest == VM_GUEST_XENPVH && xendomain_is_dom0())
@@ -197,8 +208,7 @@ consinit(void)
#endif /* XENPVHVM */
fbinfo = lookup_bootinfo(BTINFO_FRAMEBUFFER);
#endif
-
- if (!strcmp(consinfo->devname, "pc")) {
+ if (!strcmp(console_devname, "pc")) {
int error;
#if (NGENFB > 0)
if (fbinfo && fbinfo->physaddr > 0) {
@@ -254,7 +264,7 @@ dokbd:
return;
}
#if (NCOM > 0)
- if (!strcmp(consinfo->devname, "com")) {
+ if (!strcmp(console_devname, "com")) {
int addr = consinfo->addr;
int speed = consinfo->speed;
@@ -278,14 +288,14 @@ dokbd:
}
#endif
#if (NNULLCONS > 0)
- if (!strcmp(consinfo->devname, "nullcons")) {
+ if (!strcmp(console_devname, "nullcons")) {
void nullcninit(struct consdev *cn);
nullcninit(0);
return;
}
#endif
- panic("invalid console device %s", consinfo->devname);
+ panic("invalid console device %s", console_devname);
}
#ifdef KGDB
diff --git a/sys/arch/x86/x86/identcpu.c b/sys/arch/x86/x86/identcpu.c
index ec2fa0620b7..9856cf9fb15 100644
# Add VM_GUEST_GENPVH as a valid hypervisor
--- a/sys/arch/x86/x86/identcpu.c
+++ b/sys/arch/x86/x86/identcpu.c
@@ -1044,6 +1044,7 @@ static const struct vm_name_guest vm_bios_vendors[] = {
{ "BHYVE", VM_GUEST_VM }, /* bhyve */
{ "Seabios", VM_GUEST_VM }, /* KVM */
{ "innotek GmbH", VM_GUEST_VIRTUALBOX }, /* Oracle VirtualBox */
+ { "Generic PVH", VM_GUEST_GENPVH}, /* Generic PVH */
};
static const struct vm_name_guest vm_system_products[] = {
@@ -1065,6 +1066,7 @@ identify_hypervisor(void)
switch (vm_guest) {
case VM_GUEST_XENPV:
case VM_GUEST_XENPVH:
+ case VM_GUEST_GENPVH:
/* guest type already known, no bios info */
return;
default:
diff --git a/sys/arch/x86/x86/x86_autoconf.c b/sys/arch/x86/x86/x86_autoconf.c
index fd8688fe9cc..ba4b698d067 100644
# When booting in GENPVH mode, informations about the boot device are
# read from command line parameters in start_info, xen_bootconf must be
# called in order to get the right device.
--- a/sys/arch/x86/x86/x86_autoconf.c
+++ b/sys/arch/x86/x86/x86_autoconf.c
@@ -540,7 +540,7 @@ void
cpu_bootconf(void)
{
#ifdef XEN
- if (vm_guest == VM_GUEST_XENPVH) {
+ if (vm_guest == VM_GUEST_XENPVH || vm_guest == VM_GUEST_GENPVH) {
xen_bootconf();
return;
}
diff --git a/sys/arch/x86/x86/x86_machdep.c b/sys/arch/x86/x86/x86_machdep.c
index fabd76b1155..8a2364b6ecc 100644
# When booting in GENPVH mode, memory mapping information is located in
# start_info, x86_add_xen_clusters() must be called in order to read
# them.
--- a/sys/arch/x86/x86/x86_machdep.c
+++ b/sys/arch/x86/x86/x86_machdep.c
@@ -911,7 +911,7 @@ init_x86_clusters(void)
* the boot program).
*/
#ifdef XEN
- if (vm_guest == VM_GUEST_XENPVH) {
+ if (vm_guest == VM_GUEST_XENPVH || vm_guest == VM_GUEST_GENPVH) {
x86_add_xen_clusters();
}
#endif /* XEN */
diff --git a/sys/arch/xen/xen/hypervisor.c b/sys/arch/xen/xen/hypervisor.c
index 47b7dc3ec0a..7e26b7696a6 100644
# GENPVH doesn't use Xen hypercalls, this patch sets up start_info and
# returns immediately after that.
--- a/sys/arch/xen/xen/hypervisor.c
+++ b/sys/arch/xen/xen/hypervisor.c
@@ -241,10 +241,25 @@ void
init_xen_early(void)
{
const char *cmd_line;
+ if (vm_guest != VM_GUEST_XENPVH && vm_guest != VM_GUEST_GENPVH)
+ return;
+
+ hvm_start_info = (void *)((uintptr_t)hvm_start_paddr + KERNBASE);
+
+ if (hvm_start_info->cmdline_paddr != 0) {
+ cmd_line =
+ (void *)((uintptr_t)hvm_start_info->cmdline_paddr + KERNBASE);
+ strlcpy(xen_start_info.cmd_line, cmd_line,
+ sizeof(xen_start_info.cmd_line));
+ } else {
+ xen_start_info.cmd_line[0] = '\0';
+ }
+ xen_start_info.flags = hvm_start_info->flags;
+
if (vm_guest != VM_GUEST_XENPVH)
return;
+
xen_init_hypercall_page();
- hvm_start_info = (void *)((uintptr_t)hvm_start_paddr + KERNBASE);
HYPERVISOR_shared_info = (void *)((uintptr_t)HYPERVISOR_shared_info_pa + KERNBASE);
struct xen_add_to_physmap xmap = {
@@ -262,15 +277,6 @@ init_xen_early(void)
}
delay_func = x86_delay = xen_delay;
x86_initclock_func = xen_initclocks;
- if (hvm_start_info->cmdline_paddr != 0) {
- cmd_line =
- (void *)((uintptr_t)hvm_start_info->cmdline_paddr + KERNBASE);
- strlcpy(xen_start_info.cmd_line, cmd_line,
- sizeof(xen_start_info.cmd_line));
- } else {
- xen_start_info.cmd_line[0] = '\0';
- }
- xen_start_info.flags = hvm_start_info->flags;
}