acrn-hypervisor/hypervisor/common/vm_load.c

228 lines
6.4 KiB
C

/*
* Copyright (C) 2018 Intel Corporation. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <hypervisor.h>
#include <zeropage.h>
static uint32_t create_e820_table(struct e820_entry *_e820)
{
uint32_t i;
ASSERT(e820_entries > 0,
"e820 should be inited");
for (i = 0; i < e820_entries; i++) {
_e820[i].baseaddr = e820[i].baseaddr;
_e820[i].length = e820[i].length;
_e820[i].type = e820[i].type;
}
return e820_entries;
}
static uint64_t create_zero_page(struct vm *vm)
{
struct zero_page *zeropage;
struct sw_linux *sw_linux = &(vm->sw.linux_info);
struct zero_page *hva;
uint64_t gpa;
/* Set zeropage in Linux Guest RAM region just past boot args */
hva = GPA2HVA(vm, (uint64_t)sw_linux->bootargs_load_addr);
zeropage = (struct zero_page *)((char *)hva + MEM_4K);
/* clear the zeropage */
memset(zeropage, 0, MEM_2K);
/* copy part of the header into the zero page */
hva = GPA2HVA(vm, (uint64_t)vm->sw.kernel_info.kernel_load_addr);
memcpy_s(&(zeropage->hdr), sizeof(zeropage->hdr),
&(hva->hdr), sizeof(hva->hdr));
/* See if kernel has a RAM disk */
if (sw_linux->ramdisk_src_addr) {
/* Copy ramdisk load_addr and size in zeropage header structure
*/
zeropage->hdr.ramdisk_addr =
(uint32_t)(uint64_t)sw_linux->ramdisk_load_addr;
zeropage->hdr.ramdisk_size = (uint32_t)sw_linux->ramdisk_size;
}
/* Copy bootargs load_addr in zeropage header structure */
zeropage->hdr.bootargs_addr =
(uint32_t)(uint64_t)sw_linux->bootargs_load_addr;
/* set constant arguments in zero page */
zeropage->hdr.loader_type = 0xff;
zeropage->hdr.load_flags |= (1 << 5); /* quiet */
/* Create/add e820 table entries in zeropage */
zeropage->e820_nentries = create_e820_table(zeropage->e820);
/* Get the host physical address of the zeropage */
gpa = hpa2gpa(vm, HVA2HPA((uint64_t)zeropage));
/* Return Physical Base Address of zeropage */
return gpa;
}
int load_guest(struct vm *vm, struct vcpu *vcpu)
{
int ret = 0;
void *hva;
struct run_context *cur_context =
&vcpu->arch_vcpu.contexts[vcpu->arch_vcpu.cur_context];
uint64_t lowmem_gpa_top;
hva = GPA2HVA(vm, GUEST_CFG_OFFSET);
lowmem_gpa_top = *(uint64_t *)hva;
/* hardcode vcpu entry addr(kernel entry) & rsi (zeropage)*/
memset(cur_context->guest_cpu_regs.longs,
0, sizeof(uint64_t)*NUM_GPRS);
hva = GPA2HVA(vm, lowmem_gpa_top -
MEM_4K - MEM_2K);
vcpu->entry_addr = (void *)(*((uint64_t *)hva));
cur_context->guest_cpu_regs.regs.rsi =
lowmem_gpa_top - MEM_4K;
pr_info("%s, Set config according to predefined offset:",
__func__);
pr_info("VCPU%d Entry: 0x%llx, RSI: 0x%016llx, cr3: 0x%016llx",
vcpu->vcpu_id, vcpu->entry_addr,
cur_context->guest_cpu_regs.regs.rsi,
vm->arch_vm.guest_init_pml4);
return ret;
}
int general_sw_loader(struct vm *vm, struct vcpu *vcpu)
{
int ret = 0;
void *hva;
struct run_context *cur_context =
&vcpu->arch_vcpu.contexts[vcpu->arch_vcpu.cur_context];
char dyn_bootargs[100] = {0};
uint32_t kernel_entry_offset;
struct zero_page *zeropage;
ASSERT(vm != NULL, "Incorrect argument");
pr_dbg("Loading guest to run-time location");
/* FIXME: set config according to predefined offset */
if (!is_vm0(vm))
return load_guest(vm, vcpu);
/* calculate the kernel entry point */
zeropage = (struct zero_page *)
vm->sw.kernel_info.kernel_src_addr;
kernel_entry_offset = (zeropage->hdr.setup_sects + 1) * 512;
if (vcpu->arch_vcpu.cpu_mode == CPU_MODE_64BIT)
/* 64bit entry is the 512bytes after the start */
kernel_entry_offset += 512;
vm->sw.kernel_info.kernel_entry_addr =
(void *)((unsigned long)vm->sw.kernel_info.kernel_load_addr
+ kernel_entry_offset);
if (is_vcpu_bsp(vcpu)) {
/* Set VCPU entry point to kernel entry */
vcpu->entry_addr = vm->sw.kernel_info.kernel_entry_addr;
pr_info("%s, VM *d VCPU %d Entry: 0x%016llx ",
__func__, vm->attr.id, vcpu->vcpu_id, vcpu->entry_addr);
}
/* Calculate the host-physical address where the guest will be loaded */
hva = GPA2HVA(vm, (uint64_t)vm->sw.kernel_info.kernel_load_addr);
/* Copy the guest kernel image to its run-time location */
memcpy_s((void *)hva, vm->sw.kernel_info.kernel_size,
vm->sw.kernel_info.kernel_src_addr,
vm->sw.kernel_info.kernel_size);
/* See if guest is a Linux guest */
if (vm->sw.kernel_type == VM_LINUX_GUEST) {
/* Documentation states: ebx=0, edi=0, ebp=0, esi=ptr to
* zeropage
*/
memset(cur_context->guest_cpu_regs.longs,
0, sizeof(uint64_t) * NUM_GPRS);
/* Get host-physical address for guest bootargs */
hva = GPA2HVA(vm,
(uint64_t)vm->sw.linux_info.bootargs_load_addr);
/* Copy Guest OS bootargs to its load location */
strcpy_s((char *)hva, MEM_2K,
vm->sw.linux_info.bootargs_src_addr);
#ifdef CONFIG_CMA
/* add "cma=XXXXM@0xXXXXXXXX" to cmdline*/
if (is_vm0(vm) && (e820_mem.max_ram_blk_size > 0)) {
snprintf(dyn_bootargs, 100, " cma=%dM@0x%llx",
(e820_mem.max_ram_blk_size >> 20),
e820_mem.max_ram_blk_base);
strcpy_s((char *)hva
+vm->sw.linux_info.bootargs_size,
100, dyn_bootargs);
}
#else
/* add "hugepagesz=1G hugepages=x" to cmdline for 1G hugepage
* reserving. Current strategy is "total_mem_size in Giga -
* remained 1G pages" for reserving.
*/
if (is_vm0(vm) && check_mmu_1gb_support(PTT_HOST)) {
int reserving_1g_pages;
#ifdef CONFIG_REMAIN_1G_PAGES
reserving_1g_pages = (e820_mem.total_mem_size >> 30) -
CONFIG_REMAIN_1G_PAGES;
#else
reserving_1g_pages = (e820_mem.total_mem_size >> 30) -
3;
#endif
if (reserving_1g_pages > 0) {
snprintf(dyn_bootargs, 100,
" hugepagesz=1G hugepages=%d",
reserving_1g_pages);
strcpy_s((char *)hva
+vm->sw.linux_info.bootargs_size,
100, dyn_bootargs);
}
}
#endif
/* Check if a RAM disk is present with Linux guest */
if (vm->sw.linux_info.ramdisk_src_addr) {
/* Get host-physical address for guest RAM disk */
hva = GPA2HVA(vm,
(uint64_t)vm->sw.linux_info.ramdisk_load_addr);
/* Copy RAM disk to its load location */
memcpy_s((void *)hva, vm->sw.linux_info.ramdisk_size,
vm->sw.linux_info.ramdisk_src_addr,
vm->sw.linux_info.ramdisk_size);
}
/* Create Zeropage and copy Physical Base Address of Zeropage
* in RSI
*/
cur_context->guest_cpu_regs.regs.rsi = create_zero_page(vm);
pr_info("%s, RSI pointing to zero page for VM %d at GPA %X",
__func__, vm->attr.id,
cur_context->guest_cpu_regs.regs.rsi);
} else {
pr_err("%s, Loading VM SW failed", __func__);
ret = -EINVAL;
}
return ret;
}