HV: Make AP trampoline code relocatable

V3->V4: Updated function/variable names for accurancy
V2->V3: Changed a few function/variable names to make it less confusing
V1->V2: removed the unneccesary cache flushing

- For UEFI boot, allocate memory for trampoline code in ACRN EFI,
  and pass the pointer to HV through efi_ctx
- For other boot, scan E820 to allocate memory in HV run time
- update_trampoline_code_refs() updates all the references that need the
  absolute PA with the actual load address

Signed-off-by: Zheng, Gen <gen.zheng@intel.com>
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong>
Acked-by: Xu, Anthony <anthony.xu@intel.com>

Author: ZideChen0

hypervisor/arch/x86/cpu.c

@@ -30,6 +30,7 @@
 
 #include <hypervisor.h>
 #include <hv_lib.h>
+#include <acrn_hv_defs.h>
 #include <acrn_common.h>
 #include <bsp_extern.h>
 #include <hv_arch.h>
@@ -38,9 +39,12 @@
 #include <hv_debug.h>
 
 #ifdef CONFIG_EFI_STUB
+#include <acrn_efi.h>
 extern uint32_t efi_physical_available_ap_bitmap;
 #endif
 
+uint64_t trampoline_code_paddr = 0;
+
 spinlock_t cpu_secondary_spinlock = {
 	.head = 0,
 	.tail = 0
@@ -88,6 +92,7 @@ int cpu_find_logical_id(uint32_t lapic_id);
 #ifndef CONFIG_EFI_STUB
 static void start_cpus(void);
 #endif
+static uint64_t relocate_trampoline_code(void);
 static void pcpu_sync_sleep(unsigned long *sync, int mask_bit);
 int ibrs_type;
 
@@ -548,10 +553,7 @@ void bsp_boot_init(void)
 	/* Trigger event to allow secondary CPUs to continue */
 	bitmap_set(0, &pcpu_sync);
 #else
-	memcpy_s(_ld_cpu_secondary_reset_start,
-		(unsigned long)&_ld_cpu_secondary_reset_size,
-		_ld_cpu_secondary_reset_load,
-		(unsigned long)&_ld_cpu_secondary_reset_size);
+	relocate_trampoline_code();
 #endif
 
 	ASSERT(get_cpu_id() == CPU_BOOT_ID, "");
@@ -660,12 +662,9 @@ static void start_cpus()
 {
 	uint32_t timeout;
 	uint32_t expected_up;
+	uint64_t startup_paddr;
 
-	/*Copy segment for AP initialization code below 1MB */
-	memcpy_s(_ld_cpu_secondary_reset_start,
-		(unsigned long)&_ld_cpu_secondary_reset_size,
-		_ld_cpu_secondary_reset_load,
-		(unsigned long)&_ld_cpu_secondary_reset_size);
+	startup_paddr = relocate_trampoline_code();
 
 	/* Set flag showing number of CPUs expected to be up to all
 	 * cpus
@@ -674,7 +673,7 @@ static void start_cpus()
 
 	/* Broadcast IPIs to all other CPUs */
 	send_startup_ipi(INTR_CPU_STARTUP_ALL_EX_SELF,
-		       -1U, ((uint64_t) cpu_secondary_reset));
+		       -1U, startup_paddr);
 
 	/* Wait until global count is equal to expected CPU up count or
 	 * configured time-out has expired
@@ -700,6 +699,53 @@ static void start_cpus()
 }
 #endif
 
+static void update_trampoline_code_refs(uint64_t dest_pa)
+{
+	void *ptr;
+	int i;
+
+	if (dest_pa == 0)
+		return;
+
+	/* Update temporary page tables */
+	ptr = HPA2HVA(dest_pa + (uint64_t)CPU_Boot_Page_Tables_ptr);
+	*(uint32_t *)(ptr) += dest_pa;
+
+	ptr = HPA2HVA(dest_pa + (uint64_t)CPU_Boot_Page_Tables_Start);
+	*(uint64_t *)(ptr) += dest_pa;
+
+	ptr = HPA2HVA(dest_pa + (uint64_t)cpu_secondary_pdpt_addr);
+	for (i = 0; i < 4; i++ ) {
+		*(uint64_t *)(ptr + sizeof(uint64_t) * i) += dest_pa;
+	}
+
+	/* update the gdt base pointer with relocated offset */
+	ptr = HPA2HVA(dest_pa + (uint64_t)cpu_secondary_gdt_ptr);
+	*(uint64_t *)(ptr + 2) += dest_pa;
+
+	/* update trampoline jump pointer with relocated offset */
+	ptr = HPA2HVA(dest_pa + (uint64_t)ap_long_mode_jump_ref);
+	*(uint32_t *)ptr += dest_pa;
+}
+
+static uint64_t relocate_trampoline_code(void)
+{
+	uint64_t size, dest_pa;
+
+	size = (uint64_t)_ld_cpu_secondary_reset_end - (uint64_t)cpu_secondary_reset;
+#ifndef CONFIG_EFI_STUB
+	dest_pa = e820_alloc_low_memory(CONFIG_LOW_RAM_SIZE);
+#else
+	dest_pa = (uint64_t)get_ap_trampoline_buf();
+#endif
+
+	/* printf("AP trampoline code: %llx size %x\n", dest_pa, size); */
+	memcpy_s(HPA2HVA(dest_pa), size, _ld_cpu_secondary_reset_load, size);
+	update_trampoline_code_refs(dest_pa);
+	trampoline_code_paddr = dest_pa;
+	return dest_pa;
+}
+
 void cpu_dead(uint32_t logical_id)
 {
 	/* For debug purposes, using a stack variable in the while loop enables

hypervisor/bsp/uefi/efi/boot.c

@@ -189,12 +189,6 @@ construct_mbi(struct multiboot_info **mbi_ret, struct efi_ctx *efi_ctx)
 	mmap[j].mm_type = E820_RAM;
 	j++;
 
-	/* reserve secondary memory region(0x1000 ~ 0x10000) for hv */
-	err = __emalloc(CONFIG_LOW_RAM_SIZE, CONFIG_LOW_RAM_START,
-		&addr, EfiReservedMemoryType);
-	if (err != EFI_SUCCESS)
-		goto out;
-
 	mbi->mi_flags |= MULTIBOOT_INFO_HAS_MMAP | MULTIBOOT_INFO_HAS_CMDLINE;
 	mbi->mi_mmap_length = j*sizeof(struct multiboot_mmap);
 
@@ -228,6 +222,13 @@ switch_to_guest_mode(EFI_HANDLE image)
 
 	efi_ctx = (struct efi_ctx *)(UINTN)addr;
 
+	/* reserve secondary memory region for hv */
+	err = emalloc_for_low_mem(&addr, CONFIG_LOW_RAM_SIZE);
+	if (err != EFI_SUCCESS)
+		goto out;
+
+	efi_ctx->ap_trampoline_buf = (void *)addr;
+
 	config_table = sys_table->ConfigurationTable;
 
 	for (i = 0; i < sys_table->NumberOfTableEntries; i++) {

hypervisor/bsp/uefi/efi/boot.h

@@ -96,6 +96,7 @@ struct e820_entry {
 struct efi_ctx {
 	uint64_t rip;
 	VOID *rsdp;
+	VOID *ap_trampoline_buf;
 	dt_addr_t  gdt;
 	dt_addr_t  idt;
 	uint16_t   tr_sel;

hypervisor/bsp/uefi/uefi.c

@@ -100,6 +100,11 @@ int sipi_from_efi_boot_service_exit(uint32_t dest, uint32_t mode, uint32_t vec)
 	return 0;
 }
 
+void *get_ap_trampoline_buf(void)
+{
+	return efi_ctx->ap_trampoline_buf;
+}
+
 void efi_deferred_wakeup_pcpu(int cpu_id)
 {
 	uint32_t timeout;
@@ -114,7 +119,7 @@ void efi_deferred_wakeup_pcpu(int cpu_id)
 	expected_up = up_count + 1;
 
 	send_startup_ipi(INTR_CPU_STARTUP_USE_DEST,
-		cpu_id, (uint64_t)cpu_secondary_reset);
+		cpu_id, (uint64_t)get_ap_trampoline_buf());
 
 	timeout = CPU_UP_TIMEOUT * 1000;
 

hypervisor/include/arch/x86/cpu.h

@@ -157,11 +157,17 @@ int cpu_find_logical_id(uint32_t lapic_id);
 /**********************************/
 extern const uint8_t          _ld_cpu_secondary_reset_load[];
 extern uint8_t                _ld_cpu_secondary_reset_start[];
+extern uint8_t                _ld_cpu_secondary_reset_end[];
 extern const uint64_t         _ld_cpu_secondary_reset_size;
 extern uint8_t                _ld_bss_start[];
 extern uint8_t                _ld_bss_end[];
-extern uint8_t		      _ld_cpu_data_start[];
+extern uint8_t                _ld_cpu_data_start[];
 extern uint8_t                _ld_cpu_data_end[];
+extern uint8_t                CPU_Boot_Page_Tables_Start[];
+extern uint8_t                CPU_Boot_Page_Tables_ptr[];
+extern uint8_t                cpu_secondary_pdpt_addr[];
+extern uint8_t                cpu_secondary_gdt_ptr[];
+extern uint8_t                ap_long_mode_jump_ref[];
 
 extern int ibrs_type;
 

hypervisor/include/common/acrn_efi.h

@@ -39,6 +39,7 @@ typedef struct {
 struct efi_ctx {
 	uint64_t rip;
 	void *rsdp;
+	void *ap_trampoline_buf;
 	dt_addr_t  gdt;
 	dt_addr_t  idt;
 	uint16_t   tr_sel;
@@ -74,5 +75,6 @@ struct efi_ctx {
 }__attribute__((packed));
 
 void *get_rsdp_from_uefi(void);
+void *get_ap_trampoline_buf(void);
 
 #endif /* UEFI_H*/