Fix SMP booting: x86/64 startup initialisation fixes and so on.

Signed-off-by: Keir Fraser <keir@xensource.com>

linux-2.6-xen-sparse/arch/xen/i386/kernel/cpu/common.c

@@ -19,11 +19,13 @@
 
 #include "cpu.h"
 
+#ifndef CONFIG_XEN
 DEFINE_PER_CPU(struct desc_struct, cpu_gdt_table[GDT_ENTRIES]);
 EXPORT_PER_CPU_SYMBOL(cpu_gdt_table);
 
 DEFINE_PER_CPU(unsigned char, cpu_16bit_stack[CPU_16BIT_STACK_SIZE]);
 EXPORT_PER_CPU_SYMBOL(cpu_16bit_stack);
+#endif
 
 static int cachesize_override __initdata = -1;
 static int disable_x86_fxsr __initdata = 0;

linux-2.6-xen-sparse/arch/xen/i386/kernel/smpboot.c

@@ -131,15 +131,7 @@ static void map_cpu_to_logical_apicid(void);
  */
 void __init smp_alloc_memory(void)
 {
-#if 1
-	int cpu;
-
-	for (cpu = 1; cpu < NR_CPUS; cpu++) {
-		cpu_gdt_descr[cpu].address = (unsigned long)
-			alloc_bootmem_low_pages(PAGE_SIZE);
-		/* XXX free unused pages later */
-	}
-#else
+#if 0
 	trampoline_base = (void *) alloc_bootmem_low_pages(PAGE_SIZE);
 	/*
 	 * Has to be in very low memory so we can execute
@@ -861,8 +853,8 @@ static int __init do_boot_cpu(int apicid)
 	atomic_set(&init_deasserted, 0);
 
 #if 1
-	if (cpu_gdt_descr[0].size > PAGE_SIZE)
-		BUG();
+	cpu_gdt_descr[cpu].address = __get_free_page(GFP_KERNEL);
+	BUG_ON(cpu_gdt_descr[0].size > PAGE_SIZE);
 	cpu_gdt_descr[cpu].size = cpu_gdt_descr[0].size;
 	printk("GDT: copying %d bytes from %lx to %lx\n",
 		cpu_gdt_descr[0].size, cpu_gdt_descr[0].address,

linux-2.6-xen-sparse/arch/xen/i386/kernel/traps.c

@@ -871,6 +871,7 @@ fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
 	}
 }
 
+#ifndef CONFIG_XEN
 fastcall void setup_x86_bogus_stack(unsigned char * stk)
 {
 	unsigned long *switch16_ptr, *switch32_ptr;
@@ -915,6 +916,7 @@ fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
 	memcpy(stack32, stack16, len);
 	return stack32;
 }
+#endif
 
 /*
  *  'math_state_restore()' saves the current math information in the

linux-2.6-xen-sparse/arch/xen/x86_64/kernel/early_printk.c

@@ -6,6 +6,8 @@
 #include <asm/io.h>
 #include <asm/processor.h>
 
+#ifndef CONFIG_XEN
+
 /* Simple VGA output */
 
 #ifdef __i386__
@@ -59,7 +61,6 @@ static struct console early_vga_console = {
 	.index =	-1,
 };
 
-#ifndef CONFIG_XEN
 /* Serial functions loosely based on a similar package from Klaus P. Gerlicher */ 
 
 static int early_serial_base = 0x3f8;  /* ttyS0 */
@@ -148,7 +149,8 @@ static __init void early_serial_init(char *s)
 	outb((divisor >> 8) & 0xff, early_serial_base + DLH); 
 	outb(c & ~DLAB, early_serial_base + LCR);
 }
-#else
+
+#else /* CONFIG_XEN */
 
 static void
 early_serial_write(struct console *con, const char *s, unsigned count)
@@ -167,6 +169,13 @@ early_serial_write(struct console *con, const char *s, unsigned count)
 static __init void early_serial_init(char *s)
 {
 }
+
+/*
+ * No early VGA console on Xen, as we do not have convenient ISA-space
+ * mappings. Someone should fix this for domain 0. For now, use fake serial.
+ */
+#define early_vga_console early_serial_console
+
 #endif
 
 static struct console early_serial_console = {

linux-2.6-xen-sparse/arch/xen/x86_64/kernel/head.S

@@ -206,11 +206,13 @@ ENTRY(cpu_gdt_table)
 	.quad   0,0,0			/* three TLS descriptors */ 
 	.quad	0			/* unused now?   __KERNEL16_CS - 16bit PM for S3 wakeup. */
 
-gdt_end:	
+gdt_end:
+#if 0
 	/* asm/segment.h:GDT_ENTRIES must match this */	
 	/* This should be a multiple of the cache line size */
 	/* GDTs of other CPUs: */	
 	.fill (GDT_SIZE * NR_CPUS) - (gdt_end - cpu_gdt_table)
+#endif
 
 .org 0x8000
 ENTRY(empty_zero_page)

linux-2.6-xen-sparse/arch/xen/x86_64/kernel/setup.c

@@ -536,48 +536,7 @@ static inline void copy_edd(void)
 }
 #endif
 
-#ifdef CONFIG_XEN
-#define reserve_ebda_region() void(0)
-
-static void __init print_memory_map(char *who)
-{
-        int i;
-
-        for (i = 0; i < e820.nr_map; i++) {
-                early_printk(" %s: %016Lx - %016Lx ", who,
-                        e820.map[i].addr,
-                        e820.map[i].addr + e820.map[i].size);
-                switch (e820.map[i].type) {
-                case E820_RAM:  early_printk("(usable)\n");
-                                break;
-                case E820_RESERVED:
-                                early_printk("(reserved)\n");
-                                break;
-                case E820_ACPI:
-                                early_printk("(ACPI data)\n");
-                                break;
-                case E820_NVS:
-                                early_printk("(ACPI NVS)\n");
-                                break;
-                default:        early_printk("type %u\n", e820.map[i].type);
-                                break;
-                }
-        }
-}
-
-void __init smp_alloc_memory(void)
-{
-	int cpu;
-
-	for (cpu = 1; cpu < NR_CPUS; cpu++) {
-		cpu_gdt_descr[cpu].address = (unsigned long)
-			alloc_bootmem_low_pages(PAGE_SIZE);
-		/* XXX free unused pages later */
-	}
-}
-
-
-#else
+#ifndef CONFIG_XEN
 #define EBDA_ADDR_POINTER 0x40E
 static void __init reserve_ebda_region(void)
 {
@@ -628,7 +587,6 @@ void __init setup_arch(char **cmdline_p)
 			     VMASST_TYPE_writable_pagetables);
 
         ARCH_SETUP
-        print_memory_map(machine_specific_memory_setup());
 #else
  	ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
  	drive_info = DRIVE_INFO;
@@ -744,9 +702,6 @@ void __init setup_arch(char **cmdline_p)
 		}
 	}
 #endif
-#ifdef CONFIG_SMP
-	smp_alloc_memory();
-#endif
 #else	/* CONFIG_XEN */
 #ifdef CONFIG_BLK_DEV_INITRD
 	if (LOADER_TYPE && INITRD_START) {

linux-2.6-xen-sparse/arch/xen/x86_64/kernel/setup64.c

@@ -286,10 +286,10 @@ void __init cpu_init (void)
 
 	memcpy(me->thread.tls_array, cpu_gdt_table[cpu], GDT_ENTRY_TLS_ENTRIES * 8);
 #else
- 	memcpy(me->thread.tls_array, &get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN],
+	memcpy(me->thread.tls_array, &get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN],
 	    GDT_ENTRY_TLS_ENTRIES * 8);
 
-    cpu_gdt_init(&cpu_gdt_descr[cpu]);
+	cpu_gdt_init(&cpu_gdt_descr[cpu]);
 #endif
        
 	/*

linux-2.6-xen-sparse/arch/xen/x86_64/kernel/smpboot.c

@@ -739,8 +739,8 @@ static int __cpuinit do_boot_cpu(int cpu, int apicid)
 	atomic_set(&init_deasserted, 0);
 
 #ifdef CONFIG_XEN
-	if (cpu_gdt_descr[0].size > PAGE_SIZE)
-		BUG();
+	cpu_gdt_descr[cpu].address = __get_free_page(GFP_KERNEL);
+	BUG_ON(cpu_gdt_descr[0].size > PAGE_SIZE);
 	cpu_gdt_descr[cpu].size = cpu_gdt_descr[0].size;
 	memcpy((void *)cpu_gdt_descr[cpu].address,
 		(void *)cpu_gdt_descr[0].address, cpu_gdt_descr[0].size);
@@ -798,6 +798,8 @@ static int __cpuinit do_boot_cpu(int cpu, int apicid)
 	ctxt.ctrlreg[3] = virt_to_mfn(init_level4_pgt) << PAGE_SHIFT;
 
 	boot_error = HYPERVISOR_boot_vcpu(cpu, &ctxt);
+	if (boot_error)
+		printk("boot error: %ld\n", boot_error);
 
 	if (!boot_error) {
 		/*

linux-2.6-xen-sparse/arch/xen/x86_64/mm/init.c

@@ -536,70 +536,38 @@ static void __init find_early_table_space(unsigned long end)
 	    		  round_up(ptes * 8, PAGE_SIZE); 
 }
 
-static void xen_copy_pt(void)
-{
-	unsigned long va = __START_KERNEL_map;
-	unsigned long addr, *pte_page;
-	int i;
-	pud_t *pud; pmd_t *pmd; pte_t *pte;
-	unsigned long *page = (unsigned long *) init_level4_pgt;
-
-	addr = (unsigned long) page[pgd_index(va)];
-	addr_to_page(addr, page);
-
-	pud = (pud_t *) &page[pud_index(va)];
-	addr = page[pud_index(va)];
-	addr_to_page(addr, page);
-
-	level3_kernel_pgt[pud_index(va)] = 
-		__pud(__pa_symbol(level2_kernel_pgt) | _KERNPG_TABLE | _PAGE_USER);
-
-	for (;;) {
-		pmd = (pmd_t *) &page[pmd_index(va)];
-		if (pmd_present(*pmd)) {
-			level2_kernel_pgt[pmd_index(va)] = *pmd;
-			/*
-			 * if pmd is valid, check pte.
-			 */
-			addr = page[pmd_index(va)];
-			addr_to_page(addr, pte_page);
-			
-			for (i = 0; i < PTRS_PER_PTE; i++) {
-				pte = (pte_t *) &pte_page[pte_index(va)];
-				if (pte_present(*pte))
-					va += PAGE_SIZE;
-				else
-				    break;
-			}
-
-		} else
-		    break;
-	}
-
-	init_level4_pgt[pgd_index(__START_KERNEL_map)] = 
-		mk_kernel_pgd(__pa_symbol(level3_kernel_pgt));
-}
-
 void __init xen_init_pt(void)
 {
+	unsigned long addr, *page;
 	int i;
 
 	for (i = 0; i < NR_CPUS; i++)
 		per_cpu(cur_pgd, i) = init_mm.pgd;
 
-	memcpy((void *)init_level4_pgt, 
-	       (void *)xen_start_info.pt_base, PAGE_SIZE);
-
+	memset((void *)init_level4_pgt,   0, PAGE_SIZE);
 	memset((void *)level3_kernel_pgt, 0, PAGE_SIZE);
 	memset((void *)level2_kernel_pgt, 0, PAGE_SIZE);
 
-	xen_copy_pt();
+	/* Find the initial pte page that was built for us. */
+	page = (unsigned long *)xen_start_info.pt_base;
+	addr = page[pgd_index(__START_KERNEL_map)];
+	addr_to_page(addr, page);
+	addr = page[pud_index(__START_KERNEL_map)];
+	addr_to_page(addr, page);
+
+	/* Construct mapping of initial pte page in our own directories. */
+	init_level4_pgt[pgd_index(__START_KERNEL_map)] = 
+		mk_kernel_pgd(__pa_symbol(level3_kernel_pgt));
+	level3_kernel_pgt[pud_index(__START_KERNEL_map)] = 
+		__pud(__pa_symbol(level2_kernel_pgt) |
+		      _KERNPG_TABLE | _PAGE_USER);
+        memcpy((void *)level2_kernel_pgt, page, PAGE_SIZE);
 
 	make_page_readonly(init_level4_pgt);
-	make_page_readonly(level3_kernel_pgt);
-	make_page_readonly(level2_kernel_pgt);
 	make_page_readonly(init_level4_user_pgt);
-	make_page_readonly(level3_user_pgt); /* for vsyscall stuff */
+	make_page_readonly(level3_kernel_pgt);
+	make_page_readonly(level3_user_pgt);
+	make_page_readonly(level2_kernel_pgt);
 
 	xen_pgd_pin(__pa_symbol(init_level4_pgt));
 	xen_pgd_pin(__pa_symbol(init_level4_user_pgt));
@@ -609,7 +577,6 @@ void __init xen_init_pt(void)
 
 	set_pgd((pgd_t *)(init_level4_user_pgt + 511), 
 		mk_kernel_pgd(__pa_symbol(level3_user_pgt)));
-
 }
 
 /*
@@ -617,69 +584,58 @@ void __init xen_init_pt(void)
  * mapping done by Xen is minimal (e.g. 8MB) and we need to extend the
  * mapping for early initialization.
  */
-
-#define MIN_INIT_SIZE	0x800000
 static unsigned long current_size, extended_size;
 
 void __init extend_init_mapping(void) 
 {
 	unsigned long va = __START_KERNEL_map;
-	unsigned long addr, *pte_page;
-
-	unsigned long phys;
+	unsigned long phys, addr, *pte_page;
         pmd_t *pmd;
 	pte_t *pte, new_pte;
 	unsigned long *page = (unsigned long *) init_level4_pgt;
 	int i;
 
-	addr = (unsigned long) page[pgd_index(va)];
+	addr = page[pgd_index(va)];
 	addr_to_page(addr, page);
-
 	addr = page[pud_index(va)];
 	addr_to_page(addr, page);
 
 	for (;;) {
-		pmd = (pmd_t *) &page[pmd_index(va)];
-		if (pmd_present(*pmd)) {
-			/*
-			 * if pmd is valid, check pte.
-			 */
-			addr = page[pmd_index(va)];
-			addr_to_page(addr, pte_page);
-			
-			for (i = 0; i < PTRS_PER_PTE; i++) {
-				pte = (pte_t *) &pte_page[pte_index(va)];
-				
-				if (pte_present(*pte)) {
-					va += PAGE_SIZE;
-					current_size += PAGE_SIZE;
-				} else
-				    break;
-			}
-
-		} else
-		    break;
+		pmd = (pmd_t *)&page[pmd_index(va)];
+		if (!pmd_present(*pmd))
+			break;
+		addr = page[pmd_index(va)];
+		addr_to_page(addr, pte_page);
+		for (i = 0; i < PTRS_PER_PTE; i++) {
+			pte = (pte_t *) &pte_page[pte_index(va)];
+			if (!pte_present(*pte))
+				break;
+			va += PAGE_SIZE;
+			current_size += PAGE_SIZE;
+		}
 	}
 
-	for (; va < __START_KERNEL_map + current_size + tables_space; ) {
+	while (va < __START_KERNEL_map + current_size + tables_space) {
 		pmd = (pmd_t *) &page[pmd_index(va)];
-
-		if (pmd_none(*pmd)) {
-			pte_page = (unsigned long *) alloc_static_page(&phys);
-			make_page_readonly(pte_page);
-			xen_pte_pin(phys);
-			set_pmd(pmd, __pmd(phys | _KERNPG_TABLE | _PAGE_USER));
-
-			for (i = 0; i < PTRS_PER_PTE; i++, va += PAGE_SIZE) {
-				new_pte = pfn_pte((va -  __START_KERNEL_map) >> PAGE_SHIFT, 
-						  __pgprot(_KERNPG_TABLE | _PAGE_USER));
-
-				pte = (pte_t *) &pte_page[pte_index(va)];
-				xen_l1_entry_update(pte, new_pte);
-				extended_size += PAGE_SIZE;
-			}
-		} 
+		if (!pmd_none(*pmd))
+			continue;
+		pte_page = (unsigned long *) alloc_static_page(&phys);
+		make_page_readonly(pte_page);
+		xen_pte_pin(phys);
+		set_pmd(pmd, __pmd(phys | _KERNPG_TABLE | _PAGE_USER));
+		for (i = 0; i < PTRS_PER_PTE; i++, va += PAGE_SIZE) {
+			new_pte = pfn_pte(
+				(va - __START_KERNEL_map) >> PAGE_SHIFT, 
+				__pgprot(_KERNPG_TABLE | _PAGE_USER));
+			pte = (pte_t *)&pte_page[pte_index(va)];
+			xen_l1_entry_update(pte, new_pte);
+			extended_size += PAGE_SIZE;
+		}
 	}
+
+	/* Kill mapping of low 1MB. */
+	for (va = __START_KERNEL_map; va < (unsigned long)&_text; va += PAGE_SIZE)
+		HYPERVISOR_update_va_mapping(va, __pte_ma(0), 0);
 }
 
 
@@ -720,10 +676,6 @@ void __init init_memory_mapping(unsigned long start, unsigned long end)
 
         start_pfn = ((current_size + extended_size) >> PAGE_SHIFT);
 
-        /*
-         * TBD: Need to calculate at runtime
-         */
-
 	__flush_tlb_all();
         init_mapping_done = 1;
 }