Preparation for MTRR support, code is completely untested, though.

The CPU specific MTRR code will be in modules.


git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@15520 a95241bf-73f2-0310-859d-f6bbb57e9c96

headers/private/kernel/arch/vm.h

@@ -22,6 +22,10 @@ status_t arch_vm_init_post_area(struct kernel_args *args);
 status_t arch_vm_init_end(struct kernel_args *args);
 void arch_vm_aspace_swap(vm_address_space *aspace);
 bool arch_vm_supports_protection(uint32 protection);
+void arch_vm_init_area(vm_area *area);
+
+status_t arch_vm_set_memory_type(vm_area *area, uint32 type);
+void arch_vm_unset_memory_type(vm_area *area);
 
 #ifdef __cplusplus
 }

headers/private/kernel/arch/vm_types.h

@@ -0,0 +1,10 @@
+/*
+ * Copyright 2005, Axel Dörfler, axeld@pinc-software.de.
+ * Distributed under the terms of the MIT License.
+ */
+#ifndef KERNEL_ARCH_VM_TYPES_H
+#define KERNEL_ARCH_VM_TYPES_H
+
+#include <arch_vm_types.h>
+
+#endif	/* KERNEL_ARCH_VM_TYPES_H */

headers/private/kernel/arch/x86/arch_cpu.h

@@ -9,13 +9,22 @@
 #define _KERNEL_ARCH_x86_CPU_H
 
 
-#include <SupportDefs.h>
+#include <module.h>
 #include <arch/x86/descriptors.h>
 
 
 // MSR registers (possibly Intel specific)
 #define IA32_MSR_APIC_BASE	0x1b
 
+
+typedef struct x86_cpu_module_info {
+	module_info	info;
+	uint32		(*count_mtrrs)(void);
+	status_t	(*set_mtrr)(uint32 index, addr_t base, addr_t length, uint32 type);
+	status_t	(*unset_mtrr)(uint32 index);
+} x86_cpu_module_info;
+
+
 struct tss {
 	uint16 prev_task;
 	uint16 unused0;
@@ -78,9 +87,12 @@ void i386_frstor(const void *fpu_state);
 void i386_fxrstor(const void *fpu_state);
 void i386_fsave_swap(void *old_fpu_state, const void *new_fpu_state);
 void i386_fxsave_swap(void *old_fpu_state, const void *new_fpu_state);
-uint64 x86_read_msr(uint32 register);
-void x86_write_msr(uint32 register, uint64 value);
+uint64 x86_read_msr(uint32 registerNumber);
+void x86_write_msr(uint32 registerNumber, uint64 value);
 void x86_set_task_gate(int32 n, int32 segment);
+uint32 x86_count_mtrrs(void);
+status_t x86_set_mtrr(uint32 index, addr_t base, addr_t length, uint32 type);
+status_t x86_unset_mtrr(uint32 index);
 struct tss *x86_get_main_tss(void);
 
 #define read_ebp(value) \

headers/private/kernel/arch/x86/arch_vm_types.h

@@ -0,0 +1,17 @@
+/*
+ * Copyright 2005, Axel Dörfler, axeld@pinc-software.de. All rights reserved.
+ * Distributed under the terms of the MIT License.
+ */
+#ifndef _KERNEL_ARCH_x86_VM_TYPES_H
+#define _KERNEL_ARCH_x86_VM_TYPES_H
+
+
+#include <SupportDefs.h>
+
+
+struct arch_vm_memory_type {
+	uint16	type;
+	uint16	index;
+};
+
+#endif	/* _KERNEL_ARCH_x86_VM_TYPES_H */

headers/private/kernel/cpu.h

@@ -19,18 +19,14 @@
 typedef union cpu_ent {
 	struct {
 		int cpu_num;
-		
+
 		// thread.c: used to force a reschedule at quantum expiration time
 		int preempted;
 		timer quantum_timer;
 	} info;
-	// ToDo: align manually on CPU cache lines if possible
-	uint32 align[16];
-} cpu_ent;
+} cpu_ent __attribute__((aligned(64)));
+
 
-/**
- * Defined in core/cpu.c
- */
 extern cpu_ent cpu[MAX_BOOT_CPUS];
 
 

headers/private/kernel/vm_types.h

@@ -1,5 +1,5 @@
 /* 
- * Copyright 2002-2004, Axel Dörfler, axeld@pinc-software.de.
+ * Copyright 2002-2005, Axel Dörfler, axeld@pinc-software.de.
  * Distributed under the terms of the MIT License.
  *
  * Copyright 2001-2002, Travis Geiselbrecht. All rights reserved.
@@ -11,6 +11,7 @@
 
 #include <kernel.h>
 #include <sys/uio.h>
+#include <arch/vm_types.h>
 #include <arch/vm_translation_map.h>
 
 
@@ -84,6 +85,7 @@ typedef struct vm_area {
 	uint32				protection;
 	uint32				wiring;
 	int32				ref_count;
+	struct arch_vm_memory_type memory_type;
 
 	struct vm_cache_ref	*cache_ref;
 	off_t				cache_offset;

src/system/kernel/arch/x86/arch_cpu.c

@@ -7,11 +7,13 @@
  */
 
 
-#include <arch/cpu.h>
-#include <vm.h>
+#include <boot_device.h>
 #include <smp.h>
-#include <arch/x86/selector.h>
 #include <tls.h>
+#include <vm.h>
+
+#include <arch/cpu.h>
+#include <arch/x86/selector.h>
 #include <boot/kernel_args.h>
 
 #include "interrupts.h"
@@ -34,6 +36,8 @@ segment_descriptor *gGDT = NULL;
 //static struct tss sDoubleFaultTSS;
 static uint32 sDoubleFaultStack[10240];
 
+static x86_cpu_module_info *sCpuModule;
+
 
 struct tss *
 x86_get_main_tss(void)
@@ -44,6 +48,80 @@ x86_get_main_tss(void)
 }
 
 
+static x86_cpu_module_info *
+load_cpu_module(void)
+{
+	if (sCpuModule != NULL)
+		return sCpuModule;
+
+	// find model specific CPU module
+
+	if (gBootDevice > 0) {
+		void *cookie = open_module_list("cpu");
+
+		while (true) {
+			char name[B_FILE_NAME_LENGTH];
+			size_t nameLength = sizeof(name);
+
+			if (read_next_module_name(cookie, name, &nameLength) != B_OK
+				|| get_module(name, (module_info **)&sCpuModule) == B_OK)
+				break;
+		}
+
+		close_module_list(cookie);
+	} else {
+		// we're in early boot mode, let's use get_loaded_module
+
+		uint32 cookie = 0;
+
+		while (true) {
+			char name[B_FILE_NAME_LENGTH];
+			size_t nameLength = sizeof(name);
+
+			if (get_next_loaded_module_name(&cookie, name, &nameLength) != B_OK)
+				break;
+			if (strncmp(name, "cpu", 3))
+				continue;
+
+			if (get_module(name, (module_info **)&sCpuModule) == B_OK)
+				break;
+		}
+	}
+
+	return sCpuModule;	
+}
+
+
+uint32
+x86_count_mtrrs(void)
+{
+	if (load_cpu_module() == NULL)
+		return 0;
+
+	return sCpuModule->count_mtrrs();
+}
+
+
+status_t
+x86_set_mtrr(uint32 index, addr_t base, addr_t length, uint32 type)
+{
+	if (load_cpu_module() == NULL)
+		return B_UNSUPPORTED;
+
+	return sCpuModule->set_mtrr(index, base, length, type);
+}
+
+
+status_t
+x86_unset_mtrr(uint32 index)
+{
+	if (load_cpu_module() == NULL)
+		return B_UNSUPPORTED;
+
+	return sCpuModule->unset_mtrr(index);
+}
+
+
 static void
 load_tss(void *data, int cpu)
 {

src/system/kernel/arch/x86/arch_vm.c

@@ -18,6 +18,9 @@
 
 #include <arch/x86/bios.h>
 
+#include <stdlib.h>
+#include <string.h>
+
 
 //#define TRACE_ARCH_VM
 #ifdef TRACE_ARCH_VM
@@ -27,6 +30,80 @@
 #endif
 
 
+static uint32 *sMTRRBitmap;
+static int32 sMTRRCount;
+static spinlock sMTRRLock;
+
+
+static status_t
+init_mtrr_bitmap(void)
+{
+	if (sMTRRBitmap != NULL)
+		return B_OK;
+
+	sMTRRCount = x86_count_mtrrs();
+	if (sMTRRCount == 0)
+		return B_UNSUPPORTED;
+
+	sMTRRBitmap = malloc(sMTRRCount / 8);
+	if (sMTRRBitmap == NULL)
+		return B_NO_MEMORY;
+
+	memset(sMTRRBitmap, 0, sMTRRCount / 8);
+	return B_OK;
+}
+
+
+static int32
+allocate_mtrr(void)
+{
+	int32 count = sMTRRCount / 32;
+	int32 i, j;
+
+	cpu_status state = disable_interrupts();
+	acquire_spinlock(&sMTRRLock);
+
+	for (i = 0; i < count; i++) {
+		if (sMTRRBitmap[i] == 0xffffffff)
+			continue;
+
+		// find free bit
+
+		for (j = 0; j < 32; j++) {
+			if (sMTRRBitmap[i] & (1UL << j))
+				continue;
+
+			sMTRRBitmap[i] |= 1UL << j;
+
+			release_spinlock(&sMTRRLock);
+			restore_interrupts(state);
+			return i * 32 + j;
+		}
+	}
+
+	release_spinlock(&sMTRRLock);
+	restore_interrupts(state);
+
+	return -1;
+}
+
+
+static void
+free_mtrr(int32 index)
+{
+	int32 i = index / 32;
+	int32 j = index - i * 32;
+
+	cpu_status state = disable_interrupts();
+	acquire_spinlock(&sMTRRLock);
+
+	sMTRRBitmap[i] &= ~(1UL << j);
+
+	release_spinlock(&sMTRRLock);
+	restore_interrupts(state);
+}
+
+
 status_t
 arch_vm_init(kernel_args *args)
 {
@@ -91,3 +168,72 @@ arch_vm_supports_protection(uint32 protection)
 	return true;
 }
 
+
+void
+arch_vm_init_area(vm_area *area)
+{
+	area->memory_type.type = 0;
+}
+
+
+void
+arch_vm_unset_memory_type(vm_area *area)
+{
+	if (area->memory_type.type == 0)
+		return;
+
+	x86_unset_mtrr(area->memory_type.index);
+	free_mtrr(area->memory_type.index);
+}
+
+
+status_t
+arch_vm_set_memory_type(vm_area *area, uint32 type)
+{
+	status_t status;
+	int32 index;
+
+	if (type == 0)
+		return B_OK;
+
+	switch (type) {
+		case B_MTR_UC:	// uncacheable
+			type = 0;
+			break;
+		case B_MTR_WC:	// write combining
+			type = 1;
+			break;
+		case B_MTR_WT:	// write through
+			type = 4;
+			break;
+		case B_MTR_WP:	// write protected
+			type = 5;
+			break;
+		case B_MTR_WB:	// write back
+			type = 6;
+			break;
+
+		default:
+			return B_BAD_VALUE;
+	}
+
+	if (sMTRRBitmap == NULL) {
+		status = init_mtrr_bitmap();
+		if (status < B_OK)
+			return status;
+	}
+
+	index = allocate_mtrr();
+	if (index < 0)
+		return B_ERROR;
+
+	status = x86_set_mtrr(index, area->base, area->size, type);
+	if (status != B_OK) {
+		area->memory_type.type = (uint16)type;
+		area->memory_type.index = (uint16)index;
+	} else
+		free_mtrr(index);
+
+	dprintf("memory type: %u, index: %ld\n", area->memory_type.type, index);
+	return status;
+}

src/system/kernel/vm/vm.cpp

@@ -172,6 +172,7 @@ _vm_create_area_struct(vm_address_space *aspace, const char *name, uint32 wiring
 	area->cache_next = area->cache_prev = NULL;
 	area->hash_next = NULL;
 
+	arch_vm_init_area(area);
 	return area;
 }
 
@@ -484,10 +485,11 @@ insert_area(vm_address_space *addressSpace, void **_address,
 
 	status = find_and_insert_area_slot(&addressSpace->virtual_map, searchBase, size,
 				searchEnd, addressSpec, area);
-	if (status == B_OK)
+	if (status == B_OK) {
 		// ToDo: do we have to do anything about B_ANY_KERNEL_ADDRESS
 		//		vs. B_ANY_KERNEL_BLOCK_ADDRESS here?
 		*_address = (void *)area->base;
+	}
 
 	return status;
 }
@@ -804,7 +806,8 @@ vm_create_anonymous_area(aspace_id aid, const char *name, void **address,
 	}
 
 	vm_cache_acquire_ref(cache_ref, true);
-	err = map_backing_store(aspace, store, address, 0, size, addressSpec, wiring, protection, REGION_NO_PRIVATE_MAP, &area, name);
+	err = map_backing_store(aspace, store, address, 0, size, addressSpec, wiring,
+		protection, REGION_NO_PRIVATE_MAP, &area, name);
 	vm_cache_release_ref(cache_ref);
 	if (err < 0) {
 		vm_put_aspace(aspace);
@@ -975,9 +978,17 @@ vm_map_physical_memory(aspace_id aid, const char *name, void **_address,
 	cache->scan_skip = 1;
 
 	vm_cache_acquire_ref(cache_ref, true);
-	status = map_backing_store(aspace, store, _address, 0, size, addressSpec, 0, protection, REGION_NO_PRIVATE_MAP, &area, name);
+	status = map_backing_store(aspace, store, _address, 0, size,
+		addressSpec & ~B_MTR_MASK, 0, protection, REGION_NO_PRIVATE_MAP, &area, name);
 	vm_cache_release_ref(cache_ref);
 
+	if (status >= B_OK && (addressSpec & B_MTR_MASK) != 0) {
+		// set requested memory type
+		status = arch_vm_set_memory_type(area, addressSpec & B_MTR_MASK);
+		if (status < B_OK)
+			vm_put_area(area);
+	}
+
 	if (status >= B_OK) {
 		// make sure our area is mapped in completely
 		// (even if that makes the fault routine pretty much useless)
@@ -1321,6 +1332,7 @@ _vm_put_area(vm_area *area, bool aspaceLocked)
 
 	aspace = area->aspace;
 
+	arch_vm_unset_memory_type(area);
 	remove_area_from_virtual_map(aspace, area, aspaceLocked);
 
 	vm_cache_remove_area(area->cache_ref, area);