Merge branch 'vma_kernel' into vogel

Conflicts:
	kernel/tasks.c

Makefile.example

@@ -1,8 +1,12 @@
-TOPDIR = $(shell pwd)
-ARCH = x86
-# For 64bit support, you have define BIT as 64
-BIT=32
 NAME = metalsvm
+
+# For 64bit support, you have define BIT as 64
+# Note: do not forget to 'make veryclean' after changing BIT!!!
+BIT=64
+ARCH = x86
+SMP=1
+
+TOPDIR = $(shell pwd)
 LWIPDIRS = lwip/src/arch lwip/src/api lwip/src/core lwip/src/core/ipv4 lwip/src/netif
 DRIVERDIRS = drivers/net drivers/char
 KERNDIRS = libkern kernel mm fs apps arch/$(ARCH)/kernel arch/$(ARCH)/mm arch/$(ARCH)/scc $(LWIPDIRS) $(DRIVERDIRS)
@@ -30,35 +34,56 @@ RANLIB_FOR_TARGET = $(CROSSCOMPREFIX)ranlib
 STRIP_FOR_TARGET = $(CROSSCOMPREFIX)strip
 READELF_FOR_TARGET = $(CROSSCOMPREFIX)readelf
 
+# Tools
 MAKE = make
 RM = rm -rf
 NASM = nasm
-# For 64bit code, you have to use qemu-system-x86_64
-QEMU = qemu-system-i386
 GDB = gdb
 
-# For 64bit support, you have to define -felf64 instead of -felf32
+ifeq ($(BIT), 32)
-NASMFLAGS = -felf32 -g -i$(TOPDIR)/include/metalsvm/
+	QEMU = qemu-system-i386
-INCLUDE = -I$(TOPDIR)/include -I$(TOPDIR)/arch/$(ARCH)/include -I$(TOPDIR)/lwip/src/include -I$(TOPDIR)/lwip/src/include/ipv4 -I$(TOPDIR)/drivers
+else ifeq ($(BIT), 64)
-# For 64bit support, you have to define "-m64 -mno-red-zone" instead of "-m32 -march=i586"
+	QEMU = qemu-system-x86_64
+endif
+
+
+INCLUDE = -I$(TOPDIR)/include \
+          -I$(TOPDIR)/arch/$(ARCH)/include \
+          -I$(TOPDIR)/lwip/src/include \
+          -I$(TOPDIR)/lwip/src/include/ipv4 \
+          -I$(TOPDIR)/drivers
+
+
 # Compiler options for final code
-CFLAGS = -g -m32 -march=i586 -Wall -O2 -fstrength-reduce -fomit-frame-pointer -finline-functions -ffreestanding $(INCLUDE) $(STACKPROT)
+CFLAGS = -g -O2 -m$(BIT) -Wall -fomit-frame-pointer -ffreestanding -fstrength-reduce -finline-functions  $(INCLUDE) $(STACKPROT)
+
 # Compiler options for debuging
-#CFLAGS = -g -O -m32 -march=i586 -Wall -fomit-frame-pointer -ffreestanding $(INCLUDE) $(STACKPROT)
+#CFLAGS = -g -O -m$(BIT) -Wall -fomit-frame-pointer -ffreestanding $(INCLUDE) $(STACKPROT)
+
+NASMFLAGS = -felf$(BIT) -g -i$(TOPDIR)/include/metalsvm/
 ARFLAGS = rsv
 LDFLAGS = -T link$(BIT).ld -z max-page-size=4096 --defsym __BUILD_DATE=$(shell date +'%Y%m%d') --defsym __BUILD_TIME=$(shell date +'%H%M%S')
+
 STRIP_DEBUG = --strip-debug
 KEEP_DEBUG = --only-keep-debug
+
+# Do not change to elf64!
+# The Multiboot spec can only boot elf32 binaries
 OUTPUT_FORMAT = -O elf32-i386
-# For 64bit support, you have to define -m64 instead of "-m32 -march=i586"
+
-CFLAGS_FOR_NEWLIB = -m32 -march=i586 -O2 $(STACKPROT)
+CFLAGS_FOR_NEWLIB = -m$(BIT) -O2 $(STACKPROT)
-# For 64bit support, you have to define -m64 instead of "-m32 -march=i586"
+LDFLAGS_FOR_NEWLIB = -m$(BIT)
-LDFLAGS_FOR_NEWLIB = -m32 -march=i586
+CFLAGS_FOR_TOOLS = -m$(BIT) -O2 -Wall
-# For 64bit support, you have to define -m64 instead of "-m32"
-CFLAGS_FOR_TOOLS = -m32 -O2 -Wall
 LDFLAGS_FOR_TOOLS =
-# For 64bit support, you have to define -felf64 instead of -felf32
+NASMFLAGS_FOR_NEWLIB = -felf$(BIT)
-NASMFLAGS_FOR_NEWLIB = -felf32
+
+ifeq ($(BIT), 32)
+	CFLAGS += -march=i586
+	CFLAGS_FOR_NEWLIB += -march=i586
+	LDFLAGS_FOR_NEWLIB += -march=i586
+else ifeq ($(BIT), 64)
+	CFLAGS += -mno-red-zone
+endif
 
 # Prettify output
 V = 0
@@ -68,11 +93,15 @@ ifeq ($V,0)
 endif
 
 default: all
-	
+
 all: newlib tools $(NAME).elf
-	
+
 newlib:
-	$(MAKE) ARCH=$(ARCH) BIT=$(BIT) LDFLAGS="$(LDFLAGS_FOR_NEWLIB)" CFLAGS="$(CFLAGS_FOR_NEWLIB)" NASMFLAGS="$(NASMFLAGS_FOR_NEWLIB)" CC_FOR_TARGET=$(CC_FOR_TARGET) \
+	$(MAKE) ARCH=$(ARCH) BIT=$(BIT) \
+		LDFLAGS="$(LDFLAGS_FOR_NEWLIB)" \
+		CFLAGS="$(CFLAGS_FOR_NEWLIB)" \
+		NASMFLAGS="$(NASMFLAGS_FOR_NEWLIB)" \
+		CC_FOR_TARGET=$(CC_FOR_TARGET) \
 		CXX_FOR_TARGET=$(CXX_FOR_TARGET) \
 		GCC_FOR_TARGET=$(GCC_FOR_TARGET) \
 		AR_FOR_TARGET=$(AR_FOR_TARGET) \
@@ -96,14 +125,23 @@ $(NAME).elf:
 	$Q$(OBJCOPY_FOR_TARGET) $(STRIP_DEBUG) $(OUTPUT_FORMAT) $(NAME).elf
 
 qemu: newlib tools $(NAME).elf
-	$(QEMU) -monitor stdio -smp 2 -net nic,model=rtl8139 -net user,hostfwd=tcp::12345-:4711 -net dump -kernel metalsvm.elf -initrd tools/initrd.img
+	$(QEMU) -monitor stdio -serial tcp::12346,server,nowait -smp $(SMP) -net nic,model=rtl8139 -net user,hostfwd=tcp::12345-:4711 -kernel metalsvm.elf -initrd tools/initrd.img
 
 qemudbg: newlib tools $(NAME).elf
-	$(QEMU) -s -S -smp 2 -net nic,model=rtl8139 -net user,hostfwd=tcp::12345-:4711 -net dump -kernel metalsvm.elf -initrd tools/initrd.img
+	$(QEMU) -s -S -nographic -monitor stdio -serial tcp::12346,server -smp $(SMP) -net nic,model=rtl8139 -net user,hostfwd=tcp::12345-:4711 -kernel metalsvm.elf -initrd tools/initrd.img
 
 gdb: $(NAME).elf
-	make qemudbg > /dev/null &
+	$(GDB) -q -x script.gdb
-	$(GDB) -x script.gdb
+
+debug: newlib tools $(NAME).elf
+	killall $(QEMU) || true
+	killall $(GDB) || true
+	sleep 1
+	gnome-terminal --working-directory=$(TOPDIR) \
+			--tab --title=Shell --command="bash -c 'sleep 1 && telnet localhost 12345'" \
+			--tab --title=QEmu --command="make qemudbg" \
+			--tab --title=GDB --command="make gdb" \
+			--tab --title=Debug --command="bash -c 'sleep 1 && telnet localhost 12346'"
 
 clean:
 	$Q$(RM) $(NAME).elf $(NAME).sym *~
@@ -112,7 +150,7 @@ clean:
 
 veryclean: clean
 	$Q$(MAKE) -C newlib veryclean
-	@echo Very cleaned
+	@echo Very cleaned.
 
 #depend:
 #	for i in $(SUBDIRS); do $(MAKE) -k -C $$i depend; done
@@ -124,16 +162,15 @@ veryclean: clean
 	$Q$(CPP_FOR_TARGET) -MF $*.dep -MT $*.o -MM -D__KERNEL__ $(CFLAGS) $<
 
 include/metalsvm/config.inc: include/metalsvm/config.h
-	@echo "; This file is generated automatically from the config.h file." > include/metalsvm/config.inc
+	@echo "; This file is generated automatically from the config.h file." > $@
-	@echo "; Before editing this, you should consider editing config.h." >> include/metalsvm/config.inc
+	@echo "; Before editing this, you should consider editing config.h." >> $@
-	@awk '/^#define MAX_CORES/{ print "%define MAX_CORES", $$3 }' include/metalsvm/config.h >> include/metalsvm/config.inc	
+	@sed -nre 's/^[\t ]*#define[\t ]+([a-z_0-9]+)([\t ]+.*)*/%define \1/ip' $< >> $@
-	@awk '/^#define KERNEL_STACK_SIZE/{ print "%define KERNEL_STACK_SIZE", $$3 }' include/metalsvm/config.h >> include/metalsvm/config.inc
+	@sed -nre 's/^[\t ]*#define[\t ]+([a-z_0-9]+)[\t ]+([a-z_0-9.]+)([\t ]+.*)*/%define \1 \2/ip' $< >> $@
-	@awk '/^#define CONFIG_VGA/{ print "%define CONFIG_VGA", $$3 }' include/metalsvm/config.h >> include/metalsvm/config.inc
 
 %.o : %.asm include/metalsvm/config.inc
 	@echo [ASM] $@
 	$Q$(NASM) $(NASMFLAGS) -o $@ $<
 
-.PHONY: default all clean emu gdb newlib tools
+.PHONY: default all clean qemu qemudbg gdb debug newlib tools
  
 include $(addsuffix /Makefile,$(SUBDIRS))

apps/Makefile

@@ -1,4 +1,4 @@
-C_source := tests.c echo.c netio.c jacobi.c laplace.c gfx_client.c gfx_generic.c
+C_source := tests.c echo.c netio.c jacobi.c laplace.c gfx_client.c gfx_generic.c paging.c
 MODULE := apps
 
 include $(TOPDIR)/Makefile.inc

apps/paging.c

@@ -0,0 +1,227 @@
+/* 
+ * Copyright 2011 Steffen Vogel, Chair for Operating Systems,
+ *                               RWTH Aachen University
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * This file is part of MetalSVM.
+ */
+
+#include <metalsvm/stdlib.h>
+#include <metalsvm/stdio.h>
+#include <metalsvm/stdarg.h>
+#include <metalsvm/mmu.h>
+#include <metalsvm/time.h>
+#include <metalsvm/tasks.h>
+#include <metalsvm/vma.h>
+
+#include <asm/page.h>
+#include <asm/processor.h>
+
+#define PAGE_COUNT	10
+#define SIZE		(PAGE_COUNT*PAGE_SIZE)
+#define VIRT_FROM_ADDR	0x100000000000
+#define VIRT_TO_ADDR	0x200000000000
+
+/** @brief Simple helper to format our test results */
+static void test(size_t expr, char *fmt, ...)
+{
+	void _putchar(int c, void *arg) { kputchar(c); } // for kvprintf
+
+	static int c = 1;
+
+	va_list ap;
+	va_start(ap, fmt);
+
+	kprintf("%s #%u:\t", (expr) ? "PASSED" : "FAILED", c++);
+	kvprintf(fmt, _putchar, NULL, 10, ap);
+	kputs("\n");
+
+	va_end(ap);
+
+	if (!expr)
+		abort();
+}
+
+/** @brief BSD sum algorithm ('sum' Unix command) and used by QEmu */
+uint16_t checksum(size_t start, size_t end) {
+	size_t addr;
+	uint16_t sum;
+
+	for(addr = start, sum = 0; addr < end; addr++) {
+		uint8_t val = *((uint8_t *) addr);
+		sum = (sum >> 1) | (sum << 15);
+		sum += val;
+	}
+
+	return sum;
+}
+
+static int paging_stage2(void *arg) {
+	size_t old, new;
+
+	kprintf("PAGING: entering stage 2...\n");
+
+	old = *((size_t *) arg);
+	kprintf("old sum: %lu\n", old);
+
+	new = checksum(VIRT_FROM_ADDR, VIRT_FROM_ADDR + PAGE_COUNT*PAGE_SIZE);
+	test(old == new, "checksum(%p, %p) = %lu", VIRT_FROM_ADDR, VIRT_FROM_ADDR + PAGE_COUNT*PAGE_SIZE, new);
+
+	size_t cr3 = read_cr3();
+	kprintf("cr3 new = %x\n", cr3);
+
+	return 0;
+}
+
+/** @brief Test of the paging subsystem
+ *
+ * We will map a single physical memory region to two virtual regions.
+ * When writing to the first one, we should be able to read the same contents
+ * from the second one.
+ */
+static void paging(void)
+{
+	size_t c, sum;
+	size_t *p1, *p2;
+	size_t virt_from, virt_to, virt_alloc;
+	size_t phys;
+
+	// allocate physical page frames
+	phys = get_pages(PAGE_COUNT);
+	test(phys, "get_pages(%lu) = 0x%lx", PAGE_COUNT, phys);
+
+	// create first mapping
+	virt_from = map_region(VIRT_FROM_ADDR, phys, PAGE_COUNT, 0);
+	test(virt_from, "map_region(0x%lx, 0x%lx, %lu, 0x%x) = 0x%lx", VIRT_FROM_ADDR, phys, PAGE_COUNT, 0, virt_from);
+
+	// check address translation
+	phys = virt_to_phys(virt_from);
+	test(phys, "virt_to_phys(0x%lx) = 0x%lx", virt_from, phys);
+
+	// write test data
+	p1 = (size_t *) virt_from;
+	for (c = 0; c < PAGE_COUNT*PAGE_SIZE/sizeof(size_t); c++) {
+		p1[c] = c;
+	}
+
+	// create second mapping pointing to the same page frames
+	virt_to = map_region(VIRT_TO_ADDR, phys, PAGE_COUNT, 0);
+	test(virt_to, "map_region(0x%lx, 0x%lx, %lu, 0x%x) = 0x%lx", VIRT_TO_ADDR, phys, PAGE_COUNT, 0, virt_to);
+
+	// check address translation
+	phys = virt_to_phys(virt_to);
+	test(phys, "virt_to_phys(0x%lx) = 0x%lx", virt_to, phys);
+
+	// check if both mapped areas are equal
+	p2 = (size_t *) virt_to;
+	for (c = 0; c < PAGE_COUNT*PAGE_SIZE/sizeof(size_t); c++) {
+		if (p1[c] != p2[c])
+			test(0, "data mismatch: *(%p) != *(%p)", &p1[c], &p2[c]);
+	}
+	test(1, "data is equal");
+
+	// try to remap without MAP_REMAP
+	virt_to = map_region(VIRT_TO_ADDR, phys+PAGE_SIZE, PAGE_COUNT, 0);
+	test(!virt_to, "map_region(0x%lx, 0x%lx, %lu, 0x%x) = 0x%lx (without MAP_REMAP flag)", VIRT_TO_ADDR, phys+PAGE_SIZE, PAGE_COUNT, 0, virt_to);
+
+	// try to remap with MAP_REMAP
+	virt_to = map_region(VIRT_TO_ADDR, phys+PAGE_SIZE, PAGE_COUNT, MAP_REMAP);
+	test(virt_to, "map_region(0x%lx, 0x%lx, %lu, 0x%x) = 0x%lx (with MAP_REMAP flag)", VIRT_TO_ADDR, phys+PAGE_SIZE, PAGE_COUNT, MAP_REMAP, virt_to);
+
+	// check if data is not equal anymore (we remapped with 1 page offset)
+	p2 = (size_t *) virt_to;
+	for (c = 0; c < PAGE_COUNT*PAGE_SIZE/sizeof(size_t); c++) {
+		if (p1[c] == p2[c])
+			test(0, "data match at *(%p) != *(%p)", &p1[c], &p2[c]);
+	}
+	test(1, "data is unequal");
+
+	// test vma_alloc
+	virt_alloc = map_region(0, phys, PAGE_COUNT, 0);
+	test(virt_alloc, "map_region(0x%lx, 0x%lx, %lu, 0x%x) = 0x%lx", 0, phys, PAGE_COUNT, 0, virt_alloc);
+
+	// data should match against new vm addr
+	p2 = (size_t *) virt_alloc;
+	for (c = 0; c < PAGE_COUNT*PAGE_SIZE/sizeof(size_t); c++) {
+		if (p1[c] != p2[c])
+			test(0, "data mismatch at *(%p) != *(%p)", &p1[c], &p2[c]);
+	}
+	test(1, "data is equal");
+
+	// calc checksum
+	sum = checksum(virt_alloc, virt_alloc + PAGE_COUNT*PAGE_SIZE);
+	test(sum, "checksum(%p, %p) = %lu", virt_alloc, virt_alloc + PAGE_COUNT*PAGE_SIZE, sum);
+
+	size_t cr3 = read_cr3();
+	kprintf("cr3 old = %x\n", cr3);
+
+	//create_kernel_task(0, paging_stage2, &sum, NORMAL_PRIO);
+	//sleep(3);
+}
+
+/** @brief Test of the VMA allocator */
+static void vma(void)
+{
+	int ret;
+
+	// vma_alloc
+	size_t a1 = vma_alloc(SIZE, VMA_HEAP);
+	test(a1, "vma_alloc(0x%x, 0x%x) = 0x%lx", SIZE, VMA_HEAP, a1);
+	vma_dump();
+
+	size_t a2 = vma_alloc(SIZE, VMA_HEAP|VMA_USER);
+	test(a2 != 0, "vma_alloc(0x%x, 0x%x) = 0x%lx", SIZE, VMA_HEAP|VMA_USER, a2);
+	vma_dump();
+
+	// vma_add
+	ret = vma_add(VIRT_FROM_ADDR, VIRT_FROM_ADDR+SIZE, VMA_HEAP|VMA_USER);
+	test(ret >= 0, "vma_add(0x%lx, 0x%lx, 0x%x) = %u", VIRT_FROM_ADDR, VIRT_FROM_ADDR+SIZE, VMA_HEAP|VMA_USER, ret);
+	vma_dump();
+
+	ret = vma_add(VIRT_FROM_ADDR+SIZE, VIRT_FROM_ADDR+2*SIZE, VMA_HEAP|VMA_USER);
+	test(ret >= 0, "vma_add(0x%lx, 0x%lx, 0x%x) = %u", VIRT_FROM_ADDR+SIZE, VIRT_FROM_ADDR+2*SIZE, VMA_HEAP|VMA_USER, ret);
+	vma_dump();
+
+	ret = vma_add(VIRT_FROM_ADDR-SIZE, VIRT_FROM_ADDR, VMA_HEAP|VMA_USER);
+	test(ret >= 0, "vma_add(0x%lx, 0x%lx, 0x%x) = %u", VIRT_FROM_ADDR-SIZE, VIRT_FROM_ADDR, VMA_HEAP|VMA_USER, ret);
+	vma_dump();
+
+	// vma_free
+	ret = vma_free(VIRT_FROM_ADDR-SIZE, VIRT_FROM_ADDR);
+	test(ret >= 0, "vma_free(0x%lx, 0x%lx) = %u", VIRT_FROM_ADDR-SIZE, VIRT_FROM_ADDR, ret);
+	vma_dump();
+
+	ret = vma_free(VIRT_FROM_ADDR+SIZE, VIRT_FROM_ADDR+2*SIZE);
+	test(ret >= 0, "vma_free(0x%lx, 0x%lx) = %u", VIRT_FROM_ADDR+SIZE, VIRT_FROM_ADDR+2*SIZE, ret);
+	vma_dump();
+
+	ret = vma_free(VIRT_FROM_ADDR, VIRT_FROM_ADDR+SIZE);
+	test(ret >= 0, "vma_free(0x%lx, 0x%lx) = %u", VIRT_FROM_ADDR, VIRT_FROM_ADDR+SIZE, ret);
+	vma_dump();
+}
+
+/** @brief This is a simple procedure to test memory management subsystem */
+int memory(void* arg)
+{
+	kprintf("======== PAGING: test started...\n");
+	paging();
+	kprintf("======== VMA: test started...\n");
+	vma();
+
+
+	kprintf("======== All tests finished successfull...\n");
+
+	return 0;
+}
+

apps/tests.c

@@ -43,6 +43,7 @@
 
 int laplace(void* arg);
 int jacobi(void* arg);
+int memory(void* arg);
 void echo_init(void);
 void netio_init(void);
 
@@ -744,8 +745,7 @@ int test_init(void)
 	create_user_task(NULL, "/bin/jacobi", jacobi_argv);
 	//create_user_task_on_core(NULL, "/bin/jacobi", jacobi_argv, 1);
 #endif
-#ifdef START_MMNIF_TEST
+#if defined(START_MMNIF_TEST) && defined(CONFIG_LWIP) && LWIP_SOCKET
-#if defined(CONFIG_LWIP) && LWIP_SOCKET
 	if (RCCE_IAM == 0) {
 		kprintf("Start /bin/server...\n");
 		create_user_task(NULL, "/bin/server", server_argv);
@@ -755,6 +755,8 @@ int test_init(void)
 		create_user_task(NULL, "/bin/client", client_argv);
 	}
 #endif
+#ifdef START_MEMORY
+	create_kernel_task(NULL, memory, NULL, NORMAL_PRIO);
 #endif
 
 	return 0;

apps/tests.h

@@ -46,6 +46,7 @@
 //#define START_HELLO
 //#define START_TESTS
 //#define START_JACOBI
+//#define START_MEMORY
 
 //#define START_CHIEFTEST
 

arch/x86/include/asm/SCC_API.h

@@ -34,7 +34,7 @@
 //  ____                  _           _
 // / ___| _   _ _ __ ___ | |__   ___ | |___
 // \___ \| | | | '_ ` _ \| '_ \ / _ \| / __|
-//  ___) | |_| | | | | | | |_) | (_) | \__ \
+//  ___) | |_| | | | | | | |_) | (_) | \__
 // |____/ \__, |_| |_| |_|_.__/ \___/|_|___/
 //        |___/
 // 
@@ -253,7 +253,7 @@
 //  _____                 _   _
 // |  ___|   _ _ __   ___| |_(_) ___  _ __  ___
 // | |_ | | | | '_ \ / __| __| |/ _ \| '_ \/ __|
-// |  _|| |_| | | | | (__| |_| | (_) | | | \__ \
+// |  _|| |_| | | | | (__| |_| | (_) | | | \__
 // |_|   \__,_|_| |_|\___|\__|_|\___/|_| |_|___/
 // 
 // #########################################################################################

arch/x86/include/asm/io.h

@@ -102,7 +102,7 @@ inline static void outportl(unsigned short _port, unsigned int _data)
 
 inline static void uart_putchar(unsigned char _data)
 {
-	outportb(0x2F8, _data);
+	outportb(UART_PORT, _data);
 }
 
 /**

arch/x86/include/asm/multiboot.h

@@ -35,9 +35,11 @@
 
 #ifdef CONFIG_MULTIBOOT
 
-/* are there modules to do something with? */
+/// Does the bootloader provide mem_* fields?
+#define MULTIBOOT_INFO_MEM		0x00000001
+/// Does the bootloader provide a list of modules?
 #define MULTIBOOT_INFO_MODS		0x00000008
-/* is there a full memory map? */
+/// Does the bootloader provide a full memory map?
 #define MULTIBOOT_INFO_MEM_MAP		0x00000040
 
 typedef uint16_t multiboot_uint16_t;
@@ -114,7 +116,6 @@ struct multiboot_info
 	multiboot_uint16_t vbe_interface_off;
 	multiboot_uint16_t vbe_interface_len;
 };
-
 typedef struct multiboot_info multiboot_info_t;
 
 struct multiboot_mmap_entry

arch/x86/include/asm/page.h

@@ -31,49 +31,62 @@
 #include <metalsvm/stddef.h>
 #include <metalsvm/stdlib.h>
 
-#define _PAGE_BIT_PRESENT	0	/* is present */
+// 4KB pages
-#define _PAGE_BIT_RW		1	/* writeable */
+#define PAGE_SHIFT		12
-#define _PAGE_BIT_USER		2	/* userspace addressable */
+
-#define _PAGE_BIT_PWT		3	/* page write through */
+#ifdef CONFIG_X86_32
-#define _PAGE_BIT_PCD		4	/* page cache disabled */
+ #define PAGE_MAP_LEVELS	2
-#define _PAGE_BIT_ACCESSED	5	/* was accessed (raised by CPU) */
+ #define PAGE_MAP_SHIFT		10
-#define _PAGE_BIT_DIRTY		6	/* was written to (raised by CPU) */
+#elif defined(CONFIG_X86_64)
-#define _PAGE_BIT_PSE		7	/* 4 MB (or 2MB) page */
+ #define PAGE_MAP_LEVELS	4
-#define _PAGE_BIT_PAT		7	/* on 4KB pages */
+ #define PAGE_MAP_SHIFT		9
-#define _PAGE_BIT_GLOBAL	8	/* Global TLB entry PPro+ */
+#endif
-#define _PAGE_BIT_SVM_STRONG	9	/* mark a virtual address range as used by the SVM system */
+
-#define _PAGE_BIT_SVM_LAZYRELEASE	10 /* mark a virtual address range as used by the SVM system */
+// base addresses of page map structures
-#define _PAGE_BIT_SVM_INIT	11	/* mark if the MBP proxy is used  */
+#ifdef CONFIG_X86_32
+ #define PAGE_PGD		0xFFFFF000
+ #define PAGE_PGT		0xFFC00000
+#elif defined(CONFIG_X86_64)
+ #define PAGE_PML4		0xFFFFFFFFFFFFF000
+ #define PAGE_PDPT		0xFFFFFFFFFFE00000
+ #define PAGE_PGD		0xFFFFFFFFC0000000
+ #define PAGE_PGT		0xFFFFFF8000000000
+#endif
+
+#define PAGE_MAP_ENTRIES	(1 << PAGE_MAP_SHIFT)
+#define PAGE_SIZE		(1 << PAGE_SHIFT)
+#define PAGE_MASK		~(PAGE_SIZE - 1)
+#define PAGE_ALIGN(addr)	(((addr) + PAGE_SIZE - 1) & PAGE_MASK)
 
 /// Page is present
-#define PG_PRESENT	(1 << _PAGE_BIT_PRESENT)
+#define PG_PRESENT	(1 << 0)
 /// Page is read- and writable
-#define PG_RW		(1 << _PAGE_BIT_RW)
+#define PG_RW		(1 << 1)
 /// Page is addressable from userspace
-#define PG_USER		(1 << _PAGE_BIT_USER)
+#define PG_USER		(1 << 2)
 /// Page write through is activated
-#define PG_PWT		(1 << _PAGE_BIT_PWT)
+#define PG_PWT		(1 << 3)
 /// Page cache is disabled
-#define PG_PCD		(1 << _PAGE_BIT_PCD)
+#define PG_PCD		(1 << 4)
 /// Page was recently accessed (set by CPU)
-#define PG_ACCESSED	(1 << _PAGE_BIT_ACCESSED)
+#define PG_ACCESSED	(1 << 5)
 /// Page is dirty due to recentwrite-access (set by CPU)
-#define PG_DIRTY	(1 << _PAGE_BIT_DIRTY)
+#define PG_DIRTY	(1 << 6)
 /// Big page: 4MB (or 2MB)
-#define PG_PSE		(1 << _PAGE_BIT_PSE)
+#define PG_PSE		(1 << 7)
 /// Page is part of the MPB (SCC specific entry)
 #define PG_MPE		PG_PSE
 /// Global TLB entry (Pentium Pro and later)
-#define PG_GLOBAL	(1 << _PAGE_BIT_GLOBAL)
+#define PG_GLOBAL	(1 << 8)
 /// Pattern flag
-#define PG_PAT		(1 << _PAGE_BIT_PAT)
+#define PG_PAT		(1 << 7)
 /// This virtual address range is used by SVM system as marked
-#define PG_SVM PG_SVM_STRONG
+#define PG_SVM		(1 << 9)
-#define PG_SVM_STRONG		(1 << _PAGE_BIT_SVM_STRONG)
+#define PG_SVM_STRONG	PG_SVM_STRONG
 /// This virtual address range is used by SVM system as marked
-#define PG_SVM_LAZYRELEASE	(1 << _PAGE_BIT_SVM_LAZYRELEASE)
+#define PG_SVM_LAZYRELEASE	(1 << 10)
 /// Currently, no page frame is behind this page (only the MBP proxy)
-#define PG_SVM_INIT		(1 << _PAGE_BIT_SVM_INIT)
+#define PG_SVM_INIT	(1 << 11)
 
 /// This is a whole set of flags (PRESENT,RW,ACCESSED,DIRTY) for kernelspace tables
 #define KERN_TABLE	(PG_PRESENT|PG_RW|PG_ACCESSED|PG_DIRTY)
@@ -84,33 +97,14 @@
 /// This is a whole set of flags (PRESENT,RW,USER) for userspace pages
 #define USER_PAGE	(PG_PRESENT|PG_RW|PG_USER)
 
-#if __SIZEOF_POINTER__ == 4
+/** @brief General page map structure
-#define PGT_ENTRIES	1024
-#elif __SIZEOF_POINTER__ == 8
-#define PGT_ENTRIES	512
-#endif
- 
-/** @brief Page table structure
  *
- * This structure keeps page table entries.\n
+ * This page map structure is a general type for all indirecton levels.\n
- * On a 32bit system, a page table consists normally of 1024 entries.
+ * As all page map levels containing the same amount of entries.
  */
-typedef struct page_table
+typedef struct page_map {
-{
+	size_t entries[PAGE_MAP_ENTRIES];
-	/// Page table entries are unsigned 32bit integers.
+} __attribute__ ((aligned (4096))) page_map_t;
-	size_t entries[PGT_ENTRIES];
-} page_table_t __attribute__ ((aligned (4096)));
-
-/** @brief Page directory structure
- *
- * This structure keeps page directory entries.\
- * On a 32bit system, a page directory consists normally of 1024 entries.
- */
-typedef struct page_dir
-{
-	/// Page dir entries are unsigned 32bit integers.
-	size_t entries[PGT_ENTRIES];
-} page_dir_t __attribute__ ((aligned (4096)));
 
 /** @brief Converts a virtual address to a physical
  *
@@ -192,7 +186,7 @@ int arch_paging_init(void);
  *
  * @return Returns the address of the boot task's page dir array.
  */
-page_dir_t* get_boot_pgd(void);
+page_map_t* get_boot_page_map(void);
 
 /** @brief Setup a new page directory for a new user-level task
  *
@@ -203,18 +197,18 @@ page_dir_t* get_boot_pgd(void);
  * - counter of allocated page tables
  * - -ENOMEM (-12) on failure
  */
-int create_pgd(task_t* task, int copy);
+int create_page_map(task_t* task, int copy);
 
-/** @brief Delete page directory and its page tables
+/** @brief Delete all page map structures of the current task
  *
- * Puts page tables and page directory back to buffer and
+ * Puts PML4, PDPT, PGD, PGT tables back to buffer and
- * sets the task's page directory pointer to NULL
+ * sets the task's page map pointer to NULL
  *
  * @return
  * - 0 on success
  * - -EINVAL (-22) on failure (in case PGD is still the boot-pgd).
  */
-int drop_pgd(void);
+int drop_page_map(void);
 
 /** @brief Change the page permission in the page tables of the current task
  *

arch/x86/include/asm/processor.h

@@ -273,7 +273,7 @@ int ipi_tlb_flush(void);
 /** @brief Flush a specific page entry in TLB
  * @param addr The (virtual) address of the page to flush
  */
-static inline void tlb_flush_one_page(uint32_t addr)
+static inline void tlb_flush_one_page(size_t addr)
 {
 	asm volatile("invlpg (%0)" : : "r"(addr) : "memory");
 #if MAX_CORES > 1
@@ -293,7 +293,7 @@ static inline void tlb_flush_one_page(uint32_t addr)
  */
 static inline void tlb_flush(void)
 {
-	uint32_t val = read_cr3();
+	size_t val = read_cr3();
 
 	if (val)
 		write_cr3(val);

arch/x86/kernel/apic.c

@@ -387,12 +387,14 @@ void smp_start(uint32_t id)
 
 	kprintf("Application processor %d is entering its idle task\n", apic_cpu_id());
 
-	// initialize default cpu features
+#ifdef CONFIG_X86_32
+	// initialization for x86_64 is done in smp_entry()
 	cpu_init();
+#endif
 
 	// use the same gdt like the boot processors
 	gdt_flush();
-	
+
 	// install IDT
 	idt_install();
 

arch/x86/kernel/entry32.asm

@@ -29,7 +29,6 @@
 SECTION .mboot
 global start
 start:
-    mov byte [msg], 'H'
     jmp stublet
 
 ; This part MUST be 4byte aligned, so we solve that issue using 'ALIGN 4'
@@ -38,10 +37,10 @@ mboot:
     ; Multiboot macros to make a few lines more readable later
     MULTIBOOT_PAGE_ALIGN	equ 1<<0
     MULTIBOOT_MEMORY_INFO	equ 1<<1
-    ; MULTIBOOT_AOUT_KLUDGE     equ 1<<16
+    ; MULTIBOOT_AOUT_KLUDGE	equ 1<<16
     MULTIBOOT_HEADER_MAGIC	equ 0x1BADB002
     MULTIBOOT_HEADER_FLAGS	equ MULTIBOOT_PAGE_ALIGN | MULTIBOOT_MEMORY_INFO ; | MULTIBOOT_AOUT_KLUDGE
-    MULTIBOOT_CHECKSUM	equ -(MULTIBOOT_HEADER_MAGIC + MULTIBOOT_HEADER_FLAGS)
+    MULTIBOOT_CHECKSUM		equ -(MULTIBOOT_HEADER_MAGIC + MULTIBOOT_HEADER_FLAGS)
     EXTERN code, bss, end
 
     ; This is the GRUB Multiboot header. A boot signature
@@ -49,8 +48,6 @@ mboot:
     dd MULTIBOOT_HEADER_FLAGS
     dd MULTIBOOT_CHECKSUM
 
-msg db "?ello from MetalSVM kernel!!", 0
-
 SECTION .text
 ALIGN 4
 stublet:
@@ -70,7 +67,7 @@ stublet:
 ; jump to the boot processors's C code
     extern main
     call main
-    jmp $
+    jmp $ ; infinitive loop
 
 global cpu_init
 cpu_init:
@@ -112,7 +109,7 @@ global read_ip
 read_ip:
     mov eax, [esp+4]
     pop DWORD [eax]             ; Get the return address
-    add esp, 4                  ; Dirty Hack! read_ip cleanup the stacl
+    add esp, 4                  ; Dirty Hack! read_ip cleanup the stack
     jmp [eax]                   ; Return. Can't use RET because return
                                 ; address popped off the stack.
 

arch/x86/kernel/entry64.asm

@@ -30,7 +30,7 @@ extern kernel_end
 extern apic_mp
 
 ; We use a special name to map this section at the begin of our kernel
-; =>  Multiboot needs its magic number at the begin of the kernel
+; =>  Multiboot needs its magic number at the beginning of the kernel
 SECTION .mboot
 global start
 start:
@@ -42,19 +42,19 @@ mboot:
     ; Multiboot macros to make a few lines more readable later
     MULTIBOOT_PAGE_ALIGN	equ 1<<0
     MULTIBOOT_MEMORY_INFO	equ 1<<1
-    ; MULTIBOOT_AOUT_KLUDGE     equ 1<<16
+    ; MULTIBOOT_AOUT_KLUDGE	equ 1<<16
     MULTIBOOT_HEADER_MAGIC	equ 0x1BADB002
     MULTIBOOT_HEADER_FLAGS	equ MULTIBOOT_PAGE_ALIGN | MULTIBOOT_MEMORY_INFO ; | MULTIBOOT_AOUT_KLUDGE
-    MULTIBOOT_CHECKSUM	equ -(MULTIBOOT_HEADER_MAGIC + MULTIBOOT_HEADER_FLAGS)
+    MULTIBOOT_CHECKSUM		equ -(MULTIBOOT_HEADER_MAGIC + MULTIBOOT_HEADER_FLAGS)
     EXTERN code, bss, end
 
     ; This is the GRUB Multiboot header. A boot signature
     dd MULTIBOOT_HEADER_MAGIC
     dd MULTIBOOT_HEADER_FLAGS
     dd MULTIBOOT_CHECKSUM
-    
+
 ALIGN 4
-; we need already a valid GDT to switch in the 64bit modus
+; we need already a valid GDT to switch in the 64bit mode
 GDT64:                           ; Global Descriptor Table (64-bit).
     .Null: equ $ - GDT64         ; The null descriptor.
     dw 0                         ; Limit (low).
@@ -81,47 +81,39 @@ GDT64:                           ; Global Descriptor Table (64-bit).
     dw $ - GDT64 - 1             ; Limit.
     dq GDT64                     ; Base.
 
-    times 256 DD 0
+    times 256 DD 0               ; stack for booting
 startup_stack:
 
 SECTION .data
 ; create default page tables for the 64bit kernel
-global boot_pgd ; aka PML4
+global boot_pml4
 ALIGN 4096 ; of course, the page tables have to be page aligned
-NOPTS equ 512
+
-boot_pgd times 512 DQ 0
+PAGE_MAP_ENTRIES equ (1<<9)
-boot_pdpt times 512 DQ 0
+PAGE_SIZE equ (1<<12)
-boot_pd times 512 DQ 0
+
-boot_pt times (NOPTS*512) DQ 0
+boot_pml4 times PAGE_MAP_ENTRIES DQ 0
+boot_pdpt times PAGE_MAP_ENTRIES DQ 0
+boot_pgd times PAGE_MAP_ENTRIES DQ 0
+boot_pgt times (KERNEL_SPACE/PAGE_SIZE) DQ 0
 
 SECTION .text
 ALIGN 8
 %if MAX_CORES > 1
 global smp_entry
 smp_entry:
-; enable caching, disable paging and fpu emulation
+; initialize cpu features
-    and eax, 0x1ffffffb
+    call cpu_init
-; ...and turn on FPU exceptions
+; initialize cr3 register
-    or eax, 0x22
+    mov edi, boot_pml4
-    mov cr0, eax
-; clears the current pgd entry
-    xor eax, eax
-    mov cr3, eax
-; at this stage, we disable the SSE support
-    mov eax, cr4
-    and eax, 0xfffbf9ff
-    mov cr4, eax
-
-; initialize page table
-    mov edi, boot_pgd
     mov cr3, edi
 
-; we need to enable PAE modus
+; enable PAE
     mov eax, cr4
     or eax, 1 << 5
     mov cr4, eax
 
-; switch to the compatibility mode (which is part of long mode)
+; enable longmode (compatibility mode)
     mov ecx, 0xC0000080
     rdmsr
     or eax, 1 << 8
@@ -129,9 +121,10 @@ smp_entry:
 
 ; enable paging
     mov eax, cr0
-    or eax, 1 << 31 | 1 << 0 ; Set the PG-bit, which is the 31nd bit, and the PM-bit, which is the 0th bit.
+    or eax, 1 << 31 | 1 << 0 ; Set the PG-bit, which is the 31nd bit, and the PE-bit, which is the 0th bit.
-    mov cr0, eax
+    mov cr0, eax             ; According to the multiboot spec the PE-bit has to be set by bootloader already!
 
+; jump to 64-bit longmode
     mov edi, [esp+4]           ; set argumet for smp_start
     lgdt [GDT64.Pointer]       ; Load the 64-bit global descriptor table.
     jmp GDT64.Code:smp_start64 ; Set the code segment and enter 64-bit long mode.
@@ -139,54 +132,39 @@ smp_entry:
     jmp $ ; endless loop
 %endif
 
-search_apic:
+; search MP Floating Pointer Structure
+search_mps:
     push ebp
     mov ebp, esp
     push ecx
 
     xor eax, eax
     mov ecx, [ebp+8]
-L1:
+.l1:
     cmp [ecx], DWORD 0x5f504d5f ; MP_FLT_SIGNATURE
-    jne L2
+    jne .l2
     mov al, BYTE [ecx+9]
     cmp eax, 4
-    ja L2
+    ja .l2
     mov al, BYTE [ecx+11]
     cmp eax, 0
-    jne L2
+    jne .l2
     mov eax, ecx
-    jmp L3
+    jmp .l3
 
-L2:
+.l2:
     add ecx, 4
     cmp ecx, [ebp+12]
-    jb L1
+    jb .l1
     xor eax, eax
 
-L3:
+.l3:
     pop ecx
     pop ebp
     ret
 
-ALIGN 4
+check_longmode:
-stublet:
+; check for cpuid instruction
-    mov esp, startup_stack-4
-    push ebx ; save pointer to the multiboot structure
-    mov eax, cr0
-; enable caching, disable paging and fpu emulation
-    and eax, 0x1ffffffb
-; ...and turn on FPU exceptions
-    or eax, 0x22
-    mov cr0, eax
-; clears the current pgd entry
-    xor eax, eax
-    mov cr3, eax
-; at this stage, we disable the SSE support
-    mov eax, cr4
-    and eax, 0xfffbf9ff
-    mov cr4, eax
-; do we have the instruction cpuid?
     pushfd
     pop eax
     mov ecx, eax
@@ -198,59 +176,22 @@ stublet:
     push ecx
     popfd
     xor eax, ecx
-    jz Linvalid
+    jz .unsupported
-; cpuid > 0x80000000?
+; check for extended cpu features (cpuid > 0x80000000)
     mov eax, 0x80000000
     cpuid
     cmp eax, 0x80000001
-    jb Linvalid            ; It is less, there is no long mode.
+    jb .unsupported        ; It is less, there is no long mode.
-; do we have a long mode?
+; check if longmode is supported
     mov eax, 0x80000001
     cpuid
     test edx, 1 << 29      ; Test if the LM-bit, which is bit 29, is set in the D-register.
-    jz Linvalid            ; They aren't, there is no long mode.
+    jz .unsupported        ; They aren't, there is no long mode.
+    ret
+.unsupported:
+    jmp $
 
-; initialize page table
+check_lapic:
-    mov edi, boot_pgd
-    mov cr3, edi
-
-; So lets make PML4T[0] point to the PDPT and so on:
-    mov DWORD [edi], boot_pdpt    ; Set the double word at the destination index to pdpt.
-    or DWORD [edi], 0x00000003    ; Set present and writeable bit
-    mov edi, boot_pdpt
-    mov DWORD [edi], boot_pd      ; Set the double word at the destination index to pd.
-    or DWORD [edi], 0x00000003    ; Set present and writeable bit
-    mov edi, boot_pd
-    mov ebx, boot_pt
-    mov ecx, NOPTS
-L0:
-    mov DWORD [edi], ebx          ; Set the double word at the destination index to pt.
-    or DWORD [edi], 0x00000003    ; Set present and writeable bit
-    add edi, 8
-    add ebx, 0x1000
-    loop L0
-
-%ifdef CONFIG_VGA
-    ; map the VGA address into the virtual address space
-    mov edi, 0xB8000
-    shr edi, 9   ; (edi >> 12) * 8
-    add edi, boot_pt
-    mov ebx, 0xB8000
-    or ebx, 0x00000013
-    mov DWORD [edi], ebx
-%endif
-
-    ; map multiboot structure into the virtual address space
-    mov edi, [esp]
-    and edi, 0xFFFFF000
-    shr edi, 9   ; (edi >> 12) * 8
-    add edi, boot_pt
-    mov ebx, [esp]
-    and ebx, 0xFFFFF000
-    or ebx, 0x00000003
-    mov DWORD [edi], ebx
-
-    ; check if lapic is available
     push eax
     push ebx
     push ecx
@@ -259,82 +200,149 @@ L0:
     cpuid
     and edx, 0x200
     cmp edx, 0
-    je no_lapic
+    je .unsupported
-    ; map lapic at 0xFEE00000 below the kernel
+; map lapic at 0xFEE00000 below the kernel
     mov edi, kernel_start - 0x1000
     shr edi, 9   ; (edi >> 12) * 8
-    add edi, boot_pt
+    add edi, boot_pgt
-    mov ebx, 0xFEE00000
+    mov ebx, 0xFEE00000   ; LAPIC base address
     or ebx, 0x00000013
     mov DWORD [edi], ebx
-no_lapic:
+.unsupported:
     pop edx
     pop ecx
     pop ebx
     pop eax
+    ret
 
-    ; search APIC
+cpu_init:
+    mov eax, cr0
+; enable caching, disable paging and fpu emulation
+    and eax, 0x1ffffffb
+; ...and turn on FPU exceptions
+    or eax, 0x22
+    mov cr0, eax
+; clears the current pgd entry
+    xor eax, eax
+    mov cr3, eax
+; at this stage, we disable the SSE support
+    mov eax, cr4
+    and eax, 0xfffbf9ff
+    mov cr4, eax
+    ret
+
+; identity map a single page at address eax
+identity_page:
+    push edi
+    push ebx
+
+    and eax, 0xFFFFF000
+    mov edi, eax
+    shr edi, 9                    ; (edi >> 12) * 8 (index for boot_pgt)
+    add edi, boot_pgt
+    mov ebx, eax
+    or ebx, 0x13                  ; set present, writable and cache disable bits
+    mov DWORD [edi], ebx
+
+    pop ebx
+    pop edi
+    ret
+
+ALIGN 4
+stublet:
+    mov esp, startup_stack-4
+; save pointer to the Multiboot structure
+    push ebx
+; initialize cpu features
+    call cpu_init
+; check if longmode is supported
+    call check_longmode
+; check if lapic is available
+    call check_lapic
+
+; find MP Floating Pointer structure
     push DWORD 0x100000
     push DWORD 0xF0000
-    call search_apic
+    call search_mps
     add esp, 8
 
     cmp eax, 0
-    jne La
+    jne map_mps
 
     push DWORD 0xA0000
     push DWORD 0x9F000
-    call search_apic
+    call search_mps
     add esp, 8
-   
+
     cmp eax, 0
-    je Lb 
+    je map_kernel
 
-La:
+map_mps:
-    ; map MP Floating Pointer Structure
+; map MP Floating Pointer structure
     mov DWORD [apic_mp], eax
-    mov edi, eax
+    call identity_page
-    and edi, 0xFFFFF000
-    shr edi, 9   ; (edi >> 12) * 8
-    add edi, boot_pt
-    mov ebx, eax
-    and ebx, 0xFFFFF000
-    or ebx, 0x00000013
-    mov DWORD [edi], ebx
 
-    ; map mp_config
+; map MP Configuration table
-    mov edi, [eax+4]
+    mov eax, [apic_mp+4]
-    and edi, 0xFFFFF000
+    call identity_page
-    shr edi, 9   ; (edi >> 12) * 8
-    add edi, boot_pt
-    mov ebx, [eax+4]
-    and ebx, 0xFFFFF000
-    or ebx, 0x00000013
-    mov DWORD [edi], ebx
 
-Lb:
+%ifdef CONFIG_VGA
+; map VGA textmode plane
+    mov eax, 0xB8000
+    call identity_page
+%endif
+
+; map Multiboot structure
+    mov eax, [esp]                ; pointer is still on the stack
+    call identity_page
+
+map_kernel:
     mov edi, kernel_start
-    shr edi, 9   ; (kernel_start >> 12) * 8
+    shr edi, 9                    ; (edi >> 12) * 8 (index for boot_pgt)
-    add edi, boot_pt
+    add edi, boot_pgt
     mov ebx, kernel_start
-    or ebx, 0x00000003
+    or ebx, 0x00000003            ; set present and writable flags
     mov ecx, kernel_end           ; determine kernel size in number of pages
     sub ecx, kernel_start
     shr ecx, 12
     inc ecx
-
+.l1:
-Lc:
+    mov DWORD [edi], ebx
-    mov DWORD [edi], ebx         ; Set the double word at the destination index to the B-register.
     add edi, 8
     add ebx, 0x1000
-    loop Lc
+    loop .l1
 
-; we need to enable PAE modus
+init_paging:
+    mov edi, boot_pml4
+    mov cr3, edi
+
+    mov DWORD [edi], boot_pdpt
+    or DWORD [edi], 0x07          ; set present, user and writable flags
+
+    add edi, (PAGE_MAP_ENTRIES-1)*8    ; setup recursive paging
+    mov DWORD [edi], boot_pml4    ; boot_pml4[511] -> boot_pml4
+    or DWORD [edi], 0x03          ; set present and writable flags
+
+    mov edi, boot_pdpt
+    mov DWORD [edi], boot_pgd
+    or DWORD [edi], 0x03          ; set present and writable flags
+
+    mov edi, boot_pgd
+    mov ebx, boot_pgt
+    mov ecx, PAGE_MAP_ENTRIES     ; map all boot_pgt to the kernel space
+.l1:
+    mov DWORD [edi], ebx
+    or DWORD [edi], 0x03          ; set present and writable flags
+    add edi, 8
+    add ebx, 0x1000
+    loop .l1
+
+; enable PAE
     mov eax, cr4
     or eax, 1 << 5
     mov cr4, eax
 
-; switch to the compatibility mode (which is part of long mode)
+; enable longmode (compatibility mode)
     mov ecx, 0xC0000080
     rdmsr
     or eax, 1 << 8
@@ -342,16 +350,14 @@ Lc:
 
 ; enable paging
     mov eax, cr0
-    or eax, 1 << 31 | 1 << 0 ; Set the PG-bit, which is the 31nd bit, and the PM-bit, which is the 0th bit.
+    or eax, 1 << 31 | 1 << 0 ; Set the PG-bit, which is the 31nd bit, and the PE-bit, which is the 0th bit.
     mov cr0, eax
 
+; jump to 64-bit longmode
     pop ebx                  ; restore pointer to multiboot structure
     lgdt [GDT64.Pointer]     ; Load the 64-bit global descriptor table.
     jmp GDT64.Code:start64   ; Set the code segment and enter 64-bit long mode.
 
-Linvalid:
-    jmp $
-
 [BITS 64]
 start64:
 ; initialize segment registers
@@ -389,23 +395,6 @@ smp_start64:
     jmp $
 %endif
 
-global cpu_init
-cpu_init:
-;    mov eax, cr0
-; enable caching, disable paging and fpu emulation
-;    and eax, 0x1ffffffb
-; ...and turn on FPU exceptions
-;    or eax, 0x22
-;    mov cr0, eax
-; clears the current pgd entry
-;    xor eax, eax
-;    mov cr3, eax
-; at this stage, we disable the SSE support
-;    mov eax, cr4
-;    and eax, 0xfffbf9ff
-;    mov cr4, eax
-;    ret
-
 ; This will set up our new segment registers and is declared in 
 ; C as 'extern void gdt_flush();'
 global gdt_flush

arch/x86/kernel/gdt.c

@@ -50,7 +50,7 @@ size_t* get_current_stack(void)
 #endif
 
 	// use new page table
-	write_cr3(virt_to_phys((size_t)curr_task->pgd));
+	write_cr3(virt_to_phys((size_t)curr_task->page_map));
 
 	return curr_task->last_stack_pointer;
 }

arch/x86/kernel/string64.asm

@@ -42,38 +42,4 @@ L3:
    pop rax
    ret
 
-%if 0
-; The following function is derived from JamesM's kernel development tutorials
-; (http://www.jamesmolloy.co.uk/tutorial_html/)
-global copy_page_physical
-copy_page_physical:
-    push esi              ; According to __cdecl, we must preserve the contents of ESI
-    push edi              ; and EDI.
-    pushf                 ; push EFLAGS, so we can pop it and reenable interrupts
-                          ; later, if they were enabled anyway.
-    cli                   ; Disable interrupts, so we aren't interrupted.
-                          ; Load these in BEFORE we disable paging!
-
-    mov edi, [esp+12+4]   ; Destination address
-    mov esi, [esp+12+8]   ; Source address
-  
-    mov edx, cr0          ; Get the control register...
-    and edx, 0x7fffffff   ; and...
-    mov cr0, edx          ; Disable paging.
- 
-    cld 
-    mov ecx, 0x400        ; 1024*4bytes = 4096 bytes = page size
-    rep movsd             ; copy page
-
-    mov edx, cr0          ; Get the control register again
-    or  edx, 0x80000000   ; and...
-    mov cr0, edx          ; Enable paging.
-
-    popf                  ; Pop EFLAGS back.  
-    pop edi               ; Get the original value of EDI
-    pop esi               ; and ESI back.
-    ret
-
-%endif
-
 SECTION .note.GNU-stack noalloc noexec nowrite progbits

arch/x86/mm/page32.c

@@ -46,7 +46,7 @@
  * 0x00100000 - 0x0DEADFFF: Kernel (size depends on the configuration) (221MB)
  * 0x0DEAE000 - 0x3FFFEFFF: Kernel heap (801MB)
  * 0x3FFFF000 - 0x3FFFFFFF: Page Tables are mapped in this region (4KB)
- *			   (The last 256 entries belongs to kernel space)
+ *			    (The first 256 entries belongs to kernel space)
  */
 
 /* 
@@ -57,13 +57,14 @@ extern const void kernel_start;
 extern const void kernel_end;
 
 // boot task's page directory and page directory lock
-static page_dir_t boot_pgd = {{[0 ... PGT_ENTRIES-1] = 0}};
+static page_map_t boot_pgd = {{[0 ... MAP_ENTRIES-1] = 0}};
-static page_table_t pgt_container = {{[0 ... PGT_ENTRIES-1] = 0}};
+static page_map_t boot_pgt[KERNEL_SPACE/(MAP_ENTRIES*PAGE_SIZE)];
-static page_table_t boot_pgt[KERNEL_SPACE/(1024*PAGE_SIZE)];
+static page_map_t pgt_container = {{[0 ... MAP_ENTRIES-1] = 0}};
+
 static spinlock_t kslock = SPINLOCK_INIT;
 static int paging_enabled = 0;
 
-page_dir_t* get_boot_pgd(void)
+page_map_t* get_boot_page_map(void)
 {
 	return &boot_pgd;
 }
@@ -71,26 +72,26 @@ page_dir_t* get_boot_pgd(void)
 /*
  * TODO: We create a full copy of the current task. Copy-On-Access will be the better solution.
  *
- * No PGD locking is needed because onls create_pgd use this function and holds already the
+ * No PGD locking is needed because only create_page_map use this function and holds already the
  * PGD lock.
  */
-inline static size_t copy_page_table(task_t* task, uint32_t pgd_index, page_table_t* pgt, int* counter)
+inline static size_t copy_page_table(task_t* task, uint32_t pgd_index, page_map_t* pgt, int* counter)
 {
 	uint32_t i;
-	page_table_t* new_pgt;
+	page_map_t* new_pgt;
 	size_t phyaddr;
 
 	if (BUILTIN_EXPECT(!pgt, 0))
 		return 0;
 
-	new_pgt = kmalloc(sizeof(page_table_t));
+	new_pgt = kmalloc(sizeof(page_map_t));
 	if (!new_pgt)
 		return 0;
-	memset(new_pgt, 0x00, sizeof(page_table_t));
+	memset(new_pgt, 0x00, sizeof(page_map_t));
 	if (counter)
 		(*counter)++;
 
-	for(i=0; i<PGT_ENTRIES; i++) {
+	for(i=0; i<MAP_ENTRIES; i++) {
 		if (pgt->entries[i] & PAGE_MASK) {
 			if (!(pgt->entries[i] & PG_USER)) {
 				// Kernel page => copy only page entries
@@ -117,11 +118,11 @@ inline static size_t copy_page_table(task_t* task, uint32_t pgd_index, page_tabl
 	return phyaddr;
 }
 
-int create_pgd(task_t* task, int copy)
+int create_page_map(task_t* task, int copy)
 {
-	page_dir_t* pgd;
+	page_map_t* pgd;
-	page_table_t* pgt;
+	page_map_t* pgt;
-	page_table_t* pgt_container;
+	page_map_t* pgt_container;
 	uint32_t i;
 	uint32_t index1, index2;
 	size_t  viraddr, phyaddr;
@@ -133,25 +134,26 @@ int create_pgd(task_t* task, int copy)
 
 	// we already know the virtual address of the "page table container"
 	// (see file header)
-	pgt_container = (page_table_t*) ((KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK);
+	pgt_container = (page_map_t*) ((KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK);
 
 	// create new page directory for the new task
-	pgd = kmalloc(sizeof(page_dir_t));
+	pgd = kmalloc(sizeof(page_map_t));
 	if (!pgd)
 		return -ENOMEM;
-	memset(pgd, 0x00, sizeof(page_dir_t));
+	memset(pgd, 0x00, sizeof(page_map_t));
 
 	// create a new "page table container" for the new task
-	pgt = kmalloc(sizeof(page_table_t));
+	pgt = kmalloc(sizeof(page_map_t));
 	if (!pgt) {
-		kfree(pgd, sizeof(page_dir_t));
+		kfree(pgd, sizeof(page_map_t));
 		return -ENOMEM;
 	}
-	memset(pgt, 0x00, sizeof(page_table_t));
+	memset(pgt, 0x00, sizeof(page_map_t));
 
+	// copy kernel tables
 	spinlock_lock(&kslock);
 
-	for(i=0; i<PGT_ENTRIES; i++) {
+	for(i=0; i<MAP_ENTRIES; i++) {
 		pgd->entries[i] = boot_pgd.entries[i];
 		// only kernel entries will be copied
 		if (pgd->entries[i] && !(pgd->entries[i] & PG_USER))
@@ -169,36 +171,33 @@ int create_pgd(task_t* task, int copy)
 	pgd->entries[index1] = ((size_t) virt_to_phys((size_t) pgt) & PAGE_MASK)|KERN_TABLE;
 	pgt->entries[index2] = ((size_t) virt_to_phys((size_t) pgt) & PAGE_MASK)|KERN_PAGE;
 
-	task->pgd = pgd;
+	task->page_map = pgd;
 
 	if (copy) {
-		spinlock_irqsave_lock(&curr_task->pgd_lock);
+		spinlock_irqsave_lock(&curr_task->page_lock);
 
 		for (i=KERNEL_SPACE/(1024*PAGE_SIZE); i<1024; i++) {
-			if (!(curr_task->pgd->entries[i]))
+			if (!(curr_task->page_map->entries[i]))
 				continue;
-			if (!(curr_task->pgd->entries[i] & PG_USER))
+			if (!(curr_task->page_map->entries[i] & PG_USER))
 				continue;
 
-			phyaddr = copy_page_table(task, i, (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + i*PAGE_SIZE) & PAGE_MASK), &counter);
+			phyaddr = copy_page_table(task, i, (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + i*PAGE_SIZE) & PAGE_MASK), &counter);
 			if (phyaddr) {
-				pgd->entries[i] = (phyaddr & PAGE_MASK) | (curr_task->pgd->entries[i] & 0xFFF);
+				pgd->entries[i] = (phyaddr & PAGE_MASK) | (curr_task->page_map->entries[i] & 0xFFF);
 				pgt->entries[i] = (phyaddr & PAGE_MASK) | KERN_PAGE; 
 			}
 		}
 
-		spinlock_irqsave_unlock(&curr_task->pgd_lock);
+		spinlock_irqsave_unlock(&curr_task->page_lock);
 	}
 
 	return counter;
 }
 
-/* 
+int drop_page_map(void)
- * drops all page frames and the PGD of a user task
- */
-int drop_pgd(void)
 {
-	page_dir_t* pgd = per_core(current_task)->pgd;
+	page_map_t* pgd = per_core(current_task)->page_map;
 	size_t phy_pgd = virt_to_phys((size_t) pgd);
 	task_t* task = per_core(current_task);
 	uint32_t i;
@@ -206,9 +205,9 @@ int drop_pgd(void)
 	if (BUILTIN_EXPECT(pgd == &boot_pgd, 0))
 		return -EINVAL;
 
-	spinlock_irqsave_lock(&task->pgd_lock);
+	spinlock_irqsave_lock(&task->page_lock);
 	
-	for(i=0; i<PGT_ENTRIES; i++) {
+	for(i=0; i<MAP_ENTRIES; i++) {
 		if (pgd->entries[i] & PG_USER) {
 			put_page(pgd->entries[i] & PAGE_MASK);
 			pgd->entries[i] = 0;
@@ -218,9 +217,9 @@ int drop_pgd(void)
 	// freeing the page directory
 	put_page(phy_pgd);
 
-	task->pgd = NULL;
+	task->page_map = NULL;
 
-	spinlock_irqsave_unlock(&task->pgd_lock);
+	spinlock_irqsave_unlock(&task->page_lock);
 
 	return 0;
 }
@@ -229,24 +228,24 @@ size_t virt_to_phys(size_t viraddr)
 {
 	task_t* task = per_core(current_task);
 	uint32_t index1, index2;
-	page_table_t* pgt;
+	page_map_t* pgt;
 	size_t ret = 0;
 
 	if (!paging_enabled)
 		return viraddr;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
 		return 0;
 
-	spinlock_irqsave_lock(&task->pgd_lock);
+	spinlock_irqsave_lock(&task->page_lock);
 
 	index1 = viraddr >> 22;
 	index2 = (viraddr >> 12) & 0x3FF;
 
-	if (!(task->pgd->entries[index1] & PAGE_MASK))
+	if (!(task->page_map->entries[index1] & PAGE_MASK))
 		goto out;
 
-	pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+	pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 	if (!pgt || !(pgt->entries[index2]))
 		goto out;
 
@@ -255,7 +254,7 @@ size_t virt_to_phys(size_t viraddr)
 out:
 	//kprintf("vir %p to phy %p\n", viraddr, ret);
 
-	spinlock_irqsave_unlock(&task->pgd_lock);
+	spinlock_irqsave_unlock(&task->page_lock);
 
 	return ret;
 }
@@ -263,11 +262,11 @@ out:
 size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flags)
 {
 	task_t* task = per_core(current_task);
-	page_table_t* pgt;
+	page_map_t* pgt;
 	size_t index, i;
 	size_t ret;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
 		return 0;
 
 	if (BUILTIN_EXPECT(!paging_enabled && (viraddr != phyaddr), 0))
@@ -276,7 +275,7 @@ size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flag
 	if (flags & MAP_KERNEL_SPACE)
 		spinlock_lock(&kslock);
 	else 
-		spinlock_irqsave_lock(&task->pgd_lock);
+		spinlock_irqsave_lock(&task->page_lock);
 
 	if (!viraddr) {
 		viraddr = vm_alloc(npages, flags);
@@ -292,10 +291,10 @@ size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flag
 	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE, phyaddr+=PAGE_SIZE) {
 		index = viraddr >> 22;
 
-		if (!(task->pgd->entries[index])) {
+		if (!(task->page_map->entries[index])) {
-			page_table_t* pgt_container;
+			page_map_t* pgt_container;
 
-			pgt = (page_table_t*) get_pages(1);
+			pgt = (page_map_t*) get_page();
 			if (BUILTIN_EXPECT(!pgt, 0)) {
 				kputs("map_address: out of memory\n");
 				ret = 0;
@@ -304,17 +303,17 @@ size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flag
 
 			// set the new page table into the directory
 			if (flags & MAP_USER_SPACE)
-				task->pgd->entries[index] = (uint32_t)pgt|USER_TABLE;
+				task->page_map->entries[index] = (uint32_t)pgt|USER_TABLE;
 			else
-				task->pgd->entries[index] = (uint32_t)pgt|KERN_TABLE;
+				task->page_map->entries[index] = (uint32_t)pgt|KERN_TABLE;
 
 			// if paging is already enabled, we need to use the virtual address
 			if (paging_enabled)
 				// we already know the virtual address of the "page table container"
 				// (see file header)
-				pgt_container = (page_table_t*) ((KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK);
+				pgt_container = (page_map_t*) ((KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK);
 			else
-				pgt_container = (page_table_t*) (task->pgd->entries[(KERNEL_SPACE - PAGE_SIZE) >> 22] & PAGE_MASK);
+				pgt_container = (page_map_t*) (task->page_map->entries[(KERNEL_SPACE - PAGE_SIZE) >> 22] & PAGE_MASK);
 
 			if (BUILTIN_EXPECT(!pgt_container, 0)) {
 				kputs("map_address: internal error\n");
@@ -330,11 +329,11 @@ size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flag
 				memset((void*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*PAGE_SIZE) & PAGE_MASK), 0x00, PAGE_SIZE);
 			else
 				memset(pgt, 0x00, PAGE_SIZE);
-		} else pgt = (page_table_t*) (task->pgd->entries[index] & PAGE_MASK);
+		} else pgt = (page_map_t*) (task->page_map->entries[index] & PAGE_MASK);
 
 		/* convert physical address to virtual */
 		if (paging_enabled)
-			pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*PAGE_SIZE) & PAGE_MASK);
+			pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*PAGE_SIZE) & PAGE_MASK);
 
 		index = (viraddr >> 12) & 0x3FF;
 		if (pgt->entries[index] && !(flags & MAP_REMAP)) {
@@ -382,7 +381,7 @@ out:
 	if (flags & MAP_KERNEL_SPACE)
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
 
 	return ret;
 }
@@ -392,18 +391,18 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 	uint32_t index1, index2, newflags;
 	size_t viraddr = start & 0xFFFFF000;
 	size_t phyaddr;
-	page_table_t* pgt;
+	page_map_t* pgt;
-	page_dir_t* pgd;
+	page_map_t* pgd;
 	task_t* task = per_core(current_task);
 
 	if (BUILTIN_EXPECT(!paging_enabled, 0))
 		return -EINVAL;
 
-	pgd = per_core(current_task)->pgd;
+	pgd = per_core(current_task)->page_map;
 	if (BUILTIN_EXPECT(!pgd, 0))
 		return -EINVAL;
 
-	spinlock_irqsave_lock(&task->pgd_lock);
+	spinlock_irqsave_lock(&task->page_lock);
 
 	while (viraddr < end)
 	{
@@ -411,7 +410,7 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 		index2 = (viraddr >> 12) & 0x3FF;
 
 		while ((viraddr < end) && (index2 < 1024)) {
-			pgt = (page_table_t*) (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+			pgt = (page_map_t*) (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 			if (pgt && pgt->entries[index2]) {
 				phyaddr = pgt->entries[index2] & PAGE_MASK;
 				newflags = pgt->entries[index2] & 0xFFF;  // get old flags
@@ -448,7 +447,7 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 		}
 	}
 
-	spinlock_irqsave_unlock(&task->pgd_lock);
+	spinlock_irqsave_unlock(&task->page_lock);
 
 	return 0;
 }
@@ -464,9 +463,9 @@ size_t vm_alloc(uint32_t npages, uint32_t flags)
 	uint32_t index1, index2, j;
 	size_t viraddr, i, ret = 0;
 	size_t start, end;
-	page_table_t* pgt;
+	page_map_t* pgt;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map || !paging_enabled, 0))
 		return 0;
 
 	if (flags & MAP_KERNEL_SPACE) {
@@ -483,7 +482,7 @@ size_t vm_alloc(uint32_t npages, uint32_t flags)
 	if (flags & MAP_KERNEL_SPACE)
 		spinlock_lock(&kslock);
 	else
-		spinlock_irqsave_lock(&task->pgd_lock);
+		spinlock_irqsave_lock(&task->page_lock);
 
 	viraddr = i = start;
 	j = 0;
@@ -491,7 +490,7 @@ size_t vm_alloc(uint32_t npages, uint32_t flags)
 		index1 = i >> 22;
 		index2 = (i >> 12) & 0x3FF;
 
-		pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+		pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 		if (!pgt || !(pgt->entries[index2])) {
 			i+=PAGE_SIZE;
 			j++;
@@ -509,7 +508,7 @@ size_t vm_alloc(uint32_t npages, uint32_t flags)
 	if (flags & MAP_KERNEL_SPACE) 
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
 
 	return ret;
 }
@@ -519,22 +518,22 @@ int unmap_region(size_t viraddr, uint32_t npages)
 	task_t* task = per_core(current_task);
 	uint32_t i;
 	uint32_t index1, index2;
-	page_table_t* pgt;
+	page_map_t* pgt;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map || !paging_enabled, 0))
 	return -EINVAL;
 
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_lock(&kslock);
 	else
-		spinlock_irqsave_lock(&task->pgd_lock);
+		spinlock_irqsave_lock(&task->page_lock);
 
 	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE)
 	{
 		index1 = viraddr >> 22;
 		index2 = (viraddr >> 12) & 0x3FF;
 
-		pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+		pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 		if (!pgt)
 			continue;
 		pgt->entries[index2] &= ~PG_PRESENT;
@@ -548,7 +547,7 @@ int unmap_region(size_t viraddr, uint32_t npages)
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
 
 	return 0;
 }
@@ -558,22 +557,22 @@ int vm_free(size_t viraddr, uint32_t npages)
 	task_t* task = per_core(current_task);
 	uint32_t i;
 	uint32_t index1, index2;
-	page_table_t* pgt;
+	page_map_t* pgt;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map || !paging_enabled, 0))
 		return -EINVAL;
 
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_lock(&kslock);
 	else
-		spinlock_irqsave_lock(&task->pgd_lock);
+		spinlock_irqsave_lock(&task->page_lock);
 
 	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE)
 	{
 		index1 = viraddr >> 22;
 		index2 = (viraddr >> 12) & 0x3FF;
 
-		pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+		pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 		if (!pgt)
 			continue;
 		pgt->entries[index2] = 0;
@@ -584,7 +583,7 @@ int vm_free(size_t viraddr, uint32_t npages)
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
 
 	return 0;
 }
@@ -593,8 +592,8 @@ int print_paging_tree(size_t viraddr)
 {
 	task_t* task = per_core(current_task);
 	uint32_t index1, index2;
-	page_dir_t* pgd = NULL;
+	page_map_t* pgd = NULL;
-	page_table_t* pgt = NULL;
+	page_map_t* pgt = NULL;
 
 	if (BUILTIN_EXPECT(!viraddr, 0))
 		return -EINVAL;
@@ -602,20 +601,20 @@ int print_paging_tree(size_t viraddr)
 	index1 = viraddr >> 22;
 	index2 = (viraddr >> 12) & 0x3FF;
 
-	spinlock_irqsave_lock(&task->pgd_lock);
+	spinlock_irqsave_lock(&task->page_lock);
 
 	kprintf("Paging dump of address 0x%x\n", viraddr);
-	pgd = task->pgd;
+	pgd = task->page_map;
 	kprintf("\tPage directory entry %u: ", index1);
 	if (pgd) {
 		kprintf("0x%0x\n", pgd->entries[index1]);
-		pgt = (page_table_t*) (pgd->entries[index1] & PAGE_MASK);
+		pgt = (page_map_t*) (pgd->entries[index1] & PAGE_MASK);
 	} else
 		kputs("invalid page directory\n");
 
-	/* convert physical address to virtual */
+	// convert physical address to virtual
 	if (paging_enabled && pgt)
-		pgt = (page_table_t*) (KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE);
+		pgt = (page_map_t*) (KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE);
 
 	kprintf("\tPage table entry %u: ", index2);
 	if (pgt)
@@ -623,7 +622,7 @@ int print_paging_tree(size_t viraddr)
 	else 
 		kputs("invalid page table\n");
 
-	spinlock_irqsave_unlock(&task->pgd_lock);
+	spinlock_irqsave_unlock(&task->page_lock);
 
 	return 0;
 }
@@ -631,12 +630,12 @@ int print_paging_tree(size_t viraddr)
 static void pagefault_handler(struct state *s)
 {
 	task_t* task = per_core(current_task);
-	page_dir_t* pgd = task->pgd;
-	page_table_t* pgt = NULL;
 	size_t viraddr = read_cr2();
 	size_t phyaddr;
 #ifdef CONFIG_ROCKCREEK
 	uint32_t index1, index2;
+	page_map_t* pgd = task->page_map;
+	page_map_t* pgt = NULL;
 #endif
 
 	if ((viraddr >= task->start_heap) && (viraddr <= task->end_heap) && (viraddr > KERNEL_SPACE)) {
@@ -650,7 +649,7 @@ static void pagefault_handler(struct state *s)
 			memset((void*) viraddr, 0x00, PAGE_SIZE);
 			return;
 		}
-		
+
 		kprintf("Could not map 0x%x at 0x%x\n", phyaddr, viraddr);
 		put_page(phyaddr);
 	}
@@ -661,7 +660,7 @@ static void pagefault_handler(struct state *s)
 	index2 = (viraddr >> 12) & 0x3FF;
 	if (!pgd || !(pgd->entries[index1] & PAGE_MASK))
                 goto default_handler;
-        pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+        pgt = (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 	if (!pgt || !(pgt->entries[index2]))
 		goto default_handler;
 	if (pgt->entries[index2] & PG_SVM_INIT) {
@@ -687,14 +686,14 @@ default_handler:
 int arch_paging_init(void)
 {
 	uint32_t i, npages, index1, index2;
-	page_table_t* pgt;
+	page_map_t* pgt;
 	size_t viraddr;
 
-	// uninstall default handler and install our own
+	// replace default pagefault handler
 	irq_uninstall_handler(14);
 	irq_install_handler(14, pagefault_handler);
 
-	// Create a page table to reference to the other page tables
+	// create a page table to reference to the other page tables
 	pgt = &pgt_container;
 
 	// map this table at the end of the kernel space
@@ -703,21 +702,21 @@ int arch_paging_init(void)
 	index2 = (viraddr >> 12) & 0x3FF;
 
 	// now, we create a self reference
-	per_core(current_task)->pgd->entries[index1] = (((size_t) pgt) & PAGE_MASK)|KERN_TABLE;
+	per_core(current_task)->page_map->entries[index1] = (((size_t) pgt) & PAGE_MASK)|KERN_TABLE;
-	pgt->entries[index2] = ((size_t) pgt & 0xFFFFF000)|KERN_PAGE;
+	pgt->entries[index2] = ((size_t) pgt & PAGE_MASK)|KERN_PAGE;
 
 	// create the other PGTs for the kernel space
 	for(i=0; i<KERNEL_SPACE/(1024*PAGE_SIZE)-1; i++) {
 		size_t phyaddr = boot_pgt+i;
 
-		memset((void*) phyaddr, 0x00, sizeof(page_table_t));
+		memset((void*) phyaddr, 0x00, sizeof(page_map_t));
-		per_core(current_task)->pgd->entries[i] = (phyaddr & PAGE_MASK)|KERN_TABLE;
+		per_core(current_task)->page_map->entries[i] = (phyaddr & PAGE_MASK)|KERN_TABLE;
 		pgt->entries[i] = (phyaddr & PAGE_MASK)|KERN_PAGE;
 	}
 
 	/* 
-	 * Set the page table and page directory entries for the kernel. We map the kernel's physical address 
+	 * Set the page table and page directory entries for the kernel.
-	 * to the same virtual address.
+	 * We map the kernel's physical address to the same virtual address.
 	 */
 	npages = ((size_t) &kernel_end - (size_t) &kernel_start) >> PAGE_SHIFT;
 	if ((size_t)&kernel_end & (PAGE_SIZE-1))
@@ -725,7 +724,7 @@ int arch_paging_init(void)
 	map_region((size_t)&kernel_start, (size_t)&kernel_start, npages, MAP_KERNEL_SPACE);
 
 #if MAX_CORES > 1
-	// Reserve page for smp boot code
+	// reserve page for smp boot code
 	if (!map_region(SMP_SETUP_ADDR, SMP_SETUP_ADDR, 1, MAP_KERNEL_SPACE|MAP_NO_CACHE)) {
 		kputs("could not reserve page for smp boot code\n");
 		return -ENOMEM;
@@ -738,16 +737,12 @@ int arch_paging_init(void)
 #endif
 
 #ifdef CONFIG_MULTIBOOT
-	/*
+	// map mb_info into the kernel space
-	 * of course, mb_info has to map into the kernel space
-	 */
 	if (mb_info)
 		map_region((size_t) mb_info & PAGE_MASK, (size_t) mb_info & PAGE_MASK, 1, MAP_KERNEL_SPACE);
 
 #if 0
-	/*
+	// map reserved memory regions into the kernel space
-	 * Map reserved memory regions into the kernel space
-	 */
 	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MEM_MAP)) {
 		multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) mb_info->mmap_addr;
 		multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);
@@ -805,7 +800,7 @@ int arch_paging_init(void)
 	kprintf("Map FPGA regsiters at 0x%x\n", viraddr);
 #endif
 
-	/* enable paging */
+	// enable paging
 	write_cr3((uint32_t) &boot_pgd);
 	i = read_cr0();
 	i = i | (1 << 31);
@@ -822,10 +817,7 @@ int arch_paging_init(void)
 	bootinfo->addr = viraddr;
 #endif
 
-	/*
+	// we turned on paging => now, we are able to register our task
-	 * we turned on paging
-	 * => now, we are able to register our task
-	 */
 	register_task();
 
 	// APIC registers into the kernel address space

arch/x86/mm/page64.c

@@ -31,21 +31,15 @@
 #include <asm/irq.h>
 #include <asm/multiboot.h>
 #include <asm/apic.h>
-#ifdef CONFIG_ROCKCREEK
-#include <asm/RCCE_lib.h>
-#include <asm/SCC_API.h>
-#include <asm/svm.h>
-#include <asm/icc.h>
-#endif
 
 /*
  * Virtual Memory Layout of the standard configuration
  * (1 GB kernel space)
  *
- * 0x00000000 - 0x000FFFFF: reserved for IO devices (16MB)
+ * 0x000000000000 - 0x0000000FFFFF: reserved for IO devices (16MB)
- * 0x00100000 - 0x0DEADFFF: Kernel (size depends on the configuration) (221MB)
+ * 0x000000100000 - 0x00000DEADFFF: Kernel (size depends on the configuration) (221MB)
- * 0x0DEAE000 - 0x3FFFFFFF: Kernel heap 
+ * 0x00000DEAE000 - 0x00003FFFFFFF: Kernel heap
- *
+ * 0xFF8000000000 - 0xFFFFFFFFFFFF: Paging structures are mapped in this region (max 512GB)
  */
 
 /* 
@@ -55,127 +49,248 @@
 extern const void kernel_start;
 extern const void kernel_end;
 
-// boot task's page directory and page directory lock
+// boot task's page map and page map lock
-extern page_dir_t boot_pgd;
+extern page_map_t boot_pml4;
 static spinlock_t kslock = SPINLOCK_INIT;
-static int paging_enabled = 0;
 
-page_dir_t* get_boot_pgd(void)
+page_map_t* get_boot_page_map(void)
 {
-	return &boot_pgd;
+	return &boot_pml4;
 }
 
-int create_pgd(task_t* task, int copy)
+/** @brief Copy a single page frame
-{
+ *
-	// Currently, we support only kernel tasks
+ * @param src	virtual address of source page frame
-	// => all tasks are able to use the same pgd
+ * @return	physical addr to copied page frame
-
-	if (BUILTIN_EXPECT(!paging_enabled, 0))
-		return -EINVAL;
-
-	task->pgd = get_boot_pgd();
-
-	return 0;
-}
-
-/* 
- * drops all page frames and the PGD of a user task
  */
-int drop_pgd(void)
+static size_t copy_page_frame(size_t *src)
 {
-#if 0
+	kprintf("copy_page_frame(%p)\n", src);
-	page_dir_t* pgd = per_core(current_task)->pgd;
+#if 1 // TODO: untested
-	size_t phy_pgd = virt_to_phys((size_t) pgd);
+	size_t phyaddr, viraddr;
-	task_t* task = per_core(current_task);
-	uint32_t i;
 
-	if (BUILTIN_EXPECT(pgd == &boot_pgd, 0))
+	// allocate and map an empty page
+	phyaddr = get_page();
+	if (BUILTIN_EXPECT(!phyaddr, 0))
+		return 0;
+
+	viraddr = vma_alloc(PAGE_SIZE, VMA_HEAP);
+	if (BUILTIN_EXPECT(!viraddr, 0))
+		return 0;
+
+	viraddr = map_region(viraddr, phyaddr, 1, MAP_KERNEL_SPACE);
+	if (BUILTIN_EXPECT(!viraddr, 0))
+		return 0;
+
+	// copy the whole page
+	strncpy((void*) viraddr, (void*) src, PAGE_SIZE);
+
+	// unmap and free page
+	unmap_region(viraddr, 1);
+	vma_free(viraddr, viraddr+PAGE_SIZE);
+
+	return phyaddr;
+#else
+	kprintf("TODO: copy_page_frame(%lx)\n", source);
+	return 0;
+#endif
+}
+
+static inline size_t canonicalize(size_t addr)
+{
+	if (addr & (1UL<<47))
+		return addr;
+	else
+		return addr & ((1UL<<48) - 1);
+}
+
+static inline int map_to_level(size_t addr)
+{
+	if (addr >= PAGE_PML4)
+		return 4;
+	else if (addr >= PAGE_PDPT)
+		return 3;
+	else if (addr >= PAGE_PGD) 
+		return 2;
+	else if (addr >= PAGE_PGT)
+		return 1;
+	else
 		return -EINVAL;
+}
 
-	spinlock_lock(&task->pgd_lock);
+static inline const char * map_to_lvlname(size_t addr)
-	
+{
-	for(i=0; i<1024; i++) {
+	const char* names[] = {"(none)", "PGT", "PGD", "PDPT", "PML4"};
-		if (pgd->entries[i] & PG_USER) {
+	return names[map_to_level(addr)];
-			put_page(pgd->entries[i] & PAGE_MASK);
+}
-			pgd->entries[i] = 0;
+
+static inline size_t map_to_virt(size_t addr)
+{
+	return canonicalize(addr << (map_to_level(addr) * PAGE_MAP_SHIFT));
+}
+
+/*
+ * Copy page maps using recursion
+ *
+ * @param from	pointer to virtual address of source page tables
+ * @param to	pointer to virtual address of destination page tables
+ * @param copy	flags what should be copied (see #define COPY_*)
+ * @return	number of new allocated page frames (for tables only)
+ */
+static int copy_page_map(page_map_t *src, page_map_t *dest, int copy)
+{
+	page_map_t* next_src, * next_dest;
+
+	int ret = 0;
+	uint32_t i;
+	for(i=0; i<PAGE_MAP_ENTRIES; i++) {
+		if (!(src->entries[i] & PG_PRESENT))
+			// skip empty entries
+			dest->entries[i] = 0;
+		else if (src->entries[i] & PG_USER) {
+			size_t phys;
+			kprintf("d:%p (%s: 0x%012lx) -> %p\n", &src->entries[i], map_to_lvlname((size_t) &src->entries[i]), map_to_virt((size_t) &src->entries[i]),  &dest->entries[i]);
+
+			// deep copy user tables
+			if ((size_t) src >= PAGE_PGT) {
+				phys = get_page();
+				if (BUILTIN_EXPECT(!phys, 0))
+					return -ENOMEM;
+
+				dest->entries[i] = phys|(src->entries[i] & ~PAGE_MASK);
+
+				// reuse pointers to next lower page map tables
+				next_src  = (page_map_t*) ((size_t) &src->entries[i] << 9);
+				next_dest = (page_map_t*) ((size_t) &dest->entries[i] << 9);
+
+				ret += 1 + copy_page_map(next_src, next_dest, copy);
+			}
+			// deep copy page frame
+			else {
+				if (copy) {
+					phys = copy_page_frame((size_t*) src->entries[i]);
+					dest->entries[i] = phys|(src->entries[i] & ~PAGE_MASK);
+				}
+				kprintf("c: %p (%lx)\n", &src->entries[i], src->entries[i]);
+			}
+		}
+		// shallow copy kernel only tables
+		else {
+			kprintf("s:%p (%s: 0x%012lx) -> %p\n", &src->entries[i], map_to_lvlname((size_t) &src->entries[i]), map_to_virt((size_t) &src->entries[i]),  &dest->entries[i]);
+			dest->entries[i] = src->entries[i];
 		}
 	}
 
-	// freeing the page directory
+	kputs("r\n");
-	put_page(phy_pgd);
+	return ret;
+}
 
-	task->pgd = NULL;
+int create_page_map(task_t* task, int copy)
+{
+	size_t phys;
+	uint32_t ret;
 
-	spinlock_unlock(&task->pgd_lock);
+	// fixed mapping for paging structures
-#endif
+	page_map_t *current = (page_map_t*) PAGE_PML4;
+	page_map_t *new = (page_map_t*) (PAGE_PML4 - 0x1000);
+
+	// get new pml4 table
+	phys = get_page();
+	if (!phys) return -ENOMEM;
+
+	current->entries[PAGE_MAP_ENTRIES-2] = phys|KERN_TABLE;
+	new->entries[PAGE_MAP_ENTRIES-1] = phys|KERN_TABLE;
+
+	tlb_flush(); // ouch :(
+
+	spinlock_lock(&kslock);
+	ret = copy_page_map(current, new, copy);
+	spinlock_unlock(&kslock);
+
+	new->entries[PAGE_MAP_ENTRIES-1] = phys|KERN_TABLE;
+	current->entries[PAGE_MAP_ENTRIES-2] = 0;
+
+	task->page_map = (page_map_t*) phys;
+
+	kprintf("create_page_map: allocated %u page tables\n", ret);
+	return ret;
+}
+
+int drop_page_map(void)
+{
+#if 1
+	kprintf("TODO: test drop_page_map()\n");
+	return -EINVAL; // TODO
+#else
+	task_t* task = per_core(current_task);
+	page_map_t* pml4, * pdpt, * pgd, * pgt;
+	size_t phys;
+	uint32_t i, j, k, l;
+
+	pml4 = task->page_map;
+
+	if (BUILTIN_EXPECT(pml4 == &boot_pml4, 0))
+		return -EINVAL;
+
+	spinlock_lock(&task->page_lock);
+
+	// delete all user pages and tables
+	for(i=0; i<PAGE_MAP_ENTRIES; i++) { // pml4
+		if (pml4->entries[i] & PG_USER) {
+			for(j=0; j<PAGE_MAP_ENTRIES; j++) { // pdpt
+				if (pdpt->entries[j] & PG_USER) {
+					for(k=0; k<PAGE_MAP_ENTRIES; k++) { // pgd
+						if (pgd->entries[k] & PG_USER) {
+							for(l=0; l<PAGE_MAP_ENTRIES; l++) { // pgt
+								if (pgt->entries[l] & PG_USER)
+									put_page(pgt->entries[l] & PAGE_MASK);
+							}
+							// TODO: put pgt
+						}
+					}
+					// TODO: put pgd
+				}
+			}
+			// TODO: put pdpt
+		}
+	}
+
+	put_page(virt_to_phys((size_t) pml4));
+	task->page_map = NULL;
+
+	spinlock_unlock(&task->page_lock);
 
 	return 0;
+#endif
 }
 
 size_t virt_to_phys(size_t viraddr)
 {
 	task_t* task = per_core(current_task);
-	uint16_t idx_pd4 = (viraddr >> 39) & 0x1FF;
+	size_t phyaddr;
-	uint16_t idx_dirp = (viraddr >> 30) & 0x1FF;
+	size_t* pte;
-	uint16_t idx_dir = (viraddr >> 21) & 0x1FF;
-	uint16_t idx_table = (viraddr >> 12) & 0x1FF;
-	page_table_t* pgt;
-	size_t ret = 0;
 
-	if (!paging_enabled)
+	spinlock_irqsave_lock(&task->page_lock);
-		return viraddr;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd, 0))
+	pte = (size_t *) (PAGE_PGT | (viraddr >> 9));
-		return 0;
+	phyaddr = (*pte & PAGE_MASK) | (viraddr & ~PAGE_MASK);
 
-	spinlock_irqsave_lock(&task->pgd_lock);
+	spinlock_irqsave_unlock(&task->page_lock);
 
-	// Currently, we allocate pages only in kernel space.
+	return phyaddr;
-	// => physical address of the page table is identical of the virtual address
-	pgt = (page_table_t*) (task->pgd->entries[idx_pd4] & PAGE_MASK);
-	if (!pgt)
-		goto out;
-
-	pgt = (page_table_t*) (pgt->entries[idx_dirp] & PAGE_MASK);
-	if (!pgt)
-		goto out;
-
-	pgt = (page_table_t*) (pgt->entries[idx_dir] & PAGE_MASK);
-	if (!pgt)
-		goto out;
-
-	ret = (size_t) (pgt->entries[idx_table] & PAGE_MASK);
-	if (!ret)
-		goto out;
-
-	ret = ret | (viraddr & 0xFFF);			// add page offset
-out:
-	//kprintf("vir %p to phy %p\n", viraddr, ret);
-
-	spinlock_irqsave_unlock(&task->pgd_lock);
-
-	return ret;
 }
 
 size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flags)
 {
 	task_t* task = per_core(current_task);
-	page_table_t* pgt;
 	size_t i, ret;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
 		return 0;
 
-	if (BUILTIN_EXPECT(!paging_enabled && (viraddr != phyaddr), 0))
-		return 0;
-
-	if (flags & MAP_KERNEL_SPACE)
-		spinlock_lock(&kslock);
-	else 
-		spinlock_irqsave_lock(&task->pgd_lock);
-
 	if (!viraddr) {
-		viraddr = vm_alloc(npages, flags);
+		kputs("map_region: deprecated vma_alloc() call from within map_region\n");
+		viraddr = vma_alloc(npages*PAGE_SIZE, VMA_HEAP);
 		if (BUILTIN_EXPECT(!viraddr, 0)) {
 			kputs("map_region: found no valid virtual address\n");
 			ret = 0;
@@ -183,59 +298,40 @@ size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flag
 		}
 	}
 
+	// correct alignment
+	phyaddr &= PAGE_MASK;
+	viraddr &= PAGE_MASK;
 	ret = viraddr;
+
+	if (flags & MAP_KERNEL_SPACE)
+		spinlock_lock(&kslock);
+	else 
+		spinlock_irqsave_lock(&task->page_lock);
+
+	kprintf("map_region: map %u pages from 0x%lx to 0x%lx with flags: 0x%x\n", npages, viraddr, phyaddr, flags);
 	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE, phyaddr+=PAGE_SIZE) {
-		uint16_t idx_pd4 = (viraddr >> 39) & 0x1FF;
+		// page table entry
-		uint16_t idx_dirp = (viraddr >> 30) & 0x1FF;
+		size_t* pte = (size_t *) (PAGE_PGT|(viraddr >> 9));
-		uint16_t idx_dir = (viraddr >> 21) & 0x1FF;
-		uint16_t idx_table = (viraddr >> 12) & 0x1FF;
 
-		pgt = (page_table_t*) (task->pgd->entries[idx_pd4] & PAGE_MASK);
+		if (*pte && !(flags & MAP_REMAP)) {
-		if (!pgt) {
+			kprintf("map_region: 0x%lx is already mapped\n", viraddr);
-			kputs("map_region: out of memory\n");
-			ret = 0;
-			goto out;
-		}
-	
-		pgt = (page_table_t*) (pgt->entries[idx_dirp] & PAGE_MASK);
-		if (!pgt) {
-			kputs("map_region: out of memory\n");
-			ret = 0;
-			goto out;
-		}
-
-		pgt = (page_table_t*) (pgt->entries[idx_dir] & PAGE_MASK);
-		if (!pgt) {
-			kputs("map_region: out of memory\n");
-			ret = 0;
-			goto out;
-		}
-
-		/* convert physical address to virtual */
-		// Currently, we allocate pages only in kernel space.
-		// => physical address of the page table is identical of the virtual address
-		//if (paging_enabled)
-		//	pgt = (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index*PAGE_SIZE) & PAGE_MASK);
-
-		if (pgt->entries[idx_table] && !(flags & MAP_REMAP)) {
-			kprintf("0x%x is already mapped\n", viraddr);
 			ret = 0;
 			goto out;
 		}
 
 		if (flags & MAP_USER_SPACE)
-			pgt->entries[idx_table] = USER_PAGE|(phyaddr & PAGE_MASK);
+			*pte = phyaddr|USER_PAGE;
 		else
-			pgt->entries[idx_table] = KERN_PAGE|(phyaddr & PAGE_MASK);
+			*pte = phyaddr|KERN_PAGE;
 
 		if (flags & MAP_NO_CACHE)
-			pgt->entries[idx_table] |= PG_PCD;
+			*pte |= PG_PCD;
 
 		if (flags & MAP_NO_ACCESS)
-			pgt->entries[idx_table] &= ~PG_PRESENT;
+			*pte &= ~PG_PRESENT;
 
 		if (flags & MAP_WT)
-			pgt->entries[idx_table] |= PG_PWT;
+			*pte |= PG_PWT;
 
 		if (flags & MAP_USER_SPACE)
 			atomic_int32_inc(&task->user_usage);
@@ -247,7 +343,7 @@ out:
 	if (flags & MAP_KERNEL_SPACE)
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
 
 	return ret;
 }
@@ -258,18 +354,15 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 	uint32_t index1, index2, newflags;
 	size_t viraddr = start & PAGE_MASK;
 	size_t phyaddr;
-	page_table_t* pgt;
+	page_map_t* pgt;
-	page_dir_t* pgd;
+	page_map_t* pgd;
 	task_t* task = per_core(current_task);
 
-	if (BUILTIN_EXPECT(!paging_enabled, 0))
+	pgd = per_core(current_task)->page_map;
-		return -EINVAL;
-
-	pgd = per_core(current_task)->pgd;
 	if (BUILTIN_EXPECT(!pgd, 0))
 		return -EINVAL;
 
-	spinlock_lock(&task->pgd_lock);
+	spinlock_lock(&task->page_lock);
 
 	while (viraddr < end)
 	{
@@ -277,7 +370,7 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 		index2 = (viraddr >> 12) & 0x3FF;
 
 		while ((viraddr < end) && (index2 < 1024)) {
-			pgt = (page_table_t*) (page_table_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
+			pgt = (page_map_t*) (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
 			if (pgt && pgt->entries[index2]) {
 				phyaddr = pgt->entries[index2] & PAGE_MASK;
 				newflags = pgt->entries[index2] & 0xFFF;  // get old flags
@@ -292,16 +385,8 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 				// update flags
 				if (!(flags & VMA_WRITE)) {
 					newflags &= ~PG_RW;
-#ifdef CONFIG_ROCKCREEK
-					if (newflags & (PG_SVM_STRONG|PG_SVM_LAZYRELEASE))
-						newflags &= ~PG_MPE;
-#endif
 				} else  {
 					newflags |= PG_RW;
-#ifdef CONFIG_ROCKCREEK
-					if (newflags & (PG_SVM_STRONG|PG_SVM_LAZYRELEASE))
-						newflags |= PG_MPE;
-#endif
 				}
 
 				pgt->entries[index2] = (newflags & 0xFFF) | (phyaddr & PAGE_MASK);
@@ -314,149 +399,66 @@ int change_page_permissions(size_t start, size_t end, uint32_t flags)
 		}
 	}
 
-	spinlock_unlock(&task->pgd_lock);
+	spinlock_unlock(&task->page_lock);
 #endif
 
 	return -EINVAL;
 }
 
-/* 
- * Use the first fit algorithm to find a valid address range
- *
- * TODO: O(n) => bad performance, we need a better approach
- */
-size_t vm_alloc(uint32_t npages, uint32_t flags)
-{
-	task_t* task = per_core(current_task);
-	size_t viraddr, i, j, ret = 0;
-	size_t start, end;
-	page_table_t* pgt;
-
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
-		return 0;
-
-	if (flags & MAP_KERNEL_SPACE) {
-		start = (((size_t) &kernel_end) + 10*PAGE_SIZE) & PAGE_MASK;
-		end = (KERNEL_SPACE - PAGE_SIZE) & PAGE_MASK; 
-	} else {
-		start = KERNEL_SPACE & PAGE_MASK;
-		end = PAGE_MASK;
-	}
-
-	if (BUILTIN_EXPECT(!npages, 0))
-		return 0;
-
-	if (flags & MAP_KERNEL_SPACE)
-		spinlock_lock(&kslock);
-	else
-		spinlock_irqsave_lock(&task->pgd_lock);
-
-	viraddr = i = start;
-	j = 0;
-	do {
-		uint16_t idx_pd4 = (viraddr >> 39) & 0x1FF;
-		uint16_t idx_dirp = (viraddr >> 30) & 0x1FF;
-		uint16_t idx_dir = (viraddr >> 21) & 0x1FF;
-		uint16_t idx_table = (viraddr >> 12) & 0x1FF;
-
-		// Currently, we allocate pages only in kernel space.
-		// => physical address of the page table is identical of the virtual address
-		pgt = (page_table_t*) (task->pgd->entries[idx_pd4] & PAGE_MASK);
-		if (!pgt) {
-			i += (size_t)PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
-			j += PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES;
-			continue;
-		}
-
-		pgt = (page_table_t*) (pgt->entries[idx_dirp] & PAGE_MASK);
-		if (!pgt) {
-			i += PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
-			j += PGT_ENTRIES*PGT_ENTRIES;
-			continue;
-		}
-
-		pgt = (page_table_t*) (pgt->entries[idx_dir] & PAGE_MASK);
-                if (!pgt) {
-                        i += PGT_ENTRIES*PAGE_SIZE;
-                        j += PGT_ENTRIES;
-                        continue;
-                }
-
-		if (!(pgt->entries[idx_table])) {
-			i += PAGE_SIZE;
-			j++;
-		} else {
-			// restart search
-			j = 0;
-			viraddr = i + PAGE_SIZE;
-			i = i + PAGE_SIZE;
-		}
-	} while((j < npages) && (i<=end));
-
-	if ((j >= npages) && (viraddr < end))
-		ret = viraddr;
-
-	if (flags & MAP_KERNEL_SPACE)
-		spinlock_unlock(&kslock);
-	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
-
-	return ret;
-}
 
 int unmap_region(size_t viraddr, uint32_t npages)
 {
 	task_t* task = per_core(current_task);
-	page_table_t* pgt;
+	page_map_t* pdpt, * pgd, * pgt;
 	size_t i;
-	uint16_t idx_pd4, idx_dirp;
+	uint16_t index_pml4, index_pdpt;
-	uint16_t idx_dir, idx_table;
+	uint16_t index_pgd, index_pgt;
 
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
+	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
 		return -EINVAL;
 
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_lock(&kslock);
 	else
-		spinlock_irqsave_lock(&task->pgd_lock);
+		spinlock_irqsave_lock(&task->page_lock);
 
 	i = 0;
 	while(i<npages)
 	{
-		idx_pd4 = (viraddr >> 39) & 0x1FF;
+		index_pml4 = (viraddr >> 39) & 0x1FF;
-		idx_dirp = (viraddr >> 30) & 0x1FF;
+		index_pdpt = (viraddr >> 30) & 0x1FF;
-		idx_dir = (viraddr >> 21) & 0x1FF;
+		index_pgd = (viraddr >> 21) & 0x1FF;
-		idx_table = (viraddr >> 12) & 0x1FF;
+		index_pgt = (viraddr >> 12) & 0x1FF;
 
-		// Currently, we allocate pages only in kernel space.
+		// currently, we allocate pages only in kernel space.
 		// => physical address of the page table is identical of the virtual address
-		pgt = (page_table_t*) (task->pgd->entries[idx_pd4] & PAGE_MASK);
+		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
-		if (!pgt) {
+		if (!pdpt) {
-			viraddr += (size_t) PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
+			viraddr += (size_t) PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES;
+			i += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES;
 			continue;
 		}
 
-		pgt = (page_table_t*) (pgt->entries[idx_dirp] & PAGE_MASK);
+		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
-		if (!pgt) {
+		if (!pgd) {
-			viraddr += PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
+			viraddr += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES*PGT_ENTRIES;
+			i += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES;
 			continue;
 		}
 
-		pgt = (page_table_t*) (pgt->entries[idx_dir] & PAGE_MASK);
+		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
 		if (!pgt) {
-			viraddr += PGT_ENTRIES*PAGE_SIZE;
+			viraddr += PAGE_MAP_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES;
+			i += PAGE_MAP_ENTRIES;
                         continue;
                 }
 
-		if (pgt->entries[idx_table]) 
+		if (pgt->entries[index_pgt])
-			pgt->entries[idx_table] &= ~PG_PRESENT;
+			pgt->entries[index_pgt] &= ~PG_PRESENT;
 
 		viraddr +=PAGE_SIZE;
 		i++;
-			
+
 		if (viraddr > KERNEL_SPACE)
 			atomic_int32_dec(&task->user_usage);
 
@@ -466,71 +468,7 @@ int unmap_region(size_t viraddr, uint32_t npages)
 	if (viraddr <= KERNEL_SPACE)
 		spinlock_unlock(&kslock);
 	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
+		spinlock_irqsave_unlock(&task->page_lock);
-
-	return 0;
-}
-
-int vm_free(size_t viraddr, uint32_t npages)
-{
-	task_t* task = per_core(current_task);
-	page_table_t* pgt;
-	size_t i;
-	uint16_t idx_pd4, idx_dirp;
-	uint16_t idx_dir, idx_table;
-
-	if (BUILTIN_EXPECT(!task || !task->pgd || !paging_enabled, 0))
-		return -EINVAL;
-
-	if (viraddr <= KERNEL_SPACE)
-		spinlock_lock(&kslock);
-	else
-		spinlock_irqsave_lock(&task->pgd_lock);
-
-	i = 0;
-	while(i<npages)
-	{
-		idx_pd4 = (viraddr >> 39) & 0x1FF;
-		idx_dirp = (viraddr >> 30) & 0x1FF;
-		idx_dir = (viraddr >> 21) & 0x1FF;
-		idx_table = (viraddr >> 12) & 0x1FF;
-
-		// Currently, we allocate pages only in kernel space.
-		// => physical address of the page table is identical of the virtual address
-		pgt = (page_table_t*) (task->pgd->entries[idx_pd4] & PAGE_MASK);
-		if (!pgt) {
-			viraddr += (size_t) PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES*PGT_ENTRIES*PGT_ENTRIES;
-			continue;
-		}
-
-		pgt = (page_table_t*) (pgt->entries[idx_dirp] & PAGE_MASK);
-		if (!pgt) {
-			viraddr += PGT_ENTRIES*PGT_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES*PGT_ENTRIES;
-			continue;
-		}
-
-		pgt = (page_table_t*) (pgt->entries[idx_dir] & PAGE_MASK);
-		if (!pgt) {
-			viraddr += PGT_ENTRIES*PAGE_SIZE;
-			i += PGT_ENTRIES;
-			continue;
-		}
-
-		if (pgt->entries[idx_table])
-			pgt->entries[idx_table] = 0;
-
-		viraddr +=PAGE_SIZE;
-		i++;
-
-		tlb_flush_one_page(viraddr);
-	}
-
-	if (viraddr <= KERNEL_SPACE)
-		spinlock_unlock(&kslock);
-	else
-		spinlock_irqsave_unlock(&task->pgd_lock);
 
 	return 0;
 }
@@ -538,10 +476,8 @@ int vm_free(size_t viraddr, uint32_t npages)
 static void pagefault_handler(struct state *s)
 {
 	task_t* task = per_core(current_task);
-	//page_dir_t* pgd = task->pgd;
-	//page_table_t* pgt = NULL;
 	size_t viraddr = read_cr2();
-	//size_t phyaddr;
+	size_t phyaddr;
 
 #if 0
 	if ((viraddr >= task->start_heap) && (viraddr <= task->end_heap) && (viraddr > KERNEL_SPACE)) {
@@ -549,24 +485,49 @@ static void pagefault_handler(struct state *s)
 
 		phyaddr = get_page();
 		if (BUILTIN_EXPECT(!phyaddr, 0))
-			goto default_handler;
+			goto oom;
 
 		if (map_region(viraddr, phyaddr, 1, MAP_USER_SPACE) == viraddr) {
 			memset((void*) viraddr, 0x00, PAGE_SIZE);
 			return;
 		}
-		
+
 		kprintf("Could not map 0x%x at 0x%x\n", phyaddr, viraddr);
 		put_page(phyaddr);
 	}
+	/*
+	 * handle missing paging structures for userspace
+	 * all kernel space paging structures have been initialized in entry64.asm
+	 */
+	else if (viraddr >= PAGE_PGT) {
+		kprintf("map_region: missing paging structure at: 0x%lx (%s)\n", viraddr, map_to_lvlname(viraddr));
+
+		phyaddr = get_page();
+		if (BUILTIN_EXPECT(!phyaddr, 0))
+			goto oom;
+
+		// TODO: initialize with zeros
+		// TODO: check that we are in userspace
+
+		// get pointer to parent page level entry
+		size_t *entry = (size_t *) ((int64_t) viraddr >> 9 & ~0x07);
+
+		// update entry
+		*entry = phyaddr|USER_TABLE;
+
+		return;
+	}
 #endif
 
-//default_handler:
 	kprintf("PAGE FAULT: Task %u got page fault at %p (irq %llu, cs:rip 0x%llx:0x%llx)\n", task->id, viraddr, s->int_no, s->cs, s->rip);
 	kprintf("Register state: rax = 0x%llx, rbx = 0x%llx, rcx = 0x%llx, rdx = 0x%llx, rdi = 0x%llx, rsi = 0x%llx, rbp = 0x%llx, rsp = 0x%llx\n", 
 		s->rax, s->rbx, s->rcx, s->rdx, s->rdi, s->rsi, s->rbp, s->rsp);
 
-while(1);
+	irq_enable();
+	abort();
+
+oom:
+	kputs("map_region: out of memory\n");
 	irq_enable();
 	abort();
 }
@@ -575,15 +536,17 @@ int arch_paging_init(void)
 {
 	uint32_t i, npages;
 
-	// uninstall default handler and install our own
+	// replace default pagefault handler
 	irq_uninstall_handler(14);
 	irq_install_handler(14, pagefault_handler);
 
-	// kernel is already maped into the kernel space (see entry64.asm)
+	/*
-	// this includes .data, .bss, .text, video memory and the multiboot structure
+	 * In longmode the kernel is already maped into the kernel space (see entry64.asm)
+	 * this includes .data, .bss, .text, VGA, the multiboot & multiprocessing (APIC) structures
+	 */
 
 #if MAX_CORES > 1
-	// Reserve page for smp boot code
+	// reserve page for smp boot code
 	if (!map_region(SMP_SETUP_ADDR, SMP_SETUP_ADDR, 1, MAP_KERNEL_SPACE|MAP_NO_CACHE)) {
 		kputs("could not reserve page for smp boot code\n");
 		return -ENOMEM;
@@ -592,9 +555,7 @@ int arch_paging_init(void)
 
 #ifdef CONFIG_MULTIBOOT
 #if 0
-	/*
+	// map reserved memory regions into the kernel space
-	 * Map reserved memory regions into the kernel space
-	 */
 	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MEM_MAP)) {
 		multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) mb_info->mmap_addr;
 		multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);
@@ -613,7 +574,7 @@ int arch_paging_init(void)
 
 	/* 
 	 * Modules like the init ram disk are already loaded.
-	 * Therefore, we map these moduels into the kernel space.
+	 * Therefore, we map these modules into the kernel space.
 	 */
 	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MODS)) {
 		multiboot_module_t* mmodule = (multiboot_module_t*) ((size_t) mb_info->mods_addr);
@@ -634,13 +595,7 @@ int arch_paging_init(void)
 	}
 #endif
 
-	/* signalize that we are able to use paging */
+	// we turned on paging => now, we are able to register our task
-	paging_enabled = 1;
-
-	/*
-	 * we turned on paging
-	 * => now, we are able to register our task
-	 */
 	register_task();
 
 	// APIC registers into the kernel address space

drivers/char/stdio.c

@@ -86,11 +86,12 @@ static ssize_t stdio_write(fildes_t* file, uint8_t* buffer, size_t size)
 	for (i = 0; i<size; i++, buffer++) {
 #ifdef CONFIG_VGA
 		vga_putchar(*buffer);
-#elif defined(CONFIG_UART)
-		uart_putchar(*buffer);
-#else
-		kputchar(*buffer);
 #endif
+#ifdef CONFIG_UART
+		uart_putchar(*buffer);
+#endif
+
+		kputchar(*buffer);
 	}
 
 	file->offset += size;

fs/initrd.c

@@ -253,7 +253,7 @@ static int initrd_open(fildes_t* file, const char* name)
 		/* opendir was called: */
 		if (name[0] == '\0') 
 			return 0;
-	
+
 		/* open file was called: */
 		if (!(file->flags & O_CREAT)) 
 			return -ENOENT;
@@ -264,11 +264,11 @@ static int initrd_open(fildes_t* file, const char* name)
 		vfs_node_t* new_node = kmalloc(sizeof(vfs_node_t));
 		if (BUILTIN_EXPECT(!new_node, 0))
 			return -EINVAL;
-		
+
 		blist = &file->node->block_list;
 		dir_block_t* dir_block;
 		dirent_t* dirent;
-		
+
 		memset(new_node, 0x00, sizeof(vfs_node_t));
 		new_node->type = FS_FILE;
 		new_node->read = &initrd_read;
@@ -286,7 +286,7 @@ static int initrd_open(fildes_t* file, const char* name)
 						if (!dirent->vfs_node) {
 							dirent->vfs_node = new_node;
 							strncpy(dirent->name, (char*) name, MAX_FNAME);
-							goto exit_create_file; // there might be a better Solution ***************
+							goto exit_create_file; // TODO: there might be a better Solution
 						}
  					}
 				}

include/metalsvm/config.h.example

@@ -34,14 +34,14 @@ extern "C" {
 #define PAGE_SHIFT		12
 #define CACHE_LINE		64
 #define MAILBOX_SIZE		32
-#define TIMER_FREQ		100	/* in HZ */
+#define TIMER_FREQ		100 // in HZ
-#define CLOCK_TICK_RATE		1193182	/* 8254 chip's internal oscillator frequency */
+#define CLOCK_TICK_RATE		1193182	// 8254 chip's internal oscillator frequency
 #define INT_SYSCALL		0x80
 #define KERNEL_SPACE		(1*1024*1024*1024)
-#define VIDEO_MEM_ADDR 		0xB8000 // the video memora address
+#define VIDEO_MEM_ADDR 		0xB8000 // the video memory address
 #define SMP_SETUP_ADDR		0x07000
-
+#define UART_PORT		0x3F8 // 0x2F8 for SCC
-#define BYTE_ORDER LITTLE_ENDIAN
+#define BYTE_ORDER		LITTLE_ENDIAN
 
 /* 
  * address space / (page_size * sizeof(uint8_t)) 
@@ -52,7 +52,7 @@ extern "C" {
 #define CONFIG_PCI
 #define CONFIG_LWIP
 #define CONFIG_VGA
-//#define CONFIG_UART
+#define CONFIG_UART
 #define CONFIG_KEYBOARD
 #define CONFIG_MULTIBOOT
 //#define CONFIG_ROCKCREEK
@@ -72,7 +72,7 @@ extern "C" {
 //#define SHMADD
 #define SHMDBG
 //#define SHMADD_CACHEABLE
-#define SCC_BOOTINFO            0x80000
+#define SCC_BOOTINFO		0x80000
 
 #define BUILTIN_EXPECT(exp, b) 	__builtin_expect((exp), (b))
 //#define BUILTIN_EXPECT(exp, b)	(exp)

include/metalsvm/mmu.h

@@ -31,7 +31,6 @@
 
 #include <metalsvm/stddef.h>
 #include <asm/atomic.h>
-//#include <asm/mmu.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -50,33 +49,39 @@ extern atomic_int32_t		total_available_pages;
  */
 int mmu_init(void);
 
-/** @brief get continuous pages
+/** @brief Get continuous pages
  *
- * This function finds a continuous page region (first fit algorithm)
+ * Use first fit algorithm to find a suitable, continous physical memory region
- *
- * @param no_pages Desired number of pages
  *
+ * @param npages Desired number of pages
  * @return 
  * - physical address on success
  * - 0 on failure
  */
-size_t get_pages(uint32_t no_pages);
+size_t get_pages(uint32_t npages);
 
-/** @brief get a single page
+/** @brief Get a single page
  *
  * Convenience function: uses get_pages(1);
  */
 static inline size_t get_page(void) { return get_pages(1); }
 
-/** @brief Put back a page after use
+/** @brief Put back a sequence of continous pages
  *
- * @param phyaddr Physical address to put back
+ * @param phyaddr Physical address of the first page
+ * @param npages Number of pages
  *
  * @return 
  * - 0 on success
  * - -EINVAL (-22) on failure
  */
-int put_page(size_t phyaddr);
+int put_pages(size_t phyaddr, size_t npages);
+
+/** @brief Put a single page
+ *
+ * Convenience function: uses put_pages(1);
+ */
+static inline int put_page(size_t phyaddr) { return put_pages(phyaddr, 1); }
 
 #ifdef __cplusplus
 }

include/metalsvm/page.h

@@ -29,10 +29,7 @@
 #include <metalsvm/stddef.h>
 #include <asm/page.h>
 
-/**
+/** @brief Sets up the environment, page directories etc and enables paging. */
- * Sets up the environment, page directories etc and
- * enables paging.
- */
 static inline int paging_init(void) { return arch_paging_init(); }
 
 #endif

include/metalsvm/stddef.h

@@ -28,14 +28,10 @@
 extern "C" {
 #endif
 
-#define NULL 	((void*) 0)
+#define NULL ((void*) 0)
 
 typedef unsigned int tid_t;
 
-#define PAGE_SIZE		(1 << PAGE_SHIFT)
-#define PAGE_MASK		~(PAGE_SIZE - 1)
-#define PAGE_ALIGN(addr)	(((addr) + PAGE_SIZE - 1) & PAGE_MASK)
-
 #if MAX_CORES == 1
 #define per_core(name) name
 #define DECLARE_PER_CORE(type, name) extern type name;
@@ -66,10 +62,10 @@ typedef unsigned int tid_t;
 		irq_nested_enable(flags);\
 		return ret; \
 	}
-#define CORE_ID	smp_id()
+ #define CORE_ID smp_id()
 #endif
 
-/* needed to find the task, which is currently running on this core */
+// needed to find the task, which is currently running on this core
 struct task;
 DECLARE_PER_CORE(struct task*, current_task);
 

include/metalsvm/stdlib.h

@@ -113,7 +113,7 @@ unsigned long strtoul(const char* nptr, char** endptr, int base);
  */
 static inline int atoi(const char *str)
 {
-	return (int)strtol(str, (char **)NULL, 10);
+	return (int)strtol(str, (char **) NULL, 10);
 }
 
 #ifdef __cplusplus

include/metalsvm/tasks.h

@@ -147,9 +147,7 @@ tid_t wait(int32_t* result);
  */
 void update_load(void);
 
-/** @brief Print the current cpu load
+/** @brief Print the current cpu load */
- *
- */
 void dump_load(void);
 
 #if MAX_CORES > 1
@@ -201,9 +199,7 @@ int block_current_task(void);
  */
 int set_timer(uint64_t deadline);
 
-/** @brief check is a timer is expired
+/** @brief check is a timer is expired */
- *
- */
 void check_timers(void);
 
 /** @brief Abort current task */

include/metalsvm/tasks_types.h

@@ -62,7 +62,7 @@ extern "C" {
 #define TASK_L2				(1 << 3)
 
 typedef int (*entry_point_t)(void*);
-struct page_dir;
+typedef struct page_map page_map_t;
 
 /** @brief The task_t structure */
 typedef struct task {
@@ -88,10 +88,10 @@ typedef struct task {
 	uint32_t		last_core;
 	/// usage in number of pages
 	atomic_int32_t		user_usage;
-	/// avoids concurrent access to the page directory
+	/// avoids concurrent access to the page map structures
-	spinlock_irqsave_t	pgd_lock;	
+	spinlock_irqsave_t	page_lock;
-	/// pointer to the page directory
+	/// pointer to page directory (32bit) or page map level 4 (64bit) table respectively
-	struct page_dir*	pgd;            
+	page_map_t*		page_map;
 	/// lock for the VMA_list
 	spinlock_t		vma_lock;
 	/// list of VMAs

include/metalsvm/vma.h

@@ -27,56 +27,102 @@
 #define __VMA_H__
 
 #include <metalsvm/stddef.h>
+#include <asm/page.h>
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+/// Read access to this VMA is allowed
 #define VMA_READ	(1 << 0)
+/// Write access to this VMA is allowed
 #define VMA_WRITE	(1 << 1)
+/// Instructions fetches in this VMA are allowed
 #define VMA_EXECUTE	(1 << 2)
+/// This VMA is cacheable
 #define VMA_CACHEABLE	(1 << 3)
-#define VMA_NOACCESS	(1 << 4)
+/// This VMA is not accessable
+#define VMA_NO_ACCESS	(1 << 4)
+/// This VMA should be part of the userspace
+#define VMA_USER	(1 << 5)
+/// A collection of flags used for the kernel heap (kmalloc)
+#define VMA_HEAP	(VMA_READ|VMA_WRITE|VMA_CACHEABLE)
+
+// boundaries for VAS allocation
+extern const void kernel_end;
+//#define VMA_KERN_MIN	(((size_t) &kernel_end + PAGE_SIZE) & PAGE_MASK)
+#define VMA_KERN_MAX	KERNEL_SPACE
+#define VMA_USER_MAX	(1UL << 47) // TODO
 
 struct vma;
 
-/** @brief VMA structure definition */
+/** @brief VMA structure definition
+ *
+ * Each item in this linked list marks a used part of the virtual address space.
+ * Its used by vm_alloc() to find holes between them.
+ */
 typedef struct vma {
 	/// Start address of the memory area
 	size_t start;
 	/// End address of the memory area
 	size_t end;
 	/// Type flags field
-	uint32_t type;
+	uint32_t flags;
 	/// Pointer of next VMA element in the list
 	struct vma* next;
 	/// Pointer to previous VMA element in the list
 	struct vma* prev;
 } vma_t;
 
-/** @brief Add a new virtual memory region to the list of VMAs 
+/** @brief Add a new virtual memory area to the list of VMAs 
  *
- * @param task Pointer to the task_t structure of the task
+ * @param start Start address of the new area
- * @param start Start address of the new region
+ * @param end End address of the new area
- * @param end End address of the new region
+ * @param flags Type flags the new area shall have
- * @param type Type flags the new region shall have
  *
  * @return
  * - 0 on success
  * - -EINVAL (-22) or -EINVAL (-12) on failure
  */
-int vma_add(struct task* task, size_t start, size_t end, uint32_t type);
+int vma_add(size_t start, size_t end, uint32_t flags);
 
-/** @brief Dump information about this task's VMAs into the terminal.
+/** @brief Search for a free memory area
  *
- * This will print out Start, end and flags for each VMA in the task's list
+ * @param size Size of requestes VMA in bytes
+ * @param flags
+ * @return Type flags the new area shall have
+ * - 0 on failure
+ * - the start address of a free area
+ */
+size_t vma_alloc(size_t size, uint32_t flags);
+
+/** @brief Free an allocated memory area
  *
- * @param task The task's task_t structure
+ * @param start Start address of the area to be freed
+ * @param end End address of the to be freed
  * @return
  * - 0 on success
  * - -EINVAL (-22) on failure
  */
-int vma_dump(struct task* task);
+int vma_free(size_t start, size_t end);
+
+/** @brief Free all virtual memory areas
+ *
+ * @return
+ * - 0 on success
+ */
+int drop_vma_list();
+
+/** @brief Copy the VMA list of the current task to task
+ *
+ * @param task The task where the list should be copied to
+ * @return
+ * - 0 on success
+ */
+int copy_vma_list(struct task* task);
+
+/** @brief Dump information about this task's VMAs into the terminal. */
+void vma_dump();
 
 #ifdef __cplusplus
 }

kernel/init.c

@@ -63,7 +63,7 @@ extern const void bss_end;
 int lowlevel_init(void)
 {
 	// initialize .bss section
-	memset((void*)&bss_start, 0x00, ((size_t) &bss_end - (size_t) &bss_start));
+	memset(&bss_start, 0x00, (char*) &bss_end - (char*) &bss_start);
 
 	koutput_init();
 

kernel/main.c

@@ -29,6 +29,7 @@
 #include <metalsvm/fs.h>
 #include <asm/irq.h>
 #include <asm/irqflags.h>
+#include <asm/page.h>
 #include <asm/kb.h>
 #ifdef CONFIG_ROCKCREEK
 #include <asm/icc.h>
@@ -73,6 +74,7 @@ int main(void)
 	kprintf("This is MetalSVM %s Build %u, %u\n", 
 			METALSVM_VERSION, &__BUILD_DATE, &__BUILD_TIME);
 	popbg();
+
 	system_init();
 	irq_init();
 	timer_init();
@@ -85,7 +87,7 @@ int main(void)
 	icc_init();
 	svm_init();
 #endif
-   	initrd_init();
+	initrd_init();
 
 	irq_enable();
 
@@ -101,7 +103,7 @@ int main(void)
 	disable_timer_irq();
 #endif
 
-	sleep(5);
+	sleep(2);
 	create_kernel_task(&id, initd, NULL, NORMAL_PRIO);
 	kprintf("Create initd with id %u\n", id);
 	reschedule();

kernel/syscall.c

@@ -109,7 +109,7 @@ static int sys_open(const char* name, int flags, int mode)
 		curr_task->fildes_table[fd] = NULL;
 		return check;
  	}
-	
+
 	return fd;
 }
 
@@ -403,6 +403,7 @@ static int sys_sbrk(int incr)
 
 	spinlock_lock(&task->vma_lock);
 
+	// search vma containing the heap
 	tmp = task->vma_list;
         while(tmp && !((task->end_heap >= tmp->start) && (task->end_heap <= tmp->end)))
 		tmp = tmp->next;
@@ -411,11 +412,16 @@ static int sys_sbrk(int incr)
 	task->end_heap += incr;
 	if (task->end_heap < task->start_heap)
 		task->end_heap = task->start_heap;
-	
+
 	// resize virtual memory area
 	if (tmp && (tmp->end <= task->end_heap))
 		tmp->end = task->end_heap;
 
+	// allocation and mapping of new pages for the heap
+	// is catched by the pagefault handler
+
+	//kprintf("sys_sbrk: tid=%d, start_heap=%8x, end_heap=%8x, incr=%4x\n", task->id, task->start_heap, task->end_heap, incr);
+
 	spinlock_unlock(&task->vma_lock);
 
 	return ret;

kernel/tasks.c

@@ -78,6 +78,7 @@ DEFINE_PER_CORE(task_t*, current_task, task_table+0);
 extern const void boot_stack;
 
 /** @brief helper function for the assembly code to determine the current task
+ * 
  * @return Pointer to the task_t structure of current task
  */
 task_t* get_current_task(void) {
@@ -104,7 +105,7 @@ int multitasking_init(void) {
 
 	mailbox_wait_msg_init(&task_table[0].inbox);
 	memset(task_table[0].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
-	task_table[0].pgd = get_boot_pgd();
+	task_table[0].page_map = get_boot_page_map();
 	task_table[0].flags = TASK_DEFAULT_FLAGS;
 	task_table[0].prio = IDLE_PRIO;
 	task_table[0].stack = (void*) &boot_stack;
@@ -128,7 +129,7 @@ size_t get_idle_task(uint32_t id)
 	atomic_int32_set(&task_table[id].user_usage, 0);
 	mailbox_wait_msg_init(&task_table[id].inbox);
 	memset(task_table[id].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
-	task_table[id].pgd = get_boot_pgd();
+	task_table[id].page_map = get_boot_page_map();
 	current_task[id].var = task_table+id;
 	runqueues[id].idle = task_table+id;
 
@@ -193,10 +194,8 @@ static void wakeup_blocked_tasks(int result)
 	spinlock_irqsave_unlock(&table_lock);
 }
 
-/** @brief A procedure to be called by 
+/** @brief A procedure to be called by procedures which are called by exiting tasks. */
- * procedures which are called by exiting tasks. */
 static void NORETURN do_exit(int arg) {
-	vma_t* tmp;
 	task_t* curr_task = per_core(current_task);
 	uint32_t flags, core_id, fd, status;
 
@@ -204,14 +203,14 @@ static void NORETURN do_exit(int arg) {
 		for (fd = 0; fd < NR_OPEN; fd++) {
 			if(curr_task->fildes_table[fd] != NULL) {
 				/*
-			 	* delete a descriptor from the per-process object 
+				 * Delete a descriptor from the per-process object
-			 	* reference table.  If this is not the last reference to the underlying 
+				 * reference table.  If this is not the last reference to the underlying
-			 	* object, the object will be ignored. 
+				 * object, the object will be ignored.
-			 	*/
+				 */
 				if (curr_task->fildes_table[fd]->count == 1) {
-					/* try to close the file */
+					// try to close the file
 					status = close_fs(curr_task->fildes_table[fd]);
-					/* close command failed -> return check = errno */
+					// close command failed -> return check = errno
 					if (BUILTIN_EXPECT(status < 0, 0))
 						kprintf("Task %u was not able to close file descriptor %i. close_fs returned %d", curr_task->id, fd, -status);
 					kfree(curr_task->fildes_table[fd], sizeof(fildes_t));
@@ -230,24 +229,13 @@ static void NORETURN do_exit(int arg) {
 
 	wakeup_blocked_tasks(arg);
 
-	//vma_dump(curr_task);
+	drop_vma_list(); // kfree virtual memory areas and the vma_list
-	spinlock_lock(&curr_task->vma_lock);
+	drop_page_map(); // delete page directory and its page tables
-
-	// remove memory regions
-	while((tmp = curr_task->vma_list) != NULL) {
-		kfree((void*) tmp->start, tmp->end - tmp->start + 1);
-		curr_task->vma_list = tmp->next;
-		kfree((void*) tmp, sizeof(vma_t));
-	}
-
-	spinlock_unlock(&curr_task->vma_lock);
-
-	drop_pgd(); // delete page directory and its page tables
 
 #if 0	
 	if (atomic_int32_read(&curr_task->user_usage))
 		kprintf("Memory leak! Task %d did not release %d pages\n", 
-				curr_task->id, atomic_int32_read(&curr_task->user_usage));
+			curr_task->id, atomic_int32_read(&curr_task->user_usage));
 #endif
 	curr_task->status = TASK_FINISHED;
 
@@ -262,9 +250,7 @@ static void NORETURN do_exit(int arg) {
 	reschedule();
 
 	kprintf("Kernel panic: scheduler on core %d found no valid task\n", CORE_ID);
-	while(1) {
+	while(1) HALT;
-		HALT;
-	}
 }
 
 /** @brief A procedure to be called by kernel tasks */
@@ -327,10 +313,10 @@ static int create_task(tid_t* id, entry_point_t ep, void* arg, uint8_t prio, uin
 		if (task_table[i].status == TASK_INVALID) {
 			atomic_int32_set(&task_table[i].user_usage, 0);
 
-			ret = create_pgd(task_table+i, 0);		
+			ret = create_page_map(task_table+i, 0);
 			if (ret < 0) {
 				ret = -ENOMEM;
-				goto create_task_out;
+				goto out;
 			}
 
 			task_table[i].id = i;
@@ -376,7 +362,7 @@ static int create_task(tid_t* id, entry_point_t ep, void* arg, uint8_t prio, uin
 		}
 	}
 
-create_task_out:
+out:
 	spinlock_irqsave_unlock(&table_lock);
 
 	return ret;
@@ -387,11 +373,7 @@ int sys_fork(void)
 	int ret = -ENOMEM;
 	unsigned int i, core_id, fd_i;
 	task_t* parent_task = per_core(current_task);
-	vma_t** child;
-	vma_t* parent;
-	vma_t* tmp;
 
-	spinlock_lock(&parent_task->vma_lock);
 	spinlock_irqsave_lock(&table_lock);
 
 	core_id = CORE_ID;
@@ -400,46 +382,29 @@ int sys_fork(void)
 		if (task_table[i].status == TASK_INVALID) {
 			atomic_int32_set(&task_table[i].user_usage, 0);
 
-			ret = create_pgd(task_table+i, 1);		
+			ret = create_page_map(task_table+i, 1);
 			if (ret < 0) {
 				ret = -ENOMEM;
-				goto create_task_out;
+				goto out;
+			}
+
+			ret = copy_vma_list(child_task);
+			if (BUILTIN_EXPECT(!ret, 0)) {
+				ret = -ENOMEM;
+				goto out;
 			}
 
 			task_table[i].id = i;
 			task_table[i].last_stack_pointer = NULL;
                         task_table[i].stack = create_stack();
 
-			spinlock_init(&task_table[i].vma_lock);
+			// init fildes_table
-
-			// copy VMA list
-			child = &task_table[i].vma_list;
-			parent = parent_task->vma_list;
-			tmp = NULL;
-
-			while(parent) {
-				*child = (vma_t*) kmalloc(sizeof(vma_t));
-				if (BUILTIN_EXPECT(!child, 0))
-					break;
-
-				(*child)->start = parent->start;
-				(*child)->end = parent->end;
-				(*child)->type = parent->type;
-				(*child)->prev = tmp;
-				(*child)->next = NULL;
-
-				parent = parent->next;
-				tmp = *child;
-				child = &((*child)->next);
-			}
-			
-
-			/* init fildes_table */
 			task_table[i].fildes_table = kmalloc(sizeof(filp_t)*NR_OPEN);
 			memcpy(task_table[i].fildes_table, parent_task->fildes_table, sizeof(filp_t)*NR_OPEN);
-			for (fd_i = 0; fd_i < NR_OPEN; fd_i++)
+			for (fd_i = 0; fd_i < NR_OPEN; fd_i++) {
 				if ((task_table[i].fildes_table[fd_i]) != NULL)
 					task_table[i].fildes_table[fd_i]->count++;
+			}
 
 			mailbox_wait_msg_init(&task_table[i].inbox);
 			memset(task_table[i].outbox, 0x00, sizeof(mailbox_wait_msg_t*)*MAX_TASKS);
@@ -487,9 +452,8 @@ int sys_fork(void)
 		}
 	}
 
-create_task_out:
+out:
 	spinlock_irqsave_unlock(&table_lock);
-	spinlock_unlock(&parent_task->vma_lock);
 
 	return ret;
 }
@@ -679,7 +643,7 @@ static int load_task(load_args_t* largs)
 				flags |= VMA_WRITE;
 			if (prog_header.flags & PF_X)
 				flags |= VMA_EXECUTE;
-			vma_add(curr_task, prog_header.virt_addr, prog_header.virt_addr+npages*PAGE_SIZE-1, flags);
+			vma_add(prog_header.virt_addr, prog_header.virt_addr+npages*PAGE_SIZE-1, flags);
 
 			if (!(prog_header.flags & PF_W)) 
 				change_page_permissions(prog_header.virt_addr, prog_header.virt_addr+npages*PAGE_SIZE-1, flags);
@@ -708,7 +672,7 @@ static int load_task(load_args_t* largs)
 				flags |= VMA_WRITE;
 			if (prog_header.flags & PF_X)
 				flags |= VMA_EXECUTE;
-			vma_add(curr_task, stack, stack+npages*PAGE_SIZE-1, flags);
+			vma_add(stack, stack+npages*PAGE_SIZE-1, flags);
 			break;
 		 }
 	}
@@ -871,13 +835,11 @@ int create_user_task_on_core(tid_t* id, const char* fname, char** argv, uint32_t
 int sys_execve(const char* fname, char** argv, char** env)
 {
 	vfs_node_t* node;
-	vma_t* tmp;
 	size_t i, buffer_size = 0;
 	load_args_t* load_args = NULL;
 	char *dest, *src;
 	int ret, argc = 0;
 	int envc = 0;
-	task_t* curr_task = per_core(current_task);
 
 	node = findnode_fs((char*) fname);
 	if (!node || !(node->type == FS_FILE))
@@ -920,16 +882,8 @@ int sys_execve(const char* fname, char** argv, char** env)
 		while ((*dest++ = *src++) != 0);
 	}
 
-	spinlock_lock(&curr_task->vma_lock);
-
 	// remove old program
-	while((tmp = curr_task->vma_list) != NULL) {
+	drop_vma_list();
-		kfree((void*) tmp->start, tmp->end - tmp->start + 1);
-		curr_task->vma_list = tmp->next;
-		kfree((void*) tmp, sizeof(vma_t));
-	}
-
-	spinlock_unlock(&curr_task->vma_lock);
 
 	/*
 	 * we use a trap gate to enter the kernel

libkern/stdio.c

@@ -34,13 +34,7 @@
 #define VGA_EARLY_PRINT		1
 #define UART_EARLY_PRINT	2
 
-#ifdef CONFIG_VGA
-static uint32_t early_print = VGA_EARLY_PRINT;
-#elif defined(CONFIG_UART)
-static uint32_t early_print = UART_EARLY_PRINT;
-#else
 static uint32_t early_print = NO_EARLY_PRINT;
-#endif
 static spinlock_irqsave_t olock = SPINLOCK_IRQSAVE_INIT;
 static atomic_int32_t kmsg_counter = ATOMIC_INIT(0);
 static unsigned char kmessages[KMSG_SIZE] __attribute__ ((section(".kmsg"))) = {[0 ... KMSG_SIZE-1] = 0x00};
@@ -145,6 +139,10 @@ int koutput_init(void)
 {
 #ifdef CONFIG_VGA
 	vga_init();
+	early_print |= VGA_EARLY_PRINT;
+#endif
+#ifdef CONFIG_UART
+	early_print |= UART_EARLY_PRINT;
 #endif
 
 	return 0;
@@ -161,11 +159,11 @@ int kputchar(int c)
 	kmessages[pos % KMSG_SIZE] = (unsigned char) c;
 
 #ifdef CONFIG_VGA
-	if (early_print == VGA_EARLY_PRINT)
+	if (early_print & VGA_EARLY_PRINT)
 		vga_putchar(c);
 #endif
 #ifdef CONFIG_UART
-	if (early_print == UART_EARLY_PRINT)
+	if (early_print & UART_EARLY_PRINT)
 		uart_putchar(c);
 #endif
 
@@ -186,11 +184,11 @@ int kputs(const char *str)
 		pos = atomic_int32_inc(&kmsg_counter);
 		kmessages[pos % KMSG_SIZE] = str[i];
 #ifdef CONFIG_VGA
-		if (early_print == VGA_EARLY_PRINT)
+		if (early_print & VGA_EARLY_PRINT)
 			vga_putchar(str[i]);
 #endif
 #ifdef CONFIG_UART
-		if (early_print == UART_EARLY_PRINT)
+		if (early_print & UART_EARLY_PRINT)
 			uart_putchar(str[i]);
 #endif
 	}

mm/memory.c

@@ -37,17 +37,15 @@
 #endif
 
 /* 
- * 0 => free
+ * Set whole address space as occupied:
- * 1 => occupied
+ * 0 => free, 1 => occupied
- * 
- * Set whole address space as occupied
  */
-static uint8_t	 	bitmap[BITMAP_SIZE]; // = {[0 ... BITMAP_SIZE-1] = 0xFF};
+static uint8_t bitmap[BITMAP_SIZE] = {[0 ... BITMAP_SIZE-1] = 0xFF};
-static spinlock_t	bitmap_lock = SPINLOCK_INIT;
+static spinlock_t bitmap_lock = SPINLOCK_INIT;
-static size_t		alloc_start;
+
-atomic_int32_t		total_pages = ATOMIC_INIT(0);
+atomic_int32_t total_pages = ATOMIC_INIT(0);
-atomic_int32_t		total_allocated_pages = ATOMIC_INIT(0);
+atomic_int32_t total_allocated_pages = ATOMIC_INIT(0);
-atomic_int32_t		total_available_pages = ATOMIC_INIT(0);
+atomic_int32_t total_available_pages = ATOMIC_INIT(0);
 
 /* 
  * Note that linker symbols are not variables, they have no memory allocated for
@@ -74,8 +72,8 @@ inline static void page_set_mark(size_t i)
 	size_t index = i >> 3;
 	size_t mod = i & 0x7;
 
-	//if (page_marked(i))
+	if (page_marked(i))
-	//	kprintf("page %u is alread marked\n", i);
+		kprintf("page_set_mark(%u): already marked\n", i);
 
         bitmap[index] = bitmap[index] | (1 << mod);
 }
@@ -86,56 +84,155 @@ inline static void page_clear_mark(size_t i)
 	size_t mod = i % 8;
 
 	if (page_unmarked(i))
-		kprintf("page %u is already unmarked\n", i);
+		kprintf("page_clear_mark(%u): already unmarked\n", i);
 
 	bitmap[index] = bitmap[index] & ~(1 << mod);
 }
 
+size_t get_pages(uint32_t npages)
+{
+	// skip first page
+	static size_t start = 1;
+
+	uint32_t i, j, l;
+	uint32_t k = 0;
+	size_t ret = 0;
+
+	if (BUILTIN_EXPECT(!npages, 0))
+		return ret;
+
+	if (BUILTIN_EXPECT(npages > atomic_int32_read(&total_available_pages), 0))
+		return ret;
+
+	spinlock_lock(&bitmap_lock);
+	i = start;
+next_try:
+	while((k < BITMAP_SIZE) && page_marked(i)) {
+		k++;
+		i = (i+1) & (BITMAP_SIZE-1);
+	}
+
+	if (k >= BITMAP_SIZE)
+		goto oom;
+
+	for(j=1; (j<npages) && (i+j < BITMAP_SIZE) && (k < BITMAP_SIZE); j++, k++) {
+		if (page_marked(i+j)) {
+			i = (i+j) & (BITMAP_SIZE-1);
+			goto next_try;
+		}
+	}
+
+	if (i+j >= BITMAP_SIZE) {
+		i = 1;
+		goto next_try;
+	}
+
+	if (k >= BITMAP_SIZE)
+		goto oom;
+
+	ret = i*PAGE_SIZE;
+	kprintf("get_pages: ret 0x%x, i = %d, j = %d, npages = %d\n", ret, i, j, npages); // TODO: remove
+	for(l=i; l<i+j; l++)
+		page_set_mark(l);
+
+	start = i+j;
+	spinlock_unlock(&bitmap_lock);
+
+	atomic_int32_add(&total_allocated_pages, npages);
+	atomic_int32_sub(&total_available_pages, npages);
+
+	return ret;
+
+oom:
+	spinlock_unlock(&bitmap_lock);
+
+	return ret;
+}
+
+int put_pages(size_t phyaddr, size_t npages)
+{
+	if (BUILTIN_EXPECT(!phyaddr || !npages, 0))
+		return -EINVAL;
+
+	uint32_t index;
+	uint32_t base = phyaddr >> PAGE_SHIFT;
+
+	spinlock_lock(&bitmap_lock);
+	for (index=0; index<npages; index++)
+		page_clear_mark(base+index);
+	spinlock_unlock(&bitmap_lock);
+
+	atomic_int32_sub(&total_allocated_pages, npages);
+	atomic_int32_add(&total_available_pages, npages);
+
+	return 0;
+}
+
 int mmu_init(void)
 {
-	size_t kernel_size;
 	unsigned int i; 
 	size_t addr;
 	int ret = 0;
 
-	// at first, set default value of the bitmap
-	memset(bitmap, 0xFF, sizeof(uint8_t)*BITMAP_SIZE);
-
 #ifdef CONFIG_MULTIBOOT
-	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MEM_MAP)) {
+	if (mb_info) {
-		size_t end_addr;
+		if (mb_info->flags & MULTIBOOT_INFO_MEM_MAP) {
-		multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) ((size_t) mb_info->mmap_addr);
+			multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) ((size_t) mb_info->mmap_addr);
-		multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);
+			multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);
 
-		while (mmap < mmap_end) {
+			// mark available memory as free
-			if (mmap->type == MULTIBOOT_MEMORY_AVAILABLE) {
+			while (mmap < mmap_end) {
-				/* set the available memory as "unused" */
+				if (mmap->type == MULTIBOOT_MEMORY_AVAILABLE) {
-				addr = mmap->addr;
+					for (addr=mmap->addr; addr < mmap->addr + mmap->len; addr += PAGE_SIZE) {
-				end_addr = addr + mmap->len;
+						page_clear_mark(addr >> PAGE_SHIFT);
- 
+						atomic_int32_inc(&total_pages);
-				while (addr < end_addr) {
+						atomic_int32_inc(&total_available_pages);
-					page_clear_mark(addr >> PAGE_SHIFT);
+					}
-					addr += PAGE_SIZE;
-					atomic_int32_inc(&total_pages);
-					atomic_int32_inc(&total_available_pages);
 				}
+				mmap++;
 			}
-			mmap++;
 		}
-	} else {
+		else if (mb_info->flags & MULTIBOOT_INFO_MEM) {
-		kputs("Unable to initialize the memory management subsystem\n");
+			size_t page;
-		while(1) { 
+			size_t pages_lower = mb_info->mem_lower >> 2;
-			HALT;
+			size_t pages_upper = mb_info->mem_upper >> 2;
+
+			for (page=0; page<pages_lower; page++)
+				page_clear_mark(page);
+
+			for (page=0x100000; page<pages_upper+0x100000; page++)
+				page_clear_mark(page);
+
+			atomic_int32_add(&total_pages, pages_lower + pages_upper);
+			atomic_int32_add(&total_available_pages, pages_lower + pages_upper);
+		}
+		else {
+			kputs("Unable to initialize the memory management subsystem\n");
+			while (1) HALT;
+		}
+
+		// mark mb_info as used
+		page_set_mark((size_t) mb_info >> PAGE_SHIFT);
+		atomic_int32_inc(&total_allocated_pages);
+		atomic_int32_dec(&total_available_pages);
+
+		// mark modules list as used
+		if (mb_info->flags & MULTIBOOT_INFO_MODS) {
+			for(addr=mb_info->mods_addr; addr<mb_info->mods_addr+mb_info->mods_count*sizeof(multiboot_module_t); addr+=PAGE_SIZE) {
+				page_set_mark(addr >> PAGE_SHIFT);
+				atomic_int32_inc(&total_allocated_pages);
+				atomic_int32_dec(&total_available_pages);
+			}
 		}
 	}
 #elif defined(CONFIG_ROCKCREEK)
-	/* of course, the first slots belong to the private memory */
+	// of course, the first slots belong to the private memory
 	for(addr=0x00; addr<1*0x1000000; addr+=PAGE_SIZE) {
 		page_clear_mark(addr >> PAGE_SHIFT);
 		if (addr > addr + PAGE_SIZE)
 			break;
 		atomic_int32_inc(&total_pages);
-		atomic_int32_inc(&total_available_pages);	
+		atomic_int32_inc(&total_available_pages);
 	}
 
 	// Note: The last slot belongs always to the private memory.
@@ -147,71 +244,78 @@ int mmu_init(void)
 		atomic_int32_inc(&total_available_pages);
 	}
 
-	/*
+	// mark the bootinfo as used.
-	 * Mark the bootinfo as used.
-	 */
 	page_set_mark((size_t)bootinfo >> PAGE_SHIFT);
 	atomic_int32_inc(&total_allocated_pages);
 	atomic_int32_dec(&total_available_pages);
-
 #else
 	#error Currently, MetalSVM supports only the Multiboot specification or the RockCreek processor!
 #endif
 
-	kernel_size = (size_t) &kernel_end - (size_t) &kernel_start;
+	// mark kernel as used
-	if (kernel_size & (PAGE_SIZE-1))
+	for(addr=(size_t) &kernel_start; addr<(size_t) &kernel_end; addr+=PAGE_SIZE) {
-		kernel_size += PAGE_SIZE - (kernel_size & (PAGE_SIZE-1));
+		page_set_mark(addr >> PAGE_SHIFT);
-	atomic_int32_add(&total_allocated_pages, kernel_size >> PAGE_SHIFT);
+		atomic_int32_inc(&total_allocated_pages);
-	atomic_int32_sub(&total_available_pages, kernel_size >> PAGE_SHIFT);
+		atomic_int32_dec(&total_available_pages);
-
+	}
-	/* set kernel space as used */
-	for(i=(size_t) &kernel_start >> PAGE_SHIFT; i < (size_t) &kernel_end >> PAGE_SHIFT; i++)
-		page_set_mark(i);
-	if ((size_t) &kernel_end & (PAGE_SIZE-1))
-		page_set_mark(i);
-
-	alloc_start = (size_t) &kernel_end >> PAGE_SHIFT;	
-	if ((size_t) &kernel_end & (PAGE_SIZE-1))
-		alloc_start++;
 
 #if MAX_CORES > 1
-	// reserve physical page for SMP boot code
 	page_set_mark(SMP_SETUP_ADDR >> PAGE_SHIFT);
-	atomic_int32_add(&total_allocated_pages, 1);
+	atomic_int32_inc(&total_allocated_pages);
-	atomic_int32_sub(&total_available_pages, 1);
+	atomic_int32_dec(&total_available_pages);
 #endif
+
+	// enable paging and map SMP, VGA, Multiboot modules etc.
 	ret = paging_init();
 	if (ret) {
 		kprintf("Failed to initialize paging: %d\n", ret);
 		return ret;
 	}
 
+	// add kernel to VMA list
+	vma_add((size_t) &kernel_start & PAGE_MASK,
+		PAGE_ALIGN((size_t) &kernel_end),
+		VMA_READ|VMA_WRITE|VMA_EXECUTE|VMA_CACHEABLE);
+
+	// add LAPIC tp VMA list
+	vma_add((size_t) &kernel_start - PAGE_SIZE,
+		(size_t) &kernel_start,
+		VMA_READ|VMA_WRITE);
+
+#if MAX_CORES > 1
+	// reserve page for SMP boot code
+	vma_add(SMP_SETUP_ADDR & PAGE_MASK,
+		PAGE_ALIGN(SMP_SETUP_ADDR + PAGE_SIZE),
+		VMA_READ|VMA_WRITE|VMA_EXECUTE|VMA_CACHEABLE);
+#endif
+
 #ifdef CONFIG_MULTIBOOT
 	/* 
 	 * Modules like the init ram disk are already loaded.
 	 * Therefore, we set these pages as used.
 	 */
-	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MODS)) {
+	if (mb_info) {
-		multiboot_module_t* mmodule = (multiboot_module_t*) ((size_t) mb_info->mods_addr);
+		vma_add((size_t) mb_info & PAGE_MASK,
+			PAGE_ALIGN((size_t) mb_info + sizeof(multiboot_info_t)),
+			VMA_READ|VMA_CACHEABLE);
 
-		/*
+		if (mb_info->flags & MULTIBOOT_INFO_MODS) {
-		 * Mark the mb_info as used.
+			multiboot_module_t* mmodule = (multiboot_module_t*) ((size_t) mb_info->mods_addr);
-		 */
-		page_set_mark((size_t)mb_info >> PAGE_SHIFT);
-		atomic_int32_inc(&total_allocated_pages);
-		atomic_int32_dec(&total_available_pages);
 
-		for(addr = mb_info->mods_addr; addr < mb_info->mods_addr + mb_info->mods_count * sizeof(multiboot_module_t); addr += PAGE_SIZE) {
+			vma_add((size_t) mb_info->mods_addr & PAGE_MASK,
-			page_set_mark(addr >> PAGE_SHIFT);
+				PAGE_ALIGN((size_t) mb_info->mods_addr + mb_info->mods_count*sizeof(multiboot_module_t)),
-			atomic_int32_inc(&total_allocated_pages);
+				VMA_READ|VMA_CACHEABLE);
-			atomic_int32_dec(&total_available_pages);
-		}
 
-		for(i=0; i<mb_info->mods_count; i++, mmodule++) {
+			for(i=0; i<mb_info->mods_count; i++) {
-			for(addr=mmodule->mod_start; addr<mmodule->mod_end; addr+=PAGE_SIZE) {
+				vma_add(PAGE_ALIGN(mmodule[i].mod_start),
-				page_set_mark(addr >> PAGE_SHIFT);
+					PAGE_ALIGN(mmodule[i].mod_end),
-				atomic_int32_inc(&total_allocated_pages);
+					VMA_READ|VMA_WRITE|VMA_CACHEABLE);
-				atomic_int32_dec(&total_available_pages);
+
+				for(addr=mmodule[i].mod_start; addr<mmodule[i].mod_end; addr+=PAGE_SIZE) {
+					page_set_mark(addr >> PAGE_SHIFT);
+					atomic_int32_inc(&total_allocated_pages);
+					atomic_int32_dec(&total_available_pages);
+				}
 			}
 		}
 	}
@@ -239,8 +343,8 @@ int mmu_init(void)
 	 * The init ram disk are already loaded.
 	 * Therefore, we set these pages as used.
 	 */
-	for(addr=bootinfo->addr; addr < bootinfo->addr+bootinfo->size; addr+=PAGE_SIZE) {
+	for(addr=bootinfo->addr; addr<bootinfo->addr+bootinfo->size; addr+=PAGE_SIZE) {
-		// This area is already mapped, so we need to virt_to_phys() these addresses.
+		// this area is already mapped, so we need to virt_to_phys() these addresses.
 		page_set_mark(virt_to_phys(addr) >> PAGE_SHIFT);
 		atomic_int32_inc(&total_allocated_pages);
 		atomic_int32_dec(&total_available_pages);
@@ -250,83 +354,6 @@ int mmu_init(void)
 	return ret;
 }
 
-/*
- * Use first fit algorithm to find a suitable physical memory region
- */
-size_t get_pages(uint32_t npages)
-{
-	uint32_t i, j, l;
-	uint32_t k = 0;
-	size_t ret = 0;
-
-	if (BUILTIN_EXPECT(!npages, 0))
-		return ret;
-
-	if (BUILTIN_EXPECT(npages > atomic_int32_read(&total_available_pages), 0))
-		return ret;
-
-	spinlock_lock(&bitmap_lock);
-	i = alloc_start;
-next_try:
-	while((k < BITMAP_SIZE) && page_marked(i)) {
-		k++;
-		i = (i+1) & (BITMAP_SIZE-1);
-	}
-
-	if (k >= BITMAP_SIZE)
-		goto oom;
-
-	for(j=1; (j<npages) && (i+j < BITMAP_SIZE) && (k < BITMAP_SIZE); j++, k++) {
-		if (page_marked(i+j)) {
-			i = (i+j) & (BITMAP_SIZE-1);
-			goto next_try;
-		}
-	}
-
-	if (i+j >= BITMAP_SIZE) {
-		i = 0;
-		goto next_try;
-	}
-
-	if (k >= BITMAP_SIZE)
-		goto oom;
-
-	ret = i*PAGE_SIZE;
-	//kprintf("alloc: ret 0x%x, i = %d, j = %d, npages = %d\n", ret, i, j, npages);
-	for(l=i; l<i+j; l++)
-		page_set_mark(l);
-
-	alloc_start = i+j;
-	spinlock_unlock(&bitmap_lock);
-
-	atomic_int32_add(&total_allocated_pages, npages);
-	atomic_int32_sub(&total_available_pages, npages);
-
-	return ret;
-
-oom:
-	spinlock_unlock(&bitmap_lock);
-
-	return ret;
-}
-
-int put_page(size_t phyaddr)
-{
-	uint32_t index = phyaddr >> PAGE_SHIFT;
-
-	if (BUILTIN_EXPECT(!phyaddr, 0))
-		return -EINVAL;
-
-	spinlock_lock(&bitmap_lock);
-	page_clear_mark(index);
-	spinlock_unlock(&bitmap_lock);
-
-	atomic_int32_sub(&total_allocated_pages, 1);
-	atomic_int32_add(&total_available_pages, 1);
-
-	return 0;
-}
-
 void* mem_allocation(size_t sz, uint32_t flags)
 {
 	size_t phyaddr, viraddr;
@@ -371,7 +398,6 @@ void kfree(void* addr, size_t sz)
 
 		index = phyaddr >> PAGE_SHIFT;
 		page_clear_mark(index);
-		
 	}
 	spinlock_unlock(&bitmap_lock);
 

mm/vma.c

@@ -1,5 +1,5 @@
 /* 
- * Copyright 2011 Stefan Lankes, Chair for Operating Systems,
+ * Copyright 2011 Steffen Vogel, Chair for Operating Systems,
  *                               RWTH Aachen University
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,85 +17,322 @@
  * This file is part of MetalSVM.
  */
 
+#include <metalsvm/vma.h>
 #include <metalsvm/stdlib.h>
 #include <metalsvm/stdio.h>
 #include <metalsvm/tasks_types.h>
 #include <metalsvm/spinlock.h>
-#include <metalsvm/vma.h>
 #include <metalsvm/errno.h>
 
 /*
- * add a new virtual memory region to the list of VMAs
+ * Kernel space VMA list and lock
+ * 
+ * For bootstrapping we initialize the VMA list with one empty VMA
+ * (start == end) and expand this VMA by calls to vma_alloc()
  */
-int vma_add(task_t* task, size_t start, size_t end, uint32_t type)
+static vma_t vma_boot = { VMA_KERN_MAX, VMA_KERN_MAX, VMA_HEAP };
+static vma_t* vma_list = &vma_boot;
+static spinlock_t vma_lock = SPINLOCK_INIT;
+
+size_t vma_alloc(size_t size, uint32_t flags)
 {
-	vma_t* new_vma;
+	task_t* task = per_core(current_task);
-	
+	spinlock_t* lock;
-	if (BUILTIN_EXPECT(!task || start > end, 0))
+	vma_t** list;
+	size_t ret = 0;
+
+	kprintf("vma_alloc(0x%lx, 0x%x)\n", size, flags);
+
+	size_t base, limit; // boundaries for search
+	size_t start, end;
+
+	if (BUILTIN_EXPECT(!size, 0))
+		return 0;
+
+	if (flags & VMA_USER) {
+		base = VMA_KERN_MAX;
+		limit = VMA_USER_MAX;
+		list = &task->vma_list;
+		lock = &task->vma_lock;
+	}
+	else {
+		base = 0;
+		limit = VMA_KERN_MAX;
+		list = &vma_list;
+		lock = &vma_lock;
+	}
+
+	spinlock_lock(lock);
+
+	// "last" fit search for free memory area
+	vma_t* pred = *list;  // vma before current gap
+	vma_t* succ = NULL; // vma after current gap
+	do {
+		start = (pred) ? pred->end : base;
+		end = (succ) ? succ->start : limit;
+
+		if (end > start && end - start > size)
+			break; // we found a gap
+
+		succ = pred;
+		pred = (pred) ? pred->prev : NULL;
+	} while (pred || succ);
+
+	if (BUILTIN_EXPECT(end > limit || end < start || end - start < size, 0)) {
+		spinlock_unlock(lock);
+		return 0;
+	}
+
+	// resize existing vma
+	if (succ && succ->flags == flags) {
+		succ->start -= size;
+		ret = succ->start;
+	}
+	// insert new vma
+	else {
+		vma_t* new = kmalloc(sizeof(vma_t));
+		if (BUILTIN_EXPECT(!new, 0))
+			return 0;
+
+		new->start = end-size;
+		new->end = end;
+		new->flags = flags;
+		new->next = succ;
+		new->prev = pred;
+
+		if (pred)
+			pred->next = new;
+		if (succ)
+			succ->prev = new;
+		else
+			*list = new;
+
+		ret = new->start;
+	}
+
+	spinlock_unlock(lock);
+	return ret;
+}
+
+int vma_free(size_t start, size_t end)
+{
+	task_t* task = per_core(current_task);
+	spinlock_t* lock;
+	vma_t* vma;
+	vma_t** list;
+
+	if (BUILTIN_EXPECT(start >= end, 0))
 		return -EINVAL;
 
-	new_vma = kmalloc(sizeof(new_vma));
+	if (end <= VMA_KERN_MAX) {
-	if (!new_vma)
+		lock = &vma_lock;
-		return -ENOMEM;
+		list = &vma_list;
+	}
+	else if (start >= VMA_KERN_MAX) {
+		lock = &task->vma_lock;
+		list = &task->vma_list;
+	}
+	else
+		return -EINVAL;
+
+	if (BUILTIN_EXPECT(!*list, 0))
+		return -EINVAL;
+
+	spinlock_lock(lock);
+
+	// search vma
+	vma = *list;
+	while (vma) {
+		if (start >= vma->start && end <= vma->end) break;
+		vma = vma->prev;
+	}
+
+	if (BUILTIN_EXPECT(!vma, 0)) {
+		spinlock_unlock(lock);
+		return -EINVAL;
+	}
+
+	// free/resize vma
+	if (start == vma->start && end == vma->end) {
+		if (vma == *list)
+			*list = vma->next; // update list head
+		if (vma->prev)
+			vma->prev->next = vma->next;
+		if (vma->next)
+			vma->next->prev = vma->prev;
+		kfree(vma);
+	}
+	else if (start == vma->start)
+		vma->start = end;
+	else if (end == vma->end)
+		vma->end = start;
+	else {
+		vma_t* new = kmalloc(sizeof(vma_t));
+		if (BUILTIN_EXPECT(!new, 0)) {
+			spinlock_unlock(lock);
+			return -ENOMEM;
+		}
+
+		new->start = end;
+		vma->end = start;
+
+		new->end = vma->end;
+		new->next = vma->next;
+		new->prev = vma;
+		vma->next = new;
+	}
+
+	spinlock_unlock(lock);
+
+	return 0;
+}
+
+int vma_add(size_t start, size_t end, uint32_t flags)
+{
+	task_t* task = per_core(current_task);
+	spinlock_t* lock;
+	vma_t** list;
+
+	kprintf("vma_add(0x%lx, 0x%lx, 0x%x)\n", start, end, flags);
+
+	if (BUILTIN_EXPECT(start >= end, 0))
+		return -EINVAL;
+
+	if (flags & VMA_USER) {
+		list = &task->vma_list;
+		lock = &task->vma_lock;
+
+		// check if address is in userspace
+		if (BUILTIN_EXPECT(start < VMA_KERN_MAX, 0))
+			return -EINVAL;
+	}
+	else {
+		list = &vma_list;
+		lock = &vma_lock;
+
+		// check if address is in kernelspace
+		if (BUILTIN_EXPECT(end > VMA_KERN_MAX, 0))
+			return -EINVAL;
+	}
+
+	spinlock_lock(lock);
+
+	// search gap
+	vma_t* pred = *list;
+	vma_t* succ = NULL;
+	while (pred) {
+		if ((!pred || pred->end <= start) &&
+		    (!succ || succ->start >= end))
+			break;
+
+		succ = pred;
+		pred = pred->prev;
+	}
+
+	// resize existing vma
+	if (pred && pred->end == start && pred->flags == flags)
+		pred->end = end;
+	else if (succ && succ->start == end && succ->flags == flags)
+		succ->start = start;
+	// insert new vma
+	else {
+		vma_t* new = kmalloc(sizeof(vma_t));
+		if (BUILTIN_EXPECT(!new, 0))
+			return 0;
+
+		new->start = start;
+		new->end = end;
+		new->flags = flags;
+		new->next = succ;
+		new->prev = pred;
+
+		if (pred)
+			pred->next = new;
+		if (succ)
+			succ->prev = new;
+		else
+			*list = new;
+	}
+
+	spinlock_unlock(lock);
+
+	return 0;
+}
+
+int copy_vma_list(task_t* task)
+{
+	task_t* parent_task = per_core(current_task);
+
+	spinlock_init(&task->vma_lock);
+	spinlock_lock(&parent_task->vma_lock);
+	spinlock_lock(&task->vma_lock);
+
+	int ret = 0;
+	vma_t* last = NULL;
+	vma_t* parent = parent_task->vma_list;
+
+	while (parent) {
+		vma_t *new = kmalloc(sizeof(vma_t));
+		if (BUILTIN_EXPECT(!new, 0)) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		new->start = parent->start;
+		new->end = parent->end;
+		new->flags = parent->flags;
+		new->prev = last;
+
+		if (last)
+			last->next = new;
+		else
+			task->vma_list = new;
+
+		last = new;
+		parent = parent->next;
+	}
+
+out:
+	spinlock_unlock(&task->vma_lock);
+	spinlock_unlock(&parent_task->vma_lock);
+	return ret;
+}
+
+int drop_vma_list()
+{
+	task_t* task = per_core(current_task);
 
 	spinlock_lock(&task->vma_lock);
 
-	new_vma->start = start;
+	while(task->vma_list)
-	new_vma->end = end;
+		pfree((void*) task->vma_list->start, task->vma_list->end - task->vma_list->start);
-	new_vma->type = type;
-
-	if (!(task->vma_list)) {
-		new_vma->next = new_vma->prev = NULL;
-		task->vma_list = new_vma;
-	} else {
-		vma_t* tmp = task->vma_list;
-
-		while (tmp->next && tmp->start < start)
-			tmp = tmp->next;
-
-		new_vma->next = tmp->next;
-		new_vma->prev = tmp;
-		tmp->next = new_vma;
-	}
 
 	spinlock_unlock(&task->vma_lock);
 
 	return 0;
 }
 
-int vma_dump(task_t* task)
+void vma_dump()
 {
-	vma_t* tmp;
+	void print_vma(vma_t *vma) {
-
+		while (vma) {
-	if (BUILTIN_EXPECT(!task, 0))
+			kprintf("0x%lx - 0x%lx: size=%x, flags=%c%c%c\n", vma->start, vma->end, vma->end - vma->start,
-		return -EINVAL;
+				(vma->flags & VMA_READ) ? 'r' : '-',
-
+				(vma->flags & VMA_WRITE) ? 'w' : '-',
-	spinlock_lock(&task->vma_lock);
+				(vma->flags & VMA_EXECUTE) ? 'x' : '-');
-
+			vma = vma->prev;
-	tmp = task->vma_list;
+		}
-	while (tmp) {
-		kprintf("%8x - %8x: ", tmp->start, tmp->end);
-
-		if (tmp->type & VMA_READ)
-			kputs("r");
-		else
-			kputs("-");
-
-		if (tmp->type & VMA_WRITE)
-			kputs("w");
-		else
-			kputs("-");
-
-		if (tmp->type & VMA_EXECUTE)
-			kputs("x");
-		else
-			kputs("-");
-		kputs("\n");
-
-		tmp = tmp->next;
 	}
 
-	spinlock_unlock(&task->vma_lock);
+	task_t* task = per_core(current_task);
 
-	return 0;
+	kputs("Kernelspace VMAs:\n");
+	spinlock_lock(&vma_lock);
+	print_vma(vma_list);
+	spinlock_unlock(&vma_lock);
+
+	kputs("Userspace VMAs:\n");
+	spinlock_lock(&task->vma_lock);
+	print_vma(task->vma_list);
+	spinlock_unlock(&task->vma_lock);
 }

newlib/examples/hello.c

@@ -68,7 +68,7 @@ int main(int argc, char** argv)
 		exit(1);
 	}
 	testdirent = readdir(testdir);
-	printf("1. Dirent: %s", testdirent->d_name);
+	printf("1. Dirent: %s\n", testdirent->d_name);
 	closedir(testdir);
 
 	return errno;

newlib/examples/memtest.c

@@ -20,41 +20,64 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <dirent.h>
 
 int print_usage() {
-	printf("usage: [size mb/kb/b]");
+	printf("usage: size mb/kb/b [chunks]\n");
 	exit(0);
 }
 
 int main(int argc, char** argv)
 {
-	int m = 0;
+	int multp = 0;
-	uint32_t size = 0;
+	int size = 0;
-	if(argc <= 2)
+	int chunks = 1;
-		print_usage();
+
-	if(argc == 3) {
+	void **test;
-		if(!strcmp(argv[2], "mb"))
+
-			m = 1024*1024;
+	if (argc <= 2 || argc > 4)
-		else if(!strcmp(argv[2], "kb"))
-			m = 1024;
-		else if(!strcmp(argv[2], "b"))
-			m = 0;
-		else
-			print_usage();
-	}
-	if(argc > 3)
-		print_usage();
-	
-	size = atoi(argv[1]);
-	if(size <= 0)
 		print_usage();
 
-	size *= m;
+	size = atoi(argv[1]);
-	uint8_t* test = malloc(size);
+	if (size <= 0)
-	printf("malloc(%d) - START: %p END: %p \n", size, test, test + size);
+		print_usage();
+
+	if (!strcasecmp(argv[2], "mb"))
+		multp = 1024*1024;
+	else if (!strcasecmp(argv[2], "kb"))
+		multp = 1024;
+	else if (!strcasecmp(argv[2], "b"))
+		multp = 1;
+	else
+		print_usage();
+	size *= multp;
+
+	if (argc == 4)
+		chunks = atoi(argv[3]);
+
+	test = malloc(chunks * sizeof(void *));
+	if (!test)
+		printf("malloc(%d) - FAILED!\n", chunks * sizeof(void *));
+
+	// allocate...
+	int i;
+	for (i = 0; i < chunks; i++) {
+		test[i] = malloc(size);
+		if (test[i])
+			printf("malloc(%d)\tCHUNK: %d START: %p END: %p\n", size, i, test[i], test[i] + size);
+		else {
+			printf("malloc(%d)\tFAILED! Abort allocation, start with freeing memory\n", size);
+			break;
+		}
+	}
+
+	// and release again
+	for (i = 0; i < chunks; i++) {
+		if (test[i]) {
+			free(test[i]);
+			printf("free(%p)\tCHUNK: %d\n", test[i], i);
+		}
+	}
+
+	free(test);
 	return 0;
 }

script.gdb

@@ -1,7 +1,14 @@
 # Constant part of the script
+set disassembly-flavor intel
 symbol-file metalsvm.sym
 target remote localhost:1234
 
-# Configure breakpoints and everything as you wish here.
+# Debugging 32bit code
-break main
+#set architecture i386
+#break stublet
+#continue
+
+# Debugging 64bit code
+#set architecture i386:x86-64
+#break main
 continue