29 files changed, 649 insertions, 669 deletions

diff --git a/Makefile b/Makefile
index 2a388e8..e0a2ec6 100644
--- a/Makefile
+++ b/Makefile
@@ -5,7 +5,7 @@ DEBUGFLAGS	!= [ $(RELEASE) ] && echo "-flto -O2" || echo "-O0 -ggdb3 -DDEBUG"
 CFLAGS		= -fno-pie -ffreestanding -nostdlib -std=c17 -Wall -Wextra
 DEPFLAGS	= -MT $@ -MMD -MP -MF $@.d
 
-all: apos.bin
+all: kernel.bin
 
 # default values, overwrite if/when needed
 ARCH		?= riscv
@@ -20,10 +20,8 @@ OBJCOPY		?= objcopy
 # This makes sure .bss is loaded into the binary
 OBJCOPY_FLAGS	?= -Obinary --set-section-flags .bss=alloc,load,contents
 
-COMMON_SOURCES	!= echo common/*.c lib/fdt*.c
-KERNEL_SOURCES	!= echo kernel/*.c $(COMMON_SOURCES)
-INIT_SOURCES	:= $(COMMON_SOURCES)
-CLEANUP		:= build deps.mk kernel.* init.* apos.bin
+KERNEL_SOURCES	!= echo common/*.c lib/*.c
+CLEANUP		:= build deps.mk kernel.* apos.bin
 CLEANUP_CMD	:=
 
 include arch/$(ARCH)/source.mk
@@ -43,32 +41,20 @@ KERN_SIZE	= wc -c kernel.bin | cut -d ' ' -f 1
 KERN_INFO	= sed "s/<KERNEL_SIZE>/$$($(KERN_SIZE))/"
 
 KERNEL_LINK	:= arch/$(ARCH)/conf/kernel-link
-INIT_LINK	:= arch/$(ARCH)/conf/init-link
 
-KERNEL_OBJECTS	!= ./scripts/gen-deps --kern "$(KERNEL_SOURCES)"
-INIT_OBJECTS	!= ./scripts/gen-deps --init "$(INIT_SOURCES)"
-KERNEL_LD	!= ./scripts/gen-deps --kern-link "$(KERNEL_LINK).S"
-INIT_LD		!= ./scripts/gen-deps --init-link "$(INIT_LINK).S"
+KERNEL_OBJECTS	!= ./scripts/gen-deps --compile "$(KERNEL_SOURCES)"
+KERNEL_LD	!= ./scripts/gen-deps --link "$(KERNEL_LINK).S"
 
 include deps.mk
 
-apos.bin: kernel.bin init.bin
-	cat init.bin kernel.bin > $@
-
 kernel.elf: $(KERNEL_OBJECTS) $(KERNEL_LD)
 	$(GENELF) -T $(KERNEL_LD) $(KERNEL_OBJECTS) -o $@
 
 $(INIT_LD): kernel.bin
 
-init.elf: $(INIT_OBJECTS) $(INIT_LD)
-	$(GENELF) -T $(INIT_LD) $(INIT_OBJECTS) -o $@
-
 kernel.bin: kernel.elf
 	$(CROSS_COMPILE)$(OBJCOPY) $(OBJCOPY_FLAGS) $< $@
 
-init.bin: init.elf
-	$(CROSS_COMPILE)$(OBJCOPY) $(OBJCOPY_FLAGS) $< $@
-
 clean:
 	$(RM) -r $(CLEANUP)
 
diff --git a/TODO.txt b/TODO.txt
index 3691a68..149fcfc 100644
--- a/TODO.txt
+++ b/TODO.txt
@@ -1,22 +1,6 @@
-+ Add in sensible defines (uint64_t etc.) x
-+ Improve build process (Kbuild? nah, regular config.h files)
-+ Automate booting and starting qemu etc. (make run) (sort of)
-+ Add in pmem/vmem routines (both .init and .text!)
-
-+ Test string.c
-+ Write the kernel lol
-+ Test libfdt and see if it can be used to read memory geometry etc.
-
-+ Jump to vmem x
-+ Mark fdt, reserved mem, kernel, init and initrd in vmem. Remember to remove
-init from vmem when jumping to vmem
-+ Figure out best way to tell the init how large out kernel is x~
-+ Make code more robust, error checking etc
-+ Make all physical memory addresses use the same typedef x
-
-+ should I consider passing more pointers around rather than return structures?
-+ should probably come up with some kind of general error handling scheme?
-
-+ Fix unmap/map_vmem when called from virtual memory
-
 + Investigate GCC/Clang builtin atomic operations?
+
++ Add in mapping the same memory to two different processes
++ Create sensible thread handling
++ Add in init process loading
++ Make sure vmap is in 4K increments or whatever
diff --git a/arch/riscv/common/vmem.c b/arch/riscv/common/vmem.c
deleted file mode 100644
index 6b022e1..0000000
--- a/arch/riscv/common/vmem.c
+++ /dev/null
@@ -1,104 +0,0 @@
-#include <apos/string.h>
-#include <apos/pmem.h>
-#include <apos/vmem.h>
-#include <apos/mem.h>
-#include <apos/debug.h>
-#include <pages.h>
-
-#define pte_ppn(pte) (((pm_t)(pte)) >> 10)
-#define pte_flags(pte) (((pm_t)(pte)) & 0xff)
-#define to_pte(p, f) ((pm_to_pnum(p) << 10) + (f))
-#define pte_addr(pte) (pnum_to_paddr(pte_ppn(pte)))
-#define vm_to_index(a, o) (pm_to_index(a, o))
-
-#if defined(KERNEL)
-
-static vm_t *tmp_pte = 0;
-void arch_init_vmem(struct vm_branch_t *branch, vm_t pte)
-{
-	(void)branch;
-	tmp_pte = (vm_t *)pte;
-}
-
-/* this seems like an awful idea, should probably try to come up with a smarter
- * approach? flushing the cache every time is probably pretty slow and also
- * this code doesn't take into account the cpu ASID, although that's probably
- * not relevant here */
-
-/* what I probably should do is something like Linux, where the kernel has a
- * direct (or linear, I suppose?) so that all direct memory accesses are also
- * accessible from virtual memory. Not entirely sure how I should do that,
- * wouldn't there be situations where the RAM is larger than the allocated
- * memory region?
- *
- * Or maybe some kind of walker, where I maintain a virtual page table that
- * mirrors the physical page table? Not sure yet, but this seems to work for
- * _now_.
- */
-#define QUOTE2(x) #x
-#define QUOTE(x) QUOTE2(x)
-static void *set_vptr(uint8_t flags, pm_t a)
-{
-	*tmp_pte = to_pte(a, flags);
-	__asm__ ("sfence.vma %0, %1"
-			:: "rk" (0), "rK" (TMP_PTE)
-			: "memory");
-
-	return (void *)TMP_PTE;
-}
-
-#else
-struct vm_branch_t *arch_get_tmp_pte(struct vm_branch_t *branch)
-{
-	enum mm_order_t top = __mm_max_order;
-	while(top){
-		size_t idx = vm_to_index(TMP_PTE, top);
-		branch = (struct vm_branch_t *)pte_addr(branch->leaf[idx]);
-		top--;
-	}
-
-	return branch;
-}
-
-#define set_vptr(flags, a) (a)
-#endif
-
-void map_vmem(struct vm_branch_t *branch,
-		pm_t paddr, vm_t vaddr, uint8_t flags, enum mm_order_t order)
-{
-	enum mm_order_t top = __mm_max_order;
-	while (top != order) {
-		size_t idx = vm_to_index(vaddr, top);
-
-		if (!branch->leaf[idx]) {
-			pm_t new_leaf = alloc_page(MM_KPAGE, 0);
-			branch->leaf[idx] =
-				(struct vm_branch_t *)to_pte(new_leaf, VM_V);
-
-			void *leaf_ptr = (void *)set_vptr(VM_R | VM_W | VM_V, new_leaf);
-			memset(leaf_ptr, 0, sizeof(struct vm_branch_t));
-		}
-
-		pm_t pte = (pm_t)branch->leaf[idx];
-		pm_t branch_pptr = (pm_t)pte_addr(pte);
-		branch = (struct vm_branch_t *)set_vptr(VM_R | VM_W | VM_V, branch_pptr);
-		top--;
-	}
-
-	size_t idx = vm_to_index(vaddr, top);
-	branch->leaf[idx] = (struct vm_branch_t *)to_pte(paddr, flags);
-}
-
-void unmap_vmem(struct vm_branch_t *branch, vm_t vaddr, enum mm_order_t order)
-{
-	while (order) {
-		size_t idx = pm_to_index(vaddr, order);
-		branch = (struct vm_branch_t *)pte_addr(branch->leaf[idx]);
-	}
-
-	size_t idx = pm_to_index(vaddr, order);
-	if (branch->leaf[idx])
-		free_page(order, pte_addr(branch->leaf[idx]));
-
-	branch->leaf[idx] = 0;
-}
diff --git a/arch/riscv/conf/apos.its b/arch/riscv/conf/apos.its
index 7c05044..1581025 100644
--- a/arch/riscv/conf/apos.its
+++ b/arch/riscv/conf/apos.its
@@ -7,7 +7,7 @@
 	images {
 		kernel {
 			description = "Garbaggio";
-			data = /incbin/("../../../apos.bin");
+			data = /incbin/("../../../kernel.bin");
 			type = "kernel";
 			arch = "riscv";
 			os = "apos";
diff --git a/arch/riscv/conf/init-link.S b/arch/riscv/conf/init-link.S
deleted file mode 100644
index c9a6da2..0000000
--- a/arch/riscv/conf/init-link.S
+++ /dev/null
@@ -1,31 +0,0 @@
-OUTPUT_ARCH(riscv)
-ENTRY(_start)
-
-SECTIONS {
-	. = ABSOLUTE(PM_KERN);
-	__init_start = .;
-
-	/* objcopy only copies these three sections (as far as I'm aware) into the
-	   produces binary, so __init_end should point to the correct location in the
-	   final binary
-	 */
-	.text ALIGN(4K) : AT(0) {
-		*(.init.start)
-		*(.text*);
-	}
-
-	.rodata : {
-		*(.rodata*)
-	}
-
-	.data : {
-		*(.data*)
-	}
-
-	.bss : {
-		*(.bss*) *(COMMON)
-	}
-
-	__kernel_size = <KERNEL_SIZE>;
-	__init_end = . ;
-}
diff --git a/arch/riscv/conf/kernel-link.S b/arch/riscv/conf/kernel-link.S
index d3b1223..bf0ad28 100644
--- a/arch/riscv/conf/kernel-link.S
+++ b/arch/riscv/conf/kernel-link.S
@@ -2,10 +2,10 @@ OUTPUT_ARCH(riscv)
 ENTRY(main)
 
 SECTIONS {
-	. = ABSOLUTE(VM_KERN);
+	. = ABSOLUTE(PM_KERN);
 	__kernel_start = .;
 
-	.text ALIGN(4K) : AT(0) {
+	.text : AT(0) {
 		*(.kernel.start);
 		*(.text*);
 	}
diff --git a/arch/riscv/include/csr.h b/arch/riscv/include/csr.h
index ab25bdd..70e60bb 100644
--- a/arch/riscv/include/csr.h
+++ b/arch/riscv/include/csr.h
@@ -3,11 +3,39 @@
 
 #include <apos/utils.h>
 
-#define SATP_MODE_Sv32 0x80000000
-#define SATP_MODE_Sv39 0x8000000000000000
-#define SATP_MODE_Sv48 0x9000000000000000
+#define SATP_MODE_Sv32	0x80000000
+#define SATP_MODE_Sv39	0x8000000000000000
+#define SATP_MODE_Sv48	0x9000000000000000
 
-#define CSR_SATP 0x180
+
+/* supervisor CSR registers */
+#define CSR_SSTATUS	0x100
+#define CSR_SIE		0x104
+#define CSR_STVEC	0x105
+#define CSR_SCOUNTEREN	0x106
+
+#define CSR_SENVCFG	0x10A
+
+#define CSR_SSCRATCH	0x140
+#define CSR_SEPC	0x141
+#define CSR_SCAUSE	0x142
+#define CSR_STVAL	0x143
+#define CSR_SIP		0x144
+
+#define CSR_SATP	0x180
+
+#define CSR_SCONTEXT	0x5A8
+
+/* Exception causes */
+#define EXC_INST_MISALIGNED	0
+#define EXC_INST_ACCESS		1
+#define EXC_BREAKPOINT		3
+#define EXC_LOAD_ACCESS		5
+#define EXC_STORE_ACCESS	7
+#define EXC_SYSCALL		8
+#define EXC_INST_PAGE_FAULT	12
+#define EXC_LOAD_PAGE_FAULT	13
+#define EXC_STORE_PAGE_FAULT	15
 
 /* directly lifted from Linux:/arch/riscv/include/asm/asm.h:9-13 */
 #ifdef __ASSEMBLY__
diff --git a/arch/riscv/include/vmem.h b/arch/riscv/include/vmem.h
index d17aaf1..2cd3ac7 100644
--- a/arch/riscv/include/vmem.h
+++ b/arch/riscv/include/vmem.h
@@ -3,6 +3,7 @@
 
 #include <apos/types.h>
 #include <apos/attrs.h>
+#include <apos/pmem.h>
 
 #define VM_V (1 << 0)
 #define VM_R (1 << 1)
@@ -24,4 +25,11 @@ struct __packed vm_branch_t {
 	struct vm_branch_t *leaf[512];
 };
 
+
+void map_vmem(struct vm_branch_t *b,
+		pm_t paddr, vm_t vaddr, uint8_t flags, enum mm_order_t order);
+
+void unmap_vmem(struct vm_branch_t *b,
+		vm_t vaddr, enum mm_order_t order);
+
 #endif /* APOS_RISCV_VMAP_H */
diff --git a/arch/riscv/init/head.S b/arch/riscv/init/head.S
deleted file mode 100644
index 3589ead..0000000
--- a/arch/riscv/init/head.S
+++ /dev/null
@@ -1,7 +0,0 @@
-.section .init.start
-.global _start
-.type _start, @function
-_start:
-	li sp, PM_STACK_TOP
-	/* go to proper init */
-	j init
diff --git a/arch/riscv/init/init.c b/arch/riscv/init/init.c
deleted file mode 100644
index 721712c..0000000
--- a/arch/riscv/init/init.c
+++ /dev/null
@@ -1,337 +0,0 @@
-/* TODO: cleanup :P */
-
-#include <apos/init.h>
-#include <apos/sizes.h>
-#include <apos/types.h>
-#include <apos/string.h>
-#include <apos/debug.h>
-#include <apos/pmem.h>
-#include <apos/vmem.h>
-#include <apos/utils.h>
-#include <libfdt.h>
-#include <pages.h>
-#include <csr.h>
-
-struct pm_layout_t {
-	pm_t base;
-	pm_t top;
-};
-
-struct pm_orders_t {
-	size_t max_order;
-	size_t bits[10];
-	size_t page_shift;
-};
-
-static struct pm_layout_t get_memlayout(void *fdt)
-{
-	struct cell_info_t ci = get_reginfo(fdt, "/memory");
-	int mem_offset = fdt_path_offset(fdt, "/memory");
-	uint8_t *mem_reg =
-		(uint8_t *) fdt_getprop(fdt, mem_offset, "reg", NULL);
-
-	/* if riscv128 comes around we will probably see addr_cells == 4, but
-	 * I'm not too concerned about it at the moment */
-	pm_t base = (pm_t)fdt_load_int_ptr(ci.addr_cells, mem_reg);
-
-	if(ci.addr_cells == 2)
-		mem_reg += sizeof(fdt64_t);
-	else
-		mem_reg += sizeof(fdt32_t);
-
-	/* -1 because base is a legitimate memory address */
-	pm_t top = (pm_t)fdt_load_int_ptr(ci.size_cells, mem_reg) + base - 1;
-	return (struct pm_layout_t){base, top};
-}
-
-#ifdef DEBUG
-
-static void init_debug(void *fdt)
-{
-	struct dbg_info_t dbg = dbg_from_fdt(fdt);
-	dbg_init(dbg.dbg_ptr, dbg.dev);
-}
-
-#else
-#define init_debug(...)
-#endif
-
-static pm_t get_kerneltop()
-{
-	/* interesting, for some reason if I define these to be just char
-	 * pointers I get some wacky values. Not sure why that would be, but
-	 * this works. */
-	extern char __init_end, __kernel_size;
-	return (pm_t)&__init_end + (pm_t)&__kernel_size;
-}
-
-static pm_t get_initrdtop(void *fdt)
-{
-	int chosen_offset = fdt_path_offset(fdt, "/chosen");
-	struct cell_info_t ci = get_cellinfo(fdt, chosen_offset);
-
-	void *initrd_end_ptr = (void *)fdt_getprop(fdt, chosen_offset,
-			"linux,initrd-end", NULL);
-
-	return (pm_t)fdt_load_int_ptr(ci.addr_cells, initrd_end_ptr);
-}
-
-static pm_t get_initrdbase(void *fdt)
-{
-	int chosen_offset = fdt_path_offset(fdt, "/chosen");
-	struct cell_info_t ci = get_cellinfo(fdt, chosen_offset);
-
-	void *initrd_base_ptr = (void *)fdt_getprop(fdt, chosen_offset,
-			"linux,initrd-start", NULL);
-
-	return (pm_t)fdt_load_int_ptr(ci.addr_cells, initrd_base_ptr);
-}
-
-static pm_t get_fdttop(void *fdt)
-{
-	const char *b = (const char *)fdt;
-	return (pm_t)(b + fdt_totalsize(fdt));
-}
-
-static pm_t get_fdtbase(void *fdt)
-{
-	/* lol */
-	return (pm_t)fdt;
-}
-
-static void mark_area_used(pm_t base, pm_t top)
-{
-	size_t area_left = top - base;
-	/* TODO: add in a method to make sure that we use as large mappings as
-	 * possible. */
-	while(area_left >= MM_KPAGE_SIZE){
-		mark_used(base, MM_KPAGE);
-		area_left -= MM_KPAGE_SIZE;
-		base += MM_KPAGE_SIZE;
-	}
-
-	if(area_left != 0)
-		mark_used(base, MM_KPAGE_SIZE);
-}
-
-static void mark_reserved_mem(void *fdt)
-{
-	int rmem_offset = fdt_path_offset(fdt, "/reserved-memory/mmode_resv0");
-	struct cell_info_t ci = get_reginfo(fdt, "/reserved-memory/mmode_resv0");
-	uint8_t *rmem_reg = (uint8_t *)fdt_getprop(fdt, rmem_offset, "reg", NULL);
-
-	pm_t base = (pm_t)fdt_load_int_ptr(ci.addr_cells, rmem_reg);
-
-	if(ci.addr_cells == 2)
-		rmem_reg += sizeof(fdt64_t);
-	else
-		rmem_reg += sizeof(fdt32_t);
-
-	pm_t top = (pm_t)fdt_load_int_ptr(ci.size_cells, rmem_reg) + base - 1;
-	mark_area_used(base, top);
-}
-
-static struct pm_orders_t init_pmem(void *fdt)
-{
-	enum mm_mode_t mmode = get_mmode(fdt);
-
-	size_t max_order = 0;
-	size_t order_bits = 9;
-	switch(mmode){
-		case Sv32:
-			max_order = 1;
-			order_bits = 10;
-			break;
-
-		case Sv39:
-			max_order = 2;
-			break;
-
-		case Sv48:
-			max_order = 3;
-			break;
-	};
-
-	size_t bits[10] = {0};
-	for(size_t i = 0; i <= max_order; ++i)
-		bits[i] = order_bits;
-
-	init_mem(max_order, bits, 12);
-
-	struct pm_orders_t ret = {max_order, {0}, 12};
-	for(size_t i = 0; i <= __mm_max_order; ++i)
-		ret.bits[i] = bits[i];
-
-	return ret;
-}
-
-static struct pm_layout_t setup_pmem(void *fdt)
-{
-	struct pm_layout_t pmem = get_memlayout(fdt);
-
-	pm_t initrd_top = get_initrdtop(fdt);
-	pm_t kernel_top = get_kerneltop();
-	pm_t fdt_top = get_fdttop(fdt);
-
-	pm_t top = MAX3(kernel_top, initrd_top, fdt_top);
-	dbg("initrd_top:\t%#lx\n", initrd_top);
-	dbg("kernel_top:\t%#lx\n", kernel_top);
-	dbg("fdt_top:\t%#lx\n", fdt_top);
-
-	/* TODO: check that pmap placement doesn't overwrite anything, such as
-	 * stack or go over top address of memory */
-	size_t probe_size = probe_pmap(pmem.base, pmem.top - pmem.base);
-	/* riscv handles two byte boundaries better than one byte, so align
-	 * upwards */
-	pm_t pmap_base = align_up(top + 1, 2);
-	size_t actual_size = populate_pmap(pmem.base, pmem.top - pmem.base,
-			pmap_base);
-
-	/* TODO: not entirely sure what to do about this, probably give up trying to
-	 * boot? */
-	if(probe_size != actual_size){
-		dbg("BUG! probe_size (%#lx) != actual_size (%#lx)\n",
-				probe_size, actual_size);
-	}
-
-	/* mark init stack, at the moment always mapped to 2M */
-	mark_used(PM_STACK_BASE, MM_MPAGE);
-
-	/* mark kernel, at the moment it is always mapped to a 2M partition */
-	mark_used(PM_KERN, MM_MPAGE);
-
-	/* mark fdt and initrd */
-	mark_area_used(get_initrdbase(fdt), initrd_top);
-	mark_area_used(get_fdtbase(fdt), fdt_top);
-
-	/* mark pmap */
-	mark_area_used(pmap_base, pmap_base + actual_size);
-
-	/* mark reserved mem */
-	mark_reserved_mem(fdt);
-
-	return (struct pm_layout_t){.base = pmap_base, .top = actual_size + pmap_base};
-}
-
-pm_t move_kernel()
-{
-	extern char __init_end, __kernel_size;
-	pm_t dst = alloc_page(MM_MPAGE, 0);
-	memmove((void *)dst, &__init_end, (size_t)&__kernel_size);
-
-	return dst;
-}
-
-struct vm_branch_t *prepare_vmem()
-{
-	pm_t kernel_dst = move_kernel();
-
-	/* TODO: check if this actually works */
-	struct vm_branch_t *branch = (struct vm_branch_t *)alloc_page(MM_KPAGE, 0);
-	memset(branch, 0, sizeof(struct vm_branch_t));
-
-	/* TODO: check mapping flags, also iron out possible bugs etc in
-	 * map_vmem */
-	/* map kernel */
-	map_vmem(branch, kernel_dst, VM_KERN, VM_R | VM_W | VM_G | VM_X | VM_V, MM_MPAGE);
-
-	/* map init */
-	map_vmem(branch, PM_KERN, PM_KERN, VM_R | VM_W | VM_X | VM_V, MM_MPAGE);
-
-	/* map stack */
-	map_vmem(branch, PM_STACK_BASE, PM_STACK_BASE, VM_R | VM_W | VM_V, MM_MPAGE);
-
-	/* map root pte */
-	map_vmem(branch, (pm_t)branch, ROOT_PTE, VM_R | VM_W | VM_V, MM_KPAGE);
-
-	/* TODO: map more stuff? */
-	return branch;
-}
-
-static vm_t prepare_tmp_pte(struct vm_branch_t *branch)
-{
-	/* make sure the TMP_PTE is mapped to *something* */
-	map_vmem(branch, 0, TMP_PTE, VM_R | VM_W | VM_V, MM_KPAGE);
-
-	struct vm_branch_t *tmp_pte = arch_get_tmp_pte(branch);
-
-	vm_t addr = TMP_PTE + MM_KPAGE_SIZE;
-	map_vmem(branch, (vm_t)tmp_pte, addr, VM_R | VM_W | VM_V, MM_KPAGE);
-	return addr;
-}
-
-void start_vmem(void *fdt, struct vm_branch_t *branch)
-{
-	/* TODO: get ASID from CPU id */
-
-	/* TODO: probably unnecessary optimisations but this could be cached? */
-	enum mm_mode_t m = get_mmode(fdt);
-
-	if(m == Sv32)
-		csr_write(CSR_SATP, SATP_MODE_Sv32 | pm_to_pnum((pm_t)(branch)));
-	else if (m == Sv39)
-		csr_write(CSR_SATP, SATP_MODE_Sv39 | pm_to_pnum((pm_t)(branch)));
-	else
-		csr_write(CSR_SATP, SATP_MODE_Sv48 | pm_to_pnum((pm_t)(branch)));
-
-	__asm__ ("sfence.vma" : : : "memory");
-	/* Sv57 && Sv64 in the future? */
-}
-
-struct init_data_t populate_initdata(void *fdt, struct pm_orders_t o,
-		struct pm_layout_t p,
-		struct vm_branch_t *b,
-		vm_t v)
-{
-	extern char __init_start, __init_end;
-
-	struct init_data_t d = {0};
-	d.init_base = (pm_t)&__init_start;
-	d.init_top = (pm_t)&__init_end;
-
-	d.initrd_base = get_initrdbase(fdt);
-	d.initrd_top = get_initrdtop(fdt);
-
-	d.pmap_base = p.base;
-	d.pmap_top = p.top;
-
-	d.fdt_base = get_fdtbase(fdt);
-	d.fdt_top = get_fdttop(fdt);
-
-	d.stack_base = PM_STACK_BASE;
-	d.stack_top = PM_STACK_TOP;
-
-	d.kernel_vm_base = b;
-	d.tmp_pte = v;
-
-	d.max_order = o.max_order;
-	for(size_t i = 0; i <= __mm_max_order; ++i)
-		d.bits[i] = o.bits[i];
-	d.page_shift = o.page_shift;
-
-	/* initialize pmap in vmem */
-	pm_t addr = p.base;
-	while(addr < p.top){
-		map_vmem(b, addr, addr, VM_R | VM_W | VM_V, MM_MPAGE);
-		addr += MM_MPAGE_SIZE;
-	}
-
-	return d;
-}
-
-void init(void *fdt)
-{
-	init_debug(fdt);
-	dbg_fdt(fdt);
-
-	struct pm_orders_t o = init_pmem(fdt);
-	struct pm_layout_t p = setup_pmem(fdt);
-	struct vm_branch_t *b = prepare_vmem();
-	vm_t v = prepare_tmp_pte(b);
-	struct init_data_t d = populate_initdata(fdt, o, p, b, v);
-	start_vmem(fdt, b);
-
-	/* update_pmap(TODO: figure out where to place pmap in vmem); */
-	void (*main)(struct init_data_t) = (void (*)(struct init_data_t))VM_KERN;
-	main(d);
-}
diff --git a/arch/riscv/kernel/irq.c b/arch/riscv/kernel/irq.c
new file mode 100644
index 0000000..f015411
--- /dev/null
+++ b/arch/riscv/kernel/irq.c
@@ -0,0 +1,4 @@
+void irq_handler()
+{
+
+}
diff --git a/arch/riscv/kernel/main.c b/arch/riscv/kernel/main.c
index 3f524bc..c7a1290 100644
--- a/arch/riscv/kernel/main.c
+++ b/arch/riscv/kernel/main.c
@@ -1,51 +1,305 @@
 #include <apos/debug.h>
-#include <apos/init.h>
 #include <apos/vmem.h>
+#include <apos/pmem.h>
+#include <apos/types.h>
 #include <apos/mem.h>
 #include <apos/lock.h>
+#include <apos/tcb.h>
+#include <apos/string.h>
+#include <apos/proc.h>
+#include <apos/mem_nodes.h>
+#include <apos/elf.h>
+#include <apos/initrd.h>
+#include <libfdt.h>
+#include <csr.h>
+
+struct pm_layout_t {
+	pm_t base;
+	pm_t top;
+};
+
+struct pm_orders_t {
+	size_t max_order;
+	size_t bits[10];
+	size_t page_shift;
+};
 
 #ifdef DEBUG
 
-static void kernel_dbg(void *fdt)
+static void init_dbg(void *fdt)
 {
 	struct dbg_info_t dbg = dbg_from_fdt(fdt);
-	map_vmem(ROOT_PTE, (pm_t)dbg.dbg_ptr, (vm_t)dbg.dbg_ptr,
-			VM_R | VM_W | VM_V, MM_KPAGE);
 	dbg_init(dbg.dbg_ptr, dbg.dev);
 }
 
 #else
-#define kernel_dbg(...)
+#define init_dbg(...)
 #endif
 
-static void map_fdt(struct vm_branch_t *branch, vm_t fdt_base, vm_t fdt_top)
+static struct pm_layout_t get_memlayout(void *fdt)
+{
+	struct cell_info_t ci = get_reginfo(fdt, "/memory");
+	int mem_offset = fdt_path_offset(fdt, "/memory");
+	uint8_t *mem_reg =
+		(uint8_t *) fdt_getprop(fdt, mem_offset, "reg", NULL);
+
+	/* if riscv128 comes around we will probably see addr_cells == 4, but
+	 * I'm not too concerned about it at the moment */
+	pm_t base = (pm_t)fdt_load_int_ptr(ci.addr_cells, mem_reg);
+
+	if(ci.addr_cells == 2)
+		mem_reg += sizeof(fdt64_t);
+	else
+		mem_reg += sizeof(fdt32_t);
+
+	/* -1 because base is a legitimate memory address */
+	pm_t top = (pm_t)fdt_load_int_ptr(ci.size_cells, mem_reg) + base - 1;
+	return (struct pm_layout_t){base, top};
+}
+
+static pm_t get_kerneltop()
+{
+	/* interesting, for some reason if I define these to be just char
+	 * pointers I get some wacky values. Not sure why that would be, but
+	 * this works. */
+	extern char __kernel_end;
+	return (pm_t)&__kernel_end;
+}
+
+static pm_t get_initrdtop(void *fdt)
+{
+	int chosen_offset = fdt_path_offset(fdt, "/chosen");
+	struct cell_info_t ci = get_cellinfo(fdt, chosen_offset);
+
+	void *initrd_end_ptr = (void *)fdt_getprop(fdt, chosen_offset,
+			"linux,initrd-end", NULL);
+
+	return (pm_t)fdt_load_int_ptr(ci.addr_cells, initrd_end_ptr);
+}
+
+static pm_t get_initrdbase(void *fdt)
+{
+	int chosen_offset = fdt_path_offset(fdt, "/chosen");
+	struct cell_info_t ci = get_cellinfo(fdt, chosen_offset);
+
+	void *initrd_base_ptr = (void *)fdt_getprop(fdt, chosen_offset,
+			"linux,initrd-start", NULL);
+
+	return (pm_t)fdt_load_int_ptr(ci.addr_cells, initrd_base_ptr);
+}
+
+static pm_t get_fdttop(void *fdt)
+{
+	const char *b = (const char *)fdt;
+	return (pm_t)(b + fdt_totalsize(fdt));
+}
+
+static pm_t get_fdtbase(void *fdt)
+{
+	/* lol */
+	return (pm_t)fdt;
+}
+
+
+static void mark_area_used(pm_t base, pm_t top)
+{
+	size_t area_left = top - base;
+	/* TODO: add in a method to make sure that we use as large mappings as
+	 * possible. */
+	while(area_left >= MM_KPAGE_SIZE){
+		mark_used(base, MM_KPAGE);
+		area_left -= MM_KPAGE_SIZE;
+		base += MM_KPAGE_SIZE;
+	}
+
+	if(area_left != 0)
+		mark_used(base, MM_KPAGE_SIZE);
+}
+
+static void mark_reserved_mem(void *fdt)
+{
+	int rmem_offset = fdt_path_offset(fdt, "/reserved-memory/mmode_resv0");
+	struct cell_info_t ci = get_reginfo(fdt, "/reserved-memory/mmode_resv0");
+	uint8_t *rmem_reg = (uint8_t *)fdt_getprop(fdt, rmem_offset, "reg", NULL);
+
+	pm_t base = (pm_t)fdt_load_int_ptr(ci.addr_cells, rmem_reg);
+
+	if(ci.addr_cells == 2)
+		rmem_reg += sizeof(fdt64_t);
+	else
+		rmem_reg += sizeof(fdt32_t);
+
+	pm_t top = (pm_t)fdt_load_int_ptr(ci.size_cells, rmem_reg) + base - 1;
+	mark_area_used(base, top);
+}
+
+static struct pm_layout_t setup_pmem(void *fdt)
+{
+	struct pm_layout_t pmem = get_memlayout(fdt);
+
+	pm_t initrd_top = get_initrdtop(fdt);
+	pm_t kernel_top = get_kerneltop();
+	pm_t fdt_top = get_fdttop(fdt);
+
+	pm_t top = MAX3(kernel_top, initrd_top, fdt_top);
+	dbg("initrd_top:\t%#lx\n", initrd_top);
+	dbg("kernel_top:\t%#lx\n", kernel_top);
+	dbg("fdt_top:\t%#lx\n", fdt_top);
+
+	/* TODO: check that pmap placement doesn't overwrite anything, such as
+	 * stack or go over top address of memory */
+	size_t probe_size = probe_pmap(pmem.base, pmem.top - pmem.base);
+	/* riscv handles two byte boundaries better than one byte, so align
+	 * upwards */
+	pm_t pmap_base = align_up(top + 1, 2);
+	size_t actual_size = populate_pmap(pmem.base, pmem.top - pmem.base,
+			pmap_base);
+
+	/* TODO: not entirely sure what to do about this, probably give up trying to
+	 * boot? */
+	if(probe_size != actual_size){
+		dbg("BUG! probe_size (%#lx) != actual_size (%#lx)\n",
+				probe_size, actual_size);
+	}
+
+	/* mark init stack, at the moment always mapped to 2M */
+	mark_used(PM_STACK_BASE, MM_MPAGE);
+
+	/* mark kernel, at the moment it is always mapped to a 2M partition */
+	mark_used(PM_KERN, MM_MPAGE);
+
+	/* mark fdt and initrd */
+	mark_area_used(get_initrdbase(fdt), initrd_top);
+	mark_area_used(get_fdtbase(fdt), fdt_top);
+
+	/* mark pmap */
+	mark_area_used(pmap_base, pmap_base + actual_size);
+
+	/* mark reserved mem */
+	mark_reserved_mem(fdt);
+
+	return (struct pm_layout_t){.base = pmap_base, .top = actual_size + pmap_base};
+}
+
+static struct pm_orders_t init_pmem(void *fdt)
 {
-	/* TODO: fix this shit */
-	map_vregion(branch, fdt_base, fdt_base, fdt_top - fdt_base, VM_R | VM_W |VM_V);
+	enum mm_mode_t mmode = get_mmode(fdt);
+
+	size_t max_order = 0;
+	size_t order_bits = 9;
+	switch(mmode){
+		case Sv32:
+			max_order = 1;
+			order_bits = 10;
+			break;
+
+		case Sv39:
+			max_order = 2;
+			break;
+
+		case Sv48:
+			max_order = 3;
+			break;
+	};
+
+	size_t bits[10] = {0};
+	for(size_t i = 0; i <= max_order; ++i)
+		bits[i] = order_bits;
+
+	init_mem(max_order, bits, 12);
+
+	struct pm_orders_t ret = {max_order, {0}, 12};
+	for(size_t i = 0; i <= __mm_max_order; ++i)
+		ret.bits[i] = bits[i];
+
+	return ret;
+}
+
+static void populate_root_branch(struct vm_branch_t *b, size_t mul)
+{
+	for(size_t i = 0; i <= (MM_KPAGE_SIZE / sizeof(size_t))/2; ++i)
+		b->leaf[i] = (struct vm_branch_t *)(mul * i | VM_V | VM_R | VM_W | VM_X | VM_G);
+}
+
+static void start_vmem(struct vm_branch_t *branch, enum mm_mode_t m)
+{
+	/* TODO: get ASID from CPU id */
+
+	if(m == Sv32)
+		csr_write(CSR_SATP, SATP_MODE_Sv32 | pm_to_pnum((pm_t)(branch)));
+	else if (m == Sv39)
+		csr_write(CSR_SATP, SATP_MODE_Sv39 | pm_to_pnum((pm_t)(branch)));
+	else
+		csr_write(CSR_SATP, SATP_MODE_Sv48 | pm_to_pnum((pm_t)(branch)));
+
+	__asm__ ("sfence.vma" : : : "memory");
+	/* Sv57 && Sv64 in the future? */
+}
+
+static struct vm_branch_t *init_vmem(void *fdt)
+{
+	struct pm_orders_t o = init_pmem(fdt);
+	setup_pmem(fdt);
+
+	struct vm_branch_t *b = (struct vm_branch_t *)alloc_page(MM_KPAGE, 0);
+	memset(b, 0, sizeof(struct vm_branch_t));
+
+	enum mm_mode_t mm = Sv48;
+	size_t mul = 0;
+	switch(o.max_order){
+		case 3: /* Sv48 */
+			mm = Sv48;
+			mul = 1UL << 37;
+			break;
+
+		case 2: /* Sv39 */
+			mm = Sv39;
+			mul = 1UL << 28;
+			break;
+
+		case 1: /* Sv32 */
+			mm = Sv32;
+			mul = 1UL << 20;
+			break;
+	}
+
+	set_uvmem_size(mul);
+	populate_root_branch(b, uvmem_size());
+	/* jump to vmem */
+	start_vmem(b, mm);
+	return b;
+}
+
+static void init_irq(void *fdt)
+{
+}
+
+static void init_proc(void *fdt, struct vm_branch_t *b)
+{
+	struct tcb *t = (struct tcb *)alloc_page(MM_KPAGE, 0);
+	t->b_r = b;
+	t->pid = 0;
+	t->tid = 0;
+	threads_insert(t);
+	sp_mem_init(&t->sp_r, uvmem_size());
+
+	/* binary itself */
+	size_t sz = align_up(get_init_size(fdt), BASE_PAGE_SIZE);
+	t->bin = alloc_uvmem(t, sz, VM_V | VM_X);
+	/* stack */
+	t->stack = alloc_uvmem(t, SZ_2M, VM_V | VM_R | VM_W);
+	/* if it needs heap, it'll ask for it */
+
+	move_init(fdt, (void *)t->bin, sz);
+	jump_to_userspace(t, 0, 0);
 }
 
-void __main main(struct init_data_t d)
+void __main main(void *fdt)
 {
-	init_pmap((void *)d.pmap_base);
-	init_mem(d.max_order, d.bits, d.page_shift);
-	init_vmem(ROOT_PTE, d.tmp_pte);
-	/* TODO: move pmap into vmem somewhere */
-	/* TODO: mark kernel area as used */
-	map_fdt(ROOT_PTE, d.fdt_base, d.fdt_top);
-	kernel_dbg((void *)d.fdt_base);
-	dbg("test\n");
-	/* TODO: approximate order of business:
-	 * setup debugging in vmem (requires mapping fdt)
-	 * free unnecessary init
-	 * setup interrupts (should this be done in init?)
-	 * load init from initrd
-	 * setup init environment (thread control blocks etc.)
-	 * jumpstart init (free stack init setup)
-	 */
+	init_dbg(fdt);
+	struct vm_branch_t *b = init_vmem(fdt);
 
-	/* functionality that should be implemented:
-	 * figure out best continuous run of memory (pmap etc)
-	 * arbitrary user and kernel mapping (memory)
-	 * vm to pm
-	 */
+	init_mem_blocks();
+	init_irq(fdt);
+	init_proc(fdt, b);
 }
diff --git a/arch/riscv/kernel/proc.c b/arch/riscv/kernel/proc.c
new file mode 100644
index 0000000..3a39dd3
--- /dev/null
+++ b/arch/riscv/kernel/proc.c
@@ -0,0 +1,15 @@
+#include <apos/tcb.h>
+#include <apos/elf.h>
+#include <csr.h>
+
+void jump_to_userspace(struct tcb *t, char **argv, int argc)
+{
+	csr_write(CSR_SEPC, bin_entry(t->bin));
+	__asm__("mv sp, %0\n" : "=r" (t->stack) :: "memory");
+	__asm__("sret\n" ::: "memory");
+}
+
+void return_to_userspace(struct tcb *t)
+{
+	/* lol */
+}
diff --git a/arch/riscv/kernel/vmem.c b/arch/riscv/kernel/vmem.c
new file mode 100644
index 0000000..ffe8731
--- /dev/null
+++ b/arch/riscv/kernel/vmem.c
@@ -0,0 +1,52 @@
+#include <apos/string.h>
+#include <apos/pmem.h>
+#include <apos/vmem.h>
+#include <apos/mem.h>
+#include <apos/debug.h>
+#include <pages.h>
+
+#define pte_ppn(pte) (((pm_t)(pte)) >> 10)
+#define pte_flags(pte) (((pm_t)(pte)) & 0xff)
+#define to_pte(p, f) ((pm_to_pnum(p) << 10) + (f))
+#define pte_addr(pte) (pnum_to_paddr(pte_ppn(pte)))
+#define vm_to_index(a, o) (pm_to_index(a, o))
+
+void map_vmem(struct vm_branch_t *branch,
+		pm_t paddr, vm_t vaddr, uint8_t flags, enum mm_order_t order)
+{
+	enum mm_order_t top = __mm_max_order;
+	while (top != order) {
+		size_t idx = vm_to_index(vaddr, top);
+
+		if (!branch->leaf[idx]) {
+			pm_t new_leaf = alloc_page(MM_KPAGE, 0);
+			branch->leaf[idx] =
+				(struct vm_branch_t *)to_pte(new_leaf, VM_V);
+
+			void *leaf_ptr = (void *)new_leaf;
+			memset(leaf_ptr, 0, sizeof(struct vm_branch_t));
+		}
+
+		pm_t pte = (pm_t)branch->leaf[idx];
+		pm_t branch_pptr = (pm_t)pte_addr(pte);
+		branch = (struct vm_branch_t *)branch_pptr;
+		top--;
+	}
+
+	size_t idx = vm_to_index(vaddr, top);
+	branch->leaf[idx] = (struct vm_branch_t *)to_pte(paddr, flags);
+}
+
+void unmap_vmem(struct vm_branch_t *branch, vm_t vaddr, enum mm_order_t order)
+{
+	while (order) {
+		size_t idx = pm_to_index(vaddr, order);
+		branch = (struct vm_branch_t *)pte_addr(branch->leaf[idx]);
+	}
+
+	size_t idx = pm_to_index(vaddr, order);
+	if (branch->leaf[idx])
+		free_page(order, pte_addr(branch->leaf[idx]));
+
+	branch->leaf[idx] = 0;
+}
diff --git a/arch/riscv/proc.txt b/arch/riscv/proc.txt
new file mode 100644
index 0000000..29b2e92
--- /dev/null
+++ b/arch/riscv/proc.txt
@@ -0,0 +1,3 @@
+Each thread should obviously have a thread control block (tcb) which lives in
+kernelspace and has pointers to virtual memory allocations, along with PID and
+other required data. That's about what I can come up with at the moment.
diff --git a/arch/riscv/source.mk b/arch/riscv/source.mk
index b1dd361..ecc22bd 100644
--- a/arch/riscv/source.mk
+++ b/arch/riscv/source.mk
@@ -1,9 +1,6 @@
-KERNEL_LOCAL	!= echo arch/riscv/kernel/*.[cS] arch/riscv/common/*.[cS]
+KERNEL_LOCAL	!= echo arch/riscv/kernel/*.[cS]
 KERNEL_SOURCES	+= $(KERNEL_LOCAL)
 
-INIT_LOCAL	!= echo arch/riscv/init/*.[cS] arch/riscv/common/*.[cS]
-INIT_SOURCES	+= $(INIT_LOCAL)
-
 CLEANUP_CMD	:= ./arch/riscv/conf/rmimage.sh
 
 ARCH_FLAGS	:= -mcmodel=medany
diff --git a/common/elf.c b/common/elf.c
new file mode 100644
index 0000000..7531bd1
--- /dev/null
+++ b/common/elf.c
@@ -0,0 +1,9 @@
+#include <apos/elf.h>
+#include <apos/vmem.h>
+
+vm_t bin_entry(vm_t bin)
+{
+	struct elf_header *e = (struct elf_header *)bin;
+	/* TODO: check that address points to a valid ELF executable */
+	return (vm_t)e->e_entry;
+}
diff --git a/kernel/temp.c b/common/initrd.c
index e69de29..e69de29 100644
--- a/kernel/temp.c
+++ b/common/initrd.c
diff --git a/common/mem_nodes.c b/common/mem_nodes.c
index 1779276..dba714f 100644
--- a/common/mem_nodes.c
+++ b/common/mem_nodes.c
@@ -138,7 +138,7 @@ struct sp_mem *get_mem_node()
 			return __find_free_block(r);
 		}
 
-		n = n->left;
+		n = sp_left(n);
 	}
 
 	/* we need to allocate a new region */
diff --git a/common/tcb.c b/common/tcb.c
new file mode 100644
index 0000000..5b00524
--- /dev/null
+++ b/common/tcb.c
@@ -0,0 +1,55 @@
+#include <apos/tcb.h>
+#include <apos/utils.h>
+#include <apos/sp_tree.h>
+
+static struct sp_root t_root = (struct sp_root){0};
+
+#define tcb_container(x) \
+	container_of(x, struct tcb, sp_n)
+
+void threads_insert(struct tcb *t)
+{
+	if(!sp_root(t_root)){
+		sp_root(t_root) = &t->sp_n;
+		return;
+	}
+
+	struct sp_node *n = sp_root(t_root), *p = NULL;
+	enum sp_dir d = LEFT;
+
+	while(n){
+		struct tcb *tc = tcb_container(n);
+		p = n;
+
+		if(t->tid < tc->tid){
+			n = sp_left(n);
+			d = LEFT;
+		}
+
+		else {
+			n = sp_right(n);
+			d = RIGHT;
+		}
+	}
+
+	sp_insert(&sp_root(t_root), p, &t->sp_n, d);
+}
+
+struct tcb *threads_find(id_t tid)
+{
+	struct sp_node *n = sp_root(t_root);
+
+	while(n){
+		struct tcb *t = tcb_container(n);
+
+		if(t->tid == tid)
+			return t;
+
+		if(t->tid < tid)
+			n = sp_left(n);
+		else
+			n = sp_right(n);
+	}
+
+	return 0;
+}
diff --git a/common/vmem.c b/common/vmem.c
index 267fbb7..25a8617 100644
--- a/common/vmem.c
+++ b/common/vmem.c
@@ -1,3 +1,4 @@
+#include <vmem.h>
 #include <apos/vmem.h>
 #include <apos/mem_nodes.h>
 
@@ -5,8 +6,7 @@
 #define mark_region_unused(r) ((r) = 0)
 #define is_region_used(r) (r)
 
-static struct sp_root free_regions = (struct sp_root){0};
-static struct sp_root used_regions = (struct sp_root){0};
+static size_t __uvmem_size = 0;
 
 /* pretty major slowdown when we get to some really massive numbers, not
  * entirely sure why. Will need to check up on this at some point, have I
@@ -24,9 +24,9 @@ static struct sp_root used_regions = (struct sp_root){0};
  * maybe not even anything with sp_trees but more a weakness of binary trees in
  * general?
  */
-static struct sp_mem *sp_free_insert_region(struct sp_mem *m)
+static struct sp_mem *sp_free_insert_region(struct sp_reg_root *r, struct sp_mem *m)
 {
-	struct sp_node *n = sp_root(free_regions), *p = NULL;
+	struct sp_node *n = sp_root(r->free_regions), *p = NULL;
 	size_t start = m->start;
 	size_t size = m->end - m->start;
 	enum sp_dir d = LEFT;
@@ -59,17 +59,17 @@ static struct sp_mem *sp_free_insert_region(struct sp_mem *m)
 		}
 	}
 
-	if(sp_root(free_regions))
-		sp_insert(&sp_root(free_regions), p, &m->sp_n, d);
+	if(sp_root(r->free_regions))
+		sp_insert(&sp_root(r->free_regions), p, &m->sp_n, d);
 	else
-		sp_root(free_regions) = &m->sp_n;
+		sp_root(r->free_regions) = &m->sp_n;
 
 	return m;
 }
 
-static struct sp_mem *sp_used_insert_region(struct sp_mem *m)
+static struct sp_mem *sp_used_insert_region(struct sp_reg_root *r, struct sp_mem *m)
 {
-	struct sp_node *n = sp_root(used_regions), *p = NULL;
+	struct sp_node *n = sp_root(r->used_regions), *p = NULL;
 	vm_t start = m->start;
 	enum sp_dir d = LEFT;
 
@@ -93,19 +93,19 @@ static struct sp_mem *sp_used_insert_region(struct sp_mem *m)
 		}
 	}
 
-	if(sp_root(used_regions))
-		sp_insert(&sp_root(used_regions), p, &m->sp_n, d);
+	if(sp_root(r->used_regions))
+		sp_insert(&sp_root(r->used_regions), p, &m->sp_n, d);
 	else
-		sp_root(used_regions) = &m->sp_n;
+		sp_root(r->used_regions) = &m->sp_n;
 
 	return m;
 }
 
-int sp_mem_init(size_t arena_size)
+int sp_mem_init(struct sp_reg_root *r, size_t arena_size)
 {
 	struct sp_mem *m = get_mem_node();
 	m->end = arena_size;
-	sp_free_insert_region(m);
+	sp_free_insert_region(r, m);
 
 	return 0;
 }
@@ -122,10 +122,10 @@ static void __sp_mem_destroy(struct sp_node *n)
 	free_mem_node(m);
 }
 
-void sp_mem_destroy()
+void sp_mem_destroy(struct sp_reg_root *r)
 {
-	__sp_mem_destroy(sp_root(free_regions));
-	__sp_mem_destroy(sp_root(used_regions));
+	__sp_mem_destroy(sp_root(r->free_regions));
+	__sp_mem_destroy(sp_root(r->used_regions));
 }
 
 /* interestingly this is now the main bottleneck :D
@@ -133,9 +133,9 @@ void sp_mem_destroy()
  * eh, it's not a massive thing I guess, maybe the code could be a bit quicker
  * but I mean 10 000 000 memory allocations in 20 s is good enough for now
  * */
-static struct sp_mem *sp_used_find(vm_t start)
+static struct sp_mem *sp_used_find(struct sp_reg_root *r, vm_t start)
 {
-	struct sp_node *n = sp_root(used_regions);
+	struct sp_node *n = sp_root(r->used_regions);
 	while(n){
 		struct sp_mem *t = mem_container(n);
 		if(start == t->start)
@@ -161,9 +161,9 @@ static struct sp_mem *sp_mem_create_region(vm_t start, vm_t end,
 	return m;
 }
 
-static struct sp_mem *sp_free_find_first(size_t size, size_t alignment)
+static struct sp_mem *sp_free_find_first(struct sp_reg_root *r, size_t size, size_t alignment)
 {
-	struct sp_node *n = sp_root(free_regions);
+	struct sp_node *n = sp_root(r->free_regions);
 	while(n){
 		struct sp_mem *t = mem_container(n);
 		size_t nsize = t->end - align_up(t->start, alignment);
@@ -181,13 +181,13 @@ static struct sp_mem *sp_free_find_first(size_t size, size_t alignment)
  * just ignore them for now. Note that alloc_region should only be used when
  * mmap is called with MAP_ANON, all other situations should be handled in some
  * fs server */
-vm_t alloc_region(size_t size, size_t alignment)
+vm_t alloc_region(struct sp_reg_root *r, size_t size, size_t alignment)
 {
-	struct sp_mem *m = sp_free_find_first(size, alignment);
+	struct sp_mem *m = sp_free_find_first(r, size, alignment);
 	if(!m)
 		return 0;
 
-	sp_remove(&sp_root(free_regions), &m->sp_n);
+	sp_remove(&sp_root(r->free_regions), &m->sp_n);
 
 	vm_t aligned_start = align_up(m->start, alignment);
 
@@ -206,7 +206,7 @@ vm_t alloc_region(size_t size, size_t alignment)
 		if(n->prev)
 			n->prev->next = n;
 
-		sp_free_insert_region(n);
+		sp_free_insert_region(r, n);
 	}
 
 	if(post_start != post_end){
@@ -215,17 +215,17 @@ vm_t alloc_region(size_t size, size_t alignment)
 		if(n->next)
 			n->next->prev = n;
 
-		sp_free_insert_region(n);
+		sp_free_insert_region(r, n);
 	}
 
 	m->end = end;
 	m->start = start;
 	mark_region_used(m->flags);
-	sp_used_insert_region(m);
+	sp_used_insert_region(r, m);
 	return start;
 }
 
-static void __sp_try_coalesce_prev(struct sp_mem *m)
+static void __sp_try_coalesce_prev(struct sp_reg_root *r, struct sp_mem *m)
 {
 	while(m){
 		if(!m || is_region_used(m->flags))
@@ -241,14 +241,14 @@ static void __sp_try_coalesce_prev(struct sp_mem *m)
 		if(m->prev)
 			m->prev->next = m;
 
-		sp_remove(&sp_root(free_regions), &p->sp_n);
+		sp_remove(&sp_root(r->free_regions), &p->sp_n);
 		free_mem_node(p);
 
 		m = m->prev;
 	}
 }
 
-static void __sp_try_coalesce_next(struct sp_mem *m)
+static void __sp_try_coalesce_next(struct sp_reg_root *r, struct sp_mem *m)
 {
 	while(m){
 		if(!m || is_region_used(m->flags))
@@ -264,28 +264,62 @@ static void __sp_try_coalesce_next(struct sp_mem *m)
 		if(m->next)
 			m->next->prev = m;
 
-		sp_remove(&sp_root(free_regions), &n->sp_n);
+		sp_remove(&sp_root(r->free_regions), &n->sp_n);
 		free_mem_node(n);
 
 		m = m->next;
 	}
 }
 
-static void sp_mem_try_coalesce(struct sp_mem *m)
+static void sp_mem_try_coalesce(struct sp_reg_root *r, struct sp_mem *m)
 {
-	__sp_try_coalesce_prev(m);
-	__sp_try_coalesce_next(m);
+	__sp_try_coalesce_prev(r, m);
+	__sp_try_coalesce_next(r, m);
 }
 
-void free_region(vm_t start)
+void free_region(struct sp_reg_root *r, vm_t start)
 {
-	struct sp_mem *m = sp_used_find(start);
+	struct sp_mem *m = sp_used_find(r, start);
 	if(!m)
 		return;
 
-	sp_remove(&sp_root(used_regions), &m->sp_n);
+	sp_remove(&sp_root(r->used_regions), &m->sp_n);
 	mark_region_unused(m->flags);
 
-	sp_mem_try_coalesce(m);
-	sp_free_insert_region(m);
+	sp_mem_try_coalesce(r, m);
+	sp_free_insert_region(r, m);
+}
+
+void set_uvmem_size(size_t s)
+{
+	__uvmem_size = s;
+}
+
+size_t uvmem_size()
+{
+	return __uvmem_size;
+}
+
+vm_t map_fill_region(struct vm_branch_t *b, vm_t start, size_t bytes, uint8_t flags)
+{
+	size_t pages = bytes / BASE_PAGE_SIZE;
+	pm_t offset = 0;
+	for(size_t i = 0; i < pages; ++i){
+		offset = alloc_page(BASE_PAGE, offset);
+		map_vmem(b, offset, start + i * BASE_PAGE_SIZE, flags, BASE_PAGE);
+	}
+
+	return start;
+}
+
+vm_t alloc_uvmem(struct tcb *t, size_t s, uint8_t flags)
+{
+	size_t sa = align_up(s, BASE_PAGE_SIZE);
+	vm_t v = alloc_region(&t->sp_r, sa, 0);
+	return map_fill_region(t->b_r, v, sa, flags);
+}
+
+void free_uvmem(struct tcb *t, vm_t a)
+{
+	free_region(&t->sp_r, a);
 }
diff --git a/include/apos/elf.h b/include/apos/elf.h
new file mode 100644
index 0000000..492c414
--- /dev/null
+++ b/include/apos/elf.h
@@ -0,0 +1,33 @@
+#ifndef APOS_ELF_H
+#define APOS_ELF_H
+
+#include <apos/attrs.h>
+#include <apos/types.h>
+#include <apos/vmem.h>
+
+__packed struct elf_header {
+	uint64_t ei_magic;
+	uint8_t ei_class;
+	uint8_t ei_data;
+	uint8_t ei_version;
+	uint8_t ei_osabi;
+	uint8_t ei_abiversion;
+	uint8_t e_pad;
+	uint16_t e_type;
+	uint16_t e_machine;
+	uint32_t e_version;
+	void *e_entry;
+	void *e_phoff;
+	void *e_shoff;
+	uint32_t e_flags;
+	uint16_t e_ehsize;
+	uint16_t e_phentsize;
+	uint16_t e_phnum;
+	uint16_t e_shentsize;
+	uint16_t e_shnum;
+	uint16_t e_shstrndx;
+};
+
+vm_t bin_entry(vm_t bin);
+
+#endif /* APOS_ELF_H */
diff --git a/include/apos/init.h b/include/apos/init.h
deleted file mode 100644
index 2dd8a27..0000000
--- a/include/apos/init.h
+++ /dev/null
@@ -1,31 +0,0 @@
-#ifndef APOS_INIT_H
-#define APOS_INIT_H
-
-#include <apos/pmem.h>
-#include <apos/vmem.h>
-
-struct init_data_t {
-	pm_t init_base;
-	pm_t init_top;
-
-	pm_t initrd_base;
-	pm_t initrd_top;
-
-	pm_t pmap_base;
-	pm_t pmap_top;
-
-	pm_t fdt_base;
-	pm_t fdt_top;
-
-	pm_t stack_base;
-	pm_t stack_top;
-
-	struct vm_branch_t *kernel_vm_base;
-	vm_t tmp_pte;
-
-	size_t max_order;
-	size_t bits[10];
-	size_t page_shift;
-};
-
-#endif /* APOS_INIT_H */
diff --git a/include/apos/initrd.h b/include/apos/initrd.h
new file mode 100644
index 0000000..942875f
--- /dev/null
+++ b/include/apos/initrd.h
@@ -0,0 +1,9 @@
+#ifndef APOS_INITRD_H
+#define APOS_INITRD_H
+
+#include <apos/types.h>
+
+size_t get_init_size(void *fdt);
+void move_init(void *fdt, void *target, size_t sz);
+
+#endif /* APOS_INITRD_H */
diff --git a/include/apos/pmem.h b/include/apos/pmem.h
index 38dd670..01d89ce 100644
--- a/include/apos/pmem.h
+++ b/include/apos/pmem.h
@@ -25,13 +25,7 @@ void free_page(enum mm_order_t order, pm_t paddr);
 void mark_used(enum mm_order_t order, pm_t paddr);
 pm_t alloc_page(enum mm_order_t order, pm_t offset);
 
-#if defined(INIT)
 pm_t populate_pmap(pm_t ram_base, size_t ram_size, pm_t cont);
 pm_t probe_pmap(pm_t ram_base, size_t ram_size);
-#endif
-
-#if defined(KERNEL)
-void init_pmap(void *p);
-#endif
 
 #endif /* APOS_PMEM_H */
diff --git a/include/apos/proc.h b/include/apos/proc.h
new file mode 100644
index 0000000..7506b10
--- /dev/null
+++ b/include/apos/proc.h
@@ -0,0 +1,7 @@
+#ifndef APOS_PROC_H
+#define APOS_PROC_H
+
+#include <apos/tcb.h>
+void jump_to_userspace(struct tcb *t, char **argv, int argc);
+
+#endif /* APOS_PROC_H */
diff --git a/include/apos/tcb.h b/include/apos/tcb.h
new file mode 100644
index 0000000..9cef386
--- /dev/null
+++ b/include/apos/tcb.h
@@ -0,0 +1,34 @@
+#ifndef APOS_TCB_H
+#define APOS_TCB_H
+
+struct tcb;
+struct sp_reg_root;
+
+#include <vmem.h>
+#include <apos/vmem.h>
+#include <apos/types.h>
+#include <apos/sp_tree.h>
+
+typedef size_t id_t;
+
+struct sp_reg_root {
+	struct sp_root free_regions;
+	struct sp_root used_regions;
+};
+
+struct tcb {
+	struct sp_node sp_n;
+
+	id_t pid;
+	id_t tid;
+
+	vm_t stack;
+	vm_t bin;
+
+	struct vm_branch_t *b_r;
+	struct sp_reg_root sp_r;
+};
+
+void threads_insert(struct tcb *t);
+
+#endif /* APOS_TCB_H */
diff --git a/include/apos/vmem.h b/include/apos/vmem.h
index eb7ad17..fe2a76f 100644
--- a/include/apos/vmem.h
+++ b/include/apos/vmem.h
@@ -1,11 +1,14 @@
 #ifndef APOS_VMEM_H
 #define APOS_VMEM_H
 
+struct sp_mem;
+
 /* arch-specific data */
 #include <vmem.h>
 
 /* common */
 #include <apos/pmem.h>
+#include <apos/tcb.h>
 #include <apos/sp_tree.h>
 
 struct sp_mem {
@@ -24,6 +27,7 @@ struct sp_mem {
 	container_of(ptr, struct sp_mem, sp_n)
 
 
+
 /* general overview of the different functions:
  * (un)map_vmem: map one known page of physical memory to one known page of
  * virtual memory
@@ -35,16 +39,14 @@ struct sp_mem {
  */
 
 /* defined by arch */
-void map_vmem(struct vm_branch_t *branch,
-		pm_t paddr, vm_t vaddr,
-		uint8_t flags, enum mm_order_t order);
+int sp_mem_init(struct sp_reg_root *r, size_t nums);
 
-void unmap_vmem(struct vm_branch_t *branch, vm_t vaddr, enum mm_order_t order);
+vm_t alloc_uvmem(struct tcb *r, size_t s, uint8_t flags);
+void free_uvmem(struct tcb *r, vm_t a);
 
+vm_t map_fill_region(struct vm_branch_t *b, vm_t start, size_t bytes, uint8_t flags);
 
-vm_t map_vregion(struct vm_branch_t *branch, pm_t base, vm_t start, size_t size,
-		uint8_t flags);
-void unmap_vregion(struct vm_branch_t *branch, vm_t start);
+size_t uvmem_size();
+void set_uvmem_size(size_t s);
 
-void init_vmem(struct vm_branch_t *branch, vm_t tmp_pte);
 #endif /* APOS_VMEM_H */
diff --git a/scripts/gen-deps b/scripts/gen-deps
index 7a468d3..ee64022 100755
--- a/scripts/gen-deps
+++ b/scripts/gen-deps
@@ -11,43 +11,25 @@ gencommon () {
 	echo "${obj}: ${s}"		>> deps.mk
 
 }
+
 genlink () {
 	gencommon ".ld"
-	if [ "${init}" = "0" ] ; then
-		echo "	\$(GENLINK) $< | \$(STRIPLINK) > \$@"\
-		>> deps.mk;
-	else
-		echo "	\$(GENLINK) $< | \$(STRIPLINK) | \$(KERN_INFO) > \$@"\
+	echo "	\$(GENLINK) $< | \$(STRIPLINK) | \$(KERN_INFO) > \$@"\
 		>> deps.mk;
-	fi
-
 }
 
 genrule () {
 	gencommon "${1}"
-	echo "	\$(COMPILE) ${defs} -c \$< -o \$@"	>> deps.mk
+	echo "	\$(COMPILE)  -c \$< -o \$@"	>> deps.mk
 }
 
 case "${1}" in
-	--kern)
-		suffix=.k.o
-		defs=-DKERNEL
-		func=genrule
-		;;
-	--init)
-		suffix=.i.o
-		defs=-DINIT
+	--compile)
+		suffix=.o
 		func=genrule
 		;;
-	--init-link)
-		suffix=.ld
-		init=1
-		func=genlink
-		;;
-
-	--kern-link)
+	--link)
 		suffix=.ld
-		init=0
 		func=genlink
 esac