Implement argument_ptr () syscall for handling process arguments

.clang-format

@@ -50,7 +50,7 @@ AlignOperands: false
 SortIncludes: true
 
 # Comments
-ReflowComments: false
+ReflowComments: true
 CommentPragmas: '^ IWYU pragma:'
 
 # Misc

.gdbinit

@@ -0,0 +1,2 @@
+file kernel/build/kernel.elf
+target remote :1234

.gitea/workflows/docs.yaml

@@ -0,0 +1,43 @@
+name: Build documentation
+
+on:
+  push:
+    branches:
+      - master
+
+jobs:
+  build-and-deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Install software
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y doxygen make rsync
+
+      - name: Set up python3
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+
+      - name: Install mkdocs
+        run: |
+          pip install --upgrade pip
+          pip install mkdocs mkdocs-material pymdown-extensions
+
+      - name: Build
+        run: make docs
+
+      - name: Deploy
+        env:
+          SSH_KEY: ${{ secrets.DEPLOY_SSH_KEY }}
+          REMOTE_IP: ${{ vars.DEPLOY_REMOTE_IP }}
+        run: |
+          mkdir -p ~/.ssh
+          echo "$SSH_KEY" > ~/.ssh/id_rsa
+          chmod 600 ~/.ssh/id_rsa
+          ssh-keyscan -H "$REMOTE_IP" >> ~/.ssh/known_hosts
+          chmod -R 777 site
+          rsync -az --delete site/ webuser@"$REMOTE_IP":/home/webuser/mop/

.gitignore

@@ -2,3 +2,5 @@ iso_root
 mop3.iso
 bochs-log.txt
 bochs-com1.txt
+mop3dist.tar
+site/

Makefile

@@ -1,12 +1,7 @@
 platform ?= amd64
 
-all_kernel:
-	make -C kernel platform=$(platform) all
-
-clean_kernel:
-	make -C kernel platform=$(platform) clean
-
-format_kernel:
-	make -C kernel platform=$(platform) format
-
-.PHONY: all_kernel clean_kernel format_kernel
+include make/apps.mk
+include make/kernel.mk
+include make/dist.mk
+include make/docs.mk
+include make/libmsl.mk

amd64/flags.mk

@@ -0,0 +1,11 @@
+cflags += --target=x86_64-pc-none-elf \
+					-mno-sse \
+					-mno-sse2 \
+					-mno-avx \
+					-mno-mmx \
+					-mno-80387 \
+					-mno-red-zone \
+					-mcmodel=large
+
+ldflags += --target=x86_64-pc-none-elf \
+					 -Wl,-zmax-page-size=0x1000

amd64/link.ld

@@ -0,0 +1,69 @@
+OUTPUT_FORMAT(elf64-x86-64)
+
+ENTRY(_start)
+
+PHDRS {
+  text PT_LOAD;
+  rodata PT_LOAD;
+  data PT_LOAD;
+  bss PT_LOAD;
+  tls PT_TLS;
+}
+
+SECTIONS {
+  . = 0x0000500000000000;
+
+  .text : {
+    *(.text .text.*)
+    *(.ltext .ltext.*)
+  } :text
+  
+  . = ALIGN(0x1000);
+  
+  .rodata : {
+    *(.rodata .rodata.*)
+  } :rodata
+
+  . = ALIGN(0x1000);
+
+  .data : {
+    *(.data .data.*)
+    *(.ldata .ldata.*)
+  } :data
+   
+  . = ALIGN(0x1000);
+
+  __bss_start = .;
+
+  .bss : {
+    *(.bss .bss.*)
+    *(.lbss .lbss.*)
+  } :bss
+
+  __bss_end = .;
+  
+  . = ALIGN(0x1000);
+
+  __tdata_start = .;
+
+  .tdata : {
+    *(.tdata .tdata.*)
+  } :tls
+  
+  __tdata_end = .;
+  
+  __tbss_start = .;
+
+  .tbss : {
+    *(.tbss .tbss.*)
+  } :tls
+  
+  __tbss_end = .;
+  
+  __tls_size = __tbss_end - __tdata_start;
+
+  /DISCARD/ : {
+    *(.eh_frame*)
+    *(.note .note.*)
+  }
+}

aux/bochsrc.txt

@@ -1,14 +1,15 @@
-cpu: model=p4_prescott_celeron_336
+cpu: model=p4_prescott_celeron_336, ips=200000000
 
 memory: guest=4096 host=2048
 
-romimage: file=/usr/share/bochs/BIOS-bochs-latest, options=fastboot
+romimage: file=/usr/share/bochs/BIOS-bochs-latest
 vgaromimage: file=/usr/share/bochs/VGABIOS-lgpl-latest.bin
 
 ata0: enabled=1
 ata0-master: type=cdrom, path=mop3.iso, status=inserted
 com1: enabled=1, mode=file, dev=bochs-com1.txt
 pci: enabled=1, chipset=i440fx
+clock: sync=realtime, time0=local
 
 boot: cdrom
 

aux/devel.sh

@@ -0,0 +1,14 @@
+#!/bin/sh
+
+set -xe
+
+if [ "$1" = "debug" ]; then
+  make -B all_kernel buildtype=debug
+else
+  make -B all_kernel
+fi
+
+make -B all_libmsl
+make -B all_apps
+make -B all_dist
+./aux/limine_iso_amd64.sh

aux/format.sh

@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -x
+
+make -B format_kernel
+make -B format_libmsl
+make -B format_apps

aux/limine_iso_amd64.sh

@@ -10,6 +10,8 @@ cp -v boot/limine/limine-bios.sys boot/limine/limine-bios-cd.bin \
 
 cp -v boot/limine/BOOTX64.EFI boot/limine/BOOTIA32.EFI iso_root/EFI/BOOT
 
+cp -v mop3dist.tar iso_root/boot
+
 xorriso -as mkisofs -R -r -J -b boot/limine/limine-bios-cd.bin \
   -no-emul-boot -boot-load-size 4 -boot-info-table -hfsplus \
   -apm-block-size 2048 --efi-boot boot/limine/limine-uefi-cd.bin \

aux/qemu_amd64.sh

@@ -1,3 +1,5 @@
 #!/bin/sh
 
-qemu-system-x86_64 -M q35 -m 4G -serial mon:stdio -cdrom mop3.iso
+set -x
+
+qemu-system-x86_64 -M q35 -m 4G -serial stdio -enable-kvm -cdrom mop3.iso -smp 4 $@

aux/qemu_amd64_debug.sh

@@ -0,0 +1,5 @@
+#!/bin/sh
+
+set -x
+
+qemu-system-x86_64 -M q35 -m 4G -serial stdio -cdrom mop3.iso -smp 4 -s -S $@

boot/limine.conf

@@ -3,3 +3,4 @@ timeout: 10
 /mop3
   protocol: limine
   path: boot():/boot/kernel.elf
+  module_path: boot():/boot/mop3dist.tar

docs/building_kernel.md

@@ -0,0 +1,44 @@
+# Building the MOP3 kernel
+
+This article describes, how to build the kernel, how the build system works and prerequisites.
+
+## Preprequistes
+
+- POSIX host system (tested on Linux, may break on other systems)
+- Git
+- GNU make
+- LLVM toolchain/Clang C compiler
+- Xorriso
+
+## Build steps
+
+cd into root of MOP3 source tree.
+
+Build the kernel:
+```
+make -B all_kernel buildtype=<debug|release>
+```
+
+Build essential system applications:
+```
+make -B all_apps
+```
+
+Prepare the ramdisk:
+```
+make -B all_dist
+```
+
+Build ISO image:
+```
+./aux/limine_iso_amd64.sh
+```
+
+Now you have an ISO image, which can be run my QEMU or you can burn it onto a CD.
+
+For the convenience of the developer, there's a magic "do all" script located in `aux`:
+```
+./aux/devel.sh # optionally "./aux/devel.sh debug" for debugging
+```
+
+It does all the previous steps, just packed into a single script.

docs/index.md

@@ -0,0 +1,4 @@
+# MOP3 operating system documentation
+
+MOP3 is a hobby OS project of mine ;).
+

docs/processes_overview.md

@@ -0,0 +1,30 @@
+# Overview of processes in MOP3
+
+## What is a process?
+
+A process is a structure defined to represent an internal state of a user application's environment. This includes
+the necessary stacks, code, data and other resources. A process (usually) has it's own address, but in certain
+circumstances may share it with another process.
+
+## Only processes vs. processes-threads model
+
+### Overview
+
+MOP3 doesn't have a process-thread separation. Ususally in operating systems you'd have a "process", which consists
+of multiple worker threads. For eg. a single-threaded application is a process, which consists of one worker. In MOP3
+we do things a little differently. We only have processes, but some processes may work within the same pool of (generally speaking)
+"resources", such as a shared address space, shared memory allocations, mutexes and so on. An application then consists of
+not threads, but processes, which are loosely tied together via shared data.
+
+#### Processes-threads model diagram
+![Processes-threads model](assets/images/processes-threads.png)
+#### Only processes model diagram
+![Only processes model](assets/images/only-processes.png)
+
+## Scheduling
+
+MOP3 uses a round-robin based scheduler. For now priorities are left unimplemented, ie. every processes has
+equal priority, but this may change in the future.
+
+A good explaination of round-robin scheduling can be found on the OSDev wiki: [the article](https://wiki.osdev.org/Scheduling_Algorithms#Round_Robin)
+

generic/flags.mk

@@ -0,0 +1,21 @@
+cflags += -nostdinc \
+					-nostdlib \
+					-ffreestanding \
+					-fno-builtin \
+					-std=c11 \
+					-pedantic \
+					-Wall \
+					-Wextra \
+					-ffunction-sections \
+					-fdata-sections
+
+cflags += -isystem ../include
+
+ldflags += -ffreestanding \
+					 -nostdlib \
+					 -fno-builtin \
+					 -fuse-ld=lld \
+					 -static \
+					 -Wl,--gc-sections \
+					 -Wl,--strip-all \
+					 -flto

include/m/status.h

@@ -0,0 +1,13 @@
+#ifndef _M_STATUS_H
+#define _M_STATUS_H
+
+#define ST_OK                0
+#define ST_SYSCALL_NOT_FOUND 1
+#define ST_UNALIGNED         2
+#define ST_OOM_ERROR         3
+#define ST_NOT_FOUND         4
+#define ST_BAD_ADDRESS_SPACE 5
+#define ST_PERMISSION_ERROR  6
+#define ST_BAD_RESOURCE      7
+
+#endif // _M_STATUS_H

include/m/syscall_defs.h

@@ -0,0 +1,16 @@
+#ifndef _M_SYSCALL_DEFS_H
+#define _M_SYSCALL_DEFS_H
+
+#define SYS_QUIT          1
+#define SYS_TEST          2
+#define SYS_MAP           3
+#define SYS_UNMAP         4
+#define SYS_CLONE         5
+#define SYS_SCHED         6
+#define SYS_MUTEX_CREATE  7
+#define SYS_MUTEX_DELETE  8
+#define SYS_MUTEX_LOCK    9
+#define SYS_MUTEX_UNLOCK  10
+#define SYS_ARGUMENT_PTR  11
+
+#endif // _M_SYSCALL_DEFS_H

init/.gitignore

@@ -0,0 +1,2 @@
+*.o
+*.exe

init/Makefile

@@ -0,0 +1 @@
+include ../make/user.mk

init/app.mk

@@ -0,0 +1 @@
+app := init.exe

init/init.c

@@ -0,0 +1,46 @@
+#include <limits.h>
+#include <proc/local.h>
+#include <proc/proc.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string/string.h>
+
+#define MUTEX 2000
+
+LOCAL volatile char letter = 'c';
+
+void app_proc (void) {
+  char arg_letter = (char)(uintptr_t)argument_ptr ();
+
+  letter = arg_letter;
+
+  for (;;) {
+    mutex_lock (MUTEX);
+
+    for (int i = 0; i < 3; i++)
+      test (letter);
+
+    mutex_unlock (MUTEX);
+  }
+
+  process_quit ();
+}
+
+void app_main (void) {
+  mutex_create (MUTEX);
+
+  letter = 'a';
+
+  process_spawn (&app_proc, (void*)'a');
+  process_spawn (&app_proc, (void*)'b');
+  process_spawn (&app_proc, (void*)'c');
+
+  for (;;) {
+    mutex_lock (MUTEX);
+
+    for (int i = 0; i < 3; i++)
+      test (letter);
+
+    mutex_unlock (MUTEX);
+  }
+}

init/src.mk

@@ -0,0 +1,3 @@
+c += init.c
+
+o += init.o

kernel/.gitignore

@@ -0,0 +1,2 @@
+*.json
+.cache

kernel/amd64/apic.c

@@ -1,8 +1,14 @@
 #include <amd64/apic.h>
+#include <amd64/intr_defs.h>
+#include <amd64/msr-index.h>
+#include <amd64/msr.h>
 #include <libk/std.h>
 #include <limine/requests.h>
+#include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
+#include <sys/spin.h>
+#include <sys/time.h>
 #include <uacpi/acpi.h>
 #include <uacpi/status.h>
 #include <uacpi/tables.h>
@@ -11,46 +17,94 @@
 #define IOAPICS_MAX                 24
 #define INTERRUPT_SRC_OVERRIDES_MAX 24
 
-static struct acpi_madt_ioapic apics[IOAPICS_MAX];
+/* ID of Local APIC */
+#define LAPIC_ID 0x20
+/* End of interrupt register */
+#define LAPIC_EOI 0xB0
+/* Spurious interrupt vector register */
+#define LAPIC_SIVR 0xF0
+/* Interrupt command register */
+#define LAPIC_ICR 0x300
+/* LVT timer register */
+#define LAPIC_LVTTR 0x320
+/* Timer initial count register */
+#define LAPIC_TIMICT 0x380
+/* Timer current count register */
+#define LAPIC_TIMCCT 0x390
+/* Divide config register */
+#define LAPIC_DCR 0x3E0
+
+#define DIVIDER_VALUE 0x0B
+
+struct ioapic {
+  struct acpi_madt_ioapic table_data;
+  spin_lock_t lock;
+  uintptr_t mmio_base;
+};
+
+/* Table of IOAPICS */
+static struct ioapic ioapics[IOAPICS_MAX];
+/* Table of interrupt source overrides */
 /* clang-format off */
 static struct acpi_madt_interrupt_source_override intr_src_overrides[INTERRUPT_SRC_OVERRIDES_MAX];
 /* clang-format on */
+/* Count of actual IOAPIC entries */
 static size_t ioapic_entries = 0;
+/* Count of actual interrupt source overrides */
 static size_t intr_src_override_entries = 0;
 
-extern void amd64_spin (void);
+static spin_lock_t lapic_calibration_lock = SPIN_LOCK_INIT;
 
-static uint32_t amd64_ioapic_read (uintptr_t vaddr, uint32_t reg) {
-  *(volatile uint32_t*)vaddr = reg;
-  return *(volatile uint32_t*)(vaddr + 0x10);
+/* Read IOAPIC */
+static uint32_t amd64_ioapic_read (struct ioapic* ioapic, uint32_t reg) {
+  spin_lock_ctx_t ctxioar;
+
+  spin_lock (&ioapic->lock, &ctxioar);
+  *(volatile uint32_t*)ioapic->mmio_base = reg;
+  uint32_t ret = *(volatile uint32_t*)(ioapic->mmio_base + 0x10);
+  spin_unlock (&ioapic->lock, &ctxioar);
+  return ret;
 }
 
-static void amd64_ioapic_write (uintptr_t vaddr, uint32_t reg, uint32_t value) {
-  *(volatile uint32_t*)vaddr = reg;
-  *(volatile uint32_t*)(vaddr + 0x10) = value;
+/* Write IOAPIC */
+static void amd64_ioapic_write (struct ioapic* ioapic, uint32_t reg, uint32_t value) {
+  spin_lock_ctx_t ctxioaw;
+
+  spin_lock (&ioapic->lock, &ctxioaw);
+  *(volatile uint32_t*)ioapic->mmio_base = reg;
+  *(volatile uint32_t*)(ioapic->mmio_base + 0x10) = value;
+  spin_unlock (&ioapic->lock, &ctxioaw);
 }
 
-static struct acpi_madt_ioapic* amd64_ioapic_find (uint8_t irq) {
-  struct acpi_madt_ioapic* apic = NULL;
-  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+/* Find an IOAPIC corresposting to provided IRQ */
+static struct ioapic* amd64_ioapic_find (uint32_t irq) {
+  struct ioapic* ioapic = NULL;
 
   for (size_t i = 0; i < ioapic_entries; i++) {
-    apic = &apics[i];
-    uint32_t version = amd64_ioapic_read ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, 1);
-    uint32_t max = (version >> 16);
+    ioapic = &ioapics[i];
+    uint32_t version = amd64_ioapic_read (ioapic, 1);
+    uint32_t max = ((version >> 16) & 0xFF);
 
-    if ((apic->gsi_base <= irq) && ((apic->gsi_base + max) > irq))
-      break;
+    if ((irq >= ioapic->table_data.gsi_base) && (irq <= (ioapic->table_data.gsi_base + max)))
+      return ioapic;
   }
 
-  return apic;
+  return NULL;
 }
 
-void amd64_ioapic_route_irq (uint8_t vec, uint8_t irq, uint64_t flags, uint64_t lapic_id) {
-  struct acpi_madt_ioapic* apic;
+/*
+ * Route IRQ to an IDT entry of a given Local APIC.
+ *
+ * vec - Interrupt vector number, which will be delivered to the CPU.
+ * irq -Legacy IRQ number to be routed. Can be changed by an interrupt source override
+ *  into a different GSI.
+ * flags - IOAPIC redirection flags.
+ * lapic_id - Local APIC that will receive the interrupt.
+ */
+void amd64_ioapic_route_irq (uint32_t vec, uint32_t irq, uint64_t flags, uint64_t lapic_id) {
+  struct ioapic* ioapic = NULL;
   struct acpi_madt_interrupt_source_override* override;
   bool found_override = false;
-  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
 
   for (size_t i = 0; i < intr_src_override_entries; i++) {
     override = &intr_src_overrides[i];
@@ -63,57 +117,26 @@ void amd64_ioapic_route_irq (uint8_t vec, uint8_t irq, uint64_t flags, uint64_t
   uint64_t calc_flags = (lapic_id << 56) | (flags) | (vec & 0xFF);
 
   if (found_override) {
-    uint8_t polarity = ((override->flags & 0x03) == 0x03) ? 1 : 0;
-    uint8_t mode = (((override->flags >> 2) & 0x03) == 0x03) ? 1 : 0;
-    calc_flags = (lapic_id << 56) | (mode << 15) | (polarity << 14) | (vec & 0xFF) | flags;
+    uint32_t polarity = ((override->flags & 0x03) == 0x03) ? 1 : 0;
+    uint32_t mode = (((override->flags >> 2) & 0x03) == 0x03) ? 1 : 0;
+    calc_flags |= (uint64_t)mode << 15;
+    calc_flags |= (uint64_t)polarity << 13;
   }
 
-  apic = amd64_ioapic_find (irq);
+  uint32_t gsi = found_override ? override->gsi : irq;
 
-  if (apic == NULL)
+  ioapic = amd64_ioapic_find (gsi);
+
+  if (ioapic == NULL)
     return;
 
-  uint32_t irq_reg = ((irq - apic->gsi_base) * 2) + 0x10;
+  uint32_t irq_reg = ((gsi - ioapic->table_data.gsi_base) * 2) + 0x10;
 
-  amd64_ioapic_write ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg,
-                      (uint32_t)calc_flags);
-
-  amd64_ioapic_write ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg + 1,
-                      (uint32_t)(calc_flags >> 32));
-}
-
-void amd64_ioapic_mask (uint8_t irq) {
-  struct acpi_madt_ioapic* apic;
-  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-
-  apic = amd64_ioapic_find (irq);
-
-  if (apic == NULL)
-    return;
-
-  uint32_t irq_reg = ((irq - apic->gsi_base) * 2) + 0x10;
-
-  uint32_t value = amd64_ioapic_read ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg);
-  amd64_ioapic_write ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg,
-                      value | (1 << 16));
-}
-
-void amd64_ioapic_unmask (uint8_t irq) {
-  struct acpi_madt_ioapic* apic;
-  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-
-  apic = amd64_ioapic_find (irq);
-
-  if (apic == NULL)
-    return;
-
-  uint32_t irq_reg = ((irq - apic->gsi_base) * 2) + 0x10;
-
-  uint32_t value = amd64_ioapic_read ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg);
-  amd64_ioapic_write ((uintptr_t)hhdm->offset + (uintptr_t)apic->address, irq_reg,
-                      value & ~(1 << 16));
+  amd64_ioapic_write (ioapic, irq_reg + 1, (uint32_t)(calc_flags >> 32));
+  amd64_ioapic_write (ioapic, irq_reg, (uint32_t)calc_flags);
 }
 
+/* Find and initialize the IOAPIC */
 void amd64_ioapic_init (void) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
 
@@ -121,7 +144,7 @@ void amd64_ioapic_init (void) {
   uacpi_status status = uacpi_table_find_by_signature (ACPI_MADT_SIGNATURE, &apic_table);
   if (status != UACPI_STATUS_OK) {
     DEBUG ("Could not find MADT table!\n");
-    amd64_spin ();
+    spin ();
   }
 
   struct acpi_madt* apic = (struct acpi_madt*)apic_table.virt_addr;
@@ -134,11 +157,15 @@ void amd64_ioapic_init (void) {
 
     switch (current->type) {
     case ACPI_MADT_ENTRY_TYPE_IOAPIC: {
-      struct acpi_madt_ioapic* ioapic = (struct acpi_madt_ioapic*)current;
-      mm_map_kernel_page ((uintptr_t)ioapic->address,
-                          (uintptr_t)hhdm->offset + (uintptr_t)ioapic->address,
-                          MM_PG_USER | MM_PG_RW);
-      apics[ioapic_entries++] = *ioapic;
+      struct acpi_madt_ioapic* ioapic_table_data = (struct acpi_madt_ioapic*)current;
+      mm_map_kernel_page ((uintptr_t)ioapic_table_data->address,
+                          (uintptr_t)hhdm->offset + (uintptr_t)ioapic_table_data->address,
+                          MM_PG_PRESENT | MM_PG_RW);
+      ioapics[ioapic_entries++] = (struct ioapic){
+          .lock = SPIN_LOCK_INIT,
+          .table_data = *ioapic_table_data,
+          .mmio_base = ((uintptr_t)hhdm->offset + (uintptr_t)ioapic_table_data->address),
+      };
     } break;
     case ACPI_MADT_ENTRY_TYPE_INTERRUPT_SOURCE_OVERRIDE: {
       struct acpi_madt_interrupt_source_override* override =
@@ -150,3 +177,95 @@ void amd64_ioapic_init (void) {
     current = (struct acpi_entry_hdr*)((uintptr_t)current + current->length);
   }
 }
+
+/* Get MMIO base of Local APIC */
+static uintptr_t amd64_lapic_base (void) { return thiscpu->lapic_mmio_base; }
+
+/* Write Local APIC */
+static void amd64_lapic_write (uint32_t reg, uint32_t value) {
+  *(volatile uint32_t*)(amd64_lapic_base () + reg) = value;
+}
+
+/* Read Local APIC */
+static uint32_t amd64_lapic_read (uint32_t reg) {
+  return *(volatile uint32_t*)(amd64_lapic_base () + reg);
+}
+
+/* Get ID of Local APIC */
+uint32_t amd64_lapic_id (void) { return amd64_lapic_read (LAPIC_ID) >> 24; }
+
+/* Send End of interrupt command to Local APIC */
+void amd64_lapic_eoi (void) { amd64_lapic_write (LAPIC_EOI, 0); }
+
+/*
+ * Calibrate Local APIC to send interrupts in a set interval.
+ *
+ * us - Period length in microseconds
+ */
+static uint32_t amd64_lapic_calibrate (uint32_t us) {
+  spin_lock_ctx_t ctxlacb;
+
+  spin_lock (&lapic_calibration_lock, &ctxlacb);
+
+  amd64_lapic_write (LAPIC_DCR, DIVIDER_VALUE);
+
+  amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 16));
+  amd64_lapic_write (LAPIC_TIMICT, 0xFFFFFFFF);
+
+  sleep_micro (us);
+
+  amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (0 << 16));
+  uint32_t ticks = 0xFFFFFFFF - amd64_lapic_read (LAPIC_TIMCCT);
+  DEBUG ("timer ticks = %u\n", ticks);
+
+  spin_unlock (&lapic_calibration_lock, &ctxlacb);
+
+  return ticks;
+}
+
+/*
+ * Starts a Local APIC, configures LVT timer to send interrupts at SCHED_PREEMPT_TIMER.
+ *
+ * ticks - Initial tick count
+ */
+static void amd64_lapic_start (uint32_t ticks) {
+  amd64_lapic_write (LAPIC_DCR, DIVIDER_VALUE);
+  amd64_lapic_write (LAPIC_TIMICT, ticks);
+  amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 17));
+}
+
+/*
+ * Initialize Local APIC, configure to send timer interrupts at a given period. See
+ * amd64_lapic_calibrate and amd64_lapic_start.
+ */
+void amd64_lapic_init (uint32_t us) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+
+  amd64_wrmsr (MSR_APIC_BASE, amd64_rdmsr (MSR_APIC_BASE) | (1 << 11));
+
+  uintptr_t lapic_paddr = amd64_rdmsr (MSR_APIC_BASE) & 0xFFFFF000;
+  thiscpu->lapic_mmio_base = lapic_paddr + (uintptr_t)hhdm->offset;
+
+  mm_map_kernel_page (lapic_paddr, thiscpu->lapic_mmio_base, MM_PG_PRESENT | MM_PG_RW);
+
+  amd64_lapic_write (LAPIC_SIVR, 0xFF | (1 << 8));
+
+  thiscpu->lapic_ticks = amd64_lapic_calibrate (us);
+  amd64_lapic_start (thiscpu->lapic_ticks);
+}
+
+/*
+ * Send an IPI to a given Local APIC. This till invoke an IDT stub located at vec.
+ *
+ * lapic_id - Target Local APIC
+ * vec - Interrupt vector/IDT stub, which will be invoked by the IPI.
+ */
+void amd64_lapic_ipi (uint32_t lapic_id, uint32_t vec) {
+  /* wait for previous IPI to finish */
+  while (amd64_lapic_read (LAPIC_ICR) & (1 << 12)) {
+    __asm__ volatile ("pause");
+  }
+
+  amd64_lapic_write (LAPIC_ICR + 0x10, (lapic_id << 24));
+  amd64_lapic_write (LAPIC_ICR, vec | (1 << 14));
+}

kernel/amd64/apic.h

@@ -3,9 +3,12 @@
 
 #include <libk/std.h>
 
-void amd64_ioapic_route_irq (uint8_t vec, uint8_t irq, uint64_t flags, uint64_t lapic_id);
-void amd64_ioapic_mask (uint8_t irq);
-void amd64_ioapic_unmask (uint8_t irq);
+void amd64_ioapic_route_irq (uint32_t vec, uint32_t irq, uint64_t flags, uint64_t lapic_id);
 void amd64_ioapic_init (void);
 
+uint32_t amd64_lapic_id (void);
+void amd64_lapic_eoi (void);
+void amd64_lapic_ipi (uint32_t lapic_id, uint32_t vec);
+void amd64_lapic_init (uint32_t us);
+
 #endif // _KERNEL_AMD64_APIC_H

kernel/amd64/bootmain.c

@@ -1,35 +1,55 @@
 #include <amd64/apic.h>
+#include <amd64/debug.h>
 #include <amd64/hpet.h>
 #include <amd64/init.h>
+#include <amd64/intr_defs.h>
+#include <amd64/msr-index.h>
+#include <amd64/msr.h>
 #include <aux/compiler.h>
+#include <irq/irq.h>
+#include <libk/std.h>
 #include <limine/limine.h>
+#include <limine/requests.h>
 #include <mm/liballoc.h>
 #include <mm/pmm.h>
+#include <proc/proc.h>
+#include <rd/rd.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
+#include <sys/smp.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
 #include <uacpi/uacpi.h>
 
 #define UACPI_MEMORY_BUFFER_MAX 4096
 
 ALIGNED (16) static uint8_t uacpi_memory_buffer[UACPI_MEMORY_BUFFER_MAX];
 
+/*
+ * The kernel starts booting here. This is the entry point after Limine hands control. We set up all
+ * the necessary platform-dependent subsystems/drivers and jump into the init app.
+ */
 void bootmain (void) {
-  amd64_init ();
+  struct limine_mp_response* mp = limine_mp_request.response;
+
+  struct cpu* bsp_cpu = cpu_make (mp->bsp_lapic_id);
+
+  amd64_init (bsp_cpu, false);
+  syscall_init ();
+  amd64_debug_init ();
   pmm_init ();
   mm_init ();
 
+  rd_init ();
+
   uacpi_setup_early_table_access ((void*)uacpi_memory_buffer, sizeof (uacpi_memory_buffer));
 
   amd64_ioapic_init ();
   amd64_hpet_init ();
 
-  int* a = malloc (sizeof (int));
-  *a = 6969;
-  DEBUG ("a=%p, *a=%d\n", a, *a);
+  smp_init ();
 
-  amd64_hpet_sleep_micro (3000000);
-
-  DEBUG ("woke up!!!\n");
+  proc_init ();
 
   for (;;)
     ;

kernel/amd64/debug.c

@@ -3,22 +3,43 @@
 #include <libk/printf.h>
 #include <libk/std.h>
 #include <libk/string.h>
+#include <sync/spin_lock.h>
 #include <sys/debug.h>
 
-#define PORT_COM1   0x03F8
+/* Port for printing to serial */
+/* TODO: Make this configurable */
+#define PORT_COM1 0x03F8
+/* debugprintf buffer size */
 #define BUFFER_SIZE 1024
+/*
+ * Lock, which ensures that prints to the serial port are atomic (ie. one debugprintf is atomic in
+ * itself).
+ */
+static spin_lock_t serial_lock = SPIN_LOCK_INIT;
 
+static bool debug_init = false;
+
+/* Block until TX buffer is empty */
 static bool amd64_debug_serial_tx_empty (void) {
   return (bool)(amd64_io_inb (PORT_COM1 + 5) & 0x20);
 }
 
+/* Write a single character to serial */
 static void amd64_debug_serial_write (char x) {
   while (!amd64_debug_serial_tx_empty ())
     ;
   amd64_io_outb (PORT_COM1, (uint8_t)x);
 }
 
+/*
+ * Formatted printing to serial. serial_lock ensures that all prints are atomic.
+ */
 void debugprintf (const char* fmt, ...) {
+  spin_lock_ctx_t ctxdbgp;
+
+  if (!debug_init)
+    return;
+
   char buffer[BUFFER_SIZE];
   memset (buffer, 0, sizeof (buffer));
 
@@ -30,12 +51,18 @@ void debugprintf (const char* fmt, ...) {
   buffer[sizeof (buffer) - 1] = '\0';
 
   const char* p = buffer;
+
+  spin_lock (&serial_lock, &ctxdbgp);
+
   while (*p) {
     amd64_debug_serial_write (*p);
     p++;
   }
+
+  spin_unlock (&serial_lock, &ctxdbgp);
 }
 
+/* Initialize serial */
 void amd64_debug_init (void) {
   amd64_io_outb (PORT_COM1 + 1, 0x00);
   amd64_io_outb (PORT_COM1 + 3, 0x80);
@@ -44,4 +71,6 @@ void amd64_debug_init (void) {
   amd64_io_outb (PORT_COM1 + 3, 0x03);
   amd64_io_outb (PORT_COM1 + 2, 0xC7);
   amd64_io_outb (PORT_COM1 + 4, 0x0B);
+
+  debug_init = true;
 }

kernel/amd64/flags.mk

@@ -1,6 +1,11 @@
 cflags += --target=x86_64-pc-none-elf \
 					-mno-sse \
-					-mno-avx
+					-mno-sse2 \
+					-mno-avx \
+					-mno-mmx \
+					-mno-80387 \
+					-mno-red-zone \
+					-fno-omit-frame-pointer
 
 ldflags += --target=x86_64-pc-none-elf \
 					 -Wl,-zmax-page-size=0x1000

kernel/amd64/gdt.h

@@ -0,0 +1,45 @@
+#ifndef _KERNEL_AMD64_GDT_H
+#define _KERNEL_AMD64_GDT_H
+
+#include <aux/compiler.h>
+#include <libk/std.h>
+#include <proc/proc.h>
+
+#define GDT_KCODE 0x08
+#define GDT_KDATA 0x10
+#define GDT_UDATA 0x18
+#define GDT_UCODE 0x20
+#define GDT_TSS   0x28
+
+/* Size of kernel stack */
+#define KSTACK_SIZE (32 * 1024)
+
+/*
+ * 64-bit GDT structure. For more info see:
+ * - https://wiki.osdev.org/Global_Descriptor_Table
+ * - https://wiki.osdev.org/GDT_Tutorial
+ */
+
+struct gdt_entry {
+  uint16_t limitlow;
+  uint16_t baselow;
+  uint8_t basemid;
+  uint8_t access;
+  uint8_t gran;
+  uint8_t basehigh;
+} PACKED;
+
+/* Struct that gets loaded into GDTR */
+struct gdt_ptr {
+  uint16_t limit;
+  uint64_t base;
+} PACKED;
+
+/* New, extended GDT (we need to extend Limine's GDT) */
+struct gdt_extended {
+  struct gdt_entry old[5];
+  struct gdt_entry tsslow;
+  struct gdt_entry tsshigh;
+} PACKED;
+
+#endif // _KERNEL_AMD64_GDT_H

kernel/amd64/hpet.c

@@ -1,56 +1,128 @@
 #include <amd64/hpet.h>
 #include <libk/std.h>
 #include <limine/requests.h>
+#include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
+#include <sys/spin.h>
 #include <uacpi/acpi.h>
 #include <uacpi/status.h>
 #include <uacpi/tables.h>
 #include <uacpi/uacpi.h>
 
-#define HPET_MCVR  0xF0 /* Main Counter Value Register            */
-#define HPET_GCR   0x10 /* General Configuration Register         */
-#define HPET_GCIDR 0x00 /* General Capabilities and ID Register   */
+/*
+ * HPET (High Precision Event Timer) driver code. See more at https://wiki.osdev.org/HPET
+ */
 
+/* HPET Main Counter Value Register */
+#define HPET_MCVR 0xF0
+/* HPET General Configuration Register */
+#define HPET_GCR 0x10
+/* HPET General Capabilities and ID Register */
+#define HPET_GCIDR 0x00
+
+/* Set whether we sould use 32-bit or 64-bit reads/writes */
 static bool hpet_32bits = 1;
+/* Physical address for HPET MMIO */
 static uintptr_t hpet_paddr;
-static uint64_t hpet_clock_nano;
+/* HPET period in femtoseconds */
+static uint64_t hpet_period_fs;
+/* Lock, which protects concurrent access. See amd64/smp.c */
+static spin_lock_t hpet_lock = SPIN_LOCK_INIT;
 
-extern void amd64_spin (void);
-
-static uint64_t amd64_hpet_read (uint32_t reg) {
+/* Read a HPET register. Assumes caller holds hpet_lock */
+static uint64_t amd64_hpet_read64 (uint32_t reg) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   uintptr_t hpet_vaddr = hpet_paddr + (uintptr_t)hhdm->offset;
-  return (hpet_32bits ? *(volatile uint32_t*)(hpet_vaddr + reg)
-                      : *(volatile uint64_t*)(hpet_vaddr + reg));
+  return *(volatile uint64_t*)(hpet_vaddr + reg);
 }
 
-static void amd64_hpet_write (uint32_t reg, uint64_t value) {
+static uint32_t amd64_hpet_read32 (uint32_t reg) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   uintptr_t hpet_vaddr = hpet_paddr + (uintptr_t)hhdm->offset;
-  if (hpet_32bits)
-    *(volatile uint32_t*)(hpet_vaddr + reg) = (value & 0xFFFFFFFF);
-  else
-    *(volatile uint64_t*)(hpet_vaddr + reg) = value;
+  return *(volatile uint32_t*)(hpet_vaddr + reg);
 }
 
-static uint64_t amd64_hpet_timestamp (void) { return amd64_hpet_read (HPET_MCVR); }
+/* Write a HPET register. Assumes caller holds hpet_lock */
+static void amd64_hpet_write64 (uint32_t reg, uint64_t value) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  uintptr_t hpet_vaddr = hpet_paddr + (uintptr_t)hhdm->offset;
+  *(volatile uint64_t*)(hpet_vaddr + reg) = value;
+}
 
-uint64_t amd64_hpet_current_nano (void) { return amd64_hpet_timestamp () * hpet_clock_nano; }
+static void amd64_hpet_write32 (uint32_t reg, uint32_t value) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  uintptr_t hpet_vaddr = hpet_paddr + (uintptr_t)hhdm->offset;
+  *(volatile uint32_t*)(hpet_vaddr + reg) = value;
+}
 
+/* Read current value of HPET_MCVR register. */
+
+static uint64_t amd64_hpet_read_counter (void) {
+  uint64_t value;
+  spin_lock_ctx_t ctxhrc;
+
+  spin_lock (&hpet_lock, &ctxhrc);
+
+  if (!hpet_32bits)
+    value = amd64_hpet_read64 (HPET_MCVR);
+  else {
+    uint32_t hi1, lo, hi2;
+    do {
+      hi1 = amd64_hpet_read32 (HPET_MCVR + 4);
+      lo = amd64_hpet_read32 (HPET_MCVR + 0);
+      hi2 = amd64_hpet_read32 (HPET_MCVR + 4);
+    } while (hi1 != hi2);
+
+    value = ((uint64_t)hi1 << 32) | lo;
+  }
+
+  spin_unlock (&hpet_lock, &ctxhrc);
+
+  return value;
+}
+
+static void amd64_hpet_write_counter (uint64_t value) {
+  spin_lock_ctx_t ctxhwc;
+
+  spin_lock (&hpet_lock, &ctxhwc);
+
+  if (!hpet_32bits)
+    amd64_hpet_write64 (HPET_MCVR, value);
+  else {
+    amd64_hpet_write32 (HPET_MCVR, (uint32_t)value);
+    amd64_hpet_write32 (HPET_MCVR + 4, (uint32_t)(value >> 32));
+  }
+
+  spin_unlock (&hpet_lock, &ctxhwc);
+}
+
+/* Sleep for a given amount of microseconds. This time can last longer due to \ref hpet_lock being
+ * held. */
 void amd64_hpet_sleep_micro (uint64_t us) {
-  uint64_t start = amd64_hpet_timestamp ();
-  uint64_t conv = us * 1000;
-  while (((amd64_hpet_timestamp () - start) * hpet_clock_nano) < conv)
+  if (hpet_period_fs == 0)
+    return;
+
+  uint64_t ticks_to_wait = (us * 1000ULL) / (hpet_period_fs / 1000000ULL);
+  uint64_t start = amd64_hpet_read_counter ();
+
+  for (;;) {
+    uint64_t now = amd64_hpet_read_counter ();
+
+    if ((now - start) >= ticks_to_wait)
+      break;
+
     __asm__ volatile ("pause" ::: "memory");
+  }
 }
 
+/* Initialize HPET */
 void amd64_hpet_init (void) {
   struct uacpi_table hpet_table;
   uacpi_status status = uacpi_table_find_by_signature (ACPI_HPET_SIGNATURE, &hpet_table);
   if (status != UACPI_STATUS_OK) {
     DEBUG ("Could not find HPET table!\n");
-    amd64_spin ();
+    spin ();
   }
 
   struct acpi_hpet* hpet = (struct acpi_hpet*)hpet_table.virt_addr;
@@ -59,9 +131,12 @@ void amd64_hpet_init (void) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   mm_map_kernel_page (hpet_paddr, (uintptr_t)hhdm->offset + hpet_paddr, MM_PG_PRESENT | MM_PG_RW);
 
-  hpet_32bits = (amd64_hpet_read (HPET_GCIDR) & (1 << 13)) ? 0 : 1;
+  uint64_t caps = amd64_hpet_read64 (HPET_GCIDR);
+  hpet_32bits = (caps & (1 << 13)) ? 0 : 1;
 
-  amd64_hpet_write (HPET_GCR, amd64_hpet_read (HPET_GCR) | 0x01);
+  hpet_period_fs = (uint32_t)(caps >> 32);
 
-  hpet_clock_nano = amd64_hpet_read (HPET_GCIDR + 4) / 1000000;
+  amd64_hpet_write64 (HPET_GCR, 0);
+  amd64_hpet_write_counter (0);
+  amd64_hpet_write64 (HPET_GCR, 1);
 }

kernel/amd64/hpet.h

@@ -3,7 +3,6 @@
 
 #include <libk/std.h>
 
-uint64_t amd64_hpet_current_nano (void);
 void amd64_hpet_sleep_micro (uint64_t us);
 void amd64_hpet_init (void);
 

kernel/amd64/init.c

@@ -1,45 +1,15 @@
-#include <amd64/debug.h>
+#include <amd64/gdt.h>
 #include <amd64/init.h>
 #include <amd64/intr.h>
-#include <amd64/tss.h>
+#include <amd64/smp.h>
 #include <aux/compiler.h>
 #include <libk/std.h>
 #include <libk/string.h>
 
-#define GDT_KCODE 0x08
-#define GDT_KDATA 0x10
-#define GDT_UCODE 0x18
-#define GDT_UDATA 0x20
-#define GDT_TSS   0x28
-
 #define TSS         0x80
 #define TSS_PRESENT 0x89
 
-#define KSTACK_SIZE (8 * 1024)
-
-struct gdt_entry {
-  uint16_t limitlow;
-  uint16_t baselow;
-  uint8_t basemid;
-  uint8_t access;
-  uint8_t gran;
-  uint8_t basehigh;
-} PACKED;
-
-struct gdt_ptr {
-  uint16_t limit;
-  uint64_t base;
-} PACKED;
-
-struct gdt_extended {
-  struct gdt_entry old[5];
-  struct gdt_entry tsslow;
-  struct gdt_entry tsshigh;
-} PACKED;
-
-ALIGNED (16) static volatile uint8_t kernel_stack[KSTACK_SIZE];
-ALIGNED (16) static volatile struct gdt_extended gdt;
-
+/* Set a GDT entry */
 static void amd64_gdt_set (volatile struct gdt_entry* ent, uint32_t base, uint32_t limit,
                            uint8_t acc, uint8_t gran) {
   ent->baselow = (base & 0xFFFF);
@@ -50,39 +20,44 @@ static void amd64_gdt_set (volatile struct gdt_entry* ent, uint32_t base, uint32
   ent->access = acc;
 }
 
-static void amd64_gdt_init (void) {
-  volatile struct tss* tss = amd64_get_tss ();
+/* Initialize GDT and TSS structures for a given CPU */
+static void amd64_gdt_init (struct cpu* cpu) {
+  volatile struct tss* tss = &cpu->tss;
+  volatile struct gdt_extended* gdt = &cpu->gdt;
 
-  memset ((void*)&gdt, 0, sizeof (gdt));
-  memset ((void*)kernel_stack, 0, sizeof (kernel_stack));
+  memset ((void*)gdt, 0, sizeof (*gdt));
   memset ((void*)tss, 0, sizeof (*tss));
 
   tss->iopb_off = sizeof (*tss);
-  tss->rsp0 = (uint64_t)((uintptr_t)kernel_stack + sizeof (kernel_stack));
+  tss->rsp0 = (uint64_t)((uintptr_t)cpu->kernel_stack + sizeof (cpu->kernel_stack));
+  tss->ist[0] = (uint64_t)((uintptr_t)cpu->except_stack + sizeof (cpu->except_stack));
+  tss->ist[1] = (uint64_t)((uintptr_t)cpu->irq_stack + sizeof (cpu->irq_stack));
 
-  uint64_t tssbase = (uint64_t)&tss;
+  uint64_t tssbase = (uint64_t)tss;
   uint64_t tsslimit = sizeof (*tss) - 1;
 
-  amd64_gdt_set (&gdt.old[0], 0, 0, 0, 0);
-  amd64_gdt_set (&gdt.old[1], 0, 0xFFFFF, 0x9A, 0xA0);
-  amd64_gdt_set (&gdt.old[2], 0, 0xFFFFF, 0x92, 0xC0);
-  amd64_gdt_set (&gdt.old[3], 0, 0xFFFFF, 0xFA, 0xA0);
-  amd64_gdt_set (&gdt.old[4], 0, 0xFFFFF, 0xF2, 0xC0);
-  amd64_gdt_set (&gdt.tsslow, (tssbase & 0xFFFFFFFF), tsslimit, TSS_PRESENT | TSS, 0);
+  amd64_gdt_set (&gdt->old[0], 0, 0, 0, 0);
+  amd64_gdt_set (&gdt->old[1], 0, 0xFFFFF, 0x9A, 0xA0);
+  amd64_gdt_set (&gdt->old[2], 0, 0xFFFFF, 0x92, 0xC0);
+  amd64_gdt_set (&gdt->old[3], 0, 0xFFFFF, 0xF2, 0xC0);
+  amd64_gdt_set (&gdt->old[4], 0, 0xFFFFF, 0xFA, 0xA0);
+  amd64_gdt_set (&gdt->tsslow, (tssbase & 0xFFFFFFFF), tsslimit, TSS_PRESENT | TSS, 0);
 
   uint32_t tssbasehigh = (tssbase >> 32);
-  gdt.tsshigh.limitlow = (tssbasehigh & 0xFFFF);
-  gdt.tsshigh.baselow = (tssbasehigh >> 16) & 0xFFFF;
-  gdt.tsshigh.basemid = 0;
-  gdt.tsshigh.basehigh = 0;
-  gdt.tsshigh.access = 0;
-  gdt.tsshigh.gran = 0;
+  gdt->tsshigh.limitlow = (tssbasehigh & 0xFFFF);
+  gdt->tsshigh.baselow = (tssbasehigh >> 16) & 0xFFFF;
+  gdt->tsshigh.basemid = 0;
+  gdt->tsshigh.basehigh = 0;
+  gdt->tsshigh.access = 0;
+  gdt->tsshigh.gran = 0;
 
+  /* Load GDTR */
   struct gdt_ptr gdtr;
-  gdtr.limit = sizeof (gdt) - 1;
-  gdtr.base = (uint64_t)&gdt;
+  gdtr.limit = sizeof (*gdt) - 1;
+  gdtr.base = (uint64_t)gdt;
   __asm__ volatile ("lgdt %0" ::"m"(gdtr) : "memory");
 
+  /* Reload CS */
   __asm__ volatile ("pushq %[kcode]\n"
                     "lea 1f(%%rip), %%rax\n"
                     "pushq %%rax\n"
@@ -99,8 +74,16 @@ static void amd64_gdt_init (void) {
   __asm__ volatile ("ltr %0" ::"r"((uint16_t)GDT_TSS));
 }
 
-void amd64_init (void) {
-  amd64_gdt_init ();
-  amd64_debug_init ();
-  amd64_intr_init ();
+/*
+ * Initialize essentials (GDT, TSS, IDT) for a given CPU
+ *
+ * load_idt - Tell whether the IDT needs to be loaded. It only has to be loaded once on
+ *  the BSP
+ */
+void amd64_init (struct cpu* cpu, bool load_idt) {
+  amd64_gdt_init (cpu);
+  if (load_idt)
+    amd64_load_idt ();
+  else
+    amd64_intr_init ();
 }

kernel/amd64/init.h

@@ -1,6 +1,8 @@
 #ifndef _KERNEL_AMD64_INIT_H
 #define _KERNEL_AMD64_INIT_H
 
-void amd64_init (void);
+#include <amd64/smp.h>
+
+void amd64_init (struct cpu* cpu, bool load_idt);
 
 #endif // _KERNEL_AMD64_INIT_H

kernel/amd64/intr.c

@@ -1,9 +1,18 @@
+#include <amd64/apic.h>
+#include <amd64/gdt.h>
 #include <amd64/intr.h>
+#include <amd64/intr_defs.h>
 #include <amd64/io.h>
 #include <aux/compiler.h>
+#include <irq/irq.h>
 #include <libk/std.h>
 #include <libk/string.h>
+#include <m/syscall_defs.h>
 #include <sys/debug.h>
+#include <sys/irq.h>
+#include <sys/smp.h>
+#include <sys/spin.h>
+#include <syscall/syscall.h>
 
 /* 8259 PIC defs. */
 #define PIC1      0x20
@@ -32,6 +41,7 @@
 
 #define IDT_ENTRIES_MAX 256
 
+/* 64-bit <IDT entry structure: https://wiki.osdev.org/Interrupt_Descriptor_Table */
 struct idt_entry {
   uint16_t intrlow;
   uint16_t kernel_cs;
@@ -50,8 +60,6 @@ struct idt {
 ALIGNED (16) static volatile struct idt_entry idt_entries[IDT_ENTRIES_MAX];
 static volatile struct idt idt;
 
-extern void amd64_spin (void);
-
 /* Remaps and disables old 8259 PIC, since we'll be using APIC. */
 static void amd64_init_pic (void) {
 #define IO_OP(fn, ...)                                                                             \
@@ -77,79 +85,56 @@ static void amd64_init_pic (void) {
 #undef IO_OP
 }
 
-static void amd64_idt_set (volatile struct idt_entry* ent, uint64_t handler, uint8_t flags) {
+/* Set IDT entry */
+static void amd64_idt_set (volatile struct idt_entry* ent, uint64_t handler, uint8_t flags,
+                           uint8_t ist) {
   ent->intrlow = (handler & 0xFFFF);
-  ent->kernel_cs = 0x08; // GDT_KCODE (init.c)
-  ent->ist = 0;
+  ent->kernel_cs = GDT_KCODE;
+  ent->ist = ist;
   ent->attrs = flags;
   ent->intrmid = ((handler >> 16) & 0xFFFF);
   ent->intrhigh = ((handler >> 32) & 0xFFFFFFFF);
   ent->resv = 0;
 }
 
+/* Load the IDT */
+void amd64_load_idt (void) { __asm__ volatile ("lidt %0" ::"m"(idt)); }
+
+/* Initialize IDT entries */
 static void amd64_idt_init (void) {
   memset ((void*)idt_entries, 0, sizeof (idt_entries));
 
-#define IDT_ENTRY(n)                                                                               \
+#define IDT_ENTRY(n, ist)                                                                          \
   extern void amd64_intr##n (void);                                                                \
-  amd64_idt_set (&idt_entries[(n)], (uint64_t)&amd64_intr##n, 0x8E)
-  IDT_ENTRY (0);
-  IDT_ENTRY (1);
-  IDT_ENTRY (2);
-  IDT_ENTRY (3);
-  IDT_ENTRY (4);
-  IDT_ENTRY (5);
-  IDT_ENTRY (6);
-  IDT_ENTRY (7);
-  IDT_ENTRY (8);
-  IDT_ENTRY (9);
-  IDT_ENTRY (10);
-  IDT_ENTRY (11);
-  IDT_ENTRY (12);
-  IDT_ENTRY (13);
-  IDT_ENTRY (14);
-  IDT_ENTRY (15);
-  IDT_ENTRY (16);
-  IDT_ENTRY (17);
-  IDT_ENTRY (18);
-  IDT_ENTRY (19);
-  IDT_ENTRY (20);
-  IDT_ENTRY (21);
-  IDT_ENTRY (22);
-  IDT_ENTRY (23);
-  IDT_ENTRY (24);
-  IDT_ENTRY (25);
-  IDT_ENTRY (26);
-  IDT_ENTRY (27);
-  IDT_ENTRY (28);
-  IDT_ENTRY (29);
-  IDT_ENTRY (30);
-  IDT_ENTRY (31);
-  IDT_ENTRY (32);
-  IDT_ENTRY (33);
-  IDT_ENTRY (34);
-  IDT_ENTRY (35);
-  IDT_ENTRY (36);
-  IDT_ENTRY (37);
-  IDT_ENTRY (38);
-  IDT_ENTRY (39);
-  IDT_ENTRY (40);
-  IDT_ENTRY (41);
-  IDT_ENTRY (42);
-  IDT_ENTRY (43);
-  IDT_ENTRY (44);
-  IDT_ENTRY (45);
-  IDT_ENTRY (46);
-  IDT_ENTRY (47);
+  amd64_idt_set (&idt_entries[(n)], (uint64_t)&amd64_intr##n, 0x8E, (ist))
+  /* clang-format off */
+  IDT_ENTRY (0, 0); IDT_ENTRY (1, 0); IDT_ENTRY (2, 0); IDT_ENTRY (3, 0);
+  IDT_ENTRY (4, 0); IDT_ENTRY (5, 0); IDT_ENTRY (6, 0); IDT_ENTRY (7, 0);
+  IDT_ENTRY (8, 0); IDT_ENTRY (9, 0); IDT_ENTRY (10, 0); IDT_ENTRY (11, 0);
+  IDT_ENTRY (12, 0); IDT_ENTRY (13, 0); IDT_ENTRY (14, 0); IDT_ENTRY (15, 0);
+  IDT_ENTRY (16, 0); IDT_ENTRY (17, 0); IDT_ENTRY (18, 0); IDT_ENTRY (19, 0);
+  IDT_ENTRY (20, 0); IDT_ENTRY (21, 0); IDT_ENTRY (22, 0); IDT_ENTRY (23, 0);
+  IDT_ENTRY (24, 0); IDT_ENTRY (25, 0); IDT_ENTRY (26, 0); IDT_ENTRY (27, 0);
+  IDT_ENTRY (28, 0); IDT_ENTRY (29, 0); IDT_ENTRY (30, 0); IDT_ENTRY (31, 0);
+  IDT_ENTRY (32, 1); IDT_ENTRY (33, 1); IDT_ENTRY (34, 1); IDT_ENTRY (35, 1);
+  IDT_ENTRY (36, 1); IDT_ENTRY (37, 1); IDT_ENTRY (38, 1); IDT_ENTRY (39, 1);
+  IDT_ENTRY (40, 1); IDT_ENTRY (41, 1); IDT_ENTRY (42, 1); IDT_ENTRY (43, 1);
+  IDT_ENTRY (44, 1); IDT_ENTRY (45, 1); IDT_ENTRY (46, 1); IDT_ENTRY (47, 1);
+
+  IDT_ENTRY (SCHED_PREEMPT_TIMER, 1);
+  IDT_ENTRY (TLB_SHOOTDOWN, 1);
+  IDT_ENTRY (CPU_REQUEST_SCHED, 1);
+  IDT_ENTRY (CPU_SPURIOUS, 1);
+  /* clang-format on */
 #undef IDT_ENTRY
 
   idt.limit = sizeof (idt_entries) - 1;
   idt.base = (uint64_t)idt_entries;
 
-  __asm__ volatile ("lidt %0" ::"m"(idt));
-  __asm__ volatile ("sti");
+  amd64_load_idt ();
 }
 
+/* Handle CPU exception and dump registers. If incoming CS has CPL3, kill the process. */
 static void amd64_intr_exception (struct saved_regs* regs) {
   DEBUG ("cpu exception %lu (%lu)\n", regs->trap, regs->error);
 
@@ -171,20 +156,66 @@ static void amd64_intr_exception (struct saved_regs* regs) {
                regs->error, regs->rip, regs->cs, regs->rflags, regs->rsp, regs->ss, cr2, cr3,
                regs->rbx);
 
-  amd64_spin ();
+  if (regs->cs == (GDT_UCODE | 0x03)) {
+    proc_kill (thiscpu->proc_current);
+  } else {
+    spin ();
+  }
 }
 
+/* Handle incoming interrupt, dispatch IRQ handlers. */
 void amd64_intr_handler (void* stack_ptr) {
+  spin_lock_ctx_t ctxcpu, ctxpr;
+
+  amd64_load_kernel_cr3 ();
+
   struct saved_regs* regs = stack_ptr;
 
+  spin_lock (&thiscpu->lock, &ctxcpu);
+  struct proc* proc_current = thiscpu->proc_current;
+  spin_lock (&proc_current->lock, &ctxpr);
+
+  memcpy (&proc_current->pdata.regs, regs, sizeof (struct saved_regs));
+
+  spin_unlock (&proc_current->lock, &ctxpr);
+  spin_unlock (&thiscpu->lock, &ctxcpu);
+
   if (regs->trap <= 31) {
     amd64_intr_exception (regs);
   } else {
-    DEBUG ("unknown trap %lu\n", regs->trap);
+    amd64_lapic_eoi ();
+
+    struct irq* irq = irq_find (regs->trap);
+
+    if (irq != NULL) {
+      irq->func (irq->arg, stack_ptr);
+    }
   }
 }
 
+/* Initialize interrupts */
 void amd64_intr_init (void) {
   amd64_init_pic ();
   amd64_idt_init ();
 }
+
+/* Aux. */
+
+/* Save RFLAGS of the current CPU */
+static uint64_t amd64_irq_save_flags (void) {
+  uint64_t rflags;
+  __asm__ volatile ("pushfq; cli; popq %0" : "=r"(rflags)::"memory", "cc");
+  return rflags;
+}
+
+/* Restore interrupts (IF bit) from RFLAGS */
+static void amd64_irq_restore_flags (uint64_t rflags) {
+  if (rflags & (1ULL << 9))
+    __asm__ volatile ("sti");
+}
+
+/* Save current interrupt state */
+void irq_save (spin_lock_ctx_t* ctx) { *ctx = amd64_irq_save_flags (); }
+
+/* Restore interrupt state */
+void irq_restore (spin_lock_ctx_t* ctx) { amd64_irq_restore_flags (*ctx); }

kernel/amd64/intr.h

@@ -31,6 +31,7 @@ struct saved_regs {
   uint64_t ss;
 } PACKED;
 
+void amd64_load_idt (void);
 void amd64_intr_init (void);
 
 #endif // _KERNEL_AMD64_INTR_H

kernel/amd64/intr_defs.h

@@ -0,0 +1,12 @@
+#ifndef _KERNEL_AMD64_INTR_DEFS_H
+#define _KERNEL_AMD64_INTR_DEFS_H
+
+/* Definitions for custom, nonstandard IDT entries. They have to be remapped by amd64_resolve_irq
+ * into legacy IRQs. */
+
+#define SCHED_PREEMPT_TIMER 80
+#define TLB_SHOOTDOWN       81
+#define CPU_REQUEST_SCHED   82
+#define CPU_SPURIOUS        255
+
+#endif // _KERNEL_AMD64_INTR_DEFS_H

kernel/amd64/intr_stub.S

@@ -1,62 +1,47 @@
-.extern amd64_intr_handler
+#include <amd64/intr_defs.h>
+#include <amd64/regsasm.h>
 
-dupa:
-  jmp dupa
+.extern amd64_intr_handler
 
 #define err(z) \
   pushq $z;
 
 #define no_err(z) \
-  pushq $0; \
+  pushq $0;       \
   pushq $z;
 
-#define push_regs \
-  pushq %rax; \
-  pushq %rcx; \
-  pushq %rdx; \
-  pushq %rsi; \
-  pushq %rdi; \
-  pushq %rbp; \
-  pushq %rbx; \
-  pushq %r8; \
-  pushq %r9; \
-  pushq %r10; \
-  pushq %r11; \
-  pushq %r12; \
-  pushq %r13; \
-  pushq %r14; \
-  pushq %r15;
-
-#define pop_regs \
-  popq %r15; \
-  popq %r14; \
-  popq %r13; \
-  popq %r12; \
-  popq %r11; \
-  popq %r10; \
-  popq %r9; \
-  popq %r8; \
-  pushq %rbx; \
-  popq %rbp; \
-  popq %rdi; \
-  popq %rsi; \
-  popq %rdx; \
-  popq %rcx; \
-  popq %rax;
-
-#define make_intr_stub(x, n) \
-  .global amd64_intr ## n; \
-  amd64_intr ## n:; \
-  x(n); \
-  cli; \
-  push_regs; \
-  cld; \
-  movq %rsp, %rdi; \
-  andq $~0xF, %rsp; \
-  callq amd64_intr_handler; \
-  movq %rdi, %rsp; \
-  pop_regs; \
-  addq $16, %rsp; \
+#define make_intr_stub(x, n)    \
+  .global amd64_intr ## n;      \
+  amd64_intr ## n:;             \
+  x(n);                         \
+  cli;                          \
+  ;                             \
+  push_regs;                    \
+  ;                             \
+  movw $0x10, %ax;              \
+  movw %ax, %ds;                \
+  movw %ax, %es;                \
+  ;                             \
+  cld;                          \
+  ;                             \
+  movq %rsp, %rdi;              \
+  ;                             \
+  movq %cr3, %rax; pushq %rax;  \
+  ;                             \
+  movq %rsp, %rbp;              \
+  ;                             \
+  subq $8, %rsp;                \
+  andq $-16, %rsp;              \
+  ;                             \
+  callq amd64_intr_handler;     \
+  ;                             \
+  movq %rbp, %rsp;              \
+  ;                             \
+  popq %rax; movq %rax, %cr3;   \
+  ;                             \
+  pop_regs;                     \
+  addq $16, %rsp;               \
+  ;                             \
   iretq;
   
 
@@ -108,3 +93,8 @@ make_intr_stub(no_err, 44)
 make_intr_stub(no_err, 45)
 make_intr_stub(no_err, 46)
 make_intr_stub(no_err, 47)
+
+make_intr_stub(no_err, SCHED_PREEMPT_TIMER)
+make_intr_stub(no_err, TLB_SHOOTDOWN)
+make_intr_stub(no_err, CPU_REQUEST_SCHED)
+make_intr_stub(no_err, CPU_SPURIOUS)

kernel/amd64/io.c

@@ -1,42 +1,51 @@
 #include <amd64/io.h>
 #include <libk/std.h>
 
+/// Perform outb instruction (send 8-bit int)
 void amd64_io_outb (uint16_t port, uint8_t v) {
   __asm__ volatile ("outb %1, %0" ::"dN"(port), "a"(v));
 }
 
+/// Perform outw instruction (send 16-bit int)
 void amd64_io_outw (uint16_t port, uint16_t v) {
   __asm__ volatile ("outw %%ax, %%dx" ::"a"(v), "d"(port));
 }
 
+/// Perform outl instruction (send 32-bit int)
 void amd64_io_outl (uint16_t port, uint32_t v) {
   __asm__ volatile ("outl %%eax, %%dx" ::"d"(port), "a"(v));
 }
 
+/// Perform outsw instruction (send a string)
 void amd64_io_outsw (uint16_t port, const void* addr, int cnt) {
   __asm__ volatile ("cld; rep outsw" : "+S"(addr), "+c"(cnt) : "d"(port) : "memory", "cc");
 }
 
+/// Perform inb instruction (receive 8-bit int)
 uint8_t amd64_io_inb (uint16_t port) {
   uint8_t r;
   __asm__ volatile ("inb %1, %0" : "=a"(r) : "dN"(port));
   return r;
 }
 
+/// Perform inw instruction (receive 16-bit int)
 uint16_t amd64_io_inw (uint16_t port) {
   uint16_t r;
   __asm__ volatile ("inw %%dx, %%ax" : "=a"(r) : "d"(port));
   return r;
 }
 
+/// Perform inl instruction (receive 32-bit int)
 uint32_t amd64_io_inl (uint16_t port) {
   uint32_t r;
   __asm__ volatile ("inl %%dx, %%eax" : "=a"(r) : "d"(port));
   return r;
 }
 
+/// Perform insw instruction (receive a string)
 void amd64_io_insw (uint16_t port, void* addr, int cnt) {
   __asm__ volatile ("cld; rep insw" : "+D"(addr), "+c"(cnt) : "d"(port) : "memory", "cc");
 }
 
+/// output a byte on port 0x80, which does a small IO delay
 void amd64_io_wait (void) { amd64_io_outb (0x80, 0); }

kernel/amd64/mm.c

@@ -1,29 +1,53 @@
+#include <amd64/apic.h>
+#include <amd64/intr_defs.h>
 #include <aux/compiler.h>
+#include <irq/irq.h>
 #include <libk/std.h>
 #include <libk/string.h>
 #include <limine/requests.h>
 #include <mm/pmm.h>
 #include <sync/spin_lock.h>
+#include <sys/debug.h>
 #include <sys/mm.h>
+#include <sys/smp.h>
 
 #define AMD64_PG_PRESENT (1 << 0)
 #define AMD64_PG_RW      (1 << 1)
 #define AMD64_PG_USER    (1 << 2)
+#define AMD64_PG_HUGE    (1 << 7)
 
-#define AMD64_PG_TABLE_ENTRIES_MAX 512
-
+/* Auxilary struct for page directory walking */
 struct pg_index {
   uint16_t pml4, pml3, pml2, pml1;
 } PACKED;
 
-struct pd kernel_pd = {.lock = SPIN_LOCK_INIT};
+/* Kernel page directory */
+static struct pd kernel_pd;
+static spin_lock_t kernel_pd_lock;
 
+void mm_kernel_lock (spin_lock_ctx_t* ctx) { spin_lock (&kernel_pd_lock, ctx); }
+
+void mm_kernel_unlock (spin_lock_ctx_t* ctx) { spin_lock (&kernel_pd_lock, ctx); }
+
+/* Get current value of CR3 register */
 static uintptr_t amd64_current_cr3 (void) {
   uintptr_t cr3;
   __asm__ volatile ("movq %%cr3, %0" : "=r"(cr3)::"memory");
   return cr3;
 }
 
+/* Load kernel CR3 as current CR3 */
+void amd64_load_kernel_cr3 (void) {
+  uintptr_t cr3 = amd64_current_cr3 ();
+
+  if (cr3 != kernel_pd.cr3_paddr) {
+    __asm__ volatile ("movq %0, %%cr3" ::"r"(kernel_pd.cr3_paddr) : "memory");
+  }
+}
+
+struct pd* mm_get_kernel_pd (void) { return &kernel_pd; }
+
+/* Extract PML info from virtual address */
 static struct pg_index amd64_mm_page_index (uint64_t vaddr) {
   struct pg_index ret;
 
@@ -35,30 +59,43 @@ static struct pg_index amd64_mm_page_index (uint64_t vaddr) {
   return ret;
 }
 
+/* Walk paging tables and allocate necessary structures along the way */
 static uint64_t* amd64_mm_next_table (uint64_t* table, uint64_t entry_idx, bool alloc) {
   uint64_t entry = table[entry_idx];
-  uint64_t paddr;
+  physaddr_t paddr;
 
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
 
-  if (entry & AMD64_PG_PRESENT)
+  if (entry & AMD64_PG_PRESENT) {
+    if (entry & AMD64_PG_HUGE)
+      return NULL;
+
     paddr = entry & ~0xFFFULL;
-  else {
+  } else {
     if (!alloc)
       return NULL;
 
     paddr = pmm_alloc (1);
 
-    if (paddr == 0)
+    if (paddr == PMM_ALLOC_ERR)
       return NULL;
 
     memset ((void*)((uintptr_t)hhdm->offset + (uintptr_t)paddr), 0, PAGE_SIZE);
-    table[entry_idx] = paddr | AMD64_PG_PRESENT | AMD64_PG_RW;
+    table[entry_idx] = paddr | AMD64_PG_PRESENT | AMD64_PG_RW | AMD64_PG_USER;
   }
 
   return (uint64_t*)((uintptr_t)hhdm->offset + (uintptr_t)paddr);
 }
 
+static bool amd64_mm_is_table_empty (uint64_t* table) {
+  for (size_t i = 0; i < 512; i++) {
+    if (table[i] & AMD64_PG_PRESENT)
+      return false;
+  }
+  return true;
+}
+
+/* Convert generic memory management subsystem flags into AMD64-specific flags */
 static uint64_t amd64_mm_resolve_flags (uint32_t generic) {
   uint64_t flags = 0;
 
@@ -69,17 +106,15 @@ static uint64_t amd64_mm_resolve_flags (uint32_t generic) {
   return flags;
 }
 
+/* Reload the current CR3 value ON A LOCAL CPU */
 static void amd64_reload_cr3 (void) {
   uint64_t cr3;
   __asm__ volatile ("movq %%cr3, %0; movq %0, %%cr3" : "=r"(cr3)::"memory");
 }
 
+/* Map physical address to virtual address with flags. TLB needs to be flushed afterwards. */
 void mm_map_page (struct pd* pd, uintptr_t paddr, uintptr_t vaddr, uint32_t flags) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  bool do_reload = false;
-
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock);
 
   uint64_t amd64_flags = amd64_mm_resolve_flags (flags);
 
@@ -88,74 +123,199 @@ void mm_map_page (struct pd* pd, uintptr_t paddr, uintptr_t vaddr, uint32_t flag
 
   uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, true);
   if (pml3 == NULL)
-    goto done;
+    return;
 
   uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, true);
   if (pml2 == NULL)
-    goto done;
+    return;
 
   uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, true);
   if (pml1 == NULL)
-    goto done;
+    return;
 
   uint64_t* pte = &pml1[pg_index.pml1];
 
   *pte = ((paddr & ~0xFFFULL) | (amd64_flags & 0x7ULL));
-  do_reload = true;
-
-done:
-  if (do_reload)
-    amd64_reload_cr3 ();
-
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock);
 }
 
+/* Map a page into kernel page directory */
 void mm_map_kernel_page (uintptr_t paddr, uintptr_t vaddr, uint32_t flags) {
   mm_map_page (&kernel_pd, paddr, vaddr, flags);
+  amd64_reload_cr3 ();
 }
 
-void mm_unmap_page (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
+/* Unmap a virtual address. TLB needs to be flushed afterwards */
+void mm_unmap_page (struct pd* pd, uintptr_t vaddr) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  bool do_reload = false;
-
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock);
 
   uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
   struct pg_index pg_index = amd64_mm_page_index (vaddr);
 
-  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, true);
+  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, false);
   if (pml3 == NULL)
-    goto done;
+    return;
 
-  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, true);
+  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, false);
   if (pml2 == NULL)
-    goto done;
+    return;
 
-  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, true);
+  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, false);
   if (pml1 == NULL)
-    goto done;
+    return;
 
   uint64_t* pte = &pml1[pg_index.pml1];
 
-  *pte &= ~AMD64_PG_PRESENT;
-  do_reload = true;
+  if ((*pte) & AMD64_PG_PRESENT)
+    *pte = 0;
+
+  if (amd64_mm_is_table_empty (pml1)) {
+    uintptr_t pml1_phys = pml2[pg_index.pml2] & ~0xFFFULL;
+    pmm_free (pml1_phys, 1);
+    pml2[pg_index.pml2] = 0;
+
+    if (amd64_mm_is_table_empty (pml2)) {
+      uintptr_t pml2_phys = pml3[pg_index.pml3] & ~0xFFFULL;
+      pmm_free (pml2_phys, 1);
+      pml3[pg_index.pml3] = 0;
+
+      if (amd64_mm_is_table_empty (pml3)) {
+        uintptr_t pml3_phys = pml4[pg_index.pml4] & ~0xFFFULL;
+        pmm_free (pml3_phys, 1);
+        pml4[pg_index.pml4] = 0;
+      }
+    }
+  }
+}
+
+/* Unmap a page from kernel page directory */
+void mm_unmap_kernel_page (uintptr_t vaddr) {
+  mm_unmap_page (&kernel_pd, vaddr);
+  amd64_reload_cr3 ();
+}
+
+/* Allocate a userspace-ready page directory */
+uintptr_t mm_alloc_user_pd_phys (void) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+
+  physaddr_t cr3 = pmm_alloc (1);
+  if (cr3 == PMM_ALLOC_ERR)
+    return 0;
+
+  uint8_t* vu_cr3 = (uint8_t*)((uintptr_t)hhdm->offset + cr3);
+  memset ((void*)vu_cr3, 0, PAGE_SIZE / 2);
+
+  uint8_t* vk_cr3 = (uint8_t*)((uintptr_t)hhdm->offset + (uintptr_t)kernel_pd.cr3_paddr);
+
+  memcpy (&vu_cr3[PAGE_SIZE / 2], &vk_cr3[PAGE_SIZE / 2], PAGE_SIZE / 2);
+
+  return cr3;
+}
+
+bool mm_validate (struct pd* pd, uintptr_t vaddr) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  bool ret = false;
+
+  uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
+  struct pg_index pg_index = amd64_mm_page_index (vaddr);
+
+  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, false);
+  if (pml3 == NULL)
+    goto done;
+
+  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, false);
+  if (pml2 == NULL)
+    goto done;
+
+  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, false);
+  if (pml1 == NULL)
+    goto done;
+
+  uint64_t pte = pml1[pg_index.pml1];
+  ret = (pte & AMD64_PG_PRESENT) != 0;
 
 done:
-  if (do_reload)
-    amd64_reload_cr3 ();
-
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock);
+  return ret;
 }
 
-void mm_unmap_kernel_page (uintptr_t vaddr, uint32_t flags) {
-  mm_unmap_page (&kernel_pd, vaddr, flags);
+bool mm_validate_buffer (struct pd* pd, uintptr_t vaddr, size_t size) {
+  bool ok = true;
+
+  for (size_t i = 0; i < size; i++) {
+    ok = mm_validate (pd, vaddr + i);
+    if (!ok)
+      goto done;
+  }
+
+done:
+  return ok;
 }
 
-void mm_lock_kernel (void) { spin_lock (&kernel_pd.lock); }
+uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  uintptr_t ret = 0;
 
-void mm_unlock_kernel (void) { spin_unlock (&kernel_pd.lock); }
+  uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
 
+  for (size_t i4 = 0; i4 < 512; i4++) {
+    if (!(pml4[i4] & AMD64_PG_PRESENT))
+      continue;
+
+    uint64_t* pml3 = (uint64_t*)((uintptr_t)hhdm->offset + (pml4[i4] & ~0xFFFULL));
+    for (size_t i3 = 0; i3 < 512; i3++) {
+      if (!(pml3[i3] & AMD64_PG_PRESENT))
+        continue;
+
+      uint64_t* pml2 = (uint64_t*)((uintptr_t)hhdm->offset + (pml3[i3] & ~0xFFFULL));
+      for (size_t i2 = 0; i2 < 512; i2++) {
+        if (!(pml2[i2] & AMD64_PG_PRESENT))
+          continue;
+
+        uint64_t* pml1 = (uint64_t*)((uintptr_t)hhdm->offset + (pml2[i2] & ~0xFFFULL));
+        for (size_t i1 = 0; i1 < 512; i1++) {
+          if ((pml1[i1] & AMD64_PG_PRESENT) && ((pml1[i1] & ~0xFFFULL) == (paddr & ~0xFFFULL))) {
+            struct pg_index idx = {i4, i3, i2, i1};
+            ret = (((uint64_t)idx.pml4 << 39) | ((uint64_t)idx.pml3 << 30) |
+                   ((uint64_t)idx.pml2 << 21) | ((uint64_t)idx.pml1 << 12) | (paddr & 0xFFFULL));
+            goto done;
+          }
+        }
+      }
+    }
+  }
+
+done:
+  return ret;
+}
+
+uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  uintptr_t ret = 0;
+
+  uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
+  struct pg_index pg_index = amd64_mm_page_index (vaddr);
+
+  uint64_t* pml3 = amd64_mm_next_table (pml4, pg_index.pml4, false);
+  if (pml3 == NULL)
+    goto done;
+
+  uint64_t* pml2 = amd64_mm_next_table (pml3, pg_index.pml3, false);
+  if (pml2 == NULL)
+    goto done;
+
+  uint64_t* pml1 = amd64_mm_next_table (pml2, pg_index.pml2, false);
+  if (pml1 == NULL)
+    goto done;
+
+  uint64_t pte = pml1[pg_index.pml1];
+
+  if (!(pte & AMD64_PG_PRESENT))
+    goto done;
+
+  ret = ((pte & ~0xFFFULL) | (vaddr & 0xFFFULL));
+
+done:
+  return ret;
+}
+
+/* Initialize essentials for the AMD64 memory management subsystem */
 void mm_init (void) { kernel_pd.cr3_paddr = amd64_current_cr3 (); }

kernel/amd64/mm.h

@@ -7,8 +7,9 @@
 #define PAGE_SIZE 4096
 
 struct pd {
-  spin_lock_t lock;
   uintptr_t cr3_paddr;
 };
 
+void amd64_load_kernel_cr3 (void);
+
 #endif // _KERNEL_AMD64_MM_H

kernel/amd64/msr.c

@@ -1,12 +1,14 @@
 #include <amd64/msr.h>
 #include <libk/std.h>
 
+/// Read a model-specific register
 uint64_t amd64_rdmsr (uint32_t msr) {
   uint32_t low, high;
   __asm__ volatile ("rdmsr" : "=a"(low), "=d"(high) : "c"(msr));
   return ((uint64_t)high << 32 | (uint64_t)low);
 }
 
+/// Write a model-specific register
 void amd64_wrmsr (uint32_t msr, uint64_t value) {
   uint32_t low = (uint32_t)(value & 0xFFFFFFFF);
   uint32_t high = (uint32_t)(value >> 32);

kernel/amd64/proc.c

@@ -0,0 +1,138 @@
+#include <amd64/gdt.h>
+#include <amd64/proc.h>
+#include <aux/elf.h>
+#include <libk/align.h>
+#include <libk/list.h>
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <libk/string.h>
+#include <limine/requests.h>
+#include <mm/liballoc.h>
+#include <mm/pmm.h>
+#include <proc/mutex.h>
+#include <proc/proc.h>
+#include <proc/procgroup.h>
+#include <proc/resource.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/proc.h>
+
+static atomic_int pids = 0;
+
+struct proc* proc_from_elf (uint8_t* elf_contents) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+
+  struct proc* proc = malloc (sizeof (*proc));
+  if (proc == NULL)
+    return NULL;
+
+  memset (proc, 0, sizeof (*proc));
+
+  proc->lock = SPIN_LOCK_INIT;
+  atomic_store (&proc->state, PROC_READY);
+  proc->pid = atomic_fetch_add (&pids, 1);
+
+  proc->procgroup = procgroup_create ();
+  if (proc->procgroup == NULL) {
+    free (proc);
+    return NULL;
+  }
+  procgroup_attach (proc->procgroup, proc);
+
+  uintptr_t kstack_paddr = pmm_alloc (KSTACK_SIZE / PAGE_SIZE);
+  proc->pdata.kernel_stack = kstack_paddr + (uintptr_t)hhdm->offset + KSTACK_SIZE;
+
+  procgroup_map (proc->procgroup, PROC_USTACK_TOP - USTACK_SIZE, USTACK_SIZE / PAGE_SIZE,
+                 MM_PG_USER | MM_PG_PRESENT | MM_PG_RW, NULL);
+
+  proc->flags |= PROC_USTK_PREALLOC;
+
+  struct elf_aux aux = proc_load_segments (proc, elf_contents);
+
+  proc->pdata.regs.ss = GDT_UDATA | 0x03;
+  proc->pdata.regs.rsp = (uint64_t)PROC_USTACK_TOP;
+  proc->pdata.regs.rflags = 0x202;
+  proc->pdata.regs.cs = GDT_UCODE | 0x03;
+  proc->pdata.regs.rip = aux.entry;
+
+  return proc;
+}
+
+struct proc* proc_clone (struct proc* proto, uintptr_t vstack_top, uintptr_t entry,
+                         uintptr_t argument_ptr) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+  spin_lock_ctx_t ctxprt;
+
+  struct proc* proc = malloc (sizeof (*proc));
+  if (proc == NULL)
+    return NULL;
+
+  memset (proc, 0, sizeof (*proc));
+
+  proc->lock = SPIN_LOCK_INIT;
+  atomic_store (&proc->state, PROC_READY);
+  proc->pid = atomic_fetch_add (&pids, 1);
+
+  spin_lock (&proto->lock, &ctxprt);
+
+  proc->procgroup = proto->procgroup;
+  procgroup_attach (proc->procgroup, proc);
+
+  spin_unlock (&proto->lock, &ctxprt);
+
+  uintptr_t kstack_paddr = pmm_alloc (KSTACK_SIZE / PAGE_SIZE);
+  proc->pdata.kernel_stack = kstack_paddr + (uintptr_t)hhdm->offset + KSTACK_SIZE;
+
+  proc->pdata.regs.ss = GDT_UDATA | 0x03;
+  proc->pdata.regs.rsp = (uint64_t)vstack_top;
+  proc->pdata.regs.rflags = 0x202;
+  proc->pdata.regs.cs = GDT_UCODE | 0x03;
+  proc->pdata.regs.rip = (uint64_t)entry;
+
+  proc->uvaddr_argument = argument_ptr;
+
+  proc_init_tls (proc);
+
+  return proc;
+}
+
+void proc_cleanup (struct proc* proc) {
+  proc_sqs_cleanup (proc);
+  proc_mutexes_cleanup (proc);
+
+  pmm_free (proc->pdata.kernel_stack, KSTACK_SIZE / PAGE_SIZE);
+  procgroup_unmap (proc->procgroup, proc->pdata.tls_vaddr, proc->procgroup->tls.tls_tmpl_pages);
+
+  procgroup_detach (proc->procgroup, proc);
+
+  /* clean the process */
+  free (proc);
+}
+
+void proc_init_tls (struct proc* proc) {
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+
+  if (proc->procgroup->tls.tls_tmpl == NULL)
+    return;
+
+  size_t tls_size = proc->procgroup->tls.tls_tmpl_size;
+  size_t pages = proc->procgroup->tls.tls_tmpl_pages;
+
+  uintptr_t tls_paddr;
+  uint32_t flags = MM_PG_USER | MM_PG_PRESENT | MM_PG_RW;
+
+  uintptr_t tls_vaddr = procgroup_map (proc->procgroup, 0, pages, flags, &tls_paddr);
+
+  uintptr_t k_tls_addr = (uintptr_t)hhdm->offset + tls_paddr;
+
+  memset ((void*)k_tls_addr, 0, pages * PAGE_SIZE);
+  memcpy ((void*)k_tls_addr, (void*)proc->procgroup->tls.tls_tmpl, tls_size);
+
+  uintptr_t ktcb = k_tls_addr + tls_size;
+  uintptr_t utcb = tls_vaddr + tls_size;
+
+  *(uintptr_t*)ktcb = utcb;
+
+  proc->pdata.fs_base = utcb;
+  proc->pdata.tls_vaddr = tls_vaddr;
+}

kernel/amd64/proc.h

@@ -0,0 +1,22 @@
+#ifndef _KERNEL_AMD64_PROC_H
+#define _KERNEL_AMD64_PROC_H
+
+#include <amd64/intr.h>
+#include <libk/std.h>
+
+/* Top of userspace process' stack */
+#define PROC_USTACK_TOP 0x00007FFFFFFFF000ULL
+/* Size of userspace process' stack */
+#define USTACK_SIZE (256 * PAGE_SIZE)
+/* proc_map () base address */
+#define PROC_MAP_BASE 0x0000700000000000
+
+/* Platform-dependent process data */
+struct proc_platformdata {
+  struct saved_regs regs;
+  uintptr_t kernel_stack;
+  uint64_t fs_base;
+  uintptr_t tls_vaddr;
+};
+
+#endif // _KERNEL_AMD64_PROC_H

kernel/amd64/procgroup.h

@@ -0,0 +1,13 @@
+#ifndef _KERNEL_AMD64_PROCGRPUP_H
+#define _KERNEL_AMD64_PROCGRPUP_H
+
+#include <libk/std.h>
+
+struct procgroup_tls {
+  uint8_t* tls_tmpl;
+  size_t tls_tmpl_size;
+  size_t tls_tmpl_total_size;
+  size_t tls_tmpl_pages;
+};
+
+#endif // _KERNEL_AMD64_PROCGRPUP_H

kernel/amd64/regsasm.h

@@ -0,0 +1,55 @@
+#ifndef _KERNEL_AMD64_REGSASM_H
+#define _KERNEL_AMD64_REGSASM_H
+
+#define push_regs                                                                                  \
+  pushq % rax;                                                                                     \
+  pushq % rcx;                                                                                     \
+  pushq % rdx;                                                                                     \
+  pushq % rsi;                                                                                     \
+  pushq % rdi;                                                                                     \
+  pushq % rbp;                                                                                     \
+  pushq % rbx;                                                                                     \
+  pushq % r8;                                                                                      \
+  pushq % r9;                                                                                      \
+  pushq % r10;                                                                                     \
+  pushq % r11;                                                                                     \
+  pushq % r12;                                                                                     \
+  pushq % r13;                                                                                     \
+  pushq % r14;                                                                                     \
+  pushq % r15;
+
+#define pop_regs                                                                                   \
+  popq % r15;                                                                                      \
+  popq % r14;                                                                                      \
+  popq % r13;                                                                                      \
+  popq % r12;                                                                                      \
+  popq % r11;                                                                                      \
+  popq % r10;                                                                                      \
+  popq % r9;                                                                                       \
+  popq % r8;                                                                                       \
+  popq % rbx;                                                                                      \
+  popq % rbp;                                                                                      \
+  popq % rdi;                                                                                      \
+  popq % rsi;                                                                                      \
+  popq % rdx;                                                                                      \
+  popq % rcx;                                                                                      \
+  popq % rax;
+
+#define pop_regs_skip_rax                                                                          \
+  popq % r15;                                                                                      \
+  popq % r14;                                                                                      \
+  popq % r13;                                                                                      \
+  popq % r12;                                                                                      \
+  popq % r11;                                                                                      \
+  popq % r10;                                                                                      \
+  popq % r9;                                                                                       \
+  popq % r8;                                                                                       \
+  popq % rbx;                                                                                      \
+  popq % rbp;                                                                                      \
+  popq % rdi;                                                                                      \
+  popq % rsi;                                                                                      \
+  popq % rdx;                                                                                      \
+  popq % rcx;                                                                                      \
+  addq $8, % rsp
+
+#endif // _KERNEL_AMD64_REGSASM_H

kernel/amd64/sched.S

@@ -0,0 +1,9 @@
+#include <amd64/regsasm.h>
+
+.global amd64_do_sched
+amd64_do_sched:
+  movq %rsi, %cr3
+  movq %rdi, %rsp
+  pop_regs
+  addq $16, %rsp
+  iretq

kernel/amd64/sched.h

@@ -0,0 +1,7 @@
+#ifndef _KERNEL_AMD64_SCHED_H
+#define _KERNEL_AMD64_SCHED_H
+
+/// Perform process context switch
+void amd64_do_sched (void* regs, void* cr3);
+
+#endif // _KERNEL_AMD64_SCHED_H

kernel/amd64/sched1.c

@@ -0,0 +1,23 @@
+#include <amd64/msr-index.h>
+#include <amd64/msr.h>
+#include <amd64/sched.h>
+#include <libk/std.h>
+#include <proc/proc.h>
+#include <sync/spin_lock.h>
+#include <sys/mm.h>
+#include <sys/smp.h>
+
+void do_sched (struct proc* proc, spin_lock_t* cpu_lock, spin_lock_ctx_t* ctxcpu) {
+  spin_lock_ctx_t ctxpr;
+
+  spin_lock (&proc->lock, &ctxpr);
+
+  thiscpu->tss.rsp0 = proc->pdata.kernel_stack;
+  thiscpu->syscall_kernel_stack = proc->pdata.kernel_stack;
+  amd64_wrmsr (MSR_FS_BASE, proc->pdata.fs_base);
+
+  spin_unlock (&proc->lock, &ctxpr);
+  spin_unlock (cpu_lock, ctxcpu);
+
+  amd64_do_sched ((void*)&proc->pdata.regs, (void*)proc->procgroup->pd.cr3_paddr);
+}

kernel/amd64/smp.c

@@ -0,0 +1,114 @@
+#include <amd64/apic.h>
+#include <amd64/init.h>
+#include <amd64/intr_defs.h>
+#include <amd64/mm.h>
+#include <amd64/msr-index.h>
+#include <amd64/msr.h>
+#include <libk/std.h>
+#include <libk/string.h>
+#include <limine/requests.h>
+#include <mm/liballoc.h>
+#include <proc/proc.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/sched.h>
+#include <sys/smp.h>
+#include <sys/syscall.h>
+
+/// Cpu ID counter
+static atomic_uint cpu_counter = 0;
+/// The CPUs
+static struct cpu cpus[CPUS_MAX];
+
+static atomic_int cpu_init_count;
+
+/// Allocate a CPU structure
+struct cpu* cpu_make (uint64_t lapic_id) {
+  int id = atomic_fetch_add (&cpu_counter, 1);
+
+  struct cpu* cpu = &cpus[id];
+
+  memset (cpu, 0, sizeof (*cpu));
+  cpu->lock = SPIN_LOCK_INIT;
+  cpu->id = id;
+  cpu->lapic_id = lapic_id;
+
+  amd64_wrmsr (MSR_GS_BASE, (uint64_t)cpu);
+
+  return cpu;
+}
+
+struct cpu* cpu_get (void) {
+  struct cpu* ptr = (struct cpu*)amd64_rdmsr (MSR_GS_BASE);
+  return ptr;
+}
+
+void cpu_request_sched (struct cpu* cpu) {
+  if (cpu == thiscpu) {
+    proc_sched ();
+    return;
+  }
+
+  amd64_lapic_ipi (cpu->lapic_id, CPU_REQUEST_SCHED);
+}
+
+struct cpu* cpu_find_lightest (void) {
+  struct cpu* cpu = &cpus[0];
+
+  int load = atomic_load (&cpu->proc_run_q_count);
+
+  for (unsigned int i = 1; i < cpu_counter; i++) {
+    struct cpu* new_cpu = &cpus[i];
+    int new_load = atomic_load (&new_cpu->proc_run_q_count);
+    if (new_load < load) {
+      load = new_load;
+      cpu = new_cpu;
+    }
+  }
+
+  return cpu;
+}
+
+/// Bootstrap code for non-BSP CPUs
+static void amd64_smp_bootstrap (struct limine_mp_info* mp_info) {
+  amd64_load_kernel_cr3 ();
+
+  struct cpu* cpu = cpu_make (mp_info->lapic_id);
+
+  amd64_init (cpu, true); /* gdt + idt */
+  syscall_init ();
+
+  amd64_lapic_init (1000);
+
+  DEBUG ("CPU %u is online!\n", thiscpu->id);
+
+  atomic_fetch_sub (&cpu_init_count, 1);
+
+  struct proc* spin_proc = proc_spawn_rd ("spin.exe");
+  proc_register (spin_proc, thiscpu);
+
+  spin_lock_ctx_t ctxcpu;
+  spin_lock (&spin_proc->cpu->lock, &ctxcpu);
+  do_sched (spin_proc, &spin_proc->cpu->lock, &ctxcpu);
+}
+
+/// Initialize SMP subsystem for AMD64. Start AP CPUs
+void smp_init (void) {
+  amd64_lapic_init (1000);
+
+  struct limine_mp_response* mp = limine_mp_request.response;
+
+  cpu_init_count = mp->cpu_count - 1; /* Don't include BSP */
+
+  for (size_t i = 0; i < mp->cpu_count; i++) {
+    if (mp->cpus[i]->lapic_id != thiscpu->lapic_id) {
+      DEBUG ("Trying CPU %u\n", mp->cpus[i]->lapic_id);
+      mp->cpus[i]->goto_address = &amd64_smp_bootstrap;
+    }
+  }
+
+  while (atomic_load (&cpu_init_count) > 0)
+    ;
+
+  DEBUG ("All CPUs are online\n");
+}

kernel/amd64/smp.h

@@ -0,0 +1,44 @@
+#ifndef _KERNEL_AMD64_SMP_H
+#define _KERNEL_AMD64_SMP_H
+
+#include <amd64/gdt.h>
+#include <amd64/intr.h>
+#include <amd64/tss.h>
+#include <aux/compiler.h>
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <proc/proc.h>
+#include <sync/spin_lock.h>
+
+#define CPUS_MAX 32
+
+struct cpu {
+  /* for syscall instruction */
+  uintptr_t syscall_user_stack;
+  uintptr_t syscall_kernel_stack;
+  volatile uint8_t kernel_stack[KSTACK_SIZE] ALIGNED (16);
+  volatile uint8_t except_stack[KSTACK_SIZE] ALIGNED (16);
+  volatile uint8_t irq_stack[KSTACK_SIZE] ALIGNED (16);
+  volatile struct gdt_extended gdt ALIGNED (16);
+  volatile struct tss tss;
+
+  uintptr_t lapic_mmio_base;
+  uint64_t lapic_ticks;
+  uint64_t lapic_id;
+  uint32_t id;
+
+  spin_lock_t lock;
+
+  struct list_node_link* proc_run_q;
+  struct proc* proc_current;
+  atomic_int proc_run_q_count;
+};
+
+struct cpu* cpu_make (uint64_t lapic_id);
+struct cpu* cpu_get (void);
+void cpu_request_sched (struct cpu* cpu);
+struct cpu* cpu_find_lightest (void);
+
+#define thiscpu (cpu_get ())
+
+#endif // _KERNEL_AMD64_SMP_H

kernel/amd64/spin.S

@@ -1,3 +1,4 @@
 .global amd64_spin
 amd64_spin:
+  hlt
   jmp amd64_spin

kernel/amd64/spin_lock.c

@@ -1,3 +1,4 @@
 #include <sys/spin_lock.h>
 
+/// Relax the spinlock using AMD64 pause instruction
 void spin_lock_relax (void) { __asm__ volatile ("pause"); }

kernel/amd64/src.mk

@@ -1,6 +1,5 @@
 c += amd64/bootmain.c \
 		 amd64/init.c \
-		 amd64/tss.c \
 		 amd64/io.c \
 		 amd64/debug.c \
 		 amd64/spin_lock.c \
@@ -8,14 +7,20 @@ c += amd64/bootmain.c \
 		 amd64/apic.c \
 		 amd64/msr.c \
 		 amd64/hpet.c \
-		 amd64/mm.c
+		 amd64/mm.c \
+		 amd64/time.c \
+		 amd64/smp.c \
+		 amd64/sched1.c \
+		 amd64/proc.c \
+		 amd64/syscall.c
 
 S += amd64/intr_stub.S \
-		 amd64/spin.S
+		 amd64/spin.S \
+		 amd64/sched.S \
+		 amd64/syscallentry.S
 
 o += amd64/bootmain.o \
 		 amd64/init.o \
-		 amd64/tss.o \
 		 amd64/io.o \
 		 amd64/debug.o \
 		 amd64/spin_lock.o \
@@ -25,4 +30,11 @@ o += amd64/bootmain.o \
 		 amd64/apic.o \
 		 amd64/msr.o \
 		 amd64/hpet.o \
-		 amd64/mm.o
+		 amd64/mm.o \
+		 amd64/time.o \
+		 amd64/smp.o \
+		 amd64/sched.o \
+		 amd64/sched1.o \
+		 amd64/proc.o \
+		 amd64/syscall.o \
+		 amd64/syscallentry.o

kernel/amd64/syscall.c

@@ -0,0 +1,46 @@
+#include <amd64/gdt.h>
+#include <amd64/intr.h>
+#include <amd64/mm.h>
+#include <amd64/msr-index.h>
+#include <amd64/msr.h>
+#include <libk/string.h>
+#include <m/status.h>
+#include <m/syscall_defs.h>
+#include <proc/proc.h>
+#include <sys/debug.h>
+#include <sys/smp.h>
+#include <syscall/syscall.h>
+
+extern void amd64_syscall_entry (void);
+
+uintptr_t amd64_syscall_dispatch (void* stack_ptr) {
+  spin_lock_ctx_t ctxcpu, ctxpr;
+
+  amd64_load_kernel_cr3 ();
+  struct saved_regs* regs = stack_ptr;
+
+  spin_lock (&thiscpu->lock, &ctxcpu);
+  struct proc* caller = thiscpu->proc_current;
+  spin_lock (&caller->lock, &ctxpr);
+
+  memcpy (&caller->pdata.regs, regs, sizeof (struct saved_regs));
+
+  spin_unlock (&caller->lock, &ctxpr);
+  spin_unlock (&thiscpu->lock, &ctxcpu);
+
+  int syscall_num = regs->rax;
+  syscall_handler_func_t func = syscall_find_handler (syscall_num);
+
+  if (func == NULL) {
+    return -ST_SYSCALL_NOT_FOUND;
+  }
+
+  return func (caller, regs, regs->rdi, regs->rsi, regs->rdx, regs->r10, regs->r8, regs->r9);
+}
+
+void syscall_init (void) {
+  amd64_wrmsr (MSR_STAR, ((uint64_t)GDT_KCODE << 32) | ((uint64_t)(GDT_KDATA | 0x03) << 48));
+  amd64_wrmsr (MSR_LSTAR, (uint64_t)&amd64_syscall_entry);
+  amd64_wrmsr (MSR_SYSCALL_MASK, (1ULL << 9));
+  amd64_wrmsr (MSR_EFER, amd64_rdmsr (MSR_EFER) | EFER_SCE);
+}

kernel/amd64/syscallentry.S

@@ -0,0 +1,49 @@
+#include <amd64/regsasm.h>
+
+.extern amd64_syscall_dispatch
+
+.global amd64_syscall_entry
+amd64_syscall_entry:
+  cli
+
+  movq %rsp, %gs:0
+  movq %gs:8, %rsp
+ 
+  pushq $0x1b
+  pushq %gs:0
+  pushq %r11
+  pushq $0x23
+  pushq %rcx
+  pushq $0
+  pushq $0
+
+  push_regs
+
+  movw $0x10, %ax
+  movw %ax, %ds
+  movw %ax, %es
+  movw %ax, %ss
+
+  cld
+
+  movq %rsp, %rdi
+
+  movq %cr3, %rax; pushq %rax
+
+  movq %rsp, %rbp
+
+  subq $8, %rsp
+  andq $-16, %rsp
+
+  callq amd64_syscall_dispatch
+
+  movq %rbp, %rsp
+
+  popq %rbx; movq %rbx, %cr3
+
+  pop_regs_skip_rax
+  
+  addq $56, %rsp
+  movq %gs:0, %rsp
+  
+  sysretq

kernel/amd64/time.c

@@ -0,0 +1,6 @@
+#include <amd64/hpet.h>
+#include <libk/std.h>
+#include <sys/time.h>
+
+/// Sleep for given amount of microseconds
+void sleep_micro (size_t us) { amd64_hpet_sleep_micro (us); }

kernel/amd64/tss.c

@@ -1,7 +0,0 @@
-#include <amd64/tss.h>
-#include <aux/compiler.h>
-#include <libk/std.h>
-
-ALIGNED (16) static volatile struct tss tss;
-
-volatile struct tss* amd64_get_tss (void) { return &tss; }

kernel/amd64/tss.h

@@ -4,6 +4,7 @@
 #include <aux/compiler.h>
 #include <libk/std.h>
 
+/// 64-bit TSS structure: https://wiki.osdev.org/Task_State_Segment
 struct tss {
   uint32_t resv0;
   uint64_t rsp0;
@@ -16,6 +17,4 @@ struct tss {
   uint16_t iopb_off;
 } PACKED;
 
-volatile struct tss* amd64_get_tss (void);
-
 #endif // _KERNEL_AMD64_TSS_H

kernel/aux/compiler.h

@@ -4,5 +4,6 @@
 #define PACKED        __attribute__ ((packed))
 #define ALIGNED(N)    __attribute__ ((aligned ((N))))
 #define SECTION(name) __attribute__ ((section (name)))
+#define UNUSED        __attribute__ ((unused))
 
 #endif // _KERNEL_AUX_COMPILER_H

kernel/generic/flags.mk

@@ -8,7 +8,7 @@ cflags += -nostdinc \
 					-Wextra \
 					-mcmodel=kernel
 
-cflags += -isystem . -isystem c_headers/include
+cflags += -isystem . -isystem ../include
 
 cflags += -DPRINTF_INCLUDE_CONFIG_H=1 \
 					-D_ALLOC_SKIP_DEFINE

kernel/irq/.gitignore

@@ -0,0 +1 @@
+*.o

kernel/irq/irq.c

@@ -0,0 +1,46 @@
+#include <irq/irq.h>
+#include <libk/list.h>
+#include <libk/std.h>
+#include <mm/liballoc.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+
+#if defined(__x86_64__)
+#include <amd64/apic.h>
+#include <amd64/intr.h>
+#endif
+
+struct irq* irq_table[0x100];
+
+static spin_lock_t irqs_lock = SPIN_LOCK_INIT;
+
+bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num) {
+  spin_lock_ctx_t ctxiqa;
+
+  struct irq* irq = malloc (sizeof (*irq));
+  if (irq == NULL) {
+    return false;
+  }
+
+  irq->func = func;
+  irq->arg = arg;
+  irq->irq_num = irq_num;
+
+  spin_lock (&irqs_lock, &ctxiqa);
+  irq_table[irq_num] = irq;
+  spin_unlock (&irqs_lock, &ctxiqa);
+
+  return true;
+}
+
+struct irq* irq_find (uint32_t irq_num) {
+  spin_lock_ctx_t ctxiqa;
+
+  spin_lock (&irqs_lock, &ctxiqa);
+
+  struct irq* irq = irq_table[irq_num];
+
+  spin_unlock (&irqs_lock, &ctxiqa);
+
+  return irq;
+}

kernel/irq/irq.h

@@ -0,0 +1,20 @@
+#ifndef _KERNEL_IRQ_IRQ_H
+#define _KERNEL_IRQ_IRQ_H
+
+#include <libk/list.h>
+#include <libk/std.h>
+
+typedef void (*irq_func_t) (void* arg, void* regs);
+
+struct irq {
+  struct list_node_link irqs_link;
+
+  irq_func_t func;
+  void* arg;
+  uint32_t irq_num;
+};
+
+bool irq_attach (irq_func_t, void* arg, uint32_t irq_num);
+struct irq* irq_find (uint32_t irq_num);
+
+#endif // _KERNEL_IRQ_IRQ_H

kernel/irq/src.mk

@@ -0,0 +1,3 @@
+c += irq/irq.c
+
+o += irq/irq.o

kernel/libk/assert.h

@@ -0,0 +1,15 @@
+#ifndef _KERNEL_LIBK_ASSERT_H
+#define _KERNEL_LIBK_ASSERT_H
+
+#include <sys/spin.h>
+
+#define assert(x)                                                                                  \
+  do {                                                                                             \
+    if (!(x)) {                                                                                    \
+      DEBUG ("%s ssertion failed\n", #x);                                                          \
+      spin ();                                                                                     \
+      __builtin_unreachable ();                                                                    \
+    }                                                                                              \
+  } while (0)
+
+#endif // _KERNEL_LIBK_ASSERT_H

kernel/libk/bm.c

@@ -45,16 +45,16 @@ bool bm_test (struct bm* bm, size_t k) {
  * Set a range of bits in a bitmap. if starting bit is out of range, we fail.
  */
 bool bm_set_region (struct bm* bm, size_t k, size_t m) {
-  if ((k >= m) || (k >= bm->nbits) || (k + m >= bm->nbits))
+  if (((k + m) > bm->nbits) || (k + m) < k)
     return false;
 
-  for (size_t i = k; i < m; i++) {
+  for (size_t i = k; i < (k + m); i++) {
     bool taken = bm_test (bm, i);
     if (taken)
       return false;
   }
 
-  for (size_t i = k; i < m; i++)
+  for (size_t i = k; i < (k + m); i++)
     bm_set (bm, i);
 
   return true;
@@ -64,10 +64,10 @@ bool bm_set_region (struct bm* bm, size_t k, size_t m) {
  * Clear a range of bits in a bitmap. starting bit must be in range.
  */
 void bm_clear_region (struct bm* bm, size_t k, size_t m) {
-  if ((k >= m) || (k >= bm->nbits) || (k + m >= bm->nbits))
+  if (((k + m) > bm->nbits) || (k + m) < k)
     return;
 
-  for (size_t i = k; i < m; i++)
+  for (size_t i = k; i < (k + m); i++)
     bm_clear (bm, i);
 }
 
@@ -78,10 +78,10 @@ void bm_clear_region (struct bm* bm, size_t k, size_t m) {
  * useful for implementing the physical memory manager algorithm.
  */
 bool bm_test_region (struct bm* bm, size_t k, size_t m) {
-  if ((k >= m) || (k >= bm->nbits) || (k + m >= bm->nbits))
+  if (((k + m) > bm->nbits) || (k + m) < k)
     return true;
 
-  for (size_t i = k; i < m; i++) {
+  for (size_t i = k; i < (k + m); i++) {
     bool test = bm_test (bm, i);
     if (test)
       return true;

kernel/libk/list.h

@@ -0,0 +1,170 @@
+#ifndef _KERNEL_LIBK_LIST_H
+#define _KERNEL_LIBK_LIST_H
+
+struct list_node_link {
+  struct list_node_link* next;
+  struct list_node_link* prev;
+};
+
+#define list_entry(ptr, type, member) ((type*)((char*)(ptr) - offsetof (type, member)))
+
+#define list_append(head, new)                                                                     \
+  do {                                                                                             \
+    if ((new) != NULL) {                                                                           \
+      (new)->next = NULL;                                                                          \
+      if ((head) != NULL) {                                                                        \
+        struct list_node_link* __tmp = (head);                                                     \
+        while (__tmp->next != NULL) {                                                              \
+          __tmp = __tmp->next;                                                                     \
+        }                                                                                          \
+        __tmp->next = (new);                                                                       \
+        (new)->prev = __tmp;                                                                       \
+      } else {                                                                                     \
+        (new)->prev = NULL;                                                                        \
+        (head) = (new);                                                                            \
+      }                                                                                            \
+    }                                                                                              \
+  } while (0)
+
+#define list_prepend(head, new)                                                                    \
+  do {                                                                                             \
+    if ((new) != NULL) {                                                                           \
+      (new)->prev = NULL;                                                                          \
+      (new)->next = (head);                                                                        \
+      if ((head) != NULL) {                                                                        \
+        (head)->prev = (new);                                                                      \
+      }                                                                                            \
+      (head) = (new);                                                                              \
+    }                                                                                              \
+  } while (0)
+
+#define list_remove(head, ele)                                                                     \
+  do {                                                                                             \
+    if ((ele) != NULL) {                                                                           \
+      if ((ele)->prev != NULL) {                                                                   \
+        (ele)->prev->next = (ele)->next;                                                           \
+      } else {                                                                                     \
+        (head) = (ele)->next;                                                                      \
+      }                                                                                            \
+      if ((ele)->next != NULL) {                                                                   \
+        (ele)->next->prev = (ele)->prev;                                                           \
+      }                                                                                            \
+      (ele)->next = NULL;                                                                          \
+      (ele)->prev = NULL;                                                                          \
+    }                                                                                              \
+  } while (0)
+
+#define list_find(head, out, propname, propvalue)                                                  \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    struct list_node_link* __tmp = (head);                                                         \
+    while (__tmp) {                                                                                \
+      if (__tmp->propname == (propvalue)) {                                                        \
+        (out) = __tmp;                                                                             \
+        break;                                                                                     \
+      }                                                                                            \
+      __tmp = __tmp->next;                                                                         \
+    }                                                                                              \
+  } while (0)
+
+#define list_foreach(head, var, tmp)                                                               \
+  for (var = (head), tmp = (var ? var->next : NULL); var != NULL;                                  \
+       var = tmp, tmp = (var ? var->next : NULL))
+
+#define list_foreach_index(head, var, tmp, idx)                                                    \
+  for ((idx) = 0, var = (head), tmp = (var ? var->next : NULL); var != NULL;                       \
+       var = tmp, tmp = (var ? var->next : NULL), (idx)++)
+
+#define list_foreach_index_limit(head, var, tmp, idx, max)                                         \
+  for ((idx) = 0, var = (head), tmp = (var ? var->next : NULL); var != NULL && (idx) < (max);      \
+       var = tmp, tmp = (var ? var->next : NULL), (idx)++)
+
+#define list_back(head, out)                                                                       \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    if ((head) != NULL) {                                                                          \
+      struct list_node_link* __tmp = (head);                                                       \
+      while (__tmp->next != NULL) {                                                                \
+        __tmp = __tmp->next;                                                                       \
+      }                                                                                            \
+      (out) = __tmp;                                                                               \
+    }                                                                                              \
+  } while (0)
+
+#define list_front(head, out)                                                                      \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    if ((head) != NULL) {                                                                          \
+      struct list_node_link* __tmp = (head);                                                       \
+      while (__tmp->prev != NULL) {                                                                \
+        __tmp = __tmp->prev;                                                                       \
+      }                                                                                            \
+      (out) = __tmp;                                                                               \
+    }                                                                                              \
+  } while (0)
+
+#define list_insert_after(head, pos, new)                                                          \
+  do {                                                                                             \
+    if ((pos) != NULL && (new) != NULL) {                                                          \
+      (new)->prev = (pos);                                                                         \
+      (new)->next = (pos)->next;                                                                   \
+      if ((pos)->next != NULL) {                                                                   \
+        (pos)->next->prev = (new);                                                                 \
+      }                                                                                            \
+      (pos)->next = (new);                                                                         \
+    } else if ((pos) == NULL && (head) == NULL) {                                                  \
+      (new)->prev = NULL;                                                                          \
+      (new)->next = NULL;                                                                          \
+      (head) = (new);                                                                              \
+    }                                                                                              \
+  } while (0)
+
+#define list_insert_before(head, pos, new)                                                         \
+  do {                                                                                             \
+    if ((pos) != NULL && (new) != NULL) {                                                          \
+      (new)->next = (pos);                                                                         \
+      (new)->prev = (pos)->prev;                                                                   \
+      if ((pos)->prev != NULL) {                                                                   \
+        (pos)->prev->next = (new);                                                                 \
+      } else {                                                                                     \
+        (head) = (new);                                                                            \
+      }                                                                                            \
+      (pos)->prev = (new);                                                                         \
+    } else if ((pos) == NULL && (head) == NULL) {                                                  \
+      (new)->prev = NULL;                                                                          \
+      (new)->next = NULL;                                                                          \
+      (head) = (new);                                                                              \
+    }                                                                                              \
+  } while (0)
+
+#define list_index_of(head, ele, out_idx)                                                          \
+  do {                                                                                             \
+    (out_idx) = -1;                                                                                \
+    int __idx = 0;                                                                                 \
+    struct list_node_link* __tmp = (head);                                                         \
+    while (__tmp != NULL) {                                                                        \
+      if (__tmp == (ele)) {                                                                        \
+        (out_idx) = __idx;                                                                         \
+        break;                                                                                     \
+      }                                                                                            \
+      __tmp = __tmp->next;                                                                         \
+      __idx++;                                                                                     \
+    }                                                                                              \
+  } while (0)
+
+#define list_index_of_prop(head, propname, propvalue, out_idx)                                     \
+  do {                                                                                             \
+    (out_idx) = -1;                                                                                \
+    int __idx = 0;                                                                                 \
+    struct list_node_link* __tmp = (head);                                                         \
+    while (__tmp != NULL) {                                                                        \
+      if (__tmp->propname == (propvalue)) {                                                        \
+        (out_idx) = __idx;                                                                         \
+        break;                                                                                     \
+      }                                                                                            \
+      __tmp = __tmp->next;                                                                         \
+      __idx++;                                                                                     \
+    }                                                                                              \
+  } while (0)
+
+#endif // _KERNEL_LIBK_LIST_H

kernel/libk/rbtree.h

@@ -0,0 +1,323 @@
+#ifndef _KERNEL_LIBK_RBTREE_H
+#define _KERNEL_LIBK_RBTREE_H
+
+struct rb_node_link {
+  struct rb_node_link* left;
+  struct rb_node_link* right;
+  struct rb_node_link* parent;
+  int color;
+};
+
+#define RBTREE_RED   0
+#define RBTREE_BLACK 1
+
+#define rbtree_parent(x) ((x)->parent)
+#define rbtree_left(x)   ((x)->left)
+#define rbtree_right(x)  ((x)->right)
+#define rbtree_color(x)  ((x)->color)
+
+#define rbtree_entry(node, type, member) ((type*)((char*)(node) - offsetof (type, member)))
+
+#define rbtree_node_color(x) ((x) ? (x)->color : RBTREE_BLACK)
+
+#define rbtree_rotate_left(root_ptr, x_node)                                                       \
+  do {                                                                                             \
+    struct rb_node_link* __x = (x_node);                                                           \
+    struct rb_node_link* __y = __x->right;                                                         \
+    __x->right = __y->left;                                                                        \
+    if (__y->left)                                                                                 \
+      __y->left->parent = __x;                                                                     \
+    __y->parent = __x->parent;                                                                     \
+    if (!__x->parent)                                                                              \
+      *(root_ptr) = __y;                                                                           \
+    else if (__x == __x->parent->left)                                                             \
+      __x->parent->left = __y;                                                                     \
+    else                                                                                           \
+      __x->parent->right = __y;                                                                    \
+    __y->left = __x;                                                                               \
+    __x->parent = __y;                                                                             \
+  } while (0)
+
+#define rbtree_rotate_right(root_ptr, y_node)                                                      \
+  do {                                                                                             \
+    struct rb_node_link* __y = (y_node);                                                           \
+    struct rb_node_link* __x = __y->left;                                                          \
+    __y->left = __x->right;                                                                        \
+    if (__x->right)                                                                                \
+      __x->right->parent = __y;                                                                    \
+    __x->parent = __y->parent;                                                                     \
+    if (!__y->parent)                                                                              \
+      *(root_ptr) = __x;                                                                           \
+    else if (__y == __y->parent->right)                                                            \
+      __y->parent->right = __x;                                                                    \
+    else                                                                                           \
+      __y->parent->left = __x;                                                                     \
+    __x->right = __y;                                                                              \
+    __y->parent = __x;                                                                             \
+  } while (0)
+
+#define rbtree_insert_fixup(root_ptr, z_node)                                                      \
+  do {                                                                                             \
+    struct rb_node_link* __z = (z_node);                                                           \
+    while (__z->parent && __z->parent->color == RBTREE_RED) {                                      \
+      if (__z->parent == __z->parent->parent->left) {                                              \
+        struct rb_node_link* __y = __z->parent->parent->right;                                     \
+        if (rbtree_node_color (__y) == RBTREE_RED) {                                               \
+          __z->parent->color = RBTREE_BLACK;                                                       \
+          __y->color = RBTREE_BLACK;                                                               \
+          __z->parent->parent->color = RBTREE_RED;                                                 \
+          __z = __z->parent->parent;                                                               \
+        } else {                                                                                   \
+          if (__z == __z->parent->right) {                                                         \
+            __z = __z->parent;                                                                     \
+            rbtree_rotate_left (root_ptr, __z);                                                    \
+          }                                                                                        \
+          __z->parent->color = RBTREE_BLACK;                                                       \
+          __z->parent->parent->color = RBTREE_RED;                                                 \
+          rbtree_rotate_right (root_ptr, __z->parent->parent);                                     \
+        }                                                                                          \
+      } else {                                                                                     \
+        struct rb_node_link* __y = __z->parent->parent->left;                                      \
+        if (rbtree_node_color (__y) == RBTREE_RED) {                                               \
+          __z->parent->color = RBTREE_BLACK;                                                       \
+          __y->color = RBTREE_BLACK;                                                               \
+          __z->parent->parent->color = RBTREE_RED;                                                 \
+          __z = __z->parent->parent;                                                               \
+        } else {                                                                                   \
+          if (__z == __z->parent->left) {                                                          \
+            __z = __z->parent;                                                                     \
+            rbtree_rotate_right (root_ptr, __z);                                                   \
+          }                                                                                        \
+          __z->parent->color = RBTREE_BLACK;                                                       \
+          __z->parent->parent->color = RBTREE_RED;                                                 \
+          rbtree_rotate_left (root_ptr, __z->parent->parent);                                      \
+        }                                                                                          \
+      }                                                                                            \
+    }                                                                                              \
+    (*(root_ptr))->color = RBTREE_BLACK;                                                           \
+  } while (0)
+
+#define rbtree_insert(type, root_ptr, node, member, keyfield)                                      \
+  do {                                                                                             \
+    struct rb_node_link** __link = (root_ptr);                                                     \
+    struct rb_node_link* __parent = NULL;                                                          \
+    struct rb_node_link* __new = (node);                                                           \
+    type* __nobj = rbtree_entry (__new, type, member);                                             \
+    while (*__link) {                                                                              \
+      __parent = *__link;                                                                          \
+      type* __xobj = rbtree_entry (*__link, type, member);                                         \
+      if (__nobj->keyfield < __xobj->keyfield)                                                     \
+        __link = &((*__link)->left);                                                               \
+      else                                                                                         \
+        __link = &((*__link)->right);                                                              \
+    }                                                                                              \
+    __new->parent = __parent;                                                                      \
+    __new->left = __new->right = NULL;                                                             \
+    __new->color = RBTREE_RED;                                                                     \
+    *__link = __new;                                                                               \
+    rbtree_insert_fixup (root_ptr, __new);                                                         \
+  } while (0)
+
+#define rbtree_find(type, root_ptr, keyval, out, member, keyfield)                                 \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    struct rb_node_link* __cur = *(root_ptr);                                                      \
+    while (__cur) {                                                                                \
+      type* __obj = rbtree_entry (__cur, type, member);                                            \
+      if ((keyval) == __obj->keyfield) {                                                           \
+        (out) = rbtree_entry (__cur, type, member);                                                \
+        break;                                                                                     \
+      } else if ((keyval) < __obj->keyfield)                                                       \
+        __cur = __cur->left;                                                                       \
+      else                                                                                         \
+        __cur = __cur->right;                                                                      \
+    }                                                                                              \
+  } while (0)
+
+#define rbtree_min(node, out)                                                                      \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    struct rb_node_link* __n = (node);                                                             \
+    while (__n && __n->left)                                                                       \
+      __n = __n->left;                                                                             \
+    (out) = __n;                                                                                   \
+  } while (0)
+
+#define rbtree_max(node, out)                                                                      \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    struct rb_node_link* __n = (node);                                                             \
+    while (__n && __n->right)                                                                      \
+      __n = __n->right;                                                                            \
+    (out) = __n;                                                                                   \
+  } while (0)
+
+#define rbtree_first(root_ptr, out) rbtree_min (*(root_ptr), out)
+#define rbtree_last(root_ptr, out)  rbtree_max (*(root_ptr), out)
+
+#define rbtree_transplant(root_ptr, u_node, v_node)                                                \
+  do {                                                                                             \
+    struct rb_node_link* __u = (u_node);                                                           \
+    struct rb_node_link* __v = (v_node);                                                           \
+    if (!__u->parent)                                                                              \
+      *(root_ptr) = __v;                                                                           \
+    else if (__u == __u->parent->left)                                                             \
+      __u->parent->left = __v;                                                                     \
+    else                                                                                           \
+      __u->parent->right = __v;                                                                    \
+    if (__v)                                                                                       \
+      __v->parent = __u->parent;                                                                   \
+  } while (0)
+
+#define rbtree_delete_fixup(root_ptr, x_node, xparent_node)                                        \
+  do {                                                                                             \
+    struct rb_node_link* __rdf_x = (x_node);                                                       \
+    struct rb_node_link* __rdf_xp = (xparent_node);                                                \
+    while (__rdf_xp && (__rdf_x == NULL || __rdf_x->color == RBTREE_BLACK)) {                      \
+      if (__rdf_x == __rdf_xp->left) {                                                             \
+        struct rb_node_link* __w = __rdf_xp->right;                                                \
+        if (rbtree_node_color (__w) == RBTREE_RED) {                                               \
+          __w->color = RBTREE_BLACK;                                                               \
+          __rdf_xp->color = RBTREE_RED;                                                            \
+          rbtree_rotate_left (root_ptr, __rdf_xp);                                                 \
+          __w = __rdf_xp->right;                                                                   \
+        }                                                                                          \
+        if (rbtree_node_color (__w->left) == RBTREE_BLACK &&                                       \
+            rbtree_node_color (__w->right) == RBTREE_BLACK) {                                      \
+          if (__w)                                                                                 \
+            __w->color = RBTREE_RED;                                                               \
+          __rdf_x = __rdf_xp;                                                                      \
+          __rdf_xp = __rdf_x->parent;                                                              \
+        } else {                                                                                   \
+          if (rbtree_node_color (__w->right) == RBTREE_BLACK) {                                    \
+            if (__w->left)                                                                         \
+              __w->left->color = RBTREE_BLACK;                                                     \
+            __w->color = RBTREE_RED;                                                               \
+            rbtree_rotate_right (root_ptr, __w);                                                   \
+            __w = __rdf_xp->right;                                                                 \
+          }                                                                                        \
+          __w->color = __rdf_xp->color;                                                            \
+          __rdf_xp->color = RBTREE_BLACK;                                                          \
+          if (__w->right)                                                                          \
+            __w->right->color = RBTREE_BLACK;                                                      \
+          rbtree_rotate_left (root_ptr, __rdf_xp);                                                 \
+          __rdf_x = *(root_ptr);                                                                   \
+          break;                                                                                   \
+        }                                                                                          \
+      } else {                                                                                     \
+        struct rb_node_link* __w = __rdf_xp->left;                                                 \
+        if (rbtree_node_color (__w) == RBTREE_RED) {                                               \
+          __w->color = RBTREE_BLACK;                                                               \
+          __rdf_xp->color = RBTREE_RED;                                                            \
+          rbtree_rotate_right (root_ptr, __rdf_xp);                                                \
+          __w = __rdf_xp->left;                                                                    \
+        }                                                                                          \
+        if (rbtree_node_color (__w->right) == RBTREE_BLACK &&                                      \
+            rbtree_node_color (__w->left) == RBTREE_BLACK) {                                       \
+          if (__w)                                                                                 \
+            __w->color = RBTREE_RED;                                                               \
+          __rdf_x = __rdf_xp;                                                                      \
+          __rdf_xp = __rdf_x->parent;                                                              \
+        } else {                                                                                   \
+          if (rbtree_node_color (__w->left) == RBTREE_BLACK) {                                     \
+            if (__w->right)                                                                        \
+              __w->right->color = RBTREE_BLACK;                                                    \
+            __w->color = RBTREE_RED;                                                               \
+            rbtree_rotate_left (root_ptr, __w);                                                    \
+            __w = __rdf_xp->left;                                                                  \
+          }                                                                                        \
+          __w->color = __rdf_xp->color;                                                            \
+          __rdf_xp->color = RBTREE_BLACK;                                                          \
+          if (__w->left)                                                                           \
+            __w->left->color = RBTREE_BLACK;                                                       \
+          rbtree_rotate_right (root_ptr, __rdf_xp);                                                \
+          __rdf_x = *(root_ptr);                                                                   \
+          break;                                                                                   \
+        }                                                                                          \
+      }                                                                                            \
+    }                                                                                              \
+    if (__rdf_x)                                                                                   \
+      __rdf_x->color = RBTREE_BLACK;                                                               \
+  } while (0)
+
+#define rbtree_delete(root_ptr, z_node)                                                            \
+  do {                                                                                             \
+    struct rb_node_link* __rd_z = (z_node);                                                        \
+    struct rb_node_link* __rd_y = __rd_z;                                                          \
+    struct rb_node_link* __rd_x = NULL;                                                            \
+    struct rb_node_link* __rd_xp = NULL;                                                           \
+    int __rd_y_orig_color = __rd_y->color;                                                         \
+    if (!__rd_z->left) {                                                                           \
+      __rd_x = __rd_z->right;                                                                      \
+      __rd_xp = __rd_z->parent;                                                                    \
+      rbtree_transplant (root_ptr, __rd_z, __rd_z->right);                                         \
+    } else if (!__rd_z->right) {                                                                   \
+      __rd_x = __rd_z->left;                                                                       \
+      __rd_xp = __rd_z->parent;                                                                    \
+      rbtree_transplant (root_ptr, __rd_z, __rd_z->left);                                          \
+    } else {                                                                                       \
+      rbtree_min (__rd_z->right, __rd_y);                                                          \
+      __rd_y_orig_color = __rd_y->color;                                                           \
+      __rd_x = __rd_y->right;                                                                      \
+      if (__rd_y->parent == __rd_z) {                                                              \
+        __rd_xp = __rd_y;                                                                          \
+        if (__rd_x)                                                                                \
+          __rd_x->parent = __rd_y;                                                                 \
+      } else {                                                                                     \
+        __rd_xp = __rd_y->parent;                                                                  \
+        rbtree_transplant (root_ptr, __rd_y, __rd_y->right);                                       \
+        __rd_y->right = __rd_z->right;                                                             \
+        __rd_y->right->parent = __rd_y;                                                            \
+      }                                                                                            \
+      rbtree_transplant (root_ptr, __rd_z, __rd_y);                                                \
+      __rd_y->left = __rd_z->left;                                                                 \
+      __rd_y->left->parent = __rd_y;                                                               \
+      __rd_y->color = __rd_z->color;                                                               \
+    }                                                                                              \
+    if (__rd_y_orig_color == RBTREE_BLACK)                                                         \
+      rbtree_delete_fixup (root_ptr, __rd_x, __rd_xp);                                             \
+  } while (0)
+
+#define rbtree_next(node, out)                                                                     \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    if (node) {                                                                                    \
+      if ((node)->right) {                                                                         \
+        struct rb_node_link* __n = (node)->right;                                                  \
+        while (__n->left)                                                                          \
+          __n = __n->left;                                                                         \
+        (out) = __n;                                                                               \
+      } else {                                                                                     \
+        struct rb_node_link* __n = (node);                                                         \
+        struct rb_node_link* __p = (node)->parent;                                                 \
+        while (__p && __n == __p->right) {                                                         \
+          __n = __p;                                                                               \
+          __p = __p->parent;                                                                       \
+        }                                                                                          \
+        (out) = __p;                                                                               \
+      }                                                                                            \
+    }                                                                                              \
+  } while (0)
+
+#define rbtree_prev(node, out)                                                                     \
+  do {                                                                                             \
+    (out) = NULL;                                                                                  \
+    if (node) {                                                                                    \
+      if ((node)->left) {                                                                          \
+        struct rb_node_link* __n = (node)->left;                                                   \
+        while (__n->right)                                                                         \
+          __n = __n->right;                                                                        \
+        (out) = __n;                                                                               \
+      } else {                                                                                     \
+        struct rb_node_link* __n = (node);                                                         \
+        struct rb_node_link* __p = (node)->parent;                                                 \
+        while (__p && __n == __p->left) {                                                          \
+          __n = __p;                                                                               \
+          __p = __p->parent;                                                                       \
+        }                                                                                          \
+        (out) = __p;                                                                               \
+      }                                                                                            \
+    }                                                                                              \
+  } while (0)
+
+#endif // _KERNEL_LIBK_RBTREE_H

kernel/libk/string.h

@@ -1,6 +1,8 @@
 #ifndef _KERNEL_LIBK_STRING_H
 #define _KERNEL_LIBK_STRING_H
 
+#include <libk/std.h>
+
 size_t memset (void* dst, uint8_t b, size_t n);
 size_t memcpy (void* dst, const void* src, size_t n);
 void strncpy (char* dst, const char* src, size_t n);

kernel/limine/requests.c

@@ -18,3 +18,6 @@ volatile uint64_t limine_requests_end_marker[] = LIMINE_REQUESTS_END_MARKER;
 DECL_REQ (hhdm, HHDM);
 DECL_REQ (memmap, MEMMAP);
 DECL_REQ (rsdp, RSDP);
+DECL_REQ (mp, MP);
+DECL_REQ (module, MODULE);
+DECL_REQ (framebuffer, FRAMEBUFFER);

kernel/limine/requests.h

@@ -8,5 +8,8 @@
 EXTERN_REQ (hhdm);
 EXTERN_REQ (memmap);
 EXTERN_REQ (rsdp);
+EXTERN_REQ (mp);
+EXTERN_REQ (module);
+EXTERN_REQ (framebuffer);
 
 #endif // _KERNEL_LIMINE_REQUESTS_H

kernel/mm/liballoc.c

@@ -6,17 +6,18 @@
 #include <mm/pmm.h>
 #include <mm/types.h>
 #include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/mm.h>
 
-/* Porting */
 spin_lock_t _liballoc_lock = SPIN_LOCK_INIT;
 
-int liballoc_lock (void) {
-  spin_lock (&_liballoc_lock);
+int liballoc_lock (void* ctx) {
+  spin_lock (&_liballoc_lock, (spin_lock_ctx_t*)ctx);
   return 0;
 }
 
-int liballoc_unlock (void) {
-  spin_unlock (&_liballoc_lock);
+int liballoc_unlock (void* ctx) {
+  spin_unlock (&_liballoc_lock, (spin_lock_ctx_t*)ctx);
   return 0;
 }
 
@@ -28,6 +29,7 @@ void* liballoc_alloc (int pages) {
 
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   uintptr_t addr = (uintptr_t)(p_addr + hhdm->offset);
+
   return (void*)addr;
 }
 
@@ -43,7 +45,7 @@ int liballoc_free (void* ptr, int pages) {
 
 /**  Durand's Ridiculously Amazing Super Duper Memory functions.  */
 
-//#define DEBUG
+// #define DEBUG
 
 #define LIBALLOC_MAGIC 0xc001c0de
 #define MAXCOMPLETE    5
@@ -55,33 +57,21 @@ int liballoc_free (void* ptr, int pages) {
 
 #define MODE MODE_BEST
 
-#ifdef DEBUG
-#include <stdio.h>
-#endif
-
 struct boundary_tag* l_freePages[MAXEXP]; //< Allowing for 2^MAXEXP blocks
 int l_completePages[MAXEXP];              //< Allowing for 2^MAXEXP blocks
 
-#ifdef DEBUG
-unsigned int l_allocated = 0; //< The real amount of memory allocated.
-unsigned int l_inuse = 0;     //< The amount of memory in use (malloc'ed).
-#endif
-
 static int l_initialized = 0; //< Flag to indicate initialization.
 static int l_pageSize = 4096; //< Individual page size
 static int l_pageCount = 16;  //< Minimum number of pages to allocate.
 
 // ***********   HELPER FUNCTIONS  *******************************
 
-/** Returns the exponent required to manage 'size' amount of memory. 
+/** Returns the exponent required to manage 'size' amount of memory.
  *
  *  Returns n where  2^n <= size < 2^(n+1)
  */
 static inline int getexp (unsigned int size) {
   if (size < (1 << MINEXP)) {
-#ifdef DEBUG
-    printf ("getexp returns -1 for %i less than MINEXP\n", size);
-#endif
     return -1; // Smaller than the quantum.
   }
 
@@ -93,10 +83,6 @@ static inline int getexp (unsigned int size) {
     shift += 1;
   }
 
-#ifdef DEBUG
-  printf ("getexp returns %i (%i bytes) for %i size\n", shift - 1, (1 << (shift - 1)), size);
-#endif
-
   return shift - 1;
 }
 
@@ -130,37 +116,6 @@ static void* liballoc_memcpy (void* s1, const void* s2, size_t n) {
   return s1;
 }
 
-#ifdef DEBUG
-static void dump_array () {
-  int i = 0;
-  struct boundary_tag* tag = NULL;
-
-  printf ("------ Free pages array ---------\n");
-  printf ("System memory allocated: %i\n", l_allocated);
-  printf ("Memory in used (malloc'ed): %i\n", l_inuse);
-
-  for (i = 0; i < MAXEXP; i++) {
-    printf ("%.2i(%i): ", i, l_completePages[i]);
-
-    tag = l_freePages[i];
-    while (tag != NULL) {
-      if (tag->split_left != NULL)
-        printf ("*");
-      printf ("%i", tag->real_size);
-      if (tag->split_right != NULL)
-        printf ("*");
-
-      printf (" ");
-      tag = tag->next;
-    }
-    printf ("\n");
-  }
-
-  printf ("'*' denotes a split to the left/right of a tag\n");
-  fflush (stdout);
-}
-#endif
-
 static inline void insert_tag (struct boundary_tag* tag, int index) {
   int realIndex;
 
@@ -281,15 +236,6 @@ static struct boundary_tag* allocate_new_tag (unsigned int size) {
   tag->split_left = NULL;
   tag->split_right = NULL;
 
-#ifdef DEBUG
-  printf ("Resource allocated %x of %i pages (%i bytes) for %i size.\n", tag, pages,
-          pages * l_pageSize, size);
-
-  l_allocated += pages * l_pageSize;
-
-  printf ("Total memory usage = %i KB\n", (int)((l_allocated / (1024))));
-#endif
-
   return tag;
 }
 
@@ -297,13 +243,11 @@ void* malloc (size_t size) {
   int index;
   void* ptr;
   struct boundary_tag* tag = NULL;
+  spin_lock_ctx_t ctxliba;
 
-  liballoc_lock ();
+  liballoc_lock (&ctxliba);
 
   if (l_initialized == 0) {
-#ifdef DEBUG
-    printf ("%s\n", "liballoc initializing.");
-#endif
     for (index = 0; index < MAXEXP; index++) {
       l_freePages[index] = NULL;
       l_completePages[index] = 0;
@@ -320,10 +264,6 @@ void* malloc (size_t size) {
   while (tag != NULL) {
     // If there's enough space in this tag.
     if ((tag->real_size - sizeof (struct boundary_tag)) >= (size + sizeof (struct boundary_tag))) {
-#ifdef DEBUG
-      printf ("Tag search found %i >= %i\n", (tag->real_size - sizeof (struct boundary_tag)),
-              (size + sizeof (struct boundary_tag)));
-#endif
       break;
     }
 
@@ -333,7 +273,7 @@ void* malloc (size_t size) {
   // No page found. Make one.
   if (tag == NULL) {
     if ((tag = allocate_new_tag (size)) == NULL) {
-      liballoc_unlock ();
+      liballoc_unlock (&ctxliba);
       return NULL;
     }
 
@@ -351,13 +291,6 @@ void* malloc (size_t size) {
 
   // Removed... see if we can re-use the excess space.
 
-#ifdef DEBUG
-  printf (
-      "Found tag with %i bytes available (requested %i bytes, leaving %i), which has exponent: %i (%i bytes)\n",
-      tag->real_size - sizeof (struct boundary_tag), size,
-      tag->real_size - size - sizeof (struct boundary_tag), index, 1 << index);
-#endif
-
   unsigned int remainder =
       tag->real_size - size - sizeof (struct boundary_tag) * 2; // Support a new tag + remainder
 
@@ -365,73 +298,43 @@ void* malloc (size_t size) {
     int childIndex = getexp (remainder);
 
     if (childIndex >= 0) {
-#ifdef DEBUG
-      printf ("Seems to be splittable: %i >= 2^%i .. %i\n", remainder, childIndex,
-              (1 << childIndex));
-#endif
-
       struct boundary_tag* new_tag = split_tag (tag);
 
       (void)new_tag;
-
-#ifdef DEBUG
-      printf ("Old tag has become %i bytes, new tag is now %i bytes (%i exp)\n", tag->real_size,
-              new_tag->real_size, new_tag->index);
-#endif
     }
   }
 
   ptr = (void*)((uintptr_t)tag + sizeof (struct boundary_tag));
 
-#ifdef DEBUG
-  l_inuse += size;
-  printf ("malloc: %x,  %i, %i\n", ptr, (int)l_inuse / 1024, (int)l_allocated / 1024);
-  dump_array ();
-#endif
-
-  liballoc_unlock ();
+  liballoc_unlock (&ctxliba);
   return ptr;
 }
 
 void free (void* ptr) {
   int index;
   struct boundary_tag* tag;
+  spin_lock_ctx_t ctxliba;
 
   if (ptr == NULL)
     return;
 
-  liballoc_lock ();
+  liballoc_lock (&ctxliba);
 
   tag = (struct boundary_tag*)((uintptr_t)ptr - sizeof (struct boundary_tag));
 
   if (tag->magic != LIBALLOC_MAGIC) {
-    liballoc_unlock (); // release the lock
+    liballoc_unlock (&ctxliba); // release the lock
     return;
   }
 
-#ifdef DEBUG
-  l_inuse -= tag->size;
-  printf ("free: %x, %i, %i\n", ptr, (int)l_inuse / 1024, (int)l_allocated / 1024);
-#endif
-
   // MELT LEFT...
   while ((tag->split_left != NULL) && (tag->split_left->index >= 0)) {
-#ifdef DEBUG
-    printf ("Melting tag left into available memory. Left was %i, becomes %i (%i)\n",
-            tag->split_left->real_size, tag->split_left->real_size + tag->real_size,
-            tag->split_left->real_size);
-#endif
     tag = melt_left (tag);
     remove_tag (tag);
   }
 
   // MELT RIGHT...
   while ((tag->split_right != NULL) && (tag->split_right->index >= 0)) {
-#ifdef DEBUG
-    printf ("Melting tag right into available memory. This was was %i, becomes %i (%i)\n",
-            tag->real_size, tag->split_right->real_size + tag->real_size,
-            tag->split_right->real_size);
-#endif
     tag = absorb_right (tag);
   }
 
@@ -453,13 +356,7 @@ void free (void* ptr) {
 
       liballoc_free (tag, pages);
 
-#ifdef DEBUG
-      l_allocated -= pages * l_pageSize;
-      printf ("Resource freeing %x of %i pages\n", tag, pages);
-      dump_array ();
-#endif
-
-      liballoc_unlock ();
+      liballoc_unlock (&ctxliba);
       return;
     }
 
@@ -470,13 +367,7 @@ void free (void* ptr) {
 
   insert_tag (tag, index);
 
-#ifdef DEBUG
-  printf ("Returning tag with %i bytes (requested %i bytes), which has exponent: %i\n",
-          tag->real_size, tag->size, index);
-  dump_array ();
-#endif
-
-  liballoc_unlock ();
+  liballoc_unlock (&ctxliba);
 }
 
 void* calloc (size_t nobj, size_t size) {
@@ -496,6 +387,7 @@ void* realloc (void* p, size_t size) {
   void* ptr;
   struct boundary_tag* tag;
   int real_size;
+  spin_lock_ctx_t ctxliba;
 
   if (size == 0) {
     free (p);
@@ -505,11 +397,11 @@ void* realloc (void* p, size_t size) {
     return malloc (size);
 
   if (&liballoc_lock != NULL)
-    liballoc_lock (); // lockit
+    liballoc_lock (&ctxliba); // lockit
   tag = (struct boundary_tag*)((uintptr_t)p - sizeof (struct boundary_tag));
   real_size = tag->size;
   if (&liballoc_unlock != NULL)
-    liballoc_unlock ();
+    liballoc_unlock (&ctxliba);
 
   if ((size_t)real_size > size)
     real_size = size;

kernel/mm/liballoc.h

@@ -42,12 +42,12 @@ struct boundary_tag {
 
 /** This function is supposed to lock the memory data structures. It
  * could be as simple as disabling interrupts or acquiring a spinlock.
- * It's up to you to decide. 
+ * It's up to you to decide.
  *
  * \return 0 if the lock was acquired successfully. Anything else is
  * failure.
  */
-extern int liballoc_lock ();
+extern int liballoc_lock (void* ctx);
 
 /** This function unlocks what was previously locked by the liballoc_lock
  * function.  If it disabled interrupts, it enables interrupts. If it
@@ -55,7 +55,7 @@ extern int liballoc_lock ();
  *
  * \return 0 if the lock was successfully released.
  */
-extern int liballoc_unlock ();
+extern int liballoc_unlock (void* ctx);
 
 /** This is the hook into the local system which allocates pages. It
  * accepts an integer parameter which is the number of pages

kernel/mm/pmm.c

@@ -38,8 +38,8 @@ void pmm_init (void) {
       struct pmm_region* pmm_region = &pmm.regions[region];
 
       /*
-       * We need to calculate sizes for the pmm region and the bitmap. The bitmap MUSTN'T include it's
-       * own region within the bit range.
+       * We need to calculate sizes for the pmm region and the bitmap. The bitmap MUSTN'T include
+       * it's own region within the bit range.
        * */
 
       size_t size = align_down (entry->length, PAGE_SIZE);
@@ -100,6 +100,8 @@ static size_t pmm_find_free_space (struct pmm_region* pmm_region, size_t nblks)
 }
 
 physaddr_t pmm_alloc (size_t nblks) {
+  spin_lock_ctx_t ctxpmmr;
+
   for (size_t region = 0; region < PMM_REGIONS_MAX; region++) {
     struct pmm_region* pmm_region = &pmm.regions[region];
 
@@ -107,7 +109,7 @@ physaddr_t pmm_alloc (size_t nblks) {
     if (!(pmm_region->flags & PMM_REGION_ACTIVE))
       continue;
 
-    spin_lock (&pmm_region->lock);
+    spin_lock (&pmm_region->lock, &ctxpmmr);
 
     /* Find starting bit of the free bit range */
     size_t bit = pmm_find_free_space (pmm_region, nblks);
@@ -116,18 +118,19 @@ physaddr_t pmm_alloc (size_t nblks) {
     if (bit != (size_t)-1) {
       /* Mark it */
       bm_set_region (&pmm_region->bm, bit, nblks);
-      spin_unlock (&pmm_region->lock);
+      spin_unlock (&pmm_region->lock, &ctxpmmr);
 
       return pmm_region->membase + bit * PAGE_SIZE;
     }
 
-    spin_unlock (&pmm_region->lock);
+    spin_unlock (&pmm_region->lock, &ctxpmmr);
   }
 
   return PMM_ALLOC_ERR;
 }
 
 void pmm_free (physaddr_t p_addr, size_t nblks) {
+  spin_lock_ctx_t ctxpmmr;
   /* Round down to nearest page boundary */
   physaddr_t aligned_p_addr = align_down (p_addr, PAGE_SIZE);
 
@@ -139,16 +142,17 @@ void pmm_free (physaddr_t p_addr, size_t nblks) {
       continue;
 
     /* If aligned_p_addr is within the range if this region, it belongs to it. */
-    if (aligned_p_addr >= pmm_region->membase && aligned_p_addr < pmm_region->size) {
+    if (aligned_p_addr >= pmm_region->membase &&
+        aligned_p_addr < pmm_region->membase + pmm_region->size) {
       physaddr_t addr = aligned_p_addr - pmm_region->membase;
 
       size_t bit = div_align_up (addr, PAGE_SIZE);
 
-      spin_lock (&pmm_region->lock);
+      spin_lock (&pmm_region->lock, &ctxpmmr);
 
       bm_clear_region (&pmm_region->bm, bit, nblks);
 
-      spin_unlock (&pmm_region->lock);
+      spin_unlock (&pmm_region->lock, &ctxpmmr);
 
       break;
     }

kernel/proc/.gitignore

@@ -0,0 +1 @@
+*.o

kernel/proc/locks.txt

@@ -0,0 +1,10 @@
+Lock hierarchy for process scheduling:
+
+1. proc_tree_lock
+2. cpu->lock
+3. procgroup->lock
+4. proc->lock
+5. sq->lock
+
+1. procgroup_tree_lock
+2. procgroup->lock

kernel/proc/mutex.c

@@ -0,0 +1,130 @@
+#include <libk/assert.h>
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <libk/string.h>
+#include <mm/liballoc.h>
+#include <proc/mutex.h>
+#include <proc/proc.h>
+#include <proc/suspension_q.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/smp.h>
+#include <sys/spin_lock.h>
+
+void proc_mutexes_cleanup (struct proc* proc) {
+  spin_lock_ctx_t ctxpg, ctxrs;
+
+  spin_lock (&proc->procgroup->lock, &ctxpg);
+
+  struct rb_node_link* rnode;
+  rbtree_first (&proc->procgroup->resource_tree, rnode);
+
+  while (rnode) {
+    struct rb_node_link* next;
+    rbtree_next (rnode, next);
+
+    struct proc_resource* resource = rbtree_entry (rnode, struct proc_resource, resource_tree_link);
+
+    rnode = next;
+
+    spin_lock (&resource->lock, &ctxrs);
+
+    if (resource->type != PR_MUTEX) {
+      spin_unlock (&resource->lock, &ctxrs);
+      continue;
+    }
+
+    if (resource->u.mutex.owner == proc && resource->u.mutex.locked) {
+      spin_unlock (&resource->lock, &ctxrs);
+
+      proc_mutex_unlock (proc, &resource->u.mutex);
+    }
+  }
+
+  spin_unlock (&proc->procgroup->lock, &ctxpg);
+}
+
+bool proc_cleanup_resource_mutex (struct proc_resource* resource) {
+  struct proc_mutex* mutex = &resource->u.mutex;
+  spin_lock_ctx_t ctxmt, ctxsq;
+
+  spin_lock (&mutex->resource->lock, &ctxmt);
+  spin_lock (&mutex->suspension_q.lock, &ctxsq);
+
+  bool reschedule = PROC_NO_RESCHEDULE;
+
+  while (mutex->suspension_q.proc_list != NULL) {
+    struct list_node_link* node = mutex->suspension_q.proc_list;
+    struct proc_sq_entry* sq_entry = list_entry (node, struct proc_sq_entry, sq_link);
+    struct proc* suspended_proc = sq_entry->proc;
+
+    /* we will relock during resume */
+    spin_unlock (&mutex->suspension_q.lock, &ctxsq);
+    spin_unlock (&mutex->resource->lock, &ctxmt);
+
+    reschedule = reschedule || proc_sq_resume (suspended_proc, sq_entry);
+
+    /* reacquire */
+    spin_lock (&mutex->resource->lock, &ctxmt);
+    spin_lock (&mutex->suspension_q.lock, &ctxsq);
+  }
+
+  mutex->locked = false;
+  mutex->owner = NULL;
+
+  spin_unlock (&mutex->suspension_q.lock, &ctxsq);
+  spin_unlock (&mutex->resource->lock, &ctxmt);
+
+  return reschedule;
+}
+
+bool proc_mutex_lock (struct proc* proc, struct proc_mutex* mutex) {
+  spin_lock_ctx_t ctxmt;
+
+  spin_lock (&mutex->resource->lock, &ctxmt);
+
+  if (!mutex->locked || mutex->owner == proc) {
+    mutex->locked = true;
+    mutex->owner = proc;
+    spin_unlock (&mutex->resource->lock, &ctxmt);
+    return PROC_NO_RESCHEDULE;
+  }
+
+  return proc_sq_suspend (proc, &mutex->suspension_q, &mutex->resource->lock, &ctxmt);
+}
+
+bool proc_mutex_unlock (struct proc* proc, struct proc_mutex* mutex) {
+  spin_lock_ctx_t ctxmt, ctxsq;
+
+  spin_lock (&mutex->resource->lock, &ctxmt);
+
+  if (mutex->owner != proc) {
+    spin_unlock (&mutex->resource->lock, &ctxmt);
+    return PROC_NO_RESCHEDULE;
+  }
+
+  spin_lock (&mutex->suspension_q.lock, &ctxsq);
+
+  struct list_node_link* node = mutex->suspension_q.proc_list;
+
+  if (node) {
+    struct proc_sq_entry* sq_entry = list_entry (node, struct proc_sq_entry, sq_link);
+    struct proc* resumed_proc = sq_entry->proc;
+
+    mutex->owner = resumed_proc;
+    mutex->locked = true;
+
+    spin_unlock (&mutex->suspension_q.lock, &ctxsq);
+    spin_unlock (&mutex->resource->lock, &ctxmt);
+
+    return proc_sq_resume (resumed_proc, sq_entry);
+  }
+
+  mutex->locked = false;
+  mutex->owner = NULL;
+
+  spin_unlock (&mutex->suspension_q.lock, &ctxsq);
+  spin_unlock (&mutex->resource->lock, &ctxmt);
+
+  return PROC_NEED_RESCHEDULE;
+}

kernel/proc/mutex.h

@@ -0,0 +1,23 @@
+#ifndef _KERNEL_PROC_MUTEX_H
+#define _KERNEL_PROC_MUTEX_H
+
+#include <libk/std.h>
+#include <proc/suspension_q.h>
+
+struct proc;
+struct proc_resource;
+
+struct proc_mutex {
+  struct proc_resource* resource;
+
+  bool locked;
+  struct proc_suspension_q suspension_q;
+  struct proc* owner;
+};
+
+bool proc_cleanup_resource_mutex (struct proc_resource* resource);
+bool proc_mutex_lock (struct proc* proc, struct proc_mutex* mutex);
+bool proc_mutex_unlock (struct proc* proc, struct proc_mutex* mutex);
+void proc_mutexes_cleanup (struct proc* proc);
+
+#endif // _KERNEL_PROC_MUTEX_H

kernel/proc/proc.c

@@ -0,0 +1,283 @@
+#include <aux/compiler.h>
+#include <aux/elf.h>
+#include <irq/irq.h>
+#include <libk/align.h>
+#include <libk/list.h>
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <libk/string.h>
+#include <limine/requests.h>
+#include <mm/liballoc.h>
+#include <mm/pmm.h>
+#include <proc/proc.h>
+#include <proc/procgroup.h>
+#include <proc/resource.h>
+#include <rd/rd.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/mm.h>
+#include <sys/proc.h>
+#include <sys/sched.h>
+#include <sys/smp.h>
+#include <sys/spin.h>
+
+#if defined(__x86_64__)
+#include <amd64/intr_defs.h>
+#endif
+
+#define SCHED_REAP_FREQ 10
+
+static struct rb_node_link* proc_tree = NULL;
+static spin_lock_t proc_tree_lock = SPIN_LOCK_INIT;
+
+static atomic_int sched_cycles = 0;
+
+static bool proc_check_elf (uint8_t* elf) {
+  if (!((elf[0] == 0x7F) && (elf[1] == 'E') && (elf[2] == 'L') && (elf[3] == 'F')))
+    return false;
+  return true;
+}
+
+struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf) {
+  struct elf_aux aux;
+
+  Elf64_Ehdr* ehdr = (Elf64_Ehdr*)elf;
+  aux.entry = ehdr->e_entry;
+  aux.phnum = ehdr->e_phnum;
+  aux.phent = ehdr->e_phentsize;
+
+  struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
+
+  for (uint64_t segment = 0; segment < ehdr->e_phnum; segment++) {
+    Elf64_Phdr* phdr =
+        (Elf64_Phdr*)((uintptr_t)elf + ehdr->e_phoff + (ehdr->e_phentsize * segment));
+
+    switch (phdr->p_type) {
+    case PT_PHDR: {
+      aux.phdr = (uint64_t)phdr->p_vaddr;
+    } break;
+    case PT_LOAD: {
+      uintptr_t v_addr = align_down (phdr->p_vaddr, PAGE_SIZE);
+      uintptr_t off = phdr->p_vaddr - v_addr;
+
+      size_t blks = div_align_up (phdr->p_memsz + off, PAGE_SIZE);
+
+      uint32_t pg_flags = MM_PG_USER | MM_PG_PRESENT;
+      if (phdr->p_flags & PF_W)
+        pg_flags |= MM_PG_RW;
+
+      uintptr_t p_addr;
+      procgroup_map (proc->procgroup, v_addr, blks, pg_flags, &p_addr);
+
+      memset ((void*)((uintptr_t)hhdm->offset + p_addr), 0, blks * PAGE_SIZE);
+      memcpy ((void*)((uintptr_t)hhdm->offset + p_addr + off),
+              (void*)((uintptr_t)elf + phdr->p_offset), phdr->p_filesz);
+    } break;
+    case PT_TLS: {
+#if defined(__x86_64__)
+      if (phdr->p_memsz > 0) {
+        size_t tls_align = phdr->p_align ? phdr->p_align : sizeof (uintptr_t);
+        size_t tls_size = align_up (phdr->p_memsz, tls_align);
+        size_t tls_total_needed = tls_size + sizeof (uintptr_t);
+        size_t blks = div_align_up (tls_total_needed, PAGE_SIZE);
+        proc->procgroup->tls.tls_tmpl_pages = blks;
+        proc->procgroup->tls.tls_tmpl_size = tls_size;
+        proc->procgroup->tls.tls_tmpl_total_size = tls_total_needed;
+
+        proc->procgroup->tls.tls_tmpl = malloc (blks * PAGE_SIZE);
+        memset (proc->procgroup->tls.tls_tmpl, 0, blks * PAGE_SIZE);
+
+        memcpy (proc->procgroup->tls.tls_tmpl, (void*)((uintptr_t)elf + phdr->p_offset),
+                phdr->p_filesz);
+
+        proc_init_tls (proc);
+      }
+#endif
+    } break;
+    }
+  }
+
+  return aux;
+}
+
+struct proc* proc_spawn_rd (char* name) {
+  struct rd_file* rd_file = rd_get_file (name);
+
+  bool ok = proc_check_elf (rd_file->content);
+
+  if (!ok)
+    return NULL;
+
+  return proc_from_elf (rd_file->content);
+}
+
+struct proc* proc_find_pid (int pid) {
+  spin_lock_ctx_t ctxprtr;
+  struct proc* proc = NULL;
+
+  spin_lock (&proc_tree_lock, &ctxprtr);
+  rbtree_find (struct proc, &proc_tree, pid, proc, proc_tree_link, pid);
+  spin_unlock (&proc_tree_lock, &ctxprtr);
+
+  return proc;
+}
+
+void proc_register (struct proc* proc, struct cpu* cpu1) {
+  spin_lock_ctx_t ctxcpu, ctxprtr;
+
+  proc->cpu = cpu1 != NULL ? cpu1 : cpu_find_lightest ();
+
+  struct cpu* cpu = proc->cpu;
+
+  spin_lock (&proc_tree_lock, &ctxprtr);
+  spin_lock (&cpu->lock, &ctxcpu);
+
+  rbtree_insert (struct proc, &proc_tree, &proc->proc_tree_link, proc_tree_link, pid);
+
+  atomic_fetch_add (&cpu->proc_run_q_count, 1);
+  list_append (cpu->proc_run_q, &proc->cpu_run_q_link);
+  if (cpu->proc_current == NULL)
+    cpu->proc_current = proc;
+
+  spin_unlock (&proc_tree_lock, &ctxprtr);
+  spin_unlock (&cpu->lock, &ctxcpu);
+}
+
+/* caller holds cpu->lock */
+static struct proc* proc_find_sched (struct cpu* cpu) {
+  if (!cpu->proc_run_q)
+    return NULL;
+
+  struct list_node_link *current, *start;
+
+  if (cpu->proc_current)
+    current = cpu->proc_current->cpu_run_q_link.next;
+  else
+    current = cpu->proc_run_q;
+
+  if (!current)
+    current = cpu->proc_run_q;
+
+  start = current;
+
+  do {
+    struct proc* proc = list_entry (current, struct proc, cpu_run_q_link);
+
+    if (atomic_load (&proc->state) == PROC_READY)
+      return proc;
+
+    current = current->next ? current->next : cpu->proc_run_q;
+  } while (current != start);
+
+  return NULL;
+}
+
+static void proc_reap (void) {
+  struct proc* proc = NULL;
+  struct list_node_link* reap_list = NULL;
+  spin_lock_ctx_t ctxprtr;
+  spin_lock_ctx_t ctxpr;
+
+  spin_lock (&proc_tree_lock, &ctxprtr);
+
+  struct rb_node_link* node;
+  rbtree_first (&proc_tree, node);
+
+  while (node) {
+    struct rb_node_link* next;
+    rbtree_next (node, next);
+    proc = rbtree_entry (node, struct proc, proc_tree_link);
+
+    if (atomic_load (&proc->state) == PROC_DEAD) {
+      spin_lock (&proc->lock, &ctxpr);
+      rbtree_delete (&proc_tree, &proc->proc_tree_link);
+      list_append (reap_list, &proc->reap_link);
+      spin_unlock (&proc->lock, &ctxpr);
+    }
+
+    node = next;
+  }
+
+  spin_unlock (&proc_tree_lock, &ctxprtr);
+
+  struct list_node_link *reap_link, *reap_link_tmp;
+  list_foreach (reap_list, reap_link, reap_link_tmp) {
+    proc = list_entry (reap_link, struct proc, reap_link);
+
+    list_remove (reap_list, &proc->reap_link);
+    DEBUG ("cleanup PID %d\n", proc->pid);
+    proc_cleanup (proc);
+  }
+}
+
+void proc_sched (void) {
+  spin_lock_ctx_t ctxcpu;
+
+  int s_cycles = atomic_fetch_add (&sched_cycles, 1);
+
+  if (s_cycles % SCHED_REAP_FREQ == 0)
+    proc_reap ();
+
+  struct proc* next = NULL;
+  struct cpu* cpu = thiscpu;
+
+  spin_lock (&cpu->lock, &ctxcpu);
+
+  next = proc_find_sched (cpu);
+
+  if (next) {
+    cpu->proc_current = next;
+
+    do_sched (next, &cpu->lock, &ctxcpu);
+  } else {
+    cpu->proc_current = NULL;
+    spin_unlock (&cpu->lock, &ctxcpu);
+
+    spin ();
+  }
+}
+
+void proc_kill (struct proc* proc) {
+  spin_lock_ctx_t ctxpr, ctxcpu;
+  struct cpu* cpu = proc->cpu;
+
+  spin_lock (&proc->lock, &ctxpr);
+  atomic_store (&proc->state, PROC_DEAD);
+  proc->cpu = NULL;
+  spin_unlock (&proc->lock, &ctxpr);
+
+  spin_lock (&cpu->lock, &ctxcpu);
+
+  list_remove (cpu->proc_run_q, &proc->cpu_run_q_link);
+  atomic_fetch_sub (&cpu->proc_run_q_count, 1);
+  if (cpu->proc_current == proc)
+    cpu->proc_current = NULL;
+
+  spin_unlock (&cpu->lock, &ctxcpu);
+
+  DEBUG ("killed PID %d\n", proc->pid);
+
+  cpu_request_sched (cpu);
+}
+
+static void proc_irq_sched (void* arg, void* regs) {
+  (void)arg;
+  proc_sched ();
+}
+
+void proc_init (void) {
+#if defined(__x86_64__)
+  irq_attach (&proc_irq_sched, NULL, SCHED_PREEMPT_TIMER);
+  irq_attach (&proc_irq_sched, NULL, CPU_REQUEST_SCHED);
+#endif
+
+  struct proc* spin_proc = proc_spawn_rd ("spin.exe");
+  proc_register (spin_proc, thiscpu);
+
+  struct proc* init = proc_spawn_rd ("init.exe");
+  proc_register (init, NULL);
+
+  spin_lock_ctx_t ctxcpu;
+  spin_lock (&spin_proc->cpu->lock, &ctxcpu);
+  do_sched (spin_proc, &spin_proc->cpu->lock, &ctxcpu);
+}

kernel/proc/proc.h

@@ -0,0 +1,57 @@
+#ifndef _KERNEL_PROC_PROC_H
+#define _KERNEL_PROC_PROC_H
+
+#include <aux/compiler.h>
+#include <aux/elf.h>
+#include <libk/list.h>
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <proc/procgroup.h>
+#include <proc/resource.h>
+#include <proc/suspension_q.h>
+#include <sync/spin_lock.h>
+#include <sys/mm.h>
+
+#if defined(__x86_64__)
+#include <amd64/gdt.h>  /* KSTACK_SIZE */
+#include <amd64/proc.h> /* USTACK_SIZE */
+#endif
+
+#define PROC_NEED_RESCHEDULE true
+#define PROC_NO_RESCHEDULE   false
+
+/* process states */
+#define PROC_READY     0
+#define PROC_DEAD      1
+#define PROC_SUSPENDED 2
+
+/* process flags */
+#define PROC_USTK_PREALLOC (1 << 0)
+
+struct cpu;
+
+struct proc {
+  int pid;
+  struct rb_node_link proc_tree_link;
+  struct rb_node_link procgroup_memb_tree_link;
+  struct list_node_link cpu_run_q_link;
+  struct list_node_link reap_link;
+  struct list_node_link* sq_entries;
+  struct procgroup* procgroup;
+  struct proc_platformdata pdata;
+  uint32_t flags;
+  spin_lock_t lock;
+  struct cpu* cpu;
+  atomic_int state;
+  uintptr_t uvaddr_argument;
+};
+
+void proc_sched (void);
+void proc_kill (struct proc* proc);
+struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf);
+void proc_register (struct proc* proc, struct cpu* cpu);
+struct proc* proc_find_pid (int pid);
+struct proc* proc_spawn_rd (char* name);
+void proc_init (void);
+
+#endif // _KERNEL_PROC_PROC_H

kernel/proc/procgroup.c

@@ -0,0 +1,218 @@
+#include <libk/rbtree.h>
+#include <libk/std.h>
+#include <mm/liballoc.h>
+#include <mm/pmm.h>
+#include <proc/proc.h>
+#include <proc/procgroup.h>
+#include <sync/spin_lock.h>
+#include <sys/debug.h>
+#include <sys/mm.h>
+
+static struct rb_node_link* procgroup_tree = NULL;
+static spin_lock_t procgroup_tree_lock = SPIN_LOCK_INIT;
+static atomic_int pgids = 0;
+
+uintptr_t procgroup_map (struct procgroup* procgroup, uintptr_t vaddr, size_t pages, uint32_t flags,
+                         uintptr_t* out_paddr) {
+  spin_lock_ctx_t ctxpg;
+
+  spin_lock (&procgroup->lock, &ctxpg);
+
+  vaddr = (vaddr == 0) ? procgroup->map_base : vaddr;
+
+  struct proc_mapping* mapping = malloc (sizeof (*mapping));
+
+  if (mapping == NULL) {
+    spin_unlock (&procgroup->lock, &ctxpg);
+    return 0;
+  }
+
+  uintptr_t paddr = pmm_alloc (pages);
+
+  if (paddr == PMM_ALLOC_ERR) {
+    free (mapping);
+    spin_unlock (&procgroup->lock, &ctxpg);
+    return 0;
+  }
+
+  if (out_paddr != NULL)
+    *out_paddr = paddr;
+
+  mapping->paddr = paddr;
+  mapping->vaddr = vaddr;
+  mapping->size = pages * PAGE_SIZE;
+
+  procgroup->map_base += pages * PAGE_SIZE;
+
+  list_append (procgroup->mappings, &mapping->proc_mappings_link);
+
+  for (uintptr_t vpage = vaddr, ppage = paddr; vpage < vaddr + pages * PAGE_SIZE;
+       vpage += PAGE_SIZE, ppage += PAGE_SIZE) {
+    mm_map_page (&procgroup->pd, ppage, vpage, flags);
+  }
+
+  spin_unlock (&procgroup->lock, &ctxpg);
+
+  return vaddr;
+}
+
+bool procgroup_unmap (struct procgroup* procgroup, uintptr_t start_vaddr, size_t pages) {
+  size_t unmap_size = pages * PAGE_SIZE;
+  uintptr_t end_vaddr = start_vaddr + unmap_size;
+
+  struct list_node_link *mapping_link, *mapping_link_tmp;
+
+  bool used_tail_mapping = false;
+  spin_lock_ctx_t ctxpg;
+
+  struct proc_mapping* tail_mapping = malloc (sizeof (*tail_mapping));
+  if (tail_mapping == NULL)
+    return false;
+
+  spin_lock (&procgroup->lock, &ctxpg);
+
+  list_foreach (procgroup->mappings, mapping_link, mapping_link_tmp) {
+    struct proc_mapping* mapping =
+        list_entry (mapping_link, struct proc_mapping, proc_mappings_link);
+
+    uintptr_t m_start = mapping->vaddr;
+    uintptr_t m_end = mapping->vaddr + mapping->size;
+
+    /* check overlap */
+    if ((start_vaddr < m_end) && (end_vaddr > mapping->vaddr)) {
+      uintptr_t free_vstart = (start_vaddr > m_start) ? start_vaddr : m_start;
+      uintptr_t free_vend = (end_vaddr < m_end) ? end_vaddr : m_end;
+      size_t free_size = free_vend - free_vstart;
+
+      uintptr_t ppage_to_free = mapping->paddr + (free_vstart - m_start);
+      pmm_free (ppage_to_free, free_size / PAGE_SIZE);
+
+      /* split in the middle */
+      if ((start_vaddr > m_start) && (end_vaddr < m_end)) {
+        tail_mapping->vaddr = end_vaddr;
+        tail_mapping->paddr = mapping->paddr + (end_vaddr - m_start);
+        tail_mapping->size = m_end - end_vaddr;
+
+        mapping->size = start_vaddr - m_start;
+
+        list_insert_after (procgroup->mappings, &mapping->proc_mappings_link,
+                           &tail_mapping->proc_mappings_link);
+
+        used_tail_mapping = true;
+
+        break;
+      } else if ((start_vaddr <= m_start) && (end_vaddr < m_end)) { /* shrink left */
+        size_t diff = end_vaddr - m_start;
+        mapping->vaddr += diff;
+        mapping->paddr += diff;
+        mapping->size -= diff;
+      } else if ((start_vaddr > m_start) && (end_vaddr >= m_end)) { /* shrink right */
+        mapping->size = start_vaddr - m_start;
+      } else { /* full overlap */
+        list_remove (procgroup->mappings, &mapping->proc_mappings_link);
+        free (mapping);
+      }
+    }
+  }
+
+  if (!used_tail_mapping)
+    free (tail_mapping);
+
+  for (uintptr_t vpage = start_vaddr; vpage < end_vaddr; vpage += PAGE_SIZE) {
+    mm_unmap_page (&procgroup->pd, vpage);
+  }
+
+  spin_unlock (&procgroup->lock, &ctxpg);
+
+  return true;
+}
+
+struct procgroup* procgroup_create (void) {
+  spin_lock_ctx_t ctxpgtr;
+
+  struct procgroup* procgroup = malloc (sizeof (*procgroup));
+  if (procgroup == NULL) {
+    return NULL;
+  }
+
+  procgroup->refs = 0;
+  procgroup->memb_proc_tree = NULL;
+  procgroup->lock = SPIN_LOCK_INIT;
+  procgroup->pgid = atomic_fetch_add (&pgids, 1);
+  procgroup->pd.cr3_paddr = mm_alloc_user_pd_phys ();
+  procgroup->map_base = PROC_MAP_BASE;
+
+  spin_lock (&procgroup_tree_lock, &ctxpgtr);
+  rbtree_insert (struct procgroup, &procgroup_tree, &procgroup->procgroup_tree_link,
+                 procgroup_tree_link, pgid);
+  spin_unlock (&procgroup_tree_lock, &ctxpgtr);
+
+  return procgroup;
+}
+
+void procgroup_attach (struct procgroup* procgroup, struct proc* proc) {
+  spin_lock_ctx_t ctxpg, ctxpr;
+
+  spin_lock (&procgroup->lock, &ctxpg);
+  spin_lock (&proc->lock, &ctxpr);
+
+  rbtree_insert (struct proc, &procgroup->memb_proc_tree, &proc->procgroup_memb_tree_link,
+                 procgroup_memb_tree_link, pid);
+  atomic_fetch_add (&procgroup->refs, 1);
+
+  spin_unlock (&proc->lock, &ctxpr);
+  spin_unlock (&procgroup->lock, &ctxpg);
+}
+
+void procgroup_detach (struct procgroup* procgroup, struct proc* proc) {
+  spin_lock_ctx_t ctxpg, ctxpr, ctxpgtr;
+
+  spin_lock (&procgroup->lock, &ctxpg);
+  spin_lock (&proc->lock, &ctxpr);
+
+  rbtree_delete (&procgroup->memb_proc_tree, &proc->procgroup_memb_tree_link);
+  int refs = atomic_fetch_sub (&procgroup->refs, 1);
+
+  spin_unlock (&proc->lock, &ctxpr);
+  spin_unlock (&procgroup->lock, &ctxpg);
+
+  if (refs == 1) {
+    spin_lock (&procgroup_tree_lock, &ctxpgtr);
+    spin_lock (&procgroup->lock, &ctxpg);
+
+    rbtree_delete (&procgroup_tree, &procgroup->procgroup_tree_link);
+
+    spin_unlock (&procgroup->lock, &ctxpg);
+    spin_unlock (&procgroup_tree_lock, &ctxpgtr);
+
+    /* delete resources */
+    struct rb_node_link* rnode;
+    rbtree_first (&procgroup->resource_tree, rnode);
+    while (rnode) {
+      struct rb_node_link* next;
+      rbtree_next (rnode, next);
+
+      struct proc_resource* resource =
+          rbtree_entry (rnode, struct proc_resource, resource_tree_link);
+
+      rnode = next;
+
+      proc_delete_resource (resource);
+    }
+
+    struct list_node_link *mapping_link, *mapping_link_tmp;
+    list_foreach (procgroup->mappings, mapping_link, mapping_link_tmp) {
+      struct proc_mapping* mapping =
+          list_entry (mapping_link, struct proc_mapping, proc_mappings_link);
+
+      pmm_free (mapping->paddr, mapping->size / PAGE_SIZE);
+      free (mapping);
+    }
+
+    pmm_free (procgroup->pd.cr3_paddr, 1);
+
+    free (procgroup->tls.tls_tmpl);
+
+    free (procgroup);
+  }
+}