Rewrite init app in C, introduce MSL (MOP3 System Library)

Makefile

@@ -1,6 +1,7 @@
 platform ?= amd64
 
-include apps.mk
+include make/apps.mk
-include kernel.mk
+include make/kernel.mk
-include dist.mk
+include make/dist.mk
-include docs.mk
+include make/docs.mk
+include make/libc.mk

amd64/link.ld

@@ -31,10 +31,14 @@ SECTIONS {
     *(.data .data.*)
   } :data
 
+  __bss_start = .;
+
   .bss : {
     *(.bss .bss.*)
   } :data
 
+  __bss_end = .;
+
   /DISCARD/ : {
     *(.eh_frame*)
     *(.note .note.*)

aux/devel.sh

@@ -8,6 +8,7 @@ else
   make -B all_kernel
 fi
 
+make -B all_libmsl
 make -B all_apps
 make -B all_dist
 ./aux/limine_iso_amd64.sh

generic/flags.mk

@@ -5,8 +5,9 @@ cflags += -nostdinc \
 					-std=c11 \
 					-pedantic \
 					-Wall \
-					-Wextra \
+					-Wextra
-					-mcmodel=kernel
+
+cflags += -isystem ../include
 
 ldflags += -ffreestanding \
 					 -nostdlib \

include/m/syscall_defs.h

@@ -1,5 +1,5 @@
-#ifndef _KERNEL_SYSCALL_DEFS_H
+#ifndef _M_SYSCALL_DEFS_H
-#define _KERNEL_SYSCALL_DEFS_H
+#define _M_SYSCALL_DEFS_H
 
 #define SYS_PROC_QUIT 1
 #define SYS_PROC_TEST 2
@@ -7,4 +7,4 @@
 #define SR_OK                0
 #define SR_SYSCALL_NOT_FOUND 1
 
-#endif // _KERNEL_SYSCALL_DEFS_H
+#endif // _M_SYSCALL_DEFS_H

init/Makefile

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

init/init.S

@@ -1,12 +0,0 @@
-.global _start
-_start:
-  pushq $123
-  addq $8, %rsp
-  
-  movq $2, %rax
-  syscall
-
-  movq $1, %rax
-  syscall
-
-  jmp _start

init/init.c

@@ -0,0 +1,9 @@
+#include <limits.h>
+#include <m/proc.h>
+
+void app_main (void) {
+  m_proc_test ();
+  m_proc_test ();
+  m_proc_test ();
+  m_proc_test ();
+}

kernel/amd64/apic.c

@@ -235,7 +235,7 @@ void amd64_lapic_tick (uint32_t tick) { amd64_lapic_write (LAPIC_TIMICT, tick);
  * @return amount of ticsk in a given period
  */
 static uint32_t amd64_lapic_calibrate (uint32_t us) {
-  amd64_lapic_write (LAPIC_DCR, 0x03);
+  amd64_lapic_write (LAPIC_DCR, 0x0B);
 
   amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 16));
 
@@ -256,7 +256,7 @@ static uint32_t amd64_lapic_calibrate (uint32_t us) {
  *  Initial tick count
  */
 static void amd64_lapic_start (uint32_t ticks) {
-  amd64_lapic_write (LAPIC_DCR, 0x03);
+  amd64_lapic_write (LAPIC_DCR, 0x0B);
 
   amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 17));
 

kernel/amd64/bootmain.c

@@ -35,7 +35,6 @@ ALIGNED (16) static uint8_t uacpi_memory_buffer[UACPI_MEMORY_BUFFER_MAX];
  */
 void bootmain (void) {
   struct cpu* bsp_cpu = cpu_make ();
-  amd64_thiscpu_set_init ();
 
   amd64_init (bsp_cpu, false);
   syscall_init ();
@@ -52,10 +51,6 @@ void bootmain (void) {
 
   smp_init ();
 
-  /* busy wait for cpus to come online */
-  for (volatile int i = 0; i < INT_MAX; i++)
-    ;
-
   mm_init2 ();
 
   proc_init ();

kernel/amd64/hpet.c

@@ -28,8 +28,8 @@
 static bool hpet_32bits = 1;
 /// Physical address for HPET MMIO
 static uintptr_t hpet_paddr;
-/// HPET nanoseconds for conversion
+/// HPET period in femtoseconds
-static uint64_t hpet_clock_nano;
+static uint64_t hpet_period_fs;
 /// Lock, which protects concurrent access. See \ref amd64/smp.c
 static spin_lock_t hpet_lock = SPIN_LOCK_INIT;
 
@@ -54,32 +54,22 @@ static void amd64_hpet_write (uint32_t reg, uint64_t value) {
 /// Read current value of \ref HPET_MCVR register.
 static uint64_t amd64_hpet_timestamp (void) { return amd64_hpet_read (HPET_MCVR); }
 
-/** 
- * @brief Get current HPET timestamp in nanoseconds
- *
- * @param lock
- *  if true, hold \ref hpet_lock
- */
-uint64_t amd64_hpet_current_nano (bool lock) {
-  if (lock)
-    spin_lock (&hpet_lock);
-
-  uint64_t t = amd64_hpet_timestamp () * hpet_clock_nano;
-
-  if (lock)
-    spin_unlock (&hpet_lock);
-
-  return t;
-}
-
 /// 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) {
   spin_lock (&hpet_lock);
 
   uint64_t start = amd64_hpet_timestamp ();
-  uint64_t conv = us * 1000;
+  uint64_t target_fs = us * 1000000000ULL;
-  while (((amd64_hpet_timestamp () - start) * hpet_clock_nano) < conv)
+  
+  for (;;) {
+    uint64_t current = amd64_hpet_timestamp ();
+    uint64_t dt = current - start;
+
+    if ((dt * hpet_period_fs) >= target_fs)
+      break;
+
     __asm__ volatile ("pause" ::: "memory");
+  }
 
   spin_unlock (&hpet_lock);
 }
@@ -114,7 +104,5 @@ void amd64_hpet_init (void) {
     gcidr = (((uint64_t)high << 32) | low);
   }
 
-  uint64_t period_fs = (gcidr >> 32);
+  hpet_period_fs = (gcidr >> 32);
-
-  hpet_clock_nano = period_fs / 1000000;
 }

kernel/amd64/hpet.h

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

kernel/amd64/init.c

@@ -65,8 +65,6 @@ static void amd64_gdt_init (struct cpu* cpu) {
                     "movw %%ax, %%ds\n"
                     "movw %%ax, %%es\n"
                     "movw %%ax, %%ss\n"
-                    "movw %%ax, %%fs\n"
-                    "movw %%ax, %%gs\n"
                     :
                     : [kcode] "i"(GDT_KCODE), [kdata] "i"(GDT_KDATA)
                     : "rax", "memory");

kernel/amd64/intr.c

@@ -175,8 +175,6 @@ static void amd64_intr_exception (struct saved_regs* regs) {
 void amd64_intr_handler (void* stack_ptr) {
   struct saved_regs* regs = stack_ptr;
 
-  amd64_load_kernel_cr3 ();
-
   if (regs->trap <= 31) {
     amd64_intr_exception (regs);
   } else {
@@ -219,10 +217,6 @@ static void amd64_irq_restore_flags (uint64_t rflags) {
 
 /// Save current interrupt state
 void irq_save (void) {
-  /* before smp init. */
-  if (thiscpu == NULL)
-    return;
-
   int prev = atomic_fetch_add_explicit (&thiscpu->irq_ctx.nesting, 1, memory_order_acq_rel);
   if (prev == 0)
     thiscpu->irq_ctx.rflags = amd64_irq_save_flags ();
@@ -230,10 +224,6 @@ void irq_save (void) {
 
 /// Restore interrupt state
 void irq_restore (void) {
-  /* before smp init. */
-  if (thiscpu == NULL)
-    return;
-
   int prev = atomic_fetch_sub_explicit (&thiscpu->irq_ctx.nesting, 1, memory_order_acq_rel);
   if (prev == 1)
     amd64_irq_restore_flags (thiscpu->irq_ctx.rflags);

kernel/amd64/intr_stub.S

@@ -22,9 +22,6 @@
   ;\
   movq %rsp, %rdi; \
   ;\
-  movq %cr3, %rax; \
-  pushq %rax; \
-  ;\
   movq %rsp, %rbp; \
   ;\
   subq $8, %rsp; \
@@ -34,9 +31,6 @@
   ;\
   movq %rbp, %rsp; \
   ;\
-  popq %rax; \
-  movq %rax, %cr3; \
-  ;\
   pop_regs; \
   addq $16, %rsp; \
   ;\

kernel/amd64/sched1.c

@@ -9,6 +9,6 @@
 void do_sched (struct proc* proc) {
   thiscpu->tss.rsp0 = proc->pdata.kernel_stack;
   thiscpu->syscall_kernel_stack = proc->pdata.kernel_stack;
-  amd64_wrmsr (MSR_GS_BASE, (uint64_t)proc->pdata.gs_base);
+
   amd64_do_sched ((void*)&proc->pdata.regs, (void*)proc->pd.cr3_paddr);
 }

kernel/amd64/smp.c

@@ -13,38 +13,31 @@
 #include <sys/syscall.h>
 
 /// Cpu ID counter
-static uint32_t cpu_counter = 0;
+static atomic_uint cpu_counter = 0;
-/// Lock for \ref cpu_counter
-static spin_lock_t cpu_counter_lock = SPIN_LOCK_INIT;
 /// The CPUs
 static struct cpu cpus[CPUS_MAX];
 
-static bool thiscpu_init = false;
+static atomic_int cpu_init_count;
-
-void amd64_thiscpu_set_init (void) { thiscpu_init = true; }
 
 /// Allocate a CPU structure
 struct cpu* cpu_make (void) {
-  spin_lock (&cpu_counter_lock);
+  int id = atomic_fetch_add (&cpu_counter, 1);
-  int id = cpu_counter++;
-  spin_unlock (&cpu_counter_lock);
 
   struct cpu* cpu = &cpus[id];
 
   memset (cpu, 0, sizeof (*cpu));
   cpu->lock = SPIN_LOCK_INIT;
   cpu->id = id;
+  cpu->self = cpu;
 
-  amd64_wrmsr (MSR_SHADOW_GS_BASE, (uint64_t)cpu);
+  amd64_wrmsr (MSR_GS_BASE, (uint64_t)cpu);
 
   return cpu;
 }
 
 struct cpu* cpu_get (void) {
-  if (!thiscpu_init)
+  struct cpu* ptr = (struct cpu*)amd64_rdmsr (MSR_GS_BASE);
-    return NULL;
+  return ptr;
-
-  return (struct cpu*)amd64_rdmsr (MSR_SHADOW_GS_BASE);
 }
 
 /// Bootstrap code for non-BSP CPUs
@@ -56,27 +49,36 @@ static void amd64_smp_bootstrap (struct limine_mp_info* mp_info) {
   amd64_init (cpu, true); /* gdt + idt */
   syscall_init ();
 
-  thiscpu->lapic_ticks = amd64_lapic_init (2500);
+  thiscpu->lapic_ticks = amd64_lapic_init (10000);
   amd64_lapic_tick (thiscpu->lapic_ticks);
 
   DEBUG ("CPU %u is online!\n", thiscpu->id);
 
   __asm__ volatile ("sti");
 
+  atomic_fetch_sub (&cpu_init_count, 1);
+
   for (;;)
     ;
 }
 
 /// Initialize SMP subsystem for AMD64. Start AP CPUs
 void smp_init (void) {
-  thiscpu->lapic_ticks = amd64_lapic_init (2500);
+  thiscpu->lapic_ticks = amd64_lapic_init (10000);
 
   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->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

@@ -13,6 +13,7 @@ struct cpu {
   /* for syscall instruction */
   uintptr_t syscall_user_stack;
   uintptr_t syscall_kernel_stack;
+  struct cpu* self;
   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);

kernel/amd64/syscall.c

@@ -6,7 +6,7 @@
 #include <proc/proc.h>
 #include <sys/debug.h>
 #include <sys/smp.h>
-#include <syscall/defs.h>
+#include <m/syscall_defs.h>
 #include <syscall/syscall.h>
 
 extern void amd64_syscall_entry (void);
@@ -14,8 +14,6 @@ extern void amd64_syscall_entry (void);
 int amd64_syscall_dispatch (void* stack_ptr) {
   struct saved_regs* regs = stack_ptr;
 
-  amd64_load_kernel_cr3 ();
-
   int syscall_num = regs->rax;
   syscall_handler_func_t func = syscall_find_handler (syscall_num);
 

kernel/amd64/syscallentry.S

@@ -5,7 +5,6 @@
 .global amd64_syscall_entry
 amd64_syscall_entry:
   cli
-  swapgs
 
   movq %rsp, %gs:0
   movq %gs:8, %rsp
@@ -20,15 +19,10 @@ amd64_syscall_entry:
 
   push_regs
 
-  swapgs
-
   cld
 
   movq %rsp, %rdi
 
-  movq %cr3, %rax
-  pushq %rax
-
   movq %rsp, %rbp
 
   subq $8, %rsp
@@ -38,20 +32,9 @@ amd64_syscall_entry:
 
   movq %rbp, %rsp
 
-  popq %rax
-  movq %rax, %cr3
-
   pop_regs
   
-  swapgs
+  addq $56, %rsp
-
-  addq $16, %rsp
-  popq %rcx
-  addq $8, %rsp
-  popq %r11
-  addq $16, %rsp
   movq %gs:0, %rsp
   
-  swapgs
-
   sysretq

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/sys/debug.h

@@ -7,11 +7,7 @@ void debugprintf (const char* fmt, ...);
 
 #define DEBUG(fmt, ...)                                                                            \
   do {                                                                                             \
-    if (thiscpu != NULL) {                                                                         \
     debugprintf ("(CPU %u) %s:%d: " fmt, thiscpu->id, __func__, __LINE__, ##__VA_ARGS__);          \
-    } else {                                                                                       \
-      debugprintf ("%s:%d: " fmt, __func__, __LINE__, ##__VA_ARGS__);                              \
-    }                                                                                              \
   } while (0)
 
 #endif // _KERNEL_SYS_DEBUG_H

kernel/syscall/syscall.c

@@ -2,7 +2,7 @@
 #include <libk/std.h>
 #include <proc/proc.h>
 #include <sys/debug.h>
-#include <syscall/defs.h>
+#include <m/syscall_defs.h>
 #include <syscall/syscall.h>
 
 #define DEFINE_SYSCALL(name)                                                                       \

libmsl/Makefile

@@ -0,0 +1,28 @@
+cc := clang
+o :=
+c :=
+cflags  := -isystem .
+buildtype ?= release
+
+include src.mk
+include ../generic/flags.mk
+include ../$(platform)/flags.mk
+
+all: build/libmsl.a
+
+build/libmsl.a: $(o)
+	llvm-ar rcs $@ $^
+
+%.o: %.c
+	$(cc) -c -o $@ $(cflags) $<
+
+%.o: %.S
+	$(cc) -c -o $@ $(cflags) $<
+
+clean:
+	rm -f $(o) build/libmsl.a
+
+format:
+	clang-format -i $$(git ls-files '*.c' '*.h')
+
+.PHONY: all clean format

libmsl/amd64/.gitignore

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

libmsl/amd64/_start.S

@@ -0,0 +1,8 @@
+.global _start
+_start:
+  xorq %rbp, %rbp
+  movq %rsp, %rbp
+  andq $-16, %rsp
+  subq $8, %rsp
+
+  callq __premain

libmsl/amd64/src.mk

@@ -0,0 +1,6 @@
+c += amd64/syscall.c
+
+S += amd64/_start.S
+
+o += amd64/_start.o \
+		 amd64/syscall.o

libmsl/amd64/syscall.c

@@ -0,0 +1,23 @@
+#include <stddef.h>
+#include <stdint.h>
+#include <amd64/syscall.h>
+
+int msl_amd64_syscall (int syscall_num, uintptr_t a1, uintptr_t a2, uintptr_t a3, uintptr_t a4, uintptr_t a5, uintptr_t a6) {
+  uint64_t result;
+  __asm__ volatile (
+    "movq %1, %%rax\n"
+    "movq %2, %%rdi\n"
+    "movq %3, %%rsi\n"
+    "movq %4, %%rdx\n"
+    "movq %5, %%r10\n"
+    "movq %6, %%r8\n"
+    "movq %7, %%r9\n"
+    "syscall\n"
+    "movq %%rax, %0\n"
+    : "=r"(result)
+    : "r"((uint64_t)syscall_num), "r"(a1), "r"(a2), "r"(a3), "r"(a4), "r"(a5), "r"(a6)
+    : "memory", "cc", "rcx", "r11"
+  );
+  return (int)result;
+}
+

libmsl/amd64/syscall.h

@@ -0,0 +1,8 @@
+#ifndef _LIBMSL_AMD64_SYSCALL_H
+#define _LIBMSL_AMD64_SYSCALL_H
+
+#include <stdint.h>
+
+int msl_amd64_syscall (int syscall_num, uintptr_t a1, uintptr_t a2, uintptr_t a3, uintptr_t a4, uintptr_t a5, uintptr_t a6);
+
+#endif // _LIBMSL_AMD64_SYSCALL_H

libmsl/init/.gitignore

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

libmsl/init/__premain.c

@@ -0,0 +1,22 @@
+#include <stdint.h>
+#include <m/proc.h>
+
+extern volatile uint8_t __bss_start[];
+extern volatile uint8_t __bss_end[];
+
+extern void app_main (void);
+
+static void msl_clear_bss (void) {
+  uint8_t *p = (uint8_t*)__bss_start;
+  while (p < __bss_end) {
+    *p++ = 0;
+  }
+}
+
+void __premain (void) {
+  msl_clear_bss ();
+
+  app_main ();
+
+  m_proc_quit ();
+}

libmsl/init/src.mk

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

libmsl/m/.gitignore

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

libmsl/m/proc.c

@@ -0,0 +1,10 @@
+#include <m/syscall.h>
+#include <m/syscall_defs.h>
+
+int m_proc_quit (void) {
+  return m_syscall (SYS_PROC_QUIT, 0, 0, 0, 0, 0, 0);
+}
+
+int m_proc_test (void) {
+  return m_syscall (SYS_PROC_TEST, 0, 0, 0, 0, 0, 0);
+}

libmsl/m/proc.h

@@ -0,0 +1,8 @@
+#ifndef _LIBMSL_M_PROC_H
+#define _LIBMSL_M_PROC_H
+
+int m_proc_quit (void);
+
+int m_proc_test (void);
+
+#endif // _LIBMSL_M_PROC_H

libmsl/m/src.mk

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

libmsl/m/syscall.h

@@ -0,0 +1,11 @@
+#ifndef _LIBMSL_M_SYSCALL_H
+#define _LIBMSL_M_SYSCALL_H
+
+#include <m/syscall_defs.h>
+
+#if defined(__x86_64__)
+  #include <amd64/syscall.h>
+  #define m_syscall msl_amd64_syscall
+#endif
+
+#endif // _LIBMSL_M_SYSCALL_H

libmsl/src.mk

@@ -0,0 +1,3 @@
+include $(platform)/src.mk
+include init/src.mk
+include m/src.mk

make/libc.mk

@@ -0,0 +1,10 @@
+all_libmsl:
+	make -C libmsl platform=$(platform) all
+
+clean_libmsl:
+	make -C libmsl platform=$(platform) clean
+
+format_libmsl:
+	make -C libmsl platform=$(platform) format
+
+.PHONY: all_libmsl clean_libmsl

make/user.mk

@@ -1,8 +1,8 @@
 cc := clang
 o :=
 c :=
-ldflags :=
+ldflags := -L ../libmsl/build -l:libmsl.a
-cflags :=
+cflags := -isystem ../libmsl
 
 include src.mk
 include app.mk