Rename make/libc.mk -> make/libmsl.mk

Makefile

@@ -4,4 +4,4 @@ include make/apps.mk
 include make/kernel.mk
 include make/dist.mk
 include make/docs.mk
-include make/libc.mk
+include make/libmsl.mk

amd64/link.ld

@@ -6,6 +6,8 @@ PHDRS {
   text PT_LOAD;
   rodata PT_LOAD;
   data PT_LOAD;
+  bss PT_LOAD;
+  tls PT_TLS;
 }
 
 SECTIONS {
@@ -13,6 +15,7 @@ SECTIONS {
 
   .text : {
     *(.text .text.*)
+    *(.ltext .ltext.*)
   } :text
   
   . = ALIGN(CONSTANT(MAXPAGESIZE));
@@ -21,23 +24,43 @@ SECTIONS {
     *(.rodata .rodata.*)
   } :rodata
 
-  .note.gnu.build-id : {
-    *(.note.gnu.build-id)
-  } :rodata
-  
   . = ALIGN(CONSTANT(MAXPAGESIZE));
 
   .data : {
     *(.data .data.*)
+    *(.ldata .ldata.*)
   } :data
+   
+  . = ALIGN(CONSTANT(MAXPAGESIZE));
 
   __bss_start = .;
 
   .bss : {
     *(.bss .bss.*)
-  } :data
+    *(.lbss .lbss.*)
+  } :bss
 
   __bss_end = .;
+  
+  . = ALIGN(CONSTANT(MAXPAGESIZE));
+
+  __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*)

init/init.c

@@ -8,6 +8,8 @@
 
 #define MUTEX 2000
 
+__thread char letter = 'c';
+
 void app_thread1 (void);
 
 int spawn (void (*fn) (void)) {
@@ -23,24 +25,28 @@ int spawn (void (*fn) (void)) {
 void app_main (void) {
   mutex_create (MUTEX);
 
+  letter = 'a';
+
   spawn (&app_thread1);
 
   for (;;) {
     mutex_lock (MUTEX);
 
     for (int i = 0; i < 3; i++)
-      test ('a');
+      test (letter);
 
     mutex_unlock (MUTEX);
   }
 }
 
 void app_thread1 (void) {
+  letter = 'b';
+
   for (;;) {
     mutex_lock (MUTEX);
 
     for (int i = 0; i < 3; i++)
-      test ('b');
+      test (letter);
 
     mutex_unlock (MUTEX);
   }

kernel/amd64/apic.c

@@ -4,7 +4,7 @@
 #include <amd64/msr.h>
 #include <libk/std.h>
 #include <limine/requests.h>
-#include <sync/rw_spin_lock.h>
+#include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
 #include <sys/spin.h>
@@ -38,7 +38,7 @@
 
 struct ioapic {
   struct acpi_madt_ioapic table_data;
-  rw_spin_lock_t lock;
+  spin_lock_t lock;
   uintptr_t mmio_base;
 };
 
@@ -59,10 +59,10 @@ static spin_lock_t lapic_calibration_lock = SPIN_LOCK_INIT;
 static uint32_t amd64_ioapic_read (struct ioapic* ioapic, uint32_t reg) {
   spin_lock_ctx_t ctxioar;
 
-  rw_spin_read_lock (&ioapic->lock, &ctxioar);
+  spin_lock (&ioapic->lock, &ctxioar);
   *(volatile uint32_t*)ioapic->mmio_base = reg;
   uint32_t ret = *(volatile uint32_t*)(ioapic->mmio_base + 0x10);
-  rw_spin_read_unlock (&ioapic->lock, &ctxioar);
+  spin_unlock (&ioapic->lock, &ctxioar);
   return ret;
 }
 
@@ -70,10 +70,10 @@ static uint32_t amd64_ioapic_read (struct ioapic* ioapic, uint32_t reg) {
 static void amd64_ioapic_write (struct ioapic* ioapic, uint32_t reg, uint32_t value) {
   spin_lock_ctx_t ctxioaw;
 
-  rw_spin_write_lock (&ioapic->lock, &ctxioaw);
+  spin_lock (&ioapic->lock, &ctxioaw);
   *(volatile uint32_t*)ioapic->mmio_base = reg;
   *(volatile uint32_t*)(ioapic->mmio_base + 0x10) = value;
-  rw_spin_write_unlock (&ioapic->lock, &ctxioaw);
+  spin_unlock (&ioapic->lock, &ctxioaw);
 }
 
 /* Find an IOAPIC corresposting to provided IRQ */
@@ -160,9 +160,9 @@ void amd64_ioapic_init (void) {
       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 | MM_PD_RELOAD);
+                          MM_PG_PRESENT | MM_PG_RW);
       ioapics[ioapic_entries++] = (struct ioapic){
-          .lock = RW_SPIN_LOCK_INIT,
+          .lock = SPIN_LOCK_INIT,
           .table_data = *ioapic_table_data,
           .mmio_base = ((uintptr_t)hhdm->offset + (uintptr_t)ioapic_table_data->address),
       };
@@ -246,8 +246,7 @@ void amd64_lapic_init (uint32_t us) {
   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_map_kernel_page (lapic_paddr, thiscpu->lapic_mmio_base, MM_PG_PRESENT | MM_PG_RW);
-                      MM_PG_PRESENT | MM_PG_RW | MM_PD_LOCK | MM_PD_RELOAD);
 
   amd64_lapic_write (LAPIC_SIVR, 0xFF | (1 << 8));
 

kernel/amd64/bootmain.c

@@ -47,8 +47,6 @@ void bootmain (void) {
   amd64_ioapic_init ();
   amd64_hpet_init ();
 
-  mm_init2 ();
-
   smp_init ();
 
   proc_init ();

kernel/amd64/hpet.c

@@ -129,8 +129,7 @@ void amd64_hpet_init (void) {
   hpet_paddr = (uintptr_t)hpet->address.address;
 
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  mm_map_kernel_page (hpet_paddr, (uintptr_t)hhdm->offset + hpet_paddr,
+  mm_map_kernel_page (hpet_paddr, (uintptr_t)hhdm->offset + hpet_paddr, MM_PG_PRESENT | MM_PG_RW);
-                      MM_PG_PRESENT | MM_PG_RW | MM_PD_RELOAD);
 
   uint64_t caps = amd64_hpet_read64 (HPET_GCIDR);
   hpet_32bits = (caps & (1 << 13)) ? 0 : 1;

kernel/amd64/mm.c

@@ -22,10 +22,12 @@ struct pg_index {
 } PACKED;
 
 /* Kernel page directory */
-static struct pd kernel_pd = {.lock = SPIN_LOCK_INIT};
+static struct pd kernel_pd;
-static spin_lock_ctx_t ctxkpd;
+static spin_lock_t kernel_pd_lock;
-/* Lock needed to sync between map/unmap operations and TLB shootdown */
+
-static spin_lock_t mm_lock = SPIN_LOCK_INIT;
+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) {
@@ -112,15 +114,7 @@ static void amd64_reload_cr3 (void) {
 
 /* 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) {
-  spin_lock_ctx_t ctxmm, ctxpd;
-
-  spin_lock (&mm_lock, &ctxmm);
-
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  bool do_reload = false;
-
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
 
   uint64_t amd64_flags = amd64_mm_resolve_flags (flags);
 
@@ -129,69 +123,50 @@ 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 && (flags & MM_PD_RELOAD))
-    amd64_reload_cr3 ();
-
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
-  spin_unlock (&mm_lock, &ctxmm);
 }
 
 /* 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 ();
 }
 
 /* Unmap a virtual address. TLB needs to be flushed afterwards */
-void mm_unmap_page (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
+void mm_unmap_page (struct pd* pd, uintptr_t vaddr) {
-  spin_lock_ctx_t ctxmm, ctxpd;
-
-  spin_lock (&mm_lock, &ctxmm);
-
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  bool do_reload = false;
-
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
 
   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;
+    return;
 
   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, false);
   if (pml1 == NULL)
-    goto done;
+    return;
 
   uint64_t* pte = &pml1[pg_index.pml1];
 
-  if ((*pte) & AMD64_PG_PRESENT) {
+  if ((*pte) & AMD64_PG_PRESENT)
     *pte = 0;
-    do_reload = true;
-  }
 
   if (amd64_mm_is_table_empty (pml1)) {
     uintptr_t pml1_phys = pml2[pg_index.pml2] & ~0xFFFULL;
@@ -210,28 +185,14 @@ void mm_unmap_page (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
       }
     }
   }
-
-done:
-  if (do_reload && (flags & MM_PD_RELOAD))
-    amd64_reload_cr3 ();
-
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
-  spin_unlock (&mm_lock, &ctxmm);
 }
 
 /* Unmap a page from kernel page directory */
-void mm_unmap_kernel_page (uintptr_t vaddr, uint32_t flags) {
+void mm_unmap_kernel_page (uintptr_t vaddr) {
-  mm_unmap_page (&kernel_pd, vaddr, flags);
+  mm_unmap_page (&kernel_pd, vaddr);
+  amd64_reload_cr3 ();
 }
 
-/* Lock kernel page directory */
-void mm_lock_kernel (void) { spin_lock (&kernel_pd.lock, &ctxkpd); }
-
-/* Unlock kernel page directory */
-void mm_unlock_kernel (void) { spin_unlock (&kernel_pd.lock, &ctxkpd); }
-
 /* Allocate a userspace-ready page directory */
 uintptr_t mm_alloc_user_pd_phys (void) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
@@ -250,26 +211,10 @@ uintptr_t mm_alloc_user_pd_phys (void) {
   return cr3;
 }
 
-/* Reload after map/unmap operation was performed. This function does the TLB shootdown. */
+bool mm_validate (struct pd* pd, uintptr_t vaddr) {
-void mm_reload (void) {
-  struct limine_mp_response* mp = limine_mp_request.response;
-
-  for (size_t i = 0; i < mp->cpu_count; i++) {
-    amd64_lapic_ipi (mp->cpus[i]->lapic_id, TLB_SHOOTDOWN);
-  }
-}
-
-bool mm_validate (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
-  spin_lock_ctx_t ctxmm, ctxpd;
-
-  spin_lock (&mm_lock, &ctxmm);
-
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   bool ret = false;
 
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
-
   uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
   struct pg_index pg_index = amd64_mm_page_index (vaddr);
 
@@ -289,45 +234,26 @@ bool mm_validate (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
   ret = (pte & AMD64_PG_PRESENT) != 0;
 
 done:
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
-  spin_unlock (&mm_lock, &ctxmm);
-
   return ret;
 }
 
-bool mm_validate_buffer (struct pd* pd, uintptr_t vaddr, size_t size, uint32_t flags) {
+bool mm_validate_buffer (struct pd* pd, uintptr_t vaddr, size_t size) {
   bool ok = true;
-  spin_lock_ctx_t ctxpd;
-
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
 
   for (size_t i = 0; i < size; i++) {
-    ok = mm_validate (pd, vaddr + i, 0);
+    ok = mm_validate (pd, vaddr + i);
     if (!ok)
       goto done;
   }
 
 done:
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
   return ok;
 }
 
-uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr, uint32_t flags) {
+uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr) {
-  spin_lock_ctx_t ctxmm, ctxpd;
-
-  spin_lock (&mm_lock, &ctxmm);
-
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   uintptr_t ret = 0;
 
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
-
   uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
 
   for (size_t i4 = 0; i4 < 512; i4++) {
@@ -358,25 +284,13 @@ uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr, uint32_t flags) {
   }
 
 done:
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
-  spin_unlock (&mm_lock, &ctxmm);
-
   return ret;
 }
 
-uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
+uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr) {
-  spin_lock_ctx_t ctxmm, ctxpd;
-
-  spin_lock (&mm_lock, &ctxmm);
-
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
   uintptr_t ret = 0;
 
-  if (flags & MM_PD_LOCK)
-    spin_lock (&pd->lock, &ctxpd);
-
   uint64_t* pml4 = (uint64_t*)(pd->cr3_paddr + (uintptr_t)hhdm->offset);
   struct pg_index pg_index = amd64_mm_page_index (vaddr);
 
@@ -400,25 +314,8 @@ uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr, uint32_t flags) {
   ret = ((pte & ~0xFFFULL) | (vaddr & 0xFFFULL));
 
 done:
-  if (flags & MM_PD_LOCK)
-    spin_unlock (&pd->lock, &ctxpd);
-
-  spin_unlock (&mm_lock, &ctxmm);
-
   return ret;
 }
 
-/* TLB shootdown IRQ handler */
-static void amd64_tlb_shootdown_irq (void* arg, void* regs) {
-  (void)arg, (void)regs;
-
-  amd64_reload_cr3 ();
-  DEBUG ("cpu %u TLB shootdown\n", thiscpu->id);
-}
-
-/* Continue initializing memory management subsystem for AMD64 after the essential parts were
- * initialized */
-void mm_init2 (void) { irq_attach (&amd64_tlb_shootdown_irq, NULL, TLB_SHOOTDOWN); }
-
 /* Initialize essentials for the AMD64 memory management subsystem */
 void mm_init (void) { kernel_pd.cr3_paddr = amd64_current_cr3 (); }

kernel/amd64/mm.h

@@ -7,11 +7,9 @@
 #define PAGE_SIZE 4096
 
 struct pd {
-  spin_lock_t lock;
   uintptr_t cr3_paddr;
 };
 
 void amd64_load_kernel_cr3 (void);
-void mm_init2 (void);
 
 #endif // _KERNEL_AMD64_MM_H

kernel/amd64/proc.c

@@ -1,6 +1,7 @@
 #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>
@@ -11,12 +12,11 @@
 #include <proc/proc.h>
 #include <proc/procgroup.h>
 #include <proc/resource.h>
-#include <sync/rw_spin_lock.h>
 #include <sync/spin_lock.h>
 #include <sys/debug.h>
+#include <sys/proc.h>
 
-/* 0 is kpproc */
+static atomic_int pids = 0;
-static atomic_int pids = 1;
 
 struct proc* proc_from_elf (uint8_t* elf_contents) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
@@ -87,6 +87,8 @@ struct proc* proc_clone (struct proc* proto, uintptr_t vstack_top, uintptr_t ent
   proc->pdata.regs.cs = GDT_UCODE | 0x03;
   proc->pdata.regs.rip = (uint64_t)entry;
 
+  proc_init_tls (proc);
+
   return proc;
 }
 
@@ -116,10 +118,36 @@ void proc_cleanup (struct proc* proc) {
 
   spin_unlock (&proc->lock, &ctxpr);
 
-  procgroup_detach (proc->procgroup, 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;
+
+  size_t tls_size = proc->procgroup->tls.tls_tmpl_size;
+
+  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, proc->procgroup->tls.tls_tmpl_pages, flags, &tls_paddr);
+
+  uintptr_t k_tls_addr = (uintptr_t)hhdm->offset + tls_paddr;
+
+  uintptr_t ktcb = k_tls_addr + tls_size;
+  uintptr_t utcb = tls_vaddr + tls_size;
+
+  memset ((void*)k_tls_addr, 0, tls_size);
+  memcpy ((void*)k_tls_addr, (void*)proc->procgroup->tls.tls_tmpl, tls_size);
+
+  *(uintptr_t*)ktcb = utcb;
+
+  proc->pdata.fs_base = utcb;
+  proc->pdata.tls_vaddr = tls_vaddr;
+}

kernel/amd64/proc.h

@@ -15,7 +15,8 @@
 struct proc_platformdata {
   struct saved_regs regs;
   uintptr_t kernel_stack;
-  uint64_t gs_base;
+  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/sched1.c

@@ -14,6 +14,7 @@ void do_sched (struct proc* proc, spin_lock_t* cpu_lock, spin_lock_ctx_t* ctxcpu
 
   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);

kernel/irq/irq.c

@@ -2,7 +2,7 @@
 #include <libk/list.h>
 #include <libk/std.h>
 #include <mm/liballoc.h>
-#include <sync/rw_spin_lock.h>
+#include <sync/spin_lock.h>
 #include <sys/debug.h>
 
 #if defined(__x86_64__)
@@ -12,7 +12,7 @@
 
 struct irq* irq_table[0x100];
 
-static rw_spin_lock_t irqs_lock;
+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;
@@ -26,9 +26,9 @@ bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num) {
   irq->arg = arg;
   irq->irq_num = irq_num;
 
-  rw_spin_write_lock (&irqs_lock, &ctxiqa);
+  spin_lock (&irqs_lock, &ctxiqa);
   irq_table[irq_num] = irq;
-  rw_spin_write_unlock (&irqs_lock, &ctxiqa);
+  spin_unlock (&irqs_lock, &ctxiqa);
 
   return true;
 }
@@ -36,11 +36,11 @@ bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num) {
 struct irq* irq_find (uint32_t irq_num) {
   spin_lock_ctx_t ctxiqa;
 
-  rw_spin_read_lock (&irqs_lock, &ctxiqa);
+  spin_lock (&irqs_lock, &ctxiqa);
 
   struct irq* irq = irq_table[irq_num];
 
-  rw_spin_read_unlock (&irqs_lock, &ctxiqa);
+  spin_unlock (&irqs_lock, &ctxiqa);
 
   return irq;
 }

kernel/proc/proc.c

@@ -13,7 +13,6 @@
 #include <proc/procgroup.h>
 #include <proc/resource.h>
 #include <rd/rd.h>
-#include <sync/rw_spin_lock.h>
 #include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
@@ -29,7 +28,7 @@
 #define SCHED_REAP_FREQ 10
 
 static struct rb_node_link* proc_tree = NULL;
-static rw_spin_lock_t proc_tree_lock = RW_SPIN_LOCK_INIT;
+static spin_lock_t proc_tree_lock = SPIN_LOCK_INIT;
 
 static atomic_int sched_cycles = 0;
 
@@ -74,6 +73,25 @@ struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf) {
       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__)
+      size_t tls_align = phdr->p_align ? phdr->p_align : sizeof (uintptr_t);
+      size_t tls_size = phdr->p_memsz;
+      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;
     }
   }
 
@@ -84,7 +102,6 @@ struct proc* proc_spawn_rd (char* name) {
   struct rd_file* rd_file = rd_get_file (name);
 
   bool ok = proc_check_elf (rd_file->content);
-  DEBUG ("ELF magic %s\n", (ok ? "OK" : "BAD"));
 
   if (!ok)
     return NULL;
@@ -96,9 +113,9 @@ struct proc* proc_find_pid (int pid) {
   spin_lock_ctx_t ctxprtr;
   struct proc* proc = NULL;
 
-  rw_spin_read_lock (&proc_tree_lock, &ctxprtr);
+  spin_lock (&proc_tree_lock, &ctxprtr);
   rbtree_find (struct proc, &proc_tree, pid, proc, proc_tree_link, pid);
-  rw_spin_read_unlock (&proc_tree_lock, &ctxprtr);
+  spin_unlock (&proc_tree_lock, &ctxprtr);
 
   return proc;
 }
@@ -107,21 +124,20 @@ void proc_register (struct proc* proc, struct cpu* cpu1) {
   spin_lock_ctx_t ctxcpu, ctxprtr;
 
   proc->cpu = cpu1 != NULL ? cpu1 : cpu_find_lightest ();
-  DEBUG ("Assigning CPU %d to PID %d\n", proc->cpu->id, proc->pid);
 
   struct cpu* cpu = proc->cpu;
 
-  rw_spin_write_lock (&proc_tree_lock, &ctxprtr);
+  spin_lock (&proc_tree_lock, &ctxprtr);
-  rbtree_insert (struct proc, &proc_tree, &proc->proc_tree_link, proc_tree_link, pid);
-  rw_spin_write_unlock (&proc_tree_lock, &ctxprtr);
-
   spin_lock (&cpu->lock, &ctxcpu);
-  list_append (cpu->proc_run_q, &proc->cpu_run_q_link);
-  atomic_fetch_add (&cpu->proc_run_q_count, 1);
 
+  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);
 }
 
@@ -160,7 +176,7 @@ static void proc_reap (void) {
   spin_lock_ctx_t ctxprtr;
   spin_lock_ctx_t ctxpr;
 
-  rw_spin_write_lock (&proc_tree_lock, &ctxprtr);
+  spin_lock (&proc_tree_lock, &ctxprtr);
 
   struct rb_node_link* node;
   rbtree_first (&proc_tree, node);
@@ -180,7 +196,7 @@ static void proc_reap (void) {
     node = next;
   }
 
-  rw_spin_write_unlock (&proc_tree_lock, &ctxprtr);
+  spin_unlock (&proc_tree_lock, &ctxprtr);
 
   struct list_node_link *reap_link, *reap_link_tmp;
   list_foreach (reap_list, reap_link, reap_link_tmp) {

kernel/proc/proc.h

@@ -9,7 +9,6 @@
 #include <proc/procgroup.h>
 #include <proc/resource.h>
 #include <proc/suspension_q.h>
-#include <sync/rw_spin_lock.h>
 #include <sync/spin_lock.h>
 #include <sys/mm.h>
 

kernel/proc/procgroup.c

@@ -4,30 +4,34 @@
 #include <mm/pmm.h>
 #include <proc/proc.h>
 #include <proc/procgroup.h>
-#include <sync/rw_spin_lock.h>
 #include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
 
 static struct rb_node_link* procgroup_tree = NULL;
-static rw_spin_lock_t procgroup_tree_lock = RW_SPIN_LOCK_INIT;
+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 ctxprpd;
+  spin_lock_ctx_t ctxpg;
 
-  vaddr = (vaddr == 0) ? PROC_MAP_BASE : vaddr;
+  spin_lock (&procgroup->lock, &ctxpg);
+
+  vaddr = (vaddr == 0) ? procgroup->map_base : vaddr;
 
   struct proc_mapping* mapping = malloc (sizeof (*mapping));
 
-  if (mapping == NULL)
+  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;
   }
 
@@ -38,9 +42,7 @@ uintptr_t procgroup_map (struct procgroup* procgroup, uintptr_t vaddr, size_t pa
   mapping->vaddr = vaddr;
   mapping->size = pages * PAGE_SIZE;
 
-  flags &= ~(MM_PD_LOCK | MM_PD_RELOAD); /* clear LOCK flag if present, because we lock manualy */
+  procgroup->map_base += pages * PAGE_SIZE;
-
-  spin_lock (&procgroup->pd.lock, &ctxprpd);
 
   list_append (procgroup->mappings, &mapping->proc_mappings_link);
 
@@ -49,7 +51,7 @@ uintptr_t procgroup_map (struct procgroup* procgroup, uintptr_t vaddr, size_t pa
     mm_map_page (&procgroup->pd, ppage, vpage, flags);
   }
 
-  spin_unlock (&procgroup->pd.lock, &ctxprpd);
+  spin_unlock (&procgroup->lock, &ctxpg);
 
   return vaddr;
 }
@@ -57,15 +59,17 @@ uintptr_t procgroup_map (struct procgroup* procgroup, uintptr_t vaddr, size_t pa
 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 ctxprpd;
+  spin_lock_ctx_t ctxpg;
 
   struct proc_mapping* tail_mapping = malloc (sizeof (*tail_mapping));
   if (tail_mapping == NULL)
     return false;
 
-  spin_lock (&procgroup->pd.lock, &ctxprpd);
+  spin_lock (&procgroup->lock, &ctxpg);
 
   list_foreach (procgroup->mappings, mapping_link, mapping_link_tmp) {
     struct proc_mapping* mapping =
@@ -115,10 +119,10 @@ bool procgroup_unmap (struct procgroup* procgroup, uintptr_t start_vaddr, size_t
     free (tail_mapping);
 
   for (uintptr_t vpage = start_vaddr; vpage < end_vaddr; vpage += PAGE_SIZE) {
-    mm_unmap_page (&procgroup->pd, vpage, 0);
+    mm_unmap_page (&procgroup->pd, vpage);
   }
 
-  spin_unlock (&procgroup->pd.lock, &ctxprpd);
+  spin_unlock (&procgroup->lock, &ctxpg);
 
   return true;
 }
@@ -135,13 +139,13 @@ struct procgroup* procgroup_create (void) {
   procgroup->memb_proc_tree = NULL;
   procgroup->lock = SPIN_LOCK_INIT;
   procgroup->pgid = atomic_fetch_add (&pgids, 1);
-  procgroup->pd.lock = SPIN_LOCK_INIT;
   procgroup->pd.cr3_paddr = mm_alloc_user_pd_phys ();
+  procgroup->map_base = PROC_MAP_BASE;
 
-  rw_spin_write_lock (&procgroup_tree_lock, &ctxpgtr);
+  spin_lock (&procgroup_tree_lock, &ctxpgtr);
   rbtree_insert (struct procgroup, &procgroup_tree, &procgroup->procgroup_tree_link,
                  procgroup_tree_link, pgid);
-  rw_spin_write_unlock (&procgroup_tree_lock, &ctxpgtr);
+  spin_unlock (&procgroup_tree_lock, &ctxpgtr);
 
   return procgroup;
 }
@@ -155,7 +159,6 @@ void procgroup_attach (struct procgroup* procgroup, struct proc* proc) {
   rbtree_insert (struct proc, &procgroup->memb_proc_tree, &proc->procgroup_memb_tree_link,
                  procgroup_memb_tree_link, pid);
   atomic_fetch_add (&procgroup->refs, 1);
-  DEBUG ("procgrpup attach PID %d to PGID %d\n", proc->pid, procgroup->pgid);
 
   spin_unlock (&proc->lock, &ctxpr);
   spin_unlock (&procgroup->lock, &ctxpg);
@@ -169,19 +172,18 @@ void procgroup_detach (struct procgroup* procgroup, struct proc* proc) {
 
   rbtree_delete (&procgroup->memb_proc_tree, &proc->procgroup_memb_tree_link);
   int refs = atomic_fetch_sub (&procgroup->refs, 1);
-  DEBUG ("procgrpup detach PID %d to PGID %d\n", proc->pid, procgroup->pgid);
 
   spin_unlock (&proc->lock, &ctxpr);
   spin_unlock (&procgroup->lock, &ctxpg);
 
   if (refs == 1) {
-    rw_spin_write_lock (&procgroup_tree_lock, &ctxpgtr);
+    spin_lock (&procgroup_tree_lock, &ctxpgtr);
     spin_lock (&procgroup->lock, &ctxpg);
 
     rbtree_delete (&procgroup_tree, &procgroup->procgroup_tree_link);
 
     spin_unlock (&procgroup->lock, &ctxpg);
-    rw_spin_write_unlock (&procgroup_tree_lock, &ctxpgtr);
+    spin_unlock (&procgroup_tree_lock, &ctxpgtr);
 
     /* delete resources */
     struct rb_node_link* rnode;
@@ -209,6 +211,8 @@ void procgroup_detach (struct procgroup* procgroup, struct proc* proc) {
 
     pmm_free (procgroup->pd.cr3_paddr, 1);
 
+    free (procgroup->tls.tls_tmpl);
+
     free (procgroup);
   }
 }

kernel/proc/procgroup.h

@@ -7,6 +7,7 @@
 #include <proc/resource.h>
 #include <sync/spin_lock.h>
 #include <sys/mm.h>
+#include <sys/procgroup.h>
 
 struct proc;
 
@@ -27,7 +28,9 @@ struct procgroup {
   struct rb_node_link* resource_tree;
   atomic_int sys_rids;
   struct pd pd;
-  struct list_node_link* mappings; /* protected by pd.lock */
+  struct list_node_link* mappings;
+  uintptr_t map_base;
+  struct procgroup_tls tls;
 };
 
 struct procgroup* procgroup_create (void);

kernel/sync/rw_spin_lock.c

@@ -1,67 +0,0 @@
-#include <libk/assert.h>
-#include <libk/std.h>
-#include <sync/rw_spin_lock.h>
-#include <sys/debug.h>
-#include <sys/irq.h>
-#include <sys/spin_lock.h>
-
-#define WRITER_WAIT (1U << 31)
-#define READER_MASK (~WRITER_WAIT)
-
-void rw_spin_read_lock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx) {
-  uint32_t value;
-
-  irq_save (ctx);
-
-  for (;;) {
-    value = atomic_load_explicit (rw, memory_order_relaxed);
-
-    if ((value & WRITER_WAIT) == 0) {
-      if (atomic_compare_exchange_weak_explicit (rw, &value, value + 1, memory_order_acquire,
-                                                 memory_order_relaxed)) {
-        return;
-      }
-    }
-
-    spin_lock_relax ();
-  }
-}
-
-void rw_spin_read_unlock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx) {
-  uint32_t old = atomic_fetch_sub_explicit (rw, 1, memory_order_release);
-  assert ((old & READER_MASK) > 0);
-  irq_restore (ctx);
-}
-
-void rw_spin_write_lock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx) {
-  uint32_t value;
-
-  irq_save (ctx);
-
-  /* announce writer */
-  for (;;) {
-    value = atomic_load_explicit (rw, memory_order_relaxed);
-
-    if ((value & WRITER_WAIT) == 0) {
-      if (atomic_compare_exchange_weak_explicit (rw, &value, (value | WRITER_WAIT),
-                                                 memory_order_acquire, memory_order_relaxed))
-        break;
-    } else {
-      spin_lock_relax ();
-    }
-  }
-
-  /* wait for readers */
-  for (;;) {
-    value = atomic_load_explicit (rw, memory_order_acquire);
-    if ((value & READER_MASK) == 0)
-      return;
-
-    spin_lock_relax ();
-  }
-}
-
-void rw_spin_write_unlock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx) {
-  atomic_store_explicit (rw, 0, memory_order_release);
-  irq_restore (ctx);
-}

kernel/sync/rw_spin_lock.h

@@ -1,17 +0,0 @@
-#ifndef _KERNEL_SYNC_RW_SPIN_LOCK_H
-#define _KERNEL_SYNC_RW_SPIN_LOCK_H
-
-#include <libk/std.h>
-#include <sync/spin_lock.h>
-#include <sys/spin_lock.h>
-
-#define RW_SPIN_LOCK_INIT 0
-
-typedef _Atomic (uint32_t) rw_spin_lock_t;
-
-void rw_spin_read_lock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx);
-void rw_spin_read_unlock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx);
-void rw_spin_write_lock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx);
-void rw_spin_write_unlock (rw_spin_lock_t* rw, spin_lock_ctx_t* ctx);
-
-#endif // _KERNEL_SYNC_RW_SPIN_LOCK_H

kernel/sync/src.mk

@@ -1,5 +1,3 @@
-c += sync/spin_lock.c \
+c += sync/spin_lock.c
-		 sync/rw_spin_lock.c
 
-o += sync/spin_lock.o \
+o += sync/spin_lock.o
-		 sync/rw_spin_lock.o

kernel/sys/mm.h

@@ -2,6 +2,7 @@
 #define _KERNEL_SYS_MM_H
 
 #include <libk/std.h>
+#include <sync/spin_lock.h>
 
 #if defined(__x86_64__)
 #include <amd64/mm.h>
@@ -10,21 +11,18 @@
 #define MM_PG_PRESENT (1 << 0)
 #define MM_PG_RW      (1 << 1)
 #define MM_PG_USER    (1 << 2)
-#define MM_PD_RELOAD  (1 << 30)
-#define MM_PD_LOCK    (1 << 31)
 
 uintptr_t mm_alloc_user_pd_phys (void);
-void mm_reload (void);
+void mm_kernel_lock (spin_lock_ctx_t* ctx);
+void mm_kernel_unlock (spin_lock_ctx_t* ctx);
 void mm_map_page (struct pd* pd, uintptr_t paddr, uintptr_t vaddr, uint32_t flags);
 void mm_map_kernel_page (uintptr_t paddr, uintptr_t vaddr, uint32_t flags);
-void mm_unmap_page (struct pd* pd, uintptr_t vaddr, uint32_t flags);
+void mm_unmap_page (struct pd* pd, uintptr_t vaddr);
-void mm_unmap_kernel_page (uintptr_t vaddr, uint32_t flags);
+void mm_unmap_kernel_page (uintptr_t vaddr);
-void mm_lock_kernel (void);
+bool mm_validate (struct pd* pd, uintptr_t vaddr);
-void mm_unlock_kernel (void);
+bool mm_validate_buffer (struct pd* pd, uintptr_t vaddr, size_t size);
-bool mm_validate (struct pd* pd, uintptr_t vaddr, uint32_t flags);
+uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr);
-bool mm_validate_buffer (struct pd* pd, uintptr_t vaddr, size_t size, uint32_t flags);
+uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr);
-uintptr_t mm_p2v (struct pd* pd, uintptr_t paddr, uint32_t flags);
-uintptr_t mm_v2p (struct pd* pd, uintptr_t vaddr, uint32_t flags);
 struct pd* mm_get_kernel_pd (void);
 void mm_init (void);
 

kernel/sys/proc.h

@@ -8,5 +8,6 @@ struct proc;
 struct proc* proc_from_elf (uint8_t* elf_contents);
 struct proc* proc_clone (struct proc* proto, uintptr_t vstack_top, uintptr_t entry);
 void proc_cleanup (struct proc* proc);
+void proc_init_tls (struct proc* proc);
 
 #endif // _KERNEL_SYS_PROC_H

kernel/sys/procgroup.h

@@ -0,0 +1,8 @@
+#ifndef _KERNEL_SYS_PROCGROUP_H
+#define _KERNEL_SYS_PROCGROUP_H
+
+#if defined(__x86_64__)
+#include <amd64/procgroup.h>
+#endif
+
+#endif // _KERNEL_SYS_PROCGROUP_H

kernel/syscall/syscall.c

@@ -24,18 +24,18 @@
 
 static void* sys_get_user_buffer (struct proc* proc, uintptr_t uvaddr, size_t size) {
   struct limine_hhdm_response* hhdm = limine_hhdm_request.response;
-  spin_lock_ctx_t ctxprpd;
+  spin_lock_ctx_t ctxpg;
 
-  spin_lock (&proc->procgroup->pd.lock, &ctxprpd);
+  spin_lock (&proc->procgroup->lock, &ctxpg);
 
-  if (!mm_validate_buffer (&proc->procgroup->pd, (uintptr_t)uvaddr, size, 0)) {
+  if (!mm_validate_buffer (&proc->procgroup->pd, (uintptr_t)uvaddr, size)) {
-    spin_unlock (&proc->procgroup->pd.lock, &ctxprpd);
+    spin_unlock (&proc->procgroup->lock, &ctxpg);
     return NULL;
   }
 
-  uintptr_t out_paddr = mm_v2p (&proc->procgroup->pd, uvaddr, 0);
+  uintptr_t out_paddr = mm_v2p (&proc->procgroup->pd, uvaddr);
 
-  spin_unlock (&proc->procgroup->pd.lock, &ctxprpd);
+  spin_unlock (&proc->procgroup->lock, &ctxpg);
 
   uintptr_t out_kvaddr = (uintptr_t)hhdm->offset + out_paddr;
 
@@ -85,8 +85,6 @@ DEFINE_SYSCALL (sys_clone) {
 
   struct proc* new = proc_clone (proc, vstack_top, entry);
 
-  DEBUG ("new=%p\n", new);
-
   if (new == NULL) {
     return SYSRESULT (-ST_OOM_ERROR);
   }

make/libmsl.mk

@@ -7,4 +7,4 @@ clean_libmsl:
 format_libmsl:
 	make -C libmsl platform=$(platform) format
 
-.PHONY: all_libmsl clean_libmsl
+.PHONY: all_libmsl clean_libmsl format_libmsl