um: Save FPU registers between task switches
[linux-2.6.git] / arch / um / kernel / process.c
index 90c9ffa..21c1ae7 100644 (file)
@@ -1,74 +1,54 @@
 /*
- * Copyright (C) 2000, 2001, 2002 Jeff Dike (jdike@karaya.com)
+ * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Copyright 2003 PathScale, Inc.
  * Licensed under the GPL
  */
 
-#include "linux/kernel.h"
-#include "linux/sched.h"
-#include "linux/interrupt.h"
-#include "linux/string.h"
-#include "linux/mm.h"
-#include "linux/slab.h"
-#include "linux/utsname.h"
-#include "linux/fs.h"
-#include "linux/utime.h"
-#include "linux/smp_lock.h"
-#include "linux/module.h"
-#include "linux/init.h"
-#include "linux/capability.h"
-#include "linux/vmalloc.h"
-#include "linux/spinlock.h"
-#include "linux/proc_fs.h"
-#include "linux/ptrace.h"
-#include "linux/random.h"
-#include "linux/personality.h"
-#include "asm/unistd.h"
-#include "asm/mman.h"
-#include "asm/segment.h"
-#include "asm/stat.h"
-#include "asm/pgtable.h"
-#include "asm/processor.h"
-#include "asm/tlbflush.h"
-#include "asm/uaccess.h"
-#include "asm/user.h"
-#include "kern_util.h"
+#include <linux/stddef.h>
+#include <linux/err.h>
+#include <linux/hardirq.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/personality.h>
+#include <linux/proc_fs.h>
+#include <linux/ptrace.h>
+#include <linux/random.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/seq_file.h>
+#include <linux/tick.h>
+#include <linux/threads.h>
+#include <asm/current.h>
+#include <asm/pgtable.h>
+#include <asm/uaccess.h>
 #include "as-layout.h"
-#include "kern.h"
-#include "signal_kern.h"
-#include "init.h"
-#include "irq_user.h"
-#include "mem_user.h"
-#include "tlb.h"
-#include "frame_kern.h"
-#include "sigcontext.h"
+#include "kern_util.h"
 #include "os.h"
-#include "mode.h"
-#include "mode_kern.h"
-#include "choose-mode.h"
-#include "um_malloc.h"
+#include "skas.h"
+#include "tlb.h"
 
-/* This is a per-cpu array.  A processor only modifies its entry and it only
+/*
+ * This is a per-cpu array.  A processor only modifies its entry and it only
  * cares about its entry, so it's OK if another processor is modifying its
  * entry.
  */
 struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } };
 
-int external_pid(void *t)
+static inline int external_pid(void)
 {
-       struct task_struct *task = t ? t : current;
-
-       return(CHOOSE_MODE_PROC(external_pid_tt, external_pid_skas, task));
+       /* FIXME: Need to look up userspace_pid by cpu */
+       return userspace_pid[0];
 }
 
 int pid_to_processor_id(int pid)
 {
        int i;
 
-       for(i = 0; i < ncpus; i++){
-               if(cpu_tasks[i].pid == pid) return(i);
+       for (i = 0; i < ncpus; i++) {
+               if (cpu_tasks[i].pid == pid)
+                       return i;
        }
-       return(-1);
+       return -1;
 }
 
 void free_stack(unsigned long stack, int order)
@@ -84,10 +64,8 @@ unsigned long alloc_stack(int order, int atomic)
        if (atomic)
                flags = GFP_ATOMIC;
        page = __get_free_pages(flags, order);
-       if(page == 0)
-               return(0);
-       stack_protections(page);
-       return(page);
+
+       return page;
 }
 
 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
@@ -98,224 +76,186 @@ int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
        current->thread.request.u.thread.arg = arg;
        pid = do_fork(CLONE_VM | CLONE_UNTRACED | flags, 0,
                      &current->thread.regs, 0, NULL, NULL);
-       if(pid < 0)
-               panic("do_fork failed in kernel_thread, errno = %d", pid);
-       return(pid);
+       return pid;
 }
 
-void set_current(void *t)
+static inline void set_current(struct task_struct *task)
 {
-       struct task_struct *task = t;
-
        cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task)
-               { external_pid(task), task });
+               { external_pid(), task });
 }
 
+extern void arch_switch_to(struct task_struct *to);
+
 void *_switch_to(void *prev, void *next, void *last)
 {
        struct task_struct *from = prev;
-       struct task_struct *to= next;
+       struct task_struct *to = next;
 
        to->thread.prev_sched = from;
        set_current(to);
 
        do {
-               current->thread.saved_task = NULL ;
-               CHOOSE_MODE_PROC(switch_to_tt, switch_to_skas, prev, next);
-               if(current->thread.saved_task)
+               current->thread.saved_task = NULL;
+
+               switch_threads(&from->thread.switch_buf,
+                              &to->thread.switch_buf);
+
+               arch_switch_to(current);
+
+               if (current->thread.saved_task)
                        show_regs(&(current->thread.regs));
-               next= current->thread.saved_task;
-               prev= current;
-       } while(current->thread.saved_task);
+               to = current->thread.saved_task;
+               from = current;
+       } while (current->thread.saved_task);
 
-       return(current->thread.prev_sched);
+       return current->thread.prev_sched;
 
 }
 
 void interrupt_end(void)
 {
-       if(need_resched()) schedule();
-       if(test_tsk_thread_flag(current, TIF_SIGPENDING)) do_signal();
-}
-
-void release_thread(struct task_struct *task)
-{
-       CHOOSE_MODE(release_thread_tt(task), release_thread_skas(task));
+       if (need_resched())
+               schedule();
+       if (test_tsk_thread_flag(current, TIF_SIGPENDING))
+               do_signal();
 }
 
 void exit_thread(void)
 {
-       unprotect_stack((unsigned long) current_thread);
 }
 
 void *get_current(void)
 {
-       return(current);
+       return current;
 }
 
-int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
-               unsigned long stack_top, struct task_struct * p,
-               struct pt_regs *regs)
+/*
+ * This is called magically, by its address being stuffed in a jmp_buf
+ * and being longjmp-d to.
+ */
+void new_thread_handler(void)
 {
-       int ret;
+       int (*fn)(void *), n;
+       void *arg;
 
-       p->thread = (struct thread_struct) INIT_THREAD;
-       ret = CHOOSE_MODE_PROC(copy_thread_tt, copy_thread_skas, nr,
-                               clone_flags, sp, stack_top, p, regs);
-
-       if (ret || !current->thread.forking)
-               goto out;
+       if (current->thread.prev_sched != NULL)
+               schedule_tail(current->thread.prev_sched);
+       current->thread.prev_sched = NULL;
 
-       clear_flushed_tls(p);
+       fn = current->thread.request.u.thread.proc;
+       arg = current->thread.request.u.thread.arg;
 
        /*
-        * Set a new TLS for the child thread?
+        * The return value is 1 if the kernel thread execs a process,
+        * 0 if it just exits
         */
-       if (clone_flags & CLONE_SETTLS)
-               ret = arch_copy_tls(p);
-
-out:
-       return ret;
-}
-
-void initial_thread_cb(void (*proc)(void *), void *arg)
-{
-       int save_kmalloc_ok = kmalloc_ok;
-
-       kmalloc_ok = 0;
-       CHOOSE_MODE_PROC(initial_thread_cb_tt, initial_thread_cb_skas, proc,
-                        arg);
-       kmalloc_ok = save_kmalloc_ok;
-}
-
-unsigned long stack_sp(unsigned long page)
-{
-       return(page + PAGE_SIZE - sizeof(void *));
+       n = run_kernel_thread(fn, arg, &current->thread.exec_buf);
+       if (n == 1) {
+               /* Handle any immediate reschedules or signals */
+               interrupt_end();
+               userspace(&current->thread.regs.regs);
+       }
+       else do_exit(0);
 }
 
-int current_pid(void)
+/* Called magically, see new_thread_handler above */
+void fork_handler(void)
 {
-       return(current->pid);
-}
+       force_flush_all();
 
-void default_idle(void)
-{
-       CHOOSE_MODE(uml_idle_timer(), (void) 0);
+       schedule_tail(current->thread.prev_sched);
 
-       while(1){
-               /* endless idle loop with no priority at all */
+       /*
+        * XXX: if interrupt_end() calls schedule, this call to
+        * arch_switch_to isn't needed. We could want to apply this to
+        * improve performance. -bb
+        */
+       arch_switch_to(current);
 
-               /*
-                * although we are an idle CPU, we do not want to
-                * get into the scheduler unnecessarily.
-                */
-               if(need_resched())
-                       schedule();
+       current->thread.prev_sched = NULL;
 
-               idle_sleep(10);
-       }
-}
+       /* Handle any immediate reschedules or signals */
+       interrupt_end();
 
-void cpu_idle(void)
-{
-       CHOOSE_MODE(init_idle_tt(), init_idle_skas());
+       userspace(&current->thread.regs.regs);
 }
 
-int page_size(void)
+int copy_thread(unsigned long clone_flags, unsigned long sp,
+               unsigned long stack_top, struct task_struct * p,
+               struct pt_regs *regs)
 {
-       return(PAGE_SIZE);
-}
+       void (*handler)(void);
+       int ret = 0;
 
-void *um_virt_to_phys(struct task_struct *task, unsigned long addr,
-                     pte_t *pte_out)
-{
-       pgd_t *pgd;
-       pud_t *pud;
-       pmd_t *pmd;
-       pte_t *pte;
-       pte_t ptent;
+       p->thread = (struct thread_struct) INIT_THREAD;
 
-       if(task->mm == NULL)
-               return(ERR_PTR(-EINVAL));
-       pgd = pgd_offset(task->mm, addr);
-       if(!pgd_present(*pgd))
-               return(ERR_PTR(-EINVAL));
+       if (current->thread.forking) {
+               memcpy(&p->thread.regs.regs, &regs->regs,
+                      sizeof(p->thread.regs.regs));
+               REGS_SET_SYSCALL_RETURN(p->thread.regs.regs.gp, 0);
+               if (sp != 0)
+                       REGS_SP(p->thread.regs.regs.gp) = sp;
 
-       pud = pud_offset(pgd, addr);
-       if(!pud_present(*pud))
-               return(ERR_PTR(-EINVAL));
+               handler = fork_handler;
 
-       pmd = pmd_offset(pud, addr);
-       if(!pmd_present(*pmd))
-               return(ERR_PTR(-EINVAL));
+               arch_copy_thread(&current->thread.arch, &p->thread.arch);
+       }
+       else {
+               get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp);
+               p->thread.request.u.thread = current->thread.request.u.thread;
+               handler = new_thread_handler;
+       }
 
-       pte = pte_offset_kernel(pmd, addr);
-       ptent = *pte;
-       if(!pte_present(ptent))
-               return(ERR_PTR(-EINVAL));
+       new_thread(task_stack_page(p), &p->thread.switch_buf, handler);
 
-       if(pte_out != NULL)
-               *pte_out = ptent;
-       return((void *) (pte_val(ptent) & PAGE_MASK) + (addr & ~PAGE_MASK));
-}
+       if (current->thread.forking) {
+               clear_flushed_tls(p);
 
-char *current_cmd(void)
-{
-#if defined(CONFIG_SMP) || defined(CONFIG_HIGHMEM)
-       return("(Unknown)");
-#else
-       void *addr = um_virt_to_phys(current, current->mm->arg_start, NULL);
-       return IS_ERR(addr) ? "(Unknown)": __va((unsigned long) addr);
-#endif
-}
+               /*
+                * Set a new TLS for the child thread?
+                */
+               if (clone_flags & CLONE_SETTLS)
+                       ret = arch_copy_tls(p);
+       }
 
-void force_sigbus(void)
-{
-       printk(KERN_ERR "Killing pid %d because of a lack of memory\n",
-              current->pid);
-       lock_kernel();
-       sigaddset(&current->pending.signal, SIGBUS);
-       recalc_sigpending();
-       current->flags |= PF_SIGNALED;
-       do_exit(SIGBUS | 0x80);
+       return ret;
 }
 
-void dump_thread(struct pt_regs *regs, struct user *u)
+void initial_thread_cb(void (*proc)(void *), void *arg)
 {
-}
+       int save_kmalloc_ok = kmalloc_ok;
 
-void enable_hlt(void)
-{
-       panic("enable_hlt");
+       kmalloc_ok = 0;
+       initial_thread_cb_skas(proc, arg);
+       kmalloc_ok = save_kmalloc_ok;
 }
 
-EXPORT_SYMBOL(enable_hlt);
-
-void disable_hlt(void)
+void default_idle(void)
 {
-       panic("disable_hlt");
-}
+       unsigned long long nsecs;
 
-EXPORT_SYMBOL(disable_hlt);
-
-void *um_kmalloc(int size)
-{
-       return kmalloc(size, GFP_KERNEL);
-}
+       while (1) {
+               /* endless idle loop with no priority at all */
 
-void *um_kmalloc_atomic(int size)
-{
-       return kmalloc(size, GFP_ATOMIC);
-}
+               /*
+                * although we are an idle CPU, we do not want to
+                * get into the scheduler unnecessarily.
+                */
+               if (need_resched())
+                       schedule();
 
-void *um_vmalloc(int size)
-{
-       return vmalloc(size);
+               tick_nohz_stop_sched_tick(1);
+               nsecs = disable_timer();
+               idle_sleep(nsecs);
+               tick_nohz_restart_sched_tick();
+       }
 }
 
-void *um_vmalloc_atomic(int size)
+void cpu_idle(void)
 {
-       return __vmalloc(size, GFP_ATOMIC | __GFP_HIGHMEM, PAGE_KERNEL);
+       cpu_tasks[current_thread_info()->cpu].pid = os_getpid();
+       default_idle();
 }
 
 int __cant_sleep(void) {
@@ -323,26 +263,12 @@ int __cant_sleep(void) {
        /* Is in_interrupt() really needed? */
 }
 
-unsigned long get_fault_addr(void)
-{
-       return((unsigned long) current->thread.fault_addr);
-}
-
-EXPORT_SYMBOL(get_fault_addr);
-
-void not_implemented(void)
-{
-       printk(KERN_DEBUG "Something isn't implemented in here\n");
-}
-
-EXPORT_SYMBOL(not_implemented);
-
 int user_context(unsigned long sp)
 {
        unsigned long stack;
 
        stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
-       return(stack != (unsigned long) current_thread);
+       return stack != (unsigned long) current_thread_info();
 }
 
 extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;
@@ -356,45 +282,45 @@ void do_uml_exitcalls(void)
                (*call)();
 }
 
-char *uml_strdup(char *string)
+char *uml_strdup(const char *string)
 {
        return kstrdup(string, GFP_KERNEL);
 }
 
 int copy_to_user_proc(void __user *to, void *from, int size)
 {
-       return(copy_to_user(to, from, size));
+       return copy_to_user(to, from, size);
 }
 
 int copy_from_user_proc(void *to, void __user *from, int size)
 {
-       return(copy_from_user(to, from, size));
+       return copy_from_user(to, from, size);
 }
 
 int clear_user_proc(void __user *buf, int size)
 {
-       return(clear_user(buf, size));
+       return clear_user(buf, size);
 }
 
 int strlen_user_proc(char __user *str)
 {
-       return(strlen_user(str));
+       return strlen_user(str);
 }
 
 int smp_sigio_handler(void)
 {
 #ifdef CONFIG_SMP
-       int cpu = current_thread->cpu;
+       int cpu = current_thread_info()->cpu;
        IPI_handler(cpu);
-       if(cpu != 0)
-               return(1);
+       if (cpu != 0)
+               return 1;
 #endif
-       return(0);
+       return 0;
 }
 
 int cpu(void)
 {
-       return(current_thread->cpu);
+       return current_thread_info()->cpu;
 }
 
 static atomic_t using_sysemu = ATOMIC_INIT(0);
@@ -412,15 +338,19 @@ int get_using_sysemu(void)
        return atomic_read(&using_sysemu);
 }
 
-static int proc_read_sysemu(char *buf, char **start, off_t offset, int size,int *eof, void *data)
+static int sysemu_proc_show(struct seq_file *m, void *v)
 {
-       if (snprintf(buf, size, "%d\n", get_using_sysemu()) < size) /*No overflow*/
-               *eof = 1;
+       seq_printf(m, "%d\n", get_using_sysemu());
+       return 0;
+}
 
-       return strlen(buf);
+static int sysemu_proc_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, sysemu_proc_show, NULL);
 }
 
-static int proc_write_sysemu(struct file *file,const char __user *buf, unsigned long count,void *data)
+static ssize_t sysemu_proc_write(struct file *file, const char __user *buf,
+                                size_t count, loff_t *pos)
 {
        char tmp[2];
 
@@ -429,26 +359,33 @@ static int proc_write_sysemu(struct file *file,const char __user *buf, unsigned
 
        if (tmp[0] >= '0' && tmp[0] <= '2')
                set_using_sysemu(tmp[0] - '0');
-       return count; /*We use the first char, but pretend to write everything*/
+       /* We use the first char, but pretend to write everything */
+       return count;
 }
 
+static const struct file_operations sysemu_proc_fops = {
+       .owner          = THIS_MODULE,
+       .open           = sysemu_proc_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+       .write          = sysemu_proc_write,
+};
+
 int __init make_proc_sysemu(void)
 {
        struct proc_dir_entry *ent;
        if (!sysemu_supported)
                return 0;
 
-       ent = create_proc_entry("sysemu", 0600, &proc_root);
+       ent = proc_create("sysemu", 0600, NULL, &sysemu_proc_fops);
 
        if (ent == NULL)
        {
                printk(KERN_WARNING "Failed to register /proc/sysemu\n");
-               return(0);
+               return 0;
        }
 
-       ent->read_proc  = proc_read_sysemu;
-       ent->write_proc = proc_write_sysemu;
-
        return 0;
 }
 
@@ -458,11 +395,11 @@ int singlestepping(void * t)
 {
        struct task_struct *task = t ? t : current;
 
-       if ( ! (task->ptrace & PT_DTRACE) )
-               return(0);
+       if (!(task->ptrace & PT_DTRACE))
+               return 0;
 
        if (task->thread.singlestep_syscall)
-               return(1);
+               return 1;
 
        return 2;
 }
@@ -482,3 +419,46 @@ unsigned long arch_align_stack(unsigned long sp)
        return sp & ~0xf;
 }
 #endif
+
+unsigned long get_wchan(struct task_struct *p)
+{
+       unsigned long stack_page, sp, ip;
+       bool seen_sched = 0;
+
+       if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING))
+               return 0;
+
+       stack_page = (unsigned long) task_stack_page(p);
+       /* Bail if the process has no kernel stack for some reason */
+       if (stack_page == 0)
+               return 0;
+
+       sp = p->thread.switch_buf->JB_SP;
+       /*
+        * Bail if the stack pointer is below the bottom of the kernel
+        * stack for some reason
+        */
+       if (sp < stack_page)
+               return 0;
+
+       while (sp < stack_page + THREAD_SIZE) {
+               ip = *((unsigned long *) sp);
+               if (in_sched_functions(ip))
+                       /* Ignore everything until we're above the scheduler */
+                       seen_sched = 1;
+               else if (kernel_text_address(ip) && seen_sched)
+                       return ip;
+
+               sp += sizeof(unsigned long);
+       }
+
+       return 0;
+}
+
+int elf_core_copy_fpregs(struct task_struct *t, elf_fpregset_t *fpu)
+{
+       int cpu = current_thread_info()->cpu;
+
+       return save_fp_registers(userspace_pid[cpu], (unsigned long *) fpu);
+}
+