fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / arch / i386 / kernel / process-xen.c
index a9508f6..4ba058f 100644 (file)
@@ -37,6 +37,7 @@
 #include <linux/kallsyms.h>
 #include <linux/ptrace.h>
 #include <linux/random.h>
+#include <linux/personality.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
@@ -59,6 +60,7 @@
 
 #include <asm/tlbflush.h>
 #include <asm/cpu.h>
+#include <asm/pda.h>
 
 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
 
@@ -99,16 +101,19 @@ EXPORT_SYMBOL(enable_hlt);
 /* XXX XEN doesn't use default_idle(), poll_idle(). Use xen_idle() instead. */
 void xen_idle(void)
 {
-       local_irq_disable();
+       current_thread_info()->status &= ~TS_POLLING;
+       /*
+        * TS_POLLING-cleared state must be visible before we
+        * test NEED_RESCHED:
+        */
+       smp_mb();
 
-       if (need_resched())
+       local_irq_disable();
+       if (!need_resched())
+               safe_halt();    /* enables interrupts racelessly */
+       else
                local_irq_enable();
-       else {
-               current_thread_info()->status &= ~TS_POLLING;
-               smp_mb__after_clear_bit();
-               safe_halt();
-               current_thread_info()->status |= TS_POLLING;
-       }
+       current_thread_info()->status |= TS_POLLING;
 }
 #ifdef CONFIG_APM_MODULE
 EXPORT_SYMBOL(default_idle);
@@ -144,7 +149,6 @@ void cpu_idle(void)
 
        current_thread_info()->status |= TS_POLLING;
 
-
        /* endless idle loop with no priority at all */
        while (1) {
                while (!need_resched()) {
@@ -169,7 +173,7 @@ void cpu_idle(void)
 void cpu_idle_wait(void)
 {
        unsigned int cpu, this_cpu = get_cpu();
-       cpumask_t map;
+       cpumask_t map, tmp = current->cpus_allowed;
 
        set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
        put_cpu();
@@ -191,11 +195,15 @@ void cpu_idle_wait(void)
                }
                cpus_and(map, map, cpu_online_map);
        } while (!cpus_empty(map));
+
+       set_cpus_allowed(current, tmp);
 }
 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 
 /* XXX XEN doesn't use mwait_idle(), select_idle_routine(), idle_setup(). */
 /* Always use xen_idle() instead. */
+void mwait_idle_with_hints(unsigned long eax, unsigned long ecx) {}
+
 void __devinit select_idle_routine(const struct cpuinfo_x86 *c) {}
 
 void show_regs(struct pt_regs * regs)
@@ -203,22 +211,24 @@ void show_regs(struct pt_regs * regs)
        unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 
        printk("\n");
-       printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
-       printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
+       printk("Pid: %d[#%u], comm: %20s\n",
+               current->pid, current->xid, current->comm);
+       printk("EIP: %04x:[<%08lx>] CPU: %d\n",
+               0xffff & regs->xcs,regs->eip, smp_processor_id());
        print_symbol("EIP is at %s\n", regs->eip);
 
        if (user_mode_vm(regs))
                printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
        printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
-              regs->eflags, print_tainted(), system_utsname.release,
-              (int)strcspn(system_utsname.version, " "),
-              system_utsname.version);
+              regs->eflags, print_tainted(), init_utsname()->release,
+              (int)strcspn(init_utsname()->version, " "),
+              init_utsname()->version);
        printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
                regs->eax,regs->ebx,regs->ecx,regs->edx);
        printk("ESI: %08lx EDI: %08lx EBP: %08lx",
                regs->esi, regs->edi, regs->ebp);
-       printk(" DS: %04x ES: %04x\n",
-               0xffff & regs->xds,0xffff & regs->xes);
+       printk(" DS: %04x ES: %04x GS: %04x\n",
+              0xffff & regs->xds,0xffff & regs->xes, 0xffff & regs->xgs);
 
        cr0 = read_cr0();
        cr2 = read_cr2();
@@ -234,15 +244,6 @@ void show_regs(struct pt_regs * regs)
  * the "args".
  */
 extern void kernel_thread_helper(void);
-__asm__(".section .text\n"
-       ".align 4\n"
-       "kernel_thread_helper:\n\t"
-       "movl %edx,%eax\n\t"
-       "pushl %edx\n\t"
-       "call *%ebx\n\t"
-       "pushl %eax\n\t"
-       "call do_exit\n"
-       ".previous");
 
 /*
  * Create a kernel thread
@@ -258,13 +259,15 @@ int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 
        regs.xds = __USER_DS;
        regs.xes = __USER_DS;
+       regs.xgs = __KERNEL_PDA;
        regs.orig_eax = -1;
        regs.eip = (unsigned long) kernel_thread_helper;
-       regs.xcs = GET_KERNEL_CS();
+       regs.xcs = __KERNEL_CS | get_kernel_rpl();
        regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 
        /* Ok, create the new process.. */
-       return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
+       return do_fork(flags | CLONE_VM | CLONE_UNTRACED | CLONE_KTHREAD,
+               0, &regs, 0, NULL, NULL);
 }
 EXPORT_SYMBOL(kernel_thread);
 
@@ -334,17 +337,15 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
        p->thread.eip = (unsigned long) ret_from_fork;
 
        savesegment(fs,p->thread.fs);
-       savesegment(gs,p->thread.gs);
 
        tsk = current;
        if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
-               p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
+               p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
+                                               IO_BITMAP_BYTES, GFP_KERNEL);
                if (!p->thread.io_bitmap_ptr) {
                        p->thread.io_bitmap_max = 0;
                        return -ENOMEM;
                }
-               memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
-                       IO_BITMAP_BYTES);
                set_tsk_thread_flag(p, TIF_IO_BITMAP);
        }
 
@@ -414,7 +415,7 @@ void dump_thread(struct pt_regs * regs, struct user * dump)
        dump->regs.ds = regs->xds;
        dump->regs.es = regs->xes;
        savesegment(fs,dump->regs.fs);
-       savesegment(gs,dump->regs.gs);
+       dump->regs.gs = regs->xgs;
        dump->regs.orig_eax = regs->orig_eax;
        dump->regs.eip = regs->eip;
        dump->regs.cs = regs->xcs;
@@ -442,6 +443,53 @@ int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
        return 1;
 }
 
+static noinline void __switch_to_xtra(struct task_struct *next_p)
+{
+       struct thread_struct *next;
+
+       next = &next_p->thread;
+
+       if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
+               set_debugreg(next->debugreg[0], 0);
+               set_debugreg(next->debugreg[1], 1);
+               set_debugreg(next->debugreg[2], 2);
+               set_debugreg(next->debugreg[3], 3);
+               /* no 4 and 5 */
+               set_debugreg(next->debugreg[6], 6);
+               set_debugreg(next->debugreg[7], 7);
+       }
+#ifndef CONFIG_XEN
+       if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
+               /*
+                * Disable the bitmap via an invalid offset. We still cache
+                * the previous bitmap owner and the IO bitmap contents:
+                */
+               tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
+               return;
+       }
+
+       if (likely(next == tss->io_bitmap_owner)) {
+               /*
+                * Previous owner of the bitmap (hence the bitmap content)
+                * matches the next task, we dont have to do anything but
+                * to set a valid offset in the TSS:
+                */
+               tss->io_bitmap_base = IO_BITMAP_OFFSET;
+               return;
+       }
+       /*
+        * Lazy TSS's I/O bitmap copy. We set an invalid offset here
+        * and we let the task to get a GPF in case an I/O instruction
+        * is performed.  The handler of the GPF will verify that the
+        * faulting task has a valid I/O bitmap and, it true, does the
+        * real copy and restart the instruction.  This will save us
+        * redundant copies when the currently switched task does not
+        * perform any I/O during its timeslice.
+        */
+       tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
+#endif /* !CONFIG_XEN */
+}
+
 /*
  * This function selects if the context switch from prev to next
  * has to tweak the TSC disable bit in the cr4.
@@ -504,7 +552,6 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
 #ifndef CONFIG_X86_NO_TSS
        struct tss_struct *tss = &per_cpu(init_tss, cpu);
 #endif
-       struct physdev_set_iopl iopl_op;
        struct physdev_set_iobitmap iobmp_op;
        multicall_entry_t _mcl[8], *mcl = _mcl;
 
@@ -554,14 +601,6 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
        C(0); C(1); C(2);
 #undef C
 
-       if (unlikely(prev->iopl != next->iopl)) {
-               iopl_op.iopl = (next->iopl == 0) ? 1 : (next->iopl >> 12) & 3;
-               mcl->op      = __HYPERVISOR_physdev_op;
-               mcl->args[0] = PHYSDEVOP_set_iopl;
-               mcl->args[1] = (unsigned long)&iopl_op;
-               mcl++;
-       }
-
        if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
                iobmp_op.bitmap   = (char *)next->io_bitmap_ptr;
                iobmp_op.nr_ports = next->io_bitmap_ptr ? IO_BITMAP_BITS : 0;
@@ -574,32 +613,35 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
        (void)HYPERVISOR_multicall(_mcl, mcl - _mcl);
 
        /*
-        * Restore %fs and %gs if needed.
+        * Restore %fs if needed.
         *
-        * Glibc normally makes %fs be zero, and %gs is one of
-        * the TLS segments.
+        * Glibc normally makes %fs be zero.
         */
        if (unlikely(next->fs))
                loadsegment(fs, next->fs);
 
-       if (next->gs)
-               loadsegment(gs, next->gs);
+       write_pda(pcurrent, next_p);
+
+       /* we're going to use this soon, after a few expensive things */
+       if (next_p->fpu_counter > 5)
+               prefetch(&next->i387.fxsave);
 
        /*
-        * Now maybe reload the debug registers
+        * Now maybe handle debug registers and/or IO bitmaps
         */
-       if (unlikely(next->debugreg[7])) {
-               set_debugreg(next->debugreg[0], 0);
-               set_debugreg(next->debugreg[1], 1);
-               set_debugreg(next->debugreg[2], 2);
-               set_debugreg(next->debugreg[3], 3);
-               /* no 4 and 5 */
-               set_debugreg(next->debugreg[6], 6);
-               set_debugreg(next->debugreg[7], 7);
-       }
+       if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW)
+           || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)))
+               __switch_to_xtra(next_p);
 
        disable_tsc(prev_p, next_p);
 
+       /* If the task has used fpu the last 5 timeslices, just do a full
+        * restore of the math state immediately to avoid the trap; the
+        * chances of needing FPU soon are obviously high now
+        */
+       if (next_p->fpu_counter > 5)
+               math_state_restore();
+
        return prev_p;
 }
 
@@ -655,9 +697,6 @@ asmlinkage int sys_execve(struct pt_regs regs)
                        (char __user * __user *) regs.edx,
                        &regs);
        if (error == 0) {
-               task_lock(current);
-               current->ptrace &= ~PT_DTRACE;
-               task_unlock(current);
                /* Make sure we don't return using sysenter.. */
                set_thread_flag(TIF_IRET);
        }
@@ -809,7 +848,7 @@ asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 
 unsigned long arch_align_stack(unsigned long sp)
 {
-       if (randomize_va_space)
+       if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
                sp -= get_random_int() % 8192;
        return sp & ~0xf;
 }