18 files changed, 119 insertions, 86 deletions

diff --git a/arch/arm/boot/compressed/head-shark.S b/arch/arm/boot/compressed/head-shark.S
index 848f60e542..089c560e07 100644
--- a/arch/arm/boot/compressed/head-shark.S
+++ b/arch/arm/boot/compressed/head-shark.S
@@ -63,8 +63,8 @@ __beginning:	mov	r4, r0				@ save the entry to the firmware
 
                 mov     pc, r2
 
-__copy_target:  .long   0x08508000
-__copy_end:     .long   0x08608000
+__copy_target:  .long   0x08507FFC
+__copy_end:     .long   0x08607FFC
                 
                 .word   _start
                 .word   __bss_start
@@ -73,9 +73,10 @@ __copy_end:	.long	0x08608000
 __temp_stack:   .space 128
 
 __mmu_off:
-                adr     r0, __ofw_data
+                adr     r0, __ofw_data                  @ read the 1. entry of the memory map
                 ldr     r0, [r0, #4]
                 orr     r0, r0, #0x00600000
+                sub     r0, r0, #4
         
                 ldr     r1, __copy_end
                 ldr     r3, __copy_target
@@ -89,20 +90,43 @@ __mmu_off:
  * from 0x08500000 to 0x08508000 if we have only 8MB
  */
 
+/* As we get more 2.6-kernels it gets more and more
+ * uncomfortable to be bound to kernel images of 1MB only.
+ * So we add a loop here, to be able to copy some more.
+ * Alexander Schulz 2005-07-17
+ */
+
+                mov     r4, #3                          @ How many megabytes to copy
+
+
+__MoveCode:     sub     r4, r4, #1
         
 __Copy:         ldr     r2, [r0], #-4
                 str     r2, [r1], #-4
                 teq     r1, r3
                 bne     __Copy
+
+                /* The firmware maps us in blocks of 1 MB, the next block is
+                   _below_ the last one. So our decrementing source pointer
+                   ist right here, but the destination pointer must be increased
+                   by 2 MB */
+                add     r1, r1, #0x00200000
+                add     r3, r3, #0x00100000
+
+                teq     r4, #0
+                bne     __MoveCode
+
+
                 /* and jump to it */
-                adr     r2, __go_on
-                adr     r0, __ofw_data
+                adr     r2, __go_on                     @ where we want to jump
+                adr     r0, __ofw_data                  @ read the 1. entry of the memory map
                 ldr     r0, [r0, #4]
-                sub     r2, r2, r0
-                sub     r2, r2, #0x00500000
-                ldr     r0, __copy_target
+                sub     r2, r2, r0                      @ we are mapped add 0e50 now, sub that (-0e00)
+                sub     r2, r2, #0x00500000             @ -0050
+                ldr     r0, __copy_target               @ and add 0850 8000 instead
+                add     r0, r0, #4
                 add     r2, r2, r0
-                mov     pc, r2
+                mov     pc, r2                          @ and jump there
 
 __go_on:
                 adr     sp, __temp_stack
diff --git a/arch/arm/configs/bast_defconfig b/arch/arm/configs/bast_defconfig
index 2d985e9611..35e3a99bcb 100644
--- a/arch/arm/configs/bast_defconfig
+++ b/arch/arm/configs/bast_defconfig
@@ -561,7 +561,6 @@ CONFIG_SERIAL_8250_SHARE_IRQ=y
 #
 CONFIG_SERIAL_S3C2410=y
 CONFIG_SERIAL_S3C2410_CONSOLE=y
-CONFIG_SERIAL_BAST_SIO=y
 CONFIG_SERIAL_CORE=y
 CONFIG_SERIAL_CORE_CONSOLE=y
 CONFIG_UNIX98_PTYS=y
diff --git a/arch/arm/configs/s3c2410_defconfig b/arch/arm/configs/s3c2410_defconfig
index 98b72ff388..96a794d8de 100644
--- a/arch/arm/configs/s3c2410_defconfig
+++ b/arch/arm/configs/s3c2410_defconfig
@@ -570,7 +570,6 @@ CONFIG_SERIAL_8250_SHARE_IRQ=y
 #
 CONFIG_SERIAL_S3C2410=y
 CONFIG_SERIAL_S3C2410_CONSOLE=y
-CONFIG_SERIAL_BAST_SIO=y
 CONFIG_SERIAL_CORE=y
 CONFIG_SERIAL_CORE_CONSOLE=y
 CONFIG_UNIX98_PTYS=y
diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
index 7ae45c3fc8..295e0a8379 100644
--- a/arch/arm/kernel/smp.c
+++ b/arch/arm/kernel/smp.c
@@ -78,7 +78,7 @@ struct smp_call_struct {
 static struct smp_call_struct * volatile smp_call_function_data;
 static DEFINE_SPINLOCK(smp_call_function_lock);
 
-int __init __cpu_up(unsigned int cpu)
+int __cpuinit __cpu_up(unsigned int cpu)
 {
         struct task_struct *idle;
         pgd_t *pgd;
@@ -159,7 +159,7 @@ int __init __cpu_up(unsigned int cpu)
  * This is the secondary CPU boot entry.  We're using this CPUs
  * idle thread stack, but a set of temporary page tables.
  */
-asmlinkage void __init secondary_start_kernel(void)
+asmlinkage void __cpuinit secondary_start_kernel(void)
 {
         struct mm_struct *mm = &init_mm;
         unsigned int cpu = smp_processor_id();
@@ -209,7 +209,7 @@ asmlinkage void __init secondary_start_kernel(void)
  * Called by both boot and secondaries to move global data into
  * per-processor storage.
  */
-void __init smp_store_cpu_info(unsigned int cpuid)
+void __cpuinit smp_store_cpu_info(unsigned int cpuid)
 {
         struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
 
diff --git a/arch/arm/mach-integrator/platsmp.c b/arch/arm/mach-integrator/platsmp.c
index 2ba0257770..aecf47ba03 100644
--- a/arch/arm/mach-integrator/platsmp.c
+++ b/arch/arm/mach-integrator/platsmp.c
@@ -27,12 +27,12 @@ extern void integrator_secondary_startup(void);
  * control for which core is the next to come out of the secondary
  * boot "holding pen"
  */
-volatile int __initdata pen_release = -1;
-unsigned long __initdata phys_pen_release = 0;
+volatile int __cpuinitdata pen_release = -1;
+unsigned long __cpuinitdata phys_pen_release = 0;
 
 static DEFINE_SPINLOCK(boot_lock);
 
-void __init platform_secondary_init(unsigned int cpu)
+void __cpuinit platform_secondary_init(unsigned int cpu)
 {
         /*
          * the primary core may have used a "cross call" soft interrupt
@@ -61,7 +61,7 @@ void __init platform_secondary_init(unsigned int cpu)
         spin_unlock(&boot_lock);
 }
 
-int __init boot_secondary(unsigned int cpu, struct task_struct *idle)
+int __cpuinit boot_secondary(unsigned int cpu, struct task_struct *idle)
 {
         unsigned long timeout;
 
diff --git a/arch/arm/mach-s3c2410/dma.c b/arch/arm/mach-s3c2410/dma.c
index c7c28890d4..65feaf20d2 100644
--- a/arch/arm/mach-s3c2410/dma.c
+++ b/arch/arm/mach-s3c2410/dma.c
@@ -436,7 +436,7 @@ int s3c2410_dma_enqueue(unsigned int channel, void *id,
 
         buf = kmem_cache_alloc(dma_kmem, GFP_ATOMIC);
         if (buf == NULL) {
-                pr_debug("%s: out of memory (%d alloc)\n",
+                pr_debug("%s: out of memory (%ld alloc)\n",
                          __FUNCTION__, sizeof(*buf));
                 return -ENOMEM;
         }
diff --git a/arch/arm/mach-s3c2410/mach-bast.c b/arch/arm/mach-s3c2410/mach-bast.c
index ccb6bcefa4..206778ebfc 100644
--- a/arch/arm/mach-s3c2410/mach-bast.c
+++ b/arch/arm/mach-s3c2410/mach-bast.c
@@ -28,6 +28,7 @@
  *     14-Mar-2006 BJD  Updated for __iomem changes
  *     22-Jun-2006 BJD  Added DM9000 platform information
  *     28-Jun-2006 BJD  Moved pm functionality out to common code
+ *     17-Jul-2006 BJD  Changed to platform device for SuperIO 16550s
 */
 
 #include <linux/kernel.h>
@@ -64,6 +65,8 @@
 #include <linux/mtd/nand_ecc.h>
 #include <linux/mtd/partitions.h>
 
+#include <linux/serial_8250.h>
+
 #include "clock.h"
 #include "devs.h"
 #include "cpu.h"
@@ -351,6 +354,39 @@ static struct platform_device bast_device_dm9k = {
         }
 };
 
+/* serial devices */
+
+#define SERIAL_BASE  (S3C2410_CS2 + BAST_PA_SUPERIO)
+#define SERIAL_FLAGS (UPF_BOOT_AUTOCONF | UPF_IOREMAP | UPF_SHARE_IRQ)
+#define SERIAL_CLK   (1843200)
+
+static struct plat_serial8250_port bast_sio_data[] = {
+        [0] = {
+                .mapbase        = SERIAL_BASE + 0x2f8,
+                .irq            = IRQ_PCSERIAL1,
+                .flags          = SERIAL_FLAGS,
+                .iotype         = UPIO_MEM,
+                .regshift       = 0,
+                .uartclk        = SERIAL_CLK,
+        },
+        [1] = {
+                .mapbase        = SERIAL_BASE + 0x3f8,
+                .irq            = IRQ_PCSERIAL2,
+                .flags          = SERIAL_FLAGS,
+                .iotype         = UPIO_MEM,
+                .regshift       = 0,
+                .uartclk        = SERIAL_CLK,
+        },
+        { }
+};
+
+static struct platform_device bast_sio = {
+        .name                   = "serial8250",
+        .id                     = 0,
+        .dev                    = {
+                .platform_data  = &bast_sio_data,
+        },
+};
 
 /* Standard BAST devices */
 
@@ -364,6 +400,7 @@ static struct platform_device *bast_devices[] __initdata = {
         &s3c_device_nand,
         &bast_device_nor,
         &bast_device_dm9k,
+        &bast_sio,
 };
 
 static struct clk *bast_clocks[] = {
diff --git a/arch/arm/nwfpe/fpa11.h b/arch/arm/nwfpe/fpa11.h
index 45cc65426a..e4a61aea53 100644
--- a/arch/arm/nwfpe/fpa11.h
+++ b/arch/arm/nwfpe/fpa11.h
@@ -29,9 +29,7 @@
  * stack+task struct.  Use the same method as 'current' uses to
  * reach them.
  */
-register unsigned long *user_registers asm("sl");
-
-#define GET_USERREG() (user_registers)
+#define GET_USERREG() ((struct pt_regs *)(THREAD_START_SP + (unsigned long)current_thread_info()) - 1)
 
 #include <linux/config.h>
 #include <linux/thread_info.h>
diff --git a/arch/arm/nwfpe/fpmodule.c b/arch/arm/nwfpe/fpmodule.c
index a8efcf3488..12885f31d3 100644
--- a/arch/arm/nwfpe/fpmodule.c
+++ b/arch/arm/nwfpe/fpmodule.c
@@ -132,7 +132,7 @@ void float_raise(signed char flags)
         printk(KERN_DEBUG
                "NWFPE: %s[%d] takes exception %08x at %p from %08lx\n",
                current->comm, current->pid, flags,
-               __builtin_return_address(0), GET_USERREG()[15]);
+               __builtin_return_address(0), GET_USERREG()->ARM_pc);
 #endif
 
         /* Keep SoftFloat exception flags up to date.  */
diff --git a/arch/arm/nwfpe/fpmodule.inl b/arch/arm/nwfpe/fpmodule.inl
index e5f59e9a30..2c39ad408f 100644
--- a/arch/arm/nwfpe/fpmodule.inl
+++ b/arch/arm/nwfpe/fpmodule.inl
@@ -28,8 +28,8 @@ static inline unsigned long readRegister(const unsigned int nReg)
            for this in this routine.  LDF/STF instructions with Rn = PC
            depend on the PC being correct, as they use PC+8 in their
            address calculations. */
-        unsigned long *userRegisters = GET_USERREG();
-        unsigned int val = userRegisters[nReg];
+        struct pt_regs *regs = GET_USERREG();
+        unsigned int val = regs->uregs[nReg];
         if (REG_PC == nReg)
                 val -= 4;
         return val;
@@ -38,8 +38,8 @@ static inline unsigned long readRegister(const unsigned int nReg)
 static inline void
 writeRegister(const unsigned int nReg, const unsigned long val)
 {
-        unsigned long *userRegisters = GET_USERREG();
-        userRegisters[nReg] = val;
+        struct pt_regs *regs = GET_USERREG();
+        regs->uregs[nReg] = val;
 }
 
 static inline unsigned long readCPSR(void)
@@ -63,12 +63,12 @@ static inline unsigned long readConditionCodes(void)
 
 static inline void writeConditionCodes(const unsigned long val)
 {
-        unsigned long *userRegisters = GET_USERREG();
+        struct pt_regs *regs = GET_USERREG();
         unsigned long rval;
         /*
          * Operate directly on userRegisters since
          * the CPSR may be the PC register itself.
          */
-        rval = userRegisters[REG_CPSR] & ~CC_MASK;
-        userRegisters[REG_CPSR] = rval | (val & CC_MASK);
+        rval = regs->ARM_cpsr & ~CC_MASK;
+        regs->ARM_cpsr = rval | (val & CC_MASK);
 }
diff --git a/arch/i386/kernel/i387.c b/arch/i386/kernel/i387.c
index b817168d9c..d75524758d 100644
--- a/arch/i386/kernel/i387.c
+++ b/arch/i386/kernel/i387.c
@@ -82,17 +82,6 @@ void kernel_fpu_begin(void)
 }
 EXPORT_SYMBOL_GPL(kernel_fpu_begin);
 
-void restore_fpu( struct task_struct *tsk )
-{
-        if ( cpu_has_fxsr ) {
-                asm volatile( "fxrstor %0"
-                              : : "m" (tsk->thread.i387.fxsave) );
-        } else {
-                asm volatile( "frstor %0"
-                              : : "m" (tsk->thread.i387.fsave) );
-        }
-}
-
 /*
  * FPU tag word conversions.
  */
diff --git a/arch/i386/kernel/process.c b/arch/i386/kernel/process.c
index ba243a4cc1..d9492058aa 100644
--- a/arch/i386/kernel/process.c
+++ b/arch/i386/kernel/process.c
@@ -700,23 +700,27 @@ struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct tas
 
         /*
          * Restore %fs and %gs if needed.
+         *
+         * Glibc normally makes %fs be zero, and %gs is one of
+         * the TLS segments.
          */
-        if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
+        if (unlikely(prev->fs | next->fs))
                 loadsegment(fs, next->fs);
+
+        if (prev->gs | next->gs)
                 loadsegment(gs, next->gs);
-        }
 
         /*
          * Now maybe reload the debug registers
          */
         if (unlikely(next->debugreg[7])) {
-                set_debugreg(current->thread.debugreg[0], 0);
-                set_debugreg(current->thread.debugreg[1], 1);
-                set_debugreg(current->thread.debugreg[2], 2);
-                set_debugreg(current->thread.debugreg[3], 3);
+                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(current->thread.debugreg[6], 6);
-                set_debugreg(current->thread.debugreg[7], 7);
+                set_debugreg(next->debugreg[6], 6);
+                set_debugreg(next->debugreg[7], 7);
         }
 
         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
diff --git a/arch/i386/kernel/setup.c b/arch/i386/kernel/setup.c
index 7306353c52..af4de58cab 100644
--- a/arch/i386/kernel/setup.c
+++ b/arch/i386/kernel/setup.c
@@ -1414,7 +1414,7 @@ static struct nop {
    This runs before SMP is initialized to avoid SMP problems with
    self modifying code. This implies that assymetric systems where
    APs have less capabilities than the boot processor are not handled. 
-   In this case boot with "noreplacement". */ 
+   Tough. Make sure you disable such features by hand. */ 
 void apply_alternatives(void *start, void *end) 
 { 
         struct alt_instr *a; 
@@ -1442,24 +1442,12 @@ void apply_alternatives(void *start, void *end)
         }
 } 
 
-static int no_replacement __initdata = 0; 
- 
 void __init alternative_instructions(void)
 {
         extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
-        if (no_replacement) 
-                return;
         apply_alternatives(__alt_instructions, __alt_instructions_end);
 }
 
-static int __init noreplacement_setup(char *s)
-{ 
-     no_replacement = 1; 
-     return 0; 
-} 
-
-__setup("noreplacement", noreplacement_setup); 
-
 static char * __init machine_specific_memory_setup(void);
 
 #ifdef CONFIG_MCA
diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index d781f10adc..88332f0009 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -1600,11 +1600,11 @@ sys_clone:	flushw
                 ba,pt           %xcc, sparc_do_fork
                  add            %sp, PTREGS_OFF, %o2
 ret_from_syscall:
-                /* Clear SPARC_FLAG_NEWCHILD, switch_to leaves thread.flags in
-                 * %o7 for us.  Check performance counter stuff too.
+                /* Clear current_thread_info()->new_child, and
+                 * check performance counter stuff too.
                  */
-                andn            %o7, _TIF_NEWCHILD, %l0
-                stx             %l0, [%g6 + TI_FLAGS]
+                stb             %g0, [%g6 + TI_NEW_CHILD]
+                ldx             [%g6 + TI_FLAGS], %l0
                 call            schedule_tail
                  mov            %g7, %o0
                 andcc           %l0, _TIF_PERFCTR, %g0
@@ -1720,12 +1720,11 @@ ret_sys_call:
         /* Check if force_successful_syscall_return()
          * was invoked.
          */
-        ldx             [%curptr + TI_FLAGS], %l0
-        andcc           %l0, _TIF_SYSCALL_SUCCESS, %g0
-        be,pt           %icc, 1f
-         andn           %l0, _TIF_SYSCALL_SUCCESS, %l0
+        ldub            [%curptr + TI_SYS_NOERROR], %l0
+        brz,pt          %l0, 1f
+         nop
         ba,pt           %xcc, 80f
-         stx            %l0, [%curptr + TI_FLAGS]
+         stb            %g0, [%curptr + TI_SYS_NOERROR]
 
 1:
         cmp             %o0, -ERESTART_RESTARTBLOCK
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index daa2fb9305..c9b6916763 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -782,8 +782,14 @@ static void distribute_irqs(void)
 }
 #endif
 
+struct sun5_timer {
+        u64     count0;
+        u64     limit0;
+        u64     count1;
+        u64     limit1;
+};
 
-struct sun5_timer *prom_timers;
+static struct sun5_timer *prom_timers;
 static u64 prom_limit0, prom_limit1;
 
 static void map_prom_timers(void)
@@ -839,18 +845,6 @@ static void kill_prom_timer(void)
         : "g1", "g2");
 }
 
-void enable_prom_timer(void)
-{
-        if (!prom_timers)
-                return;
-
-        /* Set it to whatever was there before. */
-        prom_timers->limit1 = prom_limit1;
-        prom_timers->count1 = 0;
-        prom_timers->limit0 = prom_limit0;
-        prom_timers->count0 = 0;
-}
-
 void init_irqwork_curcpu(void)
 {
         register struct irq_work_struct *workp asm("o2");
diff --git a/arch/sparc64/kernel/process.c b/arch/sparc64/kernel/process.c
index a0cd2b2494..cffb1c8ab4 100644
--- a/arch/sparc64/kernel/process.c
+++ b/arch/sparc64/kernel/process.c
@@ -621,8 +621,8 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
         memcpy(child_trap_frame, (((struct sparc_stackf *)regs)-1), (TRACEREG_SZ+STACKFRAME_SZ));
 
         t->flags = (t->flags & ~((0xffUL << TI_FLAG_CWP_SHIFT) | (0xffUL << TI_FLAG_CURRENT_DS_SHIFT))) |
-                _TIF_NEWCHILD |
                 (((regs->tstate + 1) & TSTATE_CWP) << TI_FLAG_CWP_SHIFT);
+        t->new_child = 1;
         t->ksp = ((unsigned long) child_trap_frame) - STACK_BIAS;
         t->kregs = (struct pt_regs *)(child_trap_frame+sizeof(struct sparc_stackf));
         t->fpsaved[0] = 0;
diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c
index 7e8e2919e1..b9b42491e1 100644
--- a/arch/sparc64/kernel/smp.c
+++ b/arch/sparc64/kernel/smp.c
@@ -137,7 +137,7 @@ void __init smp_callin(void)
         /* Clear this or we will die instantly when we
          * schedule back to this idler...
          */
-        clear_thread_flag(TIF_NEWCHILD);
+        current_thread_info()->new_child = 0;
 
         /* Attach to the address space of init_task. */
         atomic_inc(&init_mm.mm_count);
diff --git a/arch/sparc64/kernel/traps.c b/arch/sparc64/kernel/traps.c
index a9f4596d7c..100b0107c4 100644
--- a/arch/sparc64/kernel/traps.c
+++ b/arch/sparc64/kernel/traps.c
@@ -2125,6 +2125,8 @@ void __init trap_init(void)
             TI_PCR != offsetof(struct thread_info, pcr_reg) ||
             TI_CEE_STUFF != offsetof(struct thread_info, cee_stuff) ||
             TI_PRE_COUNT != offsetof(struct thread_info, preempt_count) ||
+            TI_NEW_CHILD != offsetof(struct thread_info, new_child) ||
+            TI_SYS_NOERROR != offsetof(struct thread_info, syscall_noerror) ||
             TI_FPREGS != offsetof(struct thread_info, fpregs) ||
             (TI_FPREGS & (64 - 1)))
                 thread_info_offsets_are_bolixed_dave();