ia64: allocate percpu area for cpu0 like percpu areas for other cpus

[linux-2.6.git] / arch / blackfin / mach-common / head.S

diff --git a/arch/blackfin/mach-common/head.S b/arch/blackfin/mach-common/head.S
index 3069df580729d06b18cf4a47b34ec79fe540ebca..9c79dfea2a53b42e94286b0bf52a0b947c29452d 100644 (file)
--- a/arch/blackfin/mach-common/head.S
+++ b/arch/blackfin/mach-common/head.S
@@ -13,10 +13,22 @@
 #include <asm/blackfin.h>
 #include <asm/thread_info.h>
 #include <asm/trace.h>
+#include <asm/asm-offsets.h>
 
 __INIT
 
-#define INITIAL_STACK  (L1_SCRATCH_START + L1_SCRATCH_LENGTH - 12)
+ENTRY(__init_clear_bss)
+       r2 = r2 - r1;
+       cc = r2 == 0;
+       if cc jump .L_bss_done;
+       r2 >>= 2;
+       p1 = r1;
+       p2 = r2;
+       lsetup (1f, 1f) lc0 = p2;
+1:     [p1++] = r0;
+.L_bss_done:
+       rts;
+ENDPROC(__init_clear_bss)
 
 ENTRY(__start)
        /* R0: argument of command line string, passed from uboot, save it */
@@ -111,49 +123,81 @@ ENTRY(__start)
         * This happens here, since L1 gets clobbered
         * below
         */
-       p0.l = _saved_retx;
-       p0.h = _saved_retx;
+       GET_PDA(p0, r0);
+       r6 = [p0 + PDA_DF_RETX];
        p1.l = _init_saved_retx;
        p1.h = _init_saved_retx;
-       r0 = [p0];
-       [p1] = r0;
+       [p1] = r6;
 
-       p0.l = _saved_dcplb_fault_addr;
-       p0.h = _saved_dcplb_fault_addr;
+       r6 = [p0 + PDA_DF_DCPLB];
        p1.l = _init_saved_dcplb_fault_addr;
        p1.h = _init_saved_dcplb_fault_addr;
-       r0 = [p0];
-       [p1] = r0;
+       [p1] = r6;
 
-       p0.l = _saved_icplb_fault_addr;
-       p0.h = _saved_icplb_fault_addr;
+       r6 = [p0 + PDA_DF_ICPLB];
        p1.l = _init_saved_icplb_fault_addr;
        p1.h = _init_saved_icplb_fault_addr;
-       r0 = [p0];
-       [p1] = r0;
+       [p1] = r6;
 
-       p0.l = _saved_seqstat;
-       p0.h = _saved_seqstat;
+       r6 = [p0 + PDA_DF_SEQSTAT];
        p1.l = _init_saved_seqstat;
        p1.h = _init_saved_seqstat;
-       r0 = [p0];
-       [p1] = r0;
+       [p1] = r6;
 #endif
 
        /* Initialize stack pointer */
-       sp.l = lo(INITIAL_STACK);
-       sp.h = hi(INITIAL_STACK);
+       sp.l = _init_thread_union;
+       sp.h = _init_thread_union;
        fp = sp;
        usp = sp;
 
 #ifdef CONFIG_EARLY_PRINTK
        call _init_early_exception_vectors;
+       r0 = (EVT_IVHW | EVT_IRPTEN | EVT_EVX | EVT_NMI | EVT_RST | EVT_EMU);
+       sti r0;
+#endif
+
+       r0 = 0 (x);
+       /* Zero out all of the fun bss regions */
+#if L1_DATA_A_LENGTH > 0
+       r1.l = __sbss_l1;
+       r1.h = __sbss_l1;
+       r2.l = __ebss_l1;
+       r2.h = __ebss_l1;
+       call __init_clear_bss
+#endif
+#if L1_DATA_B_LENGTH > 0
+       r1.l = __sbss_b_l1;
+       r1.h = __sbss_b_l1;
+       r2.l = __ebss_b_l1;
+       r2.h = __ebss_b_l1;
+       call __init_clear_bss
+#endif
+#if L2_LENGTH > 0
+       r1.l = __sbss_l2;
+       r1.h = __sbss_l2;
+       r2.l = __ebss_l2;
+       r2.h = __ebss_l2;
+       call __init_clear_bss
 #endif
+       r1.l = ___bss_start;
+       r1.h = ___bss_start;
+       r2.l = ___bss_stop;
+       r2.h = ___bss_stop;
+       call __init_clear_bss
 
        /* Put The Code for PLL Programming and SDRAM Programming in L1 ISRAM */
        call _bfin_relocate_l1_mem;
 #ifdef CONFIG_BFIN_KERNEL_CLOCK
-       call _start_dma_code;
+       /* Only use on-chip scratch space for stack when absolutely required
+        * to avoid Anomaly 05000227 ... we know the init_clocks() func only
+        * uses L1 text and stack space and no other memory region.
+        */
+# define KERNEL_CLOCK_STACK (L1_SCRATCH_START + L1_SCRATCH_LENGTH - 12)
+       sp.l = lo(KERNEL_CLOCK_STACK);
+       sp.h = hi(KERNEL_CLOCK_STACK);
+       call _init_clocks;
+       sp = usp;       /* usp hasnt been touched, so restore from there */
 #endif
 
        /* This section keeps the processor in supervisor mode
@@ -170,16 +214,21 @@ ENTRY(__start)
        [p0] = p1;
        csync;
 
-       p0.l = lo(IMASK);
-       p0.h = hi(IMASK);
-       p1.l = IMASK_IVG15;
-       p1.h = 0x0;
-       [p0] = p1;
-       csync;
+#ifdef CONFIG_EARLY_PRINTK
+       r0 = (EVT_IVG15 | EVT_IVHW | EVT_IRPTEN | EVT_EVX | EVT_NMI | EVT_RST | EVT_EMU) (z);
+#else
+       r0 = EVT_IVG15 (z);
+#endif
+       sti r0;
 
        raise 15;
+#ifdef CONFIG_EARLY_PRINTK
+       p0.l = _early_trap;
+       p0.h = _early_trap;
+#else
        p0.l = .LWAIT_HERE;
        p0.h = .LWAIT_HERE;
+#endif
        reti = p0;
 #if ANOMALY_05000281
        nop; nop; nop;
@@ -206,45 +255,18 @@ ENTRY(_real_start)
        w[p0] = r0;
        ssync;
 
-       /* Zero out the bss region
-        * Note: this will fail if bss is 0 bytes ...
-        */
-       r0 = 0 (z);
-       r1.l = ___bss_start;
-       r1.h = ___bss_start;
-       r2.l = ___bss_stop;
-       r2.h = ___bss_stop;
-       r2 = r2 - r1;
-       r2 >>= 2;
-       p1 = r1;
-       p2 = r2;
-       lsetup (.L_clear_bss, .L_clear_bss) lc0 = p2;
-.L_clear_bss:
-       [p1++] = r0;
-
-       /* In case there is a NULL pointer reference,
-        * zero out region before stext
-        */
-       p1 = r0;
-       r2.l = __stext;
-       r2.h = __stext;
-       r2 >>= 2;
-       p2 = r2;
-       lsetup (.L_clear_zero, .L_clear_zero) lc0 = p2;
-.L_clear_zero:
-       [p1++] = r0;
-
        /* Pass the u-boot arguments to the global value command line */
        R0 = R7;
        call _cmdline_init;
 
        /* Load the current thread pointer and stack */
-       sp.l = _init_thread_union;
-       sp.h = _init_thread_union;
-       p1 = THREAD_SIZE (z);
+       p1 = THREAD_SIZE + 4 (z);       /* +4 is for reti loading */
        sp = sp + p1;
        usp = sp;
        fp = sp;
+       sp += -12;
+       call _init_pda
+       sp += 12;
        jump.l _start_kernel;
 ENDPROC(_real_start)