It was ugly.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
 #define Index_Load_Tag_D               0x05
 #define Index_Store_Tag_I              0x08
 #define Index_Store_Tag_D              0x09
-#if defined(CONFIG_CPU_LOONGSON2)
-#define Hit_Invalidate_I               0x00
-#else
 #define Hit_Invalidate_I               0x10
-#endif
 #define Hit_Invalidate_D               0x11
 #define Hit_Writeback_Inv_D            0x15
 
 #define Index_Store_Data_D             0x1d
 #define Index_Store_Data_S             0x1f
 
+/*
+ * Loongson2-specific cacheops
+ */
+#define Hit_Invalidate_I_Loongson23    0x00
+
 #endif /* __ASM_CACHEOPS_H */
 
 #include <asm/asm.h>
 #include <asm/cacheops.h>
 #include <asm/cpu-features.h>
+#include <asm/cpu-type.h>
 #include <asm/mipsmtregs.h>
 
 /*
 static inline void flush_icache_line(unsigned long addr)
 {
        __iflush_prologue
-       cache_op(Hit_Invalidate_I, addr);
+       switch (boot_cpu_type()) {
+       case CPU_LOONGSON2:
+               cache_op(Hit_Invalidate_I_Loongson23, addr);
+               break;
+
+       default:
+               cache_op(Hit_Invalidate_I, addr);
+               break;
+       }
        __iflush_epilogue
 }
 
  */
 static inline void protected_flush_icache_line(unsigned long addr)
 {
-       protected_cache_op(Hit_Invalidate_I, addr);
+       switch (boot_cpu_type()) {
+       case CPU_LOONGSON2:
+               protected_cache_op(Hit_Invalidate_I_Loongson23, addr);
+               break;
+
+       default:
+               protected_cache_op(Hit_Invalidate_I, addr);
+               break;
+       }
 }
 
 /*
 __BUILD_BLAST_CACHE(inv_s, scache, Index_Writeback_Inv_SD, Hit_Invalidate_SD, 128)
 
 /* build blast_xxx_range, protected_blast_xxx_range */
-#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot) \
-static inline void prot##blast_##pfx##cache##_range(unsigned long start, \
+#define __BUILD_BLAST_CACHE_RANGE(pfx, desc, hitop, prot, extra)       \
+static inline void prot##extra##blast_##pfx##cache##_range(unsigned long start, \
                                                    unsigned long end)  \
 {                                                                      \
        unsigned long lsize = cpu_##desc##_line_size();                 \
        __##pfx##flush_epilogue                                         \
 }
 
-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_)
-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_)
-__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_)
-__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, )
-__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, )
+__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, protected_, )
+__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, protected_, )
+__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I, protected_, )
+__BUILD_BLAST_CACHE_RANGE(i, icache, Hit_Invalidate_I_Loongson23, \
+       protected_, loongson23_)
+__BUILD_BLAST_CACHE_RANGE(d, dcache, Hit_Writeback_Inv_D, , )
+__BUILD_BLAST_CACHE_RANGE(s, scache, Hit_Writeback_Inv_SD, , )
 /* blast_inv_dcache_range */
-__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, )
-__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, )
+__BUILD_BLAST_CACHE_RANGE(inv_d, dcache, Hit_Invalidate_D, , )
+__BUILD_BLAST_CACHE_RANGE(inv_s, scache, Hit_Invalidate_SD, , )
 
 #endif /* _ASM_R4KCACHE_H */
 
 
 static inline void local_r4k___flush_cache_all(void * args)
 {
-#if defined(CONFIG_CPU_LOONGSON2)
-       r4k_blast_scache();
-       return;
-#endif
-       r4k_blast_dcache();
-       r4k_blast_icache();
-
        switch (current_cpu_type()) {
+       case CPU_LOONGSON2:
        case CPU_R4000SC:
        case CPU_R4000MC:
        case CPU_R4400SC:
        case CPU_R10000:
        case CPU_R12000:
        case CPU_R14000:
+               /*
+                * These caches are inclusive caches, that is, if something
+                * is not cached in the S-cache, we know it also won't be
+                * in one of the primary caches.
+                */
                r4k_blast_scache();
+               break;
+
+       default:
+               r4k_blast_dcache();
+               r4k_blast_icache();
+               break;
        }
 }
 
 
        if (end - start > icache_size)
                r4k_blast_icache();
-       else
-               protected_blast_icache_range(start, end);
+       else {
+               switch (boot_cpu_type()) {
+               case CPU_LOONGSON2:
+                       protected_blast_icache_range(start, end);
+                       break;
+
+               default:
+                       protected_loongson23_blast_icache_range(start, end);
+                       break;
+               }
+       }
 }
 
 static inline void local_r4k_flush_icache_range_ipi(void *args)
        case CPU_ALCHEMY:
                c->icache.flags |= MIPS_CACHE_IC_F_DC;
                break;
-       }
 
-#ifdef CONFIG_CPU_LOONGSON2
-       /*
-        * LOONGSON2 has 4 way icache, but when using indexed cache op,
-        * one op will act on all 4 ways
-        */
-       c->icache.ways = 1;
-#endif
+       case CPU_LOONGSON2:
+               /*
+                * LOONGSON2 has 4 way icache, but when using indexed cache op,
+                * one op will act on all 4 ways
+                */
+               c->icache.ways = 1;
+       }
 
        printk("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
               icache_size >> 10,
        return 1;
 }
 
-#if defined(CONFIG_CPU_LOONGSON2)
 static void __init loongson2_sc_init(void)
 {
        struct cpuinfo_mips *c = ¤t_cpu_data;
 
        c->options |= MIPS_CPU_INCLUSIVE_CACHES;
 }
-#endif
 
 extern int r5k_sc_init(void);
 extern int rm7k_sc_init(void);
 #endif
                return;
 
-#if defined(CONFIG_CPU_LOONGSON2)
        case CPU_LOONGSON2:
                loongson2_sc_init();
                return;
-#endif
+
        case CPU_XLP:
                /* don't need to worry about L2, fully coherent */
                return;
 
 
 #endif /* CONFIG_MIPS_MT_SMTC */
 
-#if defined(CONFIG_CPU_LOONGSON2)
 /*
  * LOONGSON2 has a 4 entry itlb which is a subset of dtlb,
  * unfortrunately, itlb is not totally transparent to software.
  */
-#define FLUSH_ITLB write_c0_diag(4);
-
-#define FLUSH_ITLB_VM(vma) { if ((vma)->vm_flags & VM_EXEC)  write_c0_diag(4); }
-
-#else
-
-#define FLUSH_ITLB
-#define FLUSH_ITLB_VM(vma)
+static inline void flush_itlb(void)
+{
+       switch (current_cpu_type()) {
+       case CPU_LOONGSON2:
+               write_c0_diag(4);
+               break;
+       default:
+               break;
+       }
+}
 
-#endif
+static inline void flush_itlb_vm(struct vm_area_struct *vma)
+{
+       if (vma->vm_flags & VM_EXEC)
+               flush_itlb();
+}
 
 void local_flush_tlb_all(void)
 {
        }
        tlbw_use_hazard();
        write_c0_entryhi(old_ctx);
-       FLUSH_ITLB;
+       flush_itlb();
        EXIT_CRITICAL(flags);
 }
 EXPORT_SYMBOL(local_flush_tlb_all);
                } else {
                        drop_mmu_context(mm, cpu);
                }
-               FLUSH_ITLB;
+               flush_itlb();
                EXIT_CRITICAL(flags);
        }
 }
        } else {
                local_flush_tlb_all();
        }
-       FLUSH_ITLB;
+       flush_itlb();
        EXIT_CRITICAL(flags);
 }
 
 
        finish:
                write_c0_entryhi(oldpid);
-               FLUSH_ITLB_VM(vma);
+               flush_itlb_vm(vma);
                EXIT_CRITICAL(flags);
        }
 }
                tlbw_use_hazard();
        }
        write_c0_entryhi(oldpid);
-       FLUSH_ITLB;
+       flush_itlb();
        EXIT_CRITICAL(flags);
 }
 
                        tlb_write_indexed();
        }
        tlbw_use_hazard();
-       FLUSH_ITLB_VM(vma);
+       flush_itlb_vm(vma);
        EXIT_CRITICAL(flags);
 }
 
 
         * need three, with the second nop'ed and the third being
         * unused.
         */
-       /* Loongson2 ebase is different than r4k, we have more space */
-#if defined(CONFIG_32BIT) || defined(CONFIG_CPU_LOONGSON2)
-       if ((p - tlb_handler) > 64)
-               panic("TLB refill handler space exceeded");
-#else
-       if (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 1)
-           || (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 3)
-               && uasm_insn_has_bdelay(relocs,
-                                       tlb_handler + MIPS64_REFILL_INSNS - 3)))
-               panic("TLB refill handler space exceeded");
-#endif
-
-       /*
-        * Now fold the handler in the TLB refill handler space.
-        */
-#if defined(CONFIG_32BIT) || defined(CONFIG_CPU_LOONGSON2)
-       f = final_handler;
-       /* Simplest case, just copy the handler. */
-       uasm_copy_handler(relocs, labels, tlb_handler, p, f);
-       final_len = p - tlb_handler;
-#else /* CONFIG_64BIT */
-       f = final_handler + MIPS64_REFILL_INSNS;
-       if ((p - tlb_handler) <= MIPS64_REFILL_INSNS) {
-               /* Just copy the handler. */
-               uasm_copy_handler(relocs, labels, tlb_handler, p, f);
-               final_len = p - tlb_handler;
-       } else {
-#ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
-               const enum label_id ls = label_tlb_huge_update;
-#else
-               const enum label_id ls = label_vmalloc;
-#endif
-               u32 *split;
-               int ov = 0;
-               int i;
-
-               for (i = 0; i < ARRAY_SIZE(labels) && labels[i].lab != ls; i++)
-                       ;
-               BUG_ON(i == ARRAY_SIZE(labels));
-               split = labels[i].addr;
-
-               /*
-                * See if we have overflown one way or the other.
-                */
-               if (split > tlb_handler + MIPS64_REFILL_INSNS ||
-                   split < p - MIPS64_REFILL_INSNS)
-                       ov = 1;
-
-               if (ov) {
+       switch (boot_cpu_type()) {
+       default:
+               if (sizeof(long) == 4) {
+       case CPU_LOONGSON2:
+               /* Loongson2 ebase is different than r4k, we have more space */
+                       if ((p - tlb_handler) > 64)
+                               panic("TLB refill handler space exceeded");
                        /*
-                        * Split two instructions before the end.  One
-                        * for the branch and one for the instruction
-                        * in the delay slot.
+                        * Now fold the handler in the TLB refill handler space.
                         */
-                       split = tlb_handler + MIPS64_REFILL_INSNS - 2;
-
+                       f = final_handler;
+                       /* Simplest case, just copy the handler. */
+                       uasm_copy_handler(relocs, labels, tlb_handler, p, f);
+                       final_len = p - tlb_handler;
+                       break;
+               } else {
+                       if (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 1)
+                           || (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 3)
+                               && uasm_insn_has_bdelay(relocs,
+                                                       tlb_handler + MIPS64_REFILL_INSNS - 3)))
+                               panic("TLB refill handler space exceeded");
                        /*
-                        * If the branch would fall in a delay slot,
-                        * we must back up an additional instruction
-                        * so that it is no longer in a delay slot.
+                        * Now fold the handler in the TLB refill handler space.
                         */
-                       if (uasm_insn_has_bdelay(relocs, split - 1))
-                               split--;
-               }
-               /* Copy first part of the handler. */
-               uasm_copy_handler(relocs, labels, tlb_handler, split, f);
-               f += split - tlb_handler;
-
-               if (ov) {
-                       /* Insert branch. */
-                       uasm_l_split(&l, final_handler);
-                       uasm_il_b(&f, &r, label_split);
-                       if (uasm_insn_has_bdelay(relocs, split))
-                               uasm_i_nop(&f);
-                       else {
-                               uasm_copy_handler(relocs, labels,
-                                                 split, split + 1, f);
-                               uasm_move_labels(labels, f, f + 1, -1);
-                               f++;
-                               split++;
+                       f = final_handler + MIPS64_REFILL_INSNS;
+                       if ((p - tlb_handler) <= MIPS64_REFILL_INSNS) {
+                               /* Just copy the handler. */
+                               uasm_copy_handler(relocs, labels, tlb_handler, p, f);
+                               final_len = p - tlb_handler;
+                       } else {
+#ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
+                               const enum label_id ls = label_tlb_huge_update;
+#else
+                               const enum label_id ls = label_vmalloc;
+#endif
+                               u32 *split;
+                               int ov = 0;
+                               int i;
+
+                               for (i = 0; i < ARRAY_SIZE(labels) && labels[i].lab != ls; i++)
+                                       ;
+                               BUG_ON(i == ARRAY_SIZE(labels));
+                               split = labels[i].addr;
+
+                               /*
+                                * See if we have overflown one way or the other.
+                                */
+                               if (split > tlb_handler + MIPS64_REFILL_INSNS ||
+                                   split < p - MIPS64_REFILL_INSNS)
+                                       ov = 1;
+
+                               if (ov) {
+                                       /*
+                                        * Split two instructions before the end.  One
+                                        * for the branch and one for the instruction
+                                        * in the delay slot.
+                                        */
+                                       split = tlb_handler + MIPS64_REFILL_INSNS - 2;
+
+                                       /*
+                                        * If the branch would fall in a delay slot,
+                                        * we must back up an additional instruction
+                                        * so that it is no longer in a delay slot.
+                                        */
+                                       if (uasm_insn_has_bdelay(relocs, split - 1))
+                                               split--;
+                               }
+                               /* Copy first part of the handler. */
+                               uasm_copy_handler(relocs, labels, tlb_handler, split, f);
+                               f += split - tlb_handler;
+
+                               if (ov) {
+                                       /* Insert branch. */
+                                       uasm_l_split(&l, final_handler);
+                                       uasm_il_b(&f, &r, label_split);
+                                       if (uasm_insn_has_bdelay(relocs, split))
+                                               uasm_i_nop(&f);
+                                       else {
+                                               uasm_copy_handler(relocs, labels,
+                                                                 split, split + 1, f);
+                                               uasm_move_labels(labels, f, f + 1, -1);
+                                               f++;
+                                               split++;
+                                       }
+                               }
+
+                               /* Copy the rest of the handler. */
+                               uasm_copy_handler(relocs, labels, split, p, final_handler);
+                               final_len = (f - (final_handler + MIPS64_REFILL_INSNS)) +
+                                           (p - split);
                        }
                }
-
-               /* Copy the rest of the handler. */
-               uasm_copy_handler(relocs, labels, split, p, final_handler);
-               final_len = (f - (final_handler + MIPS64_REFILL_INSNS)) +
-                           (p - split);
+               break;
        }
-#endif /* CONFIG_64BIT */
 
        uasm_resolve_relocs(relocs, labels);
        pr_debug("Wrote TLB refill handler (%u instructions).\n",