| @@ -0,0 +1,260 @@ |
| 1 | +/* |
| 2 | + * Copyright 2011 Tilera Corporation. All Rights Reserved. |
| 3 | + * |
| 4 | + * This program is free software; you can redistribute it and/or |
| 5 | + * modify it under the terms of the GNU General Public License |
| 6 | + * as published by the Free Software Foundation, version 2. |
| 7 | + * |
| 8 | + * This program is distributed in the hope that it will be useful, but |
| 9 | + * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | + * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or |
| 11 | + * NON INFRINGEMENT. See the GNU General Public License for |
| 12 | + * more details. |
| 13 | + * |
| 14 | + * copy new kernel into place and then call hv_reexec |
| 15 | + * |
| 16 | + */ |
| 17 | + |
| 18 | +#include <linux/linkage.h> |
| 19 | +#include <arch/chip.h> |
| 20 | +#include <asm/page.h> |
| 21 | +#include <hv/hypervisor.h> |
| 22 | + |
| 23 | +#undef RELOCATE_NEW_KERNEL_VERBOSE |
| 24 | + |
| 25 | +STD_ENTRY(relocate_new_kernel) |
| 26 | + |
| 27 | + move r30, r0 /* page list */ |
| 28 | + move r31, r1 /* address of page we are on */ |
| 29 | + move r32, r2 /* start address of new kernel */ |
| 30 | + |
| 31 | + shrui r1, r1, PAGE_SHIFT |
| 32 | + addi r1, r1, 1 |
| 33 | + shli sp, r1, PAGE_SHIFT |
| 34 | + addi sp, sp, -8 |
| 35 | + /* we now have a stack (whether we need one or not) */ |
| 36 | + |
| 37 | + moveli r40, hw2_last(hv_console_putc) |
| 38 | + shl16insli r40, r40, hw1(hv_console_putc) |
| 39 | + shl16insli r40, r40, hw0(hv_console_putc) |
| 40 | + |
| 41 | +#ifdef RELOCATE_NEW_KERNEL_VERBOSE |
| 42 | + moveli r0, 'r' |
| 43 | + jalr r40 |
| 44 | + |
| 45 | + moveli r0, '_' |
| 46 | + jalr r40 |
| 47 | + |
| 48 | + moveli r0, 'n' |
| 49 | + jalr r40 |
| 50 | + |
| 51 | + moveli r0, '_' |
| 52 | + jalr r40 |
| 53 | + |
| 54 | + moveli r0, 'k' |
| 55 | + jalr r40 |
| 56 | + |
| 57 | + moveli r0, '\n' |
| 58 | + jalr r40 |
| 59 | +#endif |
| 60 | + |
| 61 | + /* |
| 62 | + * Throughout this code r30 is pointer to the element of page |
| 63 | + * list we are working on. |
| 64 | + * |
| 65 | + * Normally we get to the next element of the page list by |
| 66 | + * incrementing r30 by eight. The exception is if the element |
| 67 | + * on the page list is an IND_INDIRECTION in which case we use |
| 68 | + * the element with the low bits masked off as the new value |
| 69 | + * of r30. |
| 70 | + * |
| 71 | + * To get this started, we need the value passed to us (which |
| 72 | + * will always be an IND_INDIRECTION) in memory somewhere with |
| 73 | + * r30 pointing at it. To do that, we push the value passed |
| 74 | + * to us on the stack and make r30 point to it. |
| 75 | + */ |
| 76 | + |
| 77 | + st sp, r30 |
| 78 | + move r30, sp |
| 79 | + addi sp, sp, -16 |
| 80 | + |
| 81 | +#if CHIP_HAS_CBOX_HOME_MAP() |
| 82 | + /* |
| 83 | + * On TILE-GX, we need to flush all tiles' caches, since we may |
| 84 | + * have been doing hash-for-home caching there. Note that we |
| 85 | + * must do this _after_ we're completely done modifying any memory |
| 86 | + * other than our output buffer (which we know is locally cached). |
| 87 | + * We want the caches to be fully clean when we do the reexec, |
| 88 | + * because the hypervisor is going to do this flush again at that |
| 89 | + * point, and we don't want that second flush to overwrite any memory. |
| 90 | + */ |
| 91 | + { |
| 92 | + move r0, zero /* cache_pa */ |
| 93 | + moveli r1, hw2_last(HV_FLUSH_EVICT_L2) |
| 94 | + } |
| 95 | + { |
| 96 | + shl16insli r1, r1, hw1(HV_FLUSH_EVICT_L2) |
| 97 | + movei r2, -1 /* cache_cpumask; -1 means all client tiles */ |
| 98 | + } |
| 99 | + { |
| 100 | + shl16insli r1, r1, hw0(HV_FLUSH_EVICT_L2) /* cache_control */ |
| 101 | + move r3, zero /* tlb_va */ |
| 102 | + } |
| 103 | + { |
| 104 | + move r4, zero /* tlb_length */ |
| 105 | + move r5, zero /* tlb_pgsize */ |
| 106 | + } |
| 107 | + { |
| 108 | + move r6, zero /* tlb_cpumask */ |
| 109 | + move r7, zero /* asids */ |
| 110 | + } |
| 111 | + { |
| 112 | + moveli r20, hw2_last(hv_flush_remote) |
| 113 | + move r8, zero /* asidcount */ |
| 114 | + } |
| 115 | + shl16insli r20, r20, hw1(hv_flush_remote) |
| 116 | + shl16insli r20, r20, hw0(hv_flush_remote) |
| 117 | + |
| 118 | + jalr r20 |
| 119 | +#endif |
| 120 | + |
| 121 | + /* r33 is destination pointer, default to zero */ |
| 122 | + |
| 123 | + moveli r33, 0 |
| 124 | + |
| 125 | +.Lloop: ld r10, r30 |
| 126 | + |
| 127 | + andi r9, r10, 0xf /* low 4 bits tell us what type it is */ |
| 128 | + xor r10, r10, r9 /* r10 is now value with low 4 bits stripped */ |
| 129 | + |
| 130 | + cmpeqi r0, r9, 0x1 /* IND_DESTINATION */ |
| 131 | + beqzt r0, .Ltry2 |
| 132 | + |
| 133 | + move r33, r10 |
| 134 | + |
| 135 | +#ifdef RELOCATE_NEW_KERNEL_VERBOSE |
| 136 | + moveli r0, 'd' |
| 137 | + jalr r40 |
| 138 | +#endif |
| 139 | + |
| 140 | + addi r30, r30, 8 |
| 141 | + j .Lloop |
| 142 | + |
| 143 | +.Ltry2: |
| 144 | + cmpeqi r0, r9, 0x2 /* IND_INDIRECTION */ |
| 145 | + beqzt r0, .Ltry4 |
| 146 | + |
| 147 | + move r30, r10 |
| 148 | + |
| 149 | +#ifdef RELOCATE_NEW_KERNEL_VERBOSE |
| 150 | + moveli r0, 'i' |
| 151 | + jalr r40 |
| 152 | +#endif |
| 153 | + |
| 154 | + j .Lloop |
| 155 | + |
| 156 | +.Ltry4: |
| 157 | + cmpeqi r0, r9, 0x4 /* IND_DONE */ |
| 158 | + beqzt r0, .Ltry8 |
| 159 | + |
| 160 | + mf |
| 161 | + |
| 162 | +#ifdef RELOCATE_NEW_KERNEL_VERBOSE |
| 163 | + moveli r0, 'D' |
| 164 | + jalr r40 |
| 165 | + moveli r0, '\n' |
| 166 | + jalr r40 |
| 167 | +#endif |
| 168 | + |
| 169 | + move r0, r32 |
| 170 | + |
| 171 | + moveli r41, hw2_last(hv_reexec) |
| 172 | + shl16insli r41, r41, hw1(hv_reexec) |
| 173 | + shl16insli r41, r41, hw0(hv_reexec) |
| 174 | + |
| 175 | + jalr r41 |
| 176 | + |
| 177 | + /* we should not get here */ |
| 178 | + |
| 179 | + moveli r0, '?' |
| 180 | + jalr r40 |
| 181 | + moveli r0, '\n' |
| 182 | + jalr r40 |
| 183 | + |
| 184 | + j .Lhalt |
| 185 | + |
| 186 | +.Ltry8: cmpeqi r0, r9, 0x8 /* IND_SOURCE */ |
| 187 | + beqz r0, .Lerr /* unknown type */ |
| 188 | + |
| 189 | + /* copy page at r10 to page at r33 */ |
| 190 | + |
| 191 | + move r11, r33 |
| 192 | + |
| 193 | + moveli r0, hw2_last(PAGE_SIZE) |
| 194 | + shl16insli r0, r0, hw1(PAGE_SIZE) |
| 195 | + shl16insli r0, r0, hw0(PAGE_SIZE) |
| 196 | + add r33, r33, r0 |
| 197 | + |
| 198 | + /* copy word at r10 to word at r11 until r11 equals r33 */ |
| 199 | + |
| 200 | + /* We know page size must be multiple of 8, so we can unroll |
| 201 | + * 8 times safely without any edge case checking. |
| 202 | + * |
| 203 | + * Issue a flush of the destination every 8 words to avoid |
| 204 | + * incoherence when starting the new kernel. (Now this is |
| 205 | + * just good paranoia because the hv_reexec call will also |
| 206 | + * take care of this.) |
| 207 | + */ |
| 208 | + |
| 209 | +1: |
| 210 | + { ld r0, r10; addi r10, r10, 8 } |
| 211 | + { st r11, r0; addi r11, r11, 8 } |
| 212 | + { ld r0, r10; addi r10, r10, 8 } |
| 213 | + { st r11, r0; addi r11, r11, 8 } |
| 214 | + { ld r0, r10; addi r10, r10, 8 } |
| 215 | + { st r11, r0; addi r11, r11, 8 } |
| 216 | + { ld r0, r10; addi r10, r10, 8 } |
| 217 | + { st r11, r0; addi r11, r11, 8 } |
| 218 | + { ld r0, r10; addi r10, r10, 8 } |
| 219 | + { st r11, r0; addi r11, r11, 8 } |
| 220 | + { ld r0, r10; addi r10, r10, 8 } |
| 221 | + { st r11, r0; addi r11, r11, 8 } |
| 222 | + { ld r0, r10; addi r10, r10, 8 } |
| 223 | + { st r11, r0; addi r11, r11, 8 } |
| 224 | + { ld r0, r10; addi r10, r10, 8 } |
| 225 | + { st r11, r0 } |
| 226 | + { flush r11 ; addi r11, r11, 8 } |
| 227 | + |
| 228 | + cmpeq r0, r33, r11 |
| 229 | + beqzt r0, 1b |
| 230 | + |
| 231 | +#ifdef RELOCATE_NEW_KERNEL_VERBOSE |
| 232 | + moveli r0, 's' |
| 233 | + jalr r40 |
| 234 | +#endif |
| 235 | + |
| 236 | + addi r30, r30, 8 |
| 237 | + j .Lloop |
| 238 | + |
| 239 | + |
| 240 | +.Lerr: moveli r0, 'e' |
| 241 | + jalr r40 |
| 242 | + moveli r0, 'r' |
| 243 | + jalr r40 |
| 244 | + moveli r0, 'r' |
| 245 | + jalr r40 |
| 246 | + moveli r0, '\n' |
| 247 | + jalr r40 |
| 248 | +.Lhalt: |
| 249 | + moveli r41, hw2_last(hv_halt) |
| 250 | + shl16insli r41, r41, hw1(hv_halt) |
| 251 | + shl16insli r41, r41, hw0(hv_halt) |
| 252 | + |
| 253 | + jalr r41 |
| 254 | + STD_ENDPROC(relocate_new_kernel) |
| 255 | + |
| 256 | + .section .rodata,"a" |
| 257 | + |
| 258 | + .globl relocate_new_kernel_size |
| 259 | +relocate_new_kernel_size: |
| 260 | + .long .Lend_relocate_new_kernel - relocate_new_kernel |