1 /* 2 * Copyright (C) 2010 Andreas Tobler 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 17 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 18 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 19 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 20 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 21 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 22 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 23 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 */ 25 26 #include <sys/cdefs.h> 27 __FBSDID("$FreeBSD$"); 28 29 #include <sys/param.h> 30 #include <sys/kernel.h> 31 #include <sys/ktr.h> 32 #include <sys/lock.h> 33 #include <sys/msgbuf.h> 34 #include <sys/mutex.h> 35 #include <sys/proc.h> 36 #include <sys/sysctl.h> 37 #include <sys/systm.h> 38 #include <sys/vmmeter.h> 39 40 #include <dev/ofw/openfirm.h> 41 #include <machine/ofw_machdep.h> 42 43 #include <vm/vm.h> 44 #include <vm/vm_param.h> 45 #include <vm/vm_kern.h> 46 #include <vm/vm_page.h> 47 #include <vm/vm_map.h> 48 #include <vm/vm_object.h> 49 #include <vm/vm_extern.h> 50 #include <vm/vm_pageout.h> 51 #include <vm/uma.h> 52 53 #include <powerpc/aim/mmu_oea64.h> 54 55 #include "mmu_if.h" 56 #include "moea64_if.h" 57 58 #include "phyp-hvcall.h" 59 60 extern int n_slbs; 61 62 /* 63 * Kernel MMU interface 64 */ 65 66 static void mphyp_bootstrap(mmu_t mmup, vm_offset_t kernelstart, 67 vm_offset_t kernelend); 68 static void mphyp_cpu_bootstrap(mmu_t mmup, int ap); 69 static void mphyp_pte_synch(mmu_t, uintptr_t pt, struct lpte *pvo_pt); 70 static void mphyp_pte_clear(mmu_t, uintptr_t pt, struct lpte *pvo_pt, 71 uint64_t vpn, u_int64_t ptebit); 72 static void mphyp_pte_unset(mmu_t, uintptr_t pt, struct lpte *pvo_pt, 73 uint64_t vpn); 74 static void mphyp_pte_change(mmu_t, uintptr_t pt, struct lpte *pvo_pt, 75 uint64_t vpn); 76 static int mphyp_pte_insert(mmu_t, u_int ptegidx, struct lpte *pvo_pt); 77 static uintptr_t mphyp_pvo_to_pte(mmu_t, const struct pvo_entry *pvo); 78 79 #define VSID_HASH_MASK 0x0000007fffffffffULL 80 81 82 static mmu_method_t mphyp_methods[] = { 83 MMUMETHOD(mmu_bootstrap, mphyp_bootstrap), 84 MMUMETHOD(mmu_cpu_bootstrap, mphyp_cpu_bootstrap), 85 86 MMUMETHOD(moea64_pte_synch, mphyp_pte_synch), 87 MMUMETHOD(moea64_pte_clear, mphyp_pte_clear), 88 MMUMETHOD(moea64_pte_unset, mphyp_pte_unset), 89 MMUMETHOD(moea64_pte_change, mphyp_pte_change), 90 MMUMETHOD(moea64_pte_insert, mphyp_pte_insert), 91 MMUMETHOD(moea64_pvo_to_pte, mphyp_pvo_to_pte), 92 93 { 0, 0 } 94 }; 95 96 MMU_DEF_INHERIT(pseries_mmu, "mmu_phyp", mphyp_methods, 0, oea64_mmu); 97 98 static void 99 mphyp_bootstrap(mmu_t mmup, vm_offset_t kernelstart, vm_offset_t kernelend) 100 { 101 uint64_t final_pteg_count = 0; 102 char buf[8]; 103 uint32_t prop[2]; 104 uint32_t nptlp, shift = 0, slb_encoding = 0; 105 phandle_t dev, node, root; 106 int idx, len, res; 107 108 moea64_early_bootstrap(mmup, kernelstart, kernelend); 109 110 root = OF_peer(0); 111 112 dev = OF_child(root); 113 while (dev != 0) { 114 res = OF_getprop(dev, "name", buf, sizeof(buf)); 115 if (res > 0 && strcmp(buf, "cpus") == 0) 116 break; 117 dev = OF_peer(dev); 118 } 119 120 node = OF_child(dev); 121 122 while (node != 0) { 123 res = OF_getprop(node, "device_type", buf, sizeof(buf)); 124 if (res > 0 && strcmp(buf, "cpu") == 0) 125 break; 126 node = OF_peer(node); 127 } 128 129 res = OF_getprop(node, "ibm,pft-size", prop, sizeof(prop)); 130 if (res <= 0) 131 panic("mmu_phyp: unknown PFT size"); 132 final_pteg_count = 1 << prop[1]; 133 res = OF_getprop(node, "ibm,slb-size", prop, sizeof(prop[0])); 134 if (res > 0) 135 n_slbs = prop[0]; 136 137 moea64_pteg_count = final_pteg_count / sizeof(struct lpteg); 138 139 /* 140 * Scan the large page size property for PAPR compatible machines. 141 * See PAPR D.5 Changes to Section 5.1.4, 'CPU Node Properties' 142 * for the encoding of the property. 143 */ 144 145 len = OF_getproplen(node, "ibm,segment-page-sizes"); 146 if (len > 0) { 147 /* 148 * We have to use a variable length array on the stack 149 * since we have very limited stack space. 150 */ 151 cell_t arr[len/sizeof(cell_t)]; 152 res = OF_getprop(node, "ibm,segment-page-sizes", &arr, 153 sizeof(arr)); 154 len /= 4; 155 idx = 0; 156 while (len > 0) { 157 shift = arr[idx]; 158 slb_encoding = arr[idx + 1]; 159 nptlp = arr[idx + 2]; 160 idx += 3; 161 len -= 3; 162 while (len > 0 && nptlp) { 163 idx += 2; 164 len -= 2; 165 nptlp--; 166 } 167 } 168 moea64_large_page_shift = shift; 169 moea64_large_page_size = 1ULL << shift; 170 } 171 172 moea64_mid_bootstrap(mmup, kernelstart, kernelend); 173 moea64_late_bootstrap(mmup, kernelstart, kernelend); 174 } 175 176 static void 177 mphyp_cpu_bootstrap(mmu_t mmup, int ap) 178 { 179 struct slb *slb = PCPU_GET(slb); 180 register_t seg0; 181 int i; 182 183 /* 184 * Install kernel SLB entries 185 */ 186 187 __asm __volatile ("slbia"); 188 __asm __volatile ("slbmfee %0,%1; slbie %0;" : "=r"(seg0) : "r"(0)); 189 for (i = 0; i < 64; i++) { 190 if (!(slb[i].slbe & SLBE_VALID)) 191 continue; 192 193 __asm __volatile ("slbmte %0, %1" :: 194 "r"(slb[i].slbv), "r"(slb[i].slbe)); 195 } 196 } 197 198 static void 199 mphyp_pte_synch(mmu_t mmu, uintptr_t slot, struct lpte *pvo_pt) 200 { 201 struct lpte pte; 202 uint64_t junk; 203 204 __asm __volatile("ptesync"); 205 phyp_pft_hcall(H_READ, 0, slot, 0, 0, &pte.pte_hi, &pte.pte_lo, 206 &junk); 207 208 pvo_pt->pte_lo |= pte.pte_lo & (LPTE_CHG | LPTE_REF); 209 } 210 211 static void 212 mphyp_pte_clear(mmu_t mmu, uintptr_t slot, struct lpte *pvo_pt, uint64_t vpn, 213 u_int64_t ptebit) 214 { 215 216 if (ptebit & LPTE_CHG) 217 phyp_hcall(H_CLEAR_MOD, 0, slot); 218 if (ptebit & LPTE_REF) 219 phyp_hcall(H_CLEAR_REF, 0, slot); 220 } 221 222 static void 223 mphyp_pte_unset(mmu_t mmu, uintptr_t slot, struct lpte *pvo_pt, uint64_t vpn) 224 { 225 struct lpte pte; 226 uint64_t junk; 227 int err; 228 229 err = phyp_pft_hcall(H_REMOVE, 1UL << 31, slot, 230 pvo_pt->pte_hi & LPTE_AVPN_MASK, 0, &pte.pte_hi, &pte.pte_lo, 231 &junk); 232 KASSERT(err == H_SUCCESS, ("Error removing page: %d", err)); 233 234 pvo_pt->pte_lo |= pte.pte_lo & (LPTE_CHG | LPTE_REF); 235 } 236 237 static void 238 mphyp_pte_change(mmu_t mmu, uintptr_t slot, struct lpte *pvo_pt, uint64_t vpn) 239 { 240 struct lpte evicted; 241 uint64_t index, junk; 242 int64_t result; 243 244 /* 245 * NB: this is protected by the global table lock, so this two-step 246 * is safe, except for the scratch-page case. No CPUs on which we run 247 * this code should be using scratch pages. 248 */ 249 KASSERT(!(pvo_pt->pte_hi & LPTE_LOCKED), 250 ("Locked pages not supported on PHYP")); 251 252 /* XXX: optimization using H_PROTECT for common case? */ 253 mphyp_pte_unset(mmu, slot, pvo_pt, vpn); 254 result = phyp_pft_hcall(H_ENTER, H_EXACT, slot, pvo_pt->pte_hi, 255 pvo_pt->pte_lo, &index, &evicted.pte_lo, &junk); 256 if (result != H_SUCCESS) 257 panic("mphyp_pte_change() insertion failure: %ld\n", result); 258 } 259 260 static __inline int 261 mphyp_pte_spillable_ident(u_int ptegidx, struct lpte *to_evict) 262 { 263 uint64_t slot, junk, k; 264 struct lpte pt; 265 int i, j; 266 267 /* Start at a random slot */ 268 i = mftb() % 8; 269 k = -1; 270 for (j = 0; j < 8; j++) { 271 slot = (ptegidx << 3) + (i + j) % 8; 272 phyp_pft_hcall(H_READ, 0, slot, 0, 0, &pt.pte_hi, &pt.pte_lo, 273 &junk); 274 275 if (pt.pte_hi & LPTE_SWBITS) 276 continue; 277 278 /* This is a candidate, so remember it */ 279 k = slot; 280 281 /* Try to get a page that has not been used lately */ 282 if (!(pt.pte_lo & LPTE_REF)) { 283 memcpy(to_evict, &pt, sizeof(struct lpte)); 284 return (k); 285 } 286 } 287 288 phyp_pft_hcall(H_READ, 0, slot, 0, 0, &to_evict->pte_hi, 289 &to_evict->pte_lo, &junk); 290 return (k); 291 } 292 293 static int 294 mphyp_pte_insert(mmu_t mmu, u_int ptegidx, struct lpte *pvo_pt) 295 { 296 int64_t result; 297 struct lpte evicted; 298 struct pvo_entry *pvo; 299 uint64_t index, junk; 300 u_int pteg_bktidx; 301 302 /* Check for locked pages, which we can't support on this system */ 303 KASSERT(!(pvo_pt->pte_hi & LPTE_LOCKED), 304 ("Locked pages not supported on PHYP")); 305 306 /* Initialize PTE */ 307 pvo_pt->pte_hi |= LPTE_VALID; 308 pvo_pt->pte_hi &= ~LPTE_HID; 309 evicted.pte_hi = 0; 310 311 /* 312 * First try primary hash. 313 */ 314 pteg_bktidx = ptegidx; 315 result = phyp_pft_hcall(H_ENTER, 0, pteg_bktidx << 3, pvo_pt->pte_hi, 316 pvo_pt->pte_lo, &index, &evicted.pte_lo, &junk); 317 if (result == H_SUCCESS) 318 return (index & 0x07); 319 KASSERT(result == H_PTEG_FULL, ("Page insertion error: %ld " 320 "(ptegidx: %#x/%#x, PTE %#lx/%#lx", result, ptegidx, 321 moea64_pteg_count, pvo_pt->pte_hi, pvo_pt->pte_lo)); 322 323 /* 324 * Next try secondary hash. 325 */ 326 pteg_bktidx ^= moea64_pteg_mask; 327 pvo_pt->pte_hi |= LPTE_HID; 328 result = phyp_pft_hcall(H_ENTER, 0, pteg_bktidx << 3, 329 pvo_pt->pte_hi, pvo_pt->pte_lo, &index, &evicted.pte_lo, &junk); 330 if (result == H_SUCCESS) 331 return (index & 0x07); 332 KASSERT(result == H_PTEG_FULL, ("Secondary page insertion error: %ld", 333 result)); 334 335 /* 336 * Out of luck. Find a PTE to sacrifice. 337 */ 338 pteg_bktidx = ptegidx; 339 index = mphyp_pte_spillable_ident(pteg_bktidx, &evicted); 340 if (index == -1L) { 341 pteg_bktidx ^= moea64_pteg_mask; 342 index = mphyp_pte_spillable_ident(pteg_bktidx, &evicted); 343 } 344 345 if (index == -1L) { 346 /* No freeable slots in either PTEG? We're hosed. */ 347 panic("mphyp_pte_insert: overflow"); 348 return (-1); 349 } 350 351 if (pteg_bktidx == ptegidx) 352 pvo_pt->pte_hi &= ~LPTE_HID; 353 else 354 pvo_pt->pte_hi |= LPTE_HID; 355 356 /* 357 * Synchronize the sacrifice PTE with its PVO, then mark both 358 * invalid. The PVO will be reused when/if the VM system comes 359 * here after a fault. 360 */ 361 362 if (evicted.pte_hi & LPTE_HID) 363 pteg_bktidx ^= moea64_pteg_mask; /* PTEs indexed by primary */ 364 365 LIST_FOREACH(pvo, &moea64_pvo_table[pteg_bktidx], pvo_olink) { 366 if (pvo->pvo_pte.lpte.pte_hi == evicted.pte_hi) { 367 KASSERT(pvo->pvo_pte.lpte.pte_hi & LPTE_VALID, 368 ("Invalid PVO for valid PTE!")); 369 mphyp_pte_unset(mmu, index, &pvo->pvo_pte.lpte, 370 pvo->pvo_vpn); 371 PVO_PTEGIDX_CLR(pvo); 372 moea64_pte_overflow++; 373 break; 374 } 375 } 376 377 KASSERT(pvo->pvo_pte.lpte.pte_hi == evicted.pte_hi, 378 ("Unable to find PVO for spilled PTE")); 379 380 /* 381 * Set the new PTE. 382 */ 383 result = phyp_pft_hcall(H_ENTER, H_EXACT, index, pvo_pt->pte_hi, 384 pvo_pt->pte_lo, &index, &evicted.pte_lo, &junk); 385 if (result == H_SUCCESS) 386 return (index & 0x07); 387 388 panic("Page replacement error: %ld", result); 389 return (-1); 390 } 391 392 static __inline u_int 393 va_to_pteg(uint64_t vsid, vm_offset_t addr, int large) 394 { 395 uint64_t hash; 396 int shift; 397 398 shift = large ? moea64_large_page_shift : ADDR_PIDX_SHFT; 399 hash = (vsid & VSID_HASH_MASK) ^ (((uint64_t)addr & ADDR_PIDX) >> 400 shift); 401 return (hash & moea64_pteg_mask); 402 } 403 404 static uintptr_t 405 mphyp_pvo_to_pte(mmu_t mmu, const struct pvo_entry *pvo) 406 { 407 uint64_t vsid; 408 u_int ptegidx; 409 410 /* If the PTEG index is not set, then there is no page table entry */ 411 if (!PVO_PTEGIDX_ISSET(pvo)) 412 return (-1); 413 414 vsid = PVO_VSID(pvo); 415 ptegidx = va_to_pteg(vsid, PVO_VADDR(pvo), pvo->pvo_vaddr & PVO_LARGE); 416 417 /* 418 * We can find the actual pte entry without searching by grabbing 419 * the PTEG index from 3 unused bits in pvo_vaddr and by 420 * noticing the HID bit. 421 */ 422 if (pvo->pvo_pte.lpte.pte_hi & LPTE_HID) 423 ptegidx ^= moea64_pteg_mask; 424 425 return ((ptegidx << 3) | PVO_PTEGIDX_GET(pvo)); 426 } 427 428