Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved. |
| 3 | * |
| 4 | * Author: Yu Liu, yu.liu@freescale.com |
| 5 | * Scott Wood, scottwood@freescale.com |
| 6 | * Ashish Kalra, ashish.kalra@freescale.com |
| 7 | * Varun Sethi, varun.sethi@freescale.com |
| 8 | * Alexander Graf, agraf@suse.de |
| 9 | * |
| 10 | * Description: |
| 11 | * This file is based on arch/powerpc/kvm/44x_tlb.c, |
| 12 | * by Hollis Blanchard <hollisb@us.ibm.com>. |
| 13 | * |
| 14 | * This program is free software; you can redistribute it and/or modify |
| 15 | * it under the terms of the GNU General Public License, version 2, as |
| 16 | * published by the Free Software Foundation. |
| 17 | */ |
| 18 | |
| 19 | #include <linux/kernel.h> |
| 20 | #include <linux/types.h> |
| 21 | #include <linux/slab.h> |
| 22 | #include <linux/string.h> |
| 23 | #include <linux/kvm.h> |
| 24 | #include <linux/kvm_host.h> |
| 25 | #include <linux/highmem.h> |
| 26 | #include <linux/log2.h> |
| 27 | #include <linux/uaccess.h> |
| 28 | #include <linux/sched.h> |
| 29 | #include <linux/rwsem.h> |
| 30 | #include <linux/vmalloc.h> |
| 31 | #include <linux/hugetlb.h> |
| 32 | #include <asm/kvm_ppc.h> |
| 33 | |
| 34 | #include "e500.h" |
| 35 | #include "trace.h" |
| 36 | #include "timing.h" |
| 37 | #include "e500_mmu_host.h" |
| 38 | |
| 39 | #define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1) |
| 40 | |
| 41 | static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM]; |
| 42 | |
| 43 | static inline unsigned int tlb1_max_shadow_size(void) |
| 44 | { |
| 45 | /* reserve one entry for magic page */ |
| 46 | return host_tlb_params[1].entries - tlbcam_index - 1; |
| 47 | } |
| 48 | |
| 49 | static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode) |
| 50 | { |
| 51 | /* Mask off reserved bits. */ |
| 52 | mas3 &= MAS3_ATTRIB_MASK; |
| 53 | |
| 54 | #ifndef CONFIG_KVM_BOOKE_HV |
| 55 | if (!usermode) { |
| 56 | /* Guest is in supervisor mode, |
| 57 | * so we need to translate guest |
| 58 | * supervisor permissions into user permissions. */ |
| 59 | mas3 &= ~E500_TLB_USER_PERM_MASK; |
| 60 | mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1; |
| 61 | } |
| 62 | mas3 |= E500_TLB_SUPER_PERM_MASK; |
| 63 | #endif |
| 64 | return mas3; |
| 65 | } |
| 66 | |
| 67 | static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode) |
| 68 | { |
| 69 | #ifdef CONFIG_SMP |
| 70 | return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M; |
| 71 | #else |
| 72 | return mas2 & MAS2_ATTRIB_MASK; |
| 73 | #endif |
| 74 | } |
| 75 | |
| 76 | /* |
| 77 | * writing shadow tlb entry to host TLB |
| 78 | */ |
| 79 | static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe, |
| 80 | uint32_t mas0) |
| 81 | { |
| 82 | unsigned long flags; |
| 83 | |
| 84 | local_irq_save(flags); |
| 85 | mtspr(SPRN_MAS0, mas0); |
| 86 | mtspr(SPRN_MAS1, stlbe->mas1); |
| 87 | mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2); |
| 88 | mtspr(SPRN_MAS3, (u32)stlbe->mas7_3); |
| 89 | mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32)); |
| 90 | #ifdef CONFIG_KVM_BOOKE_HV |
| 91 | mtspr(SPRN_MAS8, stlbe->mas8); |
| 92 | #endif |
| 93 | asm volatile("isync; tlbwe" : : : "memory"); |
| 94 | |
| 95 | #ifdef CONFIG_KVM_BOOKE_HV |
| 96 | /* Must clear mas8 for other host tlbwe's */ |
| 97 | mtspr(SPRN_MAS8, 0); |
| 98 | isync(); |
| 99 | #endif |
| 100 | local_irq_restore(flags); |
| 101 | |
| 102 | trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1, |
| 103 | stlbe->mas2, stlbe->mas7_3); |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * Acquire a mas0 with victim hint, as if we just took a TLB miss. |
| 108 | * |
| 109 | * We don't care about the address we're searching for, other than that it's |
| 110 | * in the right set and is not present in the TLB. Using a zero PID and a |
| 111 | * userspace address means we don't have to set and then restore MAS5, or |
| 112 | * calculate a proper MAS6 value. |
| 113 | */ |
| 114 | static u32 get_host_mas0(unsigned long eaddr) |
| 115 | { |
| 116 | unsigned long flags; |
| 117 | u32 mas0; |
| 118 | |
| 119 | local_irq_save(flags); |
| 120 | mtspr(SPRN_MAS6, 0); |
| 121 | asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET)); |
| 122 | mas0 = mfspr(SPRN_MAS0); |
| 123 | local_irq_restore(flags); |
| 124 | |
| 125 | return mas0; |
| 126 | } |
| 127 | |
| 128 | /* sesel is for tlb1 only */ |
| 129 | static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500, |
| 130 | int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe) |
| 131 | { |
| 132 | u32 mas0; |
| 133 | |
| 134 | if (tlbsel == 0) { |
| 135 | mas0 = get_host_mas0(stlbe->mas2); |
| 136 | __write_host_tlbe(stlbe, mas0); |
| 137 | } else { |
| 138 | __write_host_tlbe(stlbe, |
| 139 | MAS0_TLBSEL(1) | |
| 140 | MAS0_ESEL(to_htlb1_esel(sesel))); |
| 141 | } |
| 142 | } |
| 143 | |
| 144 | /* sesel is for tlb1 only */ |
| 145 | static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500, |
| 146 | struct kvm_book3e_206_tlb_entry *gtlbe, |
| 147 | struct kvm_book3e_206_tlb_entry *stlbe, |
| 148 | int stlbsel, int sesel) |
| 149 | { |
| 150 | int stid; |
| 151 | |
| 152 | preempt_disable(); |
| 153 | stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe); |
| 154 | |
| 155 | stlbe->mas1 |= MAS1_TID(stid); |
| 156 | write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe); |
| 157 | preempt_enable(); |
| 158 | } |
| 159 | |
| 160 | #ifdef CONFIG_KVM_E500V2 |
| 161 | /* XXX should be a hook in the gva2hpa translation */ |
| 162 | void kvmppc_map_magic(struct kvm_vcpu *vcpu) |
| 163 | { |
| 164 | struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu); |
| 165 | struct kvm_book3e_206_tlb_entry magic; |
| 166 | ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK; |
| 167 | unsigned int stid; |
| 168 | pfn_t pfn; |
| 169 | |
| 170 | pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT; |
| 171 | get_page(pfn_to_page(pfn)); |
| 172 | |
| 173 | preempt_disable(); |
| 174 | stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0); |
| 175 | |
| 176 | magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) | |
| 177 | MAS1_TSIZE(BOOK3E_PAGESZ_4K); |
| 178 | magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M; |
| 179 | magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) | |
| 180 | MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR; |
| 181 | magic.mas8 = 0; |
| 182 | |
| 183 | __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index)); |
| 184 | preempt_enable(); |
| 185 | } |
| 186 | #endif |
| 187 | |
| 188 | void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel, |
| 189 | int esel) |
| 190 | { |
| 191 | struct kvm_book3e_206_tlb_entry *gtlbe = |
| 192 | get_entry(vcpu_e500, tlbsel, esel); |
| 193 | struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref; |
| 194 | |
| 195 | /* Don't bother with unmapped entries */ |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 196 | if (!(ref->flags & E500_TLB_VALID)) { |
| 197 | WARN(ref->flags & (E500_TLB_BITMAP | E500_TLB_TLB0), |
| 198 | "%s: flags %x\n", __func__, ref->flags); |
| 199 | WARN_ON(tlbsel == 1 && vcpu_e500->g2h_tlb1_map[esel]); |
| 200 | } |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 201 | |
| 202 | if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) { |
| 203 | u64 tmp = vcpu_e500->g2h_tlb1_map[esel]; |
| 204 | int hw_tlb_indx; |
| 205 | unsigned long flags; |
| 206 | |
| 207 | local_irq_save(flags); |
| 208 | while (tmp) { |
| 209 | hw_tlb_indx = __ilog2_u64(tmp & -tmp); |
| 210 | mtspr(SPRN_MAS0, |
| 211 | MAS0_TLBSEL(1) | |
| 212 | MAS0_ESEL(to_htlb1_esel(hw_tlb_indx))); |
| 213 | mtspr(SPRN_MAS1, 0); |
| 214 | asm volatile("tlbwe"); |
| 215 | vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0; |
| 216 | tmp &= tmp - 1; |
| 217 | } |
| 218 | mb(); |
| 219 | vcpu_e500->g2h_tlb1_map[esel] = 0; |
| 220 | ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID); |
| 221 | local_irq_restore(flags); |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 222 | } |
| 223 | |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 224 | if (tlbsel == 1 && ref->flags & E500_TLB_TLB0) { |
| 225 | /* |
| 226 | * TLB1 entry is backed by 4k pages. This should happen |
| 227 | * rarely and is not worth optimizing. Invalidate everything. |
| 228 | */ |
| 229 | kvmppc_e500_tlbil_all(vcpu_e500); |
| 230 | ref->flags &= ~(E500_TLB_TLB0 | E500_TLB_VALID); |
| 231 | } |
| 232 | |
| 233 | /* Already invalidated in between */ |
| 234 | if (!(ref->flags & E500_TLB_VALID)) |
| 235 | return; |
| 236 | |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 237 | /* Guest tlbe is backed by at most one host tlbe per shadow pid. */ |
| 238 | kvmppc_e500_tlbil_one(vcpu_e500, gtlbe); |
| 239 | |
| 240 | /* Mark the TLB as not backed by the host anymore */ |
| 241 | ref->flags &= ~E500_TLB_VALID; |
| 242 | } |
| 243 | |
| 244 | static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe) |
| 245 | { |
| 246 | return tlbe->mas7_3 & (MAS3_SW|MAS3_UW); |
| 247 | } |
| 248 | |
| 249 | static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref, |
| 250 | struct kvm_book3e_206_tlb_entry *gtlbe, |
| 251 | pfn_t pfn) |
| 252 | { |
| 253 | ref->pfn = pfn; |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 254 | ref->flags |= E500_TLB_VALID; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 255 | |
| 256 | if (tlbe_is_writable(gtlbe)) |
| 257 | kvm_set_pfn_dirty(pfn); |
| 258 | } |
| 259 | |
| 260 | static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref) |
| 261 | { |
| 262 | if (ref->flags & E500_TLB_VALID) { |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 263 | /* FIXME: don't log bogus pfn for TLB1 */ |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 264 | trace_kvm_booke206_ref_release(ref->pfn, ref->flags); |
| 265 | ref->flags = 0; |
| 266 | } |
| 267 | } |
| 268 | |
Alexander Graf | 483ba97 | 2013-01-18 15:13:19 +0100 | [diff] [blame] | 269 | static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500) |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 270 | { |
| 271 | if (vcpu_e500->g2h_tlb1_map) |
| 272 | memset(vcpu_e500->g2h_tlb1_map, 0, |
| 273 | sizeof(u64) * vcpu_e500->gtlb_params[1].entries); |
| 274 | if (vcpu_e500->h2g_tlb1_rmap) |
| 275 | memset(vcpu_e500->h2g_tlb1_rmap, 0, |
| 276 | sizeof(unsigned int) * host_tlb_params[1].entries); |
| 277 | } |
| 278 | |
| 279 | static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500) |
| 280 | { |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 281 | int tlbsel; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 282 | int i; |
| 283 | |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 284 | for (tlbsel = 0; tlbsel <= 1; tlbsel++) { |
| 285 | for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) { |
| 286 | struct tlbe_ref *ref = |
| 287 | &vcpu_e500->gtlb_priv[tlbsel][i].ref; |
| 288 | kvmppc_e500_ref_release(ref); |
| 289 | } |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 290 | } |
| 291 | } |
| 292 | |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 293 | void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu) |
| 294 | { |
| 295 | struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu); |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 296 | kvmppc_e500_tlbil_all(vcpu_e500); |
| 297 | clear_tlb_privs(vcpu_e500); |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 298 | clear_tlb1_bitmap(vcpu_e500); |
| 299 | } |
| 300 | |
| 301 | /* TID must be supplied by the caller */ |
| 302 | static void kvmppc_e500_setup_stlbe( |
| 303 | struct kvm_vcpu *vcpu, |
| 304 | struct kvm_book3e_206_tlb_entry *gtlbe, |
| 305 | int tsize, struct tlbe_ref *ref, u64 gvaddr, |
| 306 | struct kvm_book3e_206_tlb_entry *stlbe) |
| 307 | { |
| 308 | pfn_t pfn = ref->pfn; |
| 309 | u32 pr = vcpu->arch.shared->msr & MSR_PR; |
| 310 | |
| 311 | BUG_ON(!(ref->flags & E500_TLB_VALID)); |
| 312 | |
| 313 | /* Force IPROT=0 for all guest mappings. */ |
| 314 | stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID; |
| 315 | stlbe->mas2 = (gvaddr & MAS2_EPN) | |
| 316 | e500_shadow_mas2_attrib(gtlbe->mas2, pr); |
| 317 | stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) | |
| 318 | e500_shadow_mas3_attrib(gtlbe->mas7_3, pr); |
| 319 | |
| 320 | #ifdef CONFIG_KVM_BOOKE_HV |
| 321 | stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid; |
| 322 | #endif |
| 323 | } |
| 324 | |
| 325 | static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500, |
| 326 | u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe, |
| 327 | int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe, |
| 328 | struct tlbe_ref *ref) |
| 329 | { |
| 330 | struct kvm_memory_slot *slot; |
| 331 | unsigned long pfn = 0; /* silence GCC warning */ |
| 332 | unsigned long hva; |
| 333 | int pfnmap = 0; |
| 334 | int tsize = BOOK3E_PAGESZ_4K; |
| 335 | |
| 336 | /* |
| 337 | * Translate guest physical to true physical, acquiring |
| 338 | * a page reference if it is normal, non-reserved memory. |
| 339 | * |
| 340 | * gfn_to_memslot() must succeed because otherwise we wouldn't |
| 341 | * have gotten this far. Eventually we should just pass the slot |
| 342 | * pointer through from the first lookup. |
| 343 | */ |
| 344 | slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn); |
| 345 | hva = gfn_to_hva_memslot(slot, gfn); |
| 346 | |
| 347 | if (tlbsel == 1) { |
| 348 | struct vm_area_struct *vma; |
| 349 | down_read(¤t->mm->mmap_sem); |
| 350 | |
| 351 | vma = find_vma(current->mm, hva); |
| 352 | if (vma && hva >= vma->vm_start && |
| 353 | (vma->vm_flags & VM_PFNMAP)) { |
| 354 | /* |
| 355 | * This VMA is a physically contiguous region (e.g. |
| 356 | * /dev/mem) that bypasses normal Linux page |
| 357 | * management. Find the overlap between the |
| 358 | * vma and the memslot. |
| 359 | */ |
| 360 | |
| 361 | unsigned long start, end; |
| 362 | unsigned long slot_start, slot_end; |
| 363 | |
| 364 | pfnmap = 1; |
| 365 | |
| 366 | start = vma->vm_pgoff; |
| 367 | end = start + |
| 368 | ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT); |
| 369 | |
| 370 | pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT); |
| 371 | |
| 372 | slot_start = pfn - (gfn - slot->base_gfn); |
| 373 | slot_end = slot_start + slot->npages; |
| 374 | |
| 375 | if (start < slot_start) |
| 376 | start = slot_start; |
| 377 | if (end > slot_end) |
| 378 | end = slot_end; |
| 379 | |
| 380 | tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >> |
| 381 | MAS1_TSIZE_SHIFT; |
| 382 | |
| 383 | /* |
| 384 | * e500 doesn't implement the lowest tsize bit, |
| 385 | * or 1K pages. |
| 386 | */ |
| 387 | tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1); |
| 388 | |
| 389 | /* |
| 390 | * Now find the largest tsize (up to what the guest |
| 391 | * requested) that will cover gfn, stay within the |
| 392 | * range, and for which gfn and pfn are mutually |
| 393 | * aligned. |
| 394 | */ |
| 395 | |
| 396 | for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) { |
| 397 | unsigned long gfn_start, gfn_end, tsize_pages; |
| 398 | tsize_pages = 1 << (tsize - 2); |
| 399 | |
| 400 | gfn_start = gfn & ~(tsize_pages - 1); |
| 401 | gfn_end = gfn_start + tsize_pages; |
| 402 | |
| 403 | if (gfn_start + pfn - gfn < start) |
| 404 | continue; |
| 405 | if (gfn_end + pfn - gfn > end) |
| 406 | continue; |
| 407 | if ((gfn & (tsize_pages - 1)) != |
| 408 | (pfn & (tsize_pages - 1))) |
| 409 | continue; |
| 410 | |
| 411 | gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1); |
| 412 | pfn &= ~(tsize_pages - 1); |
| 413 | break; |
| 414 | } |
| 415 | } else if (vma && hva >= vma->vm_start && |
| 416 | (vma->vm_flags & VM_HUGETLB)) { |
| 417 | unsigned long psize = vma_kernel_pagesize(vma); |
| 418 | |
| 419 | tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >> |
| 420 | MAS1_TSIZE_SHIFT; |
| 421 | |
| 422 | /* |
| 423 | * Take the largest page size that satisfies both host |
| 424 | * and guest mapping |
| 425 | */ |
| 426 | tsize = min(__ilog2(psize) - 10, tsize); |
| 427 | |
| 428 | /* |
| 429 | * e500 doesn't implement the lowest tsize bit, |
| 430 | * or 1K pages. |
| 431 | */ |
| 432 | tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1); |
| 433 | } |
| 434 | |
| 435 | up_read(¤t->mm->mmap_sem); |
| 436 | } |
| 437 | |
| 438 | if (likely(!pfnmap)) { |
| 439 | unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT); |
| 440 | pfn = gfn_to_pfn_memslot(slot, gfn); |
| 441 | if (is_error_noslot_pfn(pfn)) { |
| 442 | printk(KERN_ERR "Couldn't get real page for gfn %lx!\n", |
| 443 | (long)gfn); |
| 444 | return -EINVAL; |
| 445 | } |
| 446 | |
| 447 | /* Align guest and physical address to page map boundaries */ |
| 448 | pfn &= ~(tsize_pages - 1); |
| 449 | gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1); |
| 450 | } |
| 451 | |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 452 | kvmppc_e500_ref_setup(ref, gtlbe, pfn); |
| 453 | |
| 454 | kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize, |
| 455 | ref, gvaddr, stlbe); |
| 456 | |
| 457 | /* Clear i-cache for new pages */ |
| 458 | kvmppc_mmu_flush_icache(pfn); |
| 459 | |
| 460 | /* Drop refcount on page, so that mmu notifiers can clear it */ |
| 461 | kvm_release_pfn_clean(pfn); |
| 462 | |
| 463 | return 0; |
| 464 | } |
| 465 | |
| 466 | /* XXX only map the one-one case, for now use TLB0 */ |
| 467 | static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500, int esel, |
| 468 | struct kvm_book3e_206_tlb_entry *stlbe) |
| 469 | { |
| 470 | struct kvm_book3e_206_tlb_entry *gtlbe; |
| 471 | struct tlbe_ref *ref; |
| 472 | int stlbsel = 0; |
| 473 | int sesel = 0; |
| 474 | int r; |
| 475 | |
| 476 | gtlbe = get_entry(vcpu_e500, 0, esel); |
| 477 | ref = &vcpu_e500->gtlb_priv[0][esel].ref; |
| 478 | |
| 479 | r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe), |
| 480 | get_tlb_raddr(gtlbe) >> PAGE_SHIFT, |
| 481 | gtlbe, 0, stlbe, ref); |
| 482 | if (r) |
| 483 | return r; |
| 484 | |
| 485 | write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel); |
| 486 | |
| 487 | return 0; |
| 488 | } |
| 489 | |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 490 | static int kvmppc_e500_tlb1_map_tlb1(struct kvmppc_vcpu_e500 *vcpu_e500, |
| 491 | struct tlbe_ref *ref, |
| 492 | int esel) |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 493 | { |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 494 | unsigned int sesel = vcpu_e500->host_tlb1_nv++; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 495 | |
| 496 | if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size())) |
| 497 | vcpu_e500->host_tlb1_nv = 0; |
| 498 | |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 499 | if (vcpu_e500->h2g_tlb1_rmap[sesel]) { |
Scott Wood | 6b2ba1a | 2013-02-13 19:37:48 +0000 | [diff] [blame] | 500 | unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel] - 1; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 501 | vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel); |
| 502 | } |
Scott Wood | 66a5fec | 2013-02-13 19:37:49 +0000 | [diff] [blame] | 503 | |
Scott Wood | 66a5fec | 2013-02-13 19:37:49 +0000 | [diff] [blame] | 504 | vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP; |
| 505 | vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel; |
Scott Wood | 6b2ba1a | 2013-02-13 19:37:48 +0000 | [diff] [blame] | 506 | vcpu_e500->h2g_tlb1_rmap[sesel] = esel + 1; |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 507 | WARN_ON(!(ref->flags & E500_TLB_VALID)); |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 508 | |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 509 | return sesel; |
| 510 | } |
| 511 | |
| 512 | /* Caller must ensure that the specified guest TLB entry is safe to insert into |
| 513 | * the shadow TLB. */ |
| 514 | /* For both one-one and one-to-many */ |
| 515 | static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500, |
| 516 | u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe, |
| 517 | struct kvm_book3e_206_tlb_entry *stlbe, int esel) |
| 518 | { |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 519 | struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[1][esel].ref; |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 520 | int sesel; |
| 521 | int r; |
| 522 | |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 523 | r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe, |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 524 | ref); |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 525 | if (r) |
| 526 | return r; |
| 527 | |
| 528 | /* Use TLB0 when we can only map a page with 4k */ |
| 529 | if (get_tlb_tsize(stlbe) == BOOK3E_PAGESZ_4K) { |
| 530 | vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_TLB0; |
| 531 | write_stlbe(vcpu_e500, gtlbe, stlbe, 0, 0); |
| 532 | return 0; |
| 533 | } |
| 534 | |
| 535 | /* Otherwise map into TLB1 */ |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 536 | sesel = kvmppc_e500_tlb1_map_tlb1(vcpu_e500, ref, esel); |
Alexander Graf | c015c62 | 2013-01-17 17:54:36 +0100 | [diff] [blame] | 537 | write_stlbe(vcpu_e500, gtlbe, stlbe, 1, sesel); |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 538 | |
| 539 | return 0; |
| 540 | } |
| 541 | |
| 542 | void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr, |
| 543 | unsigned int index) |
| 544 | { |
| 545 | struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu); |
| 546 | struct tlbe_priv *priv; |
| 547 | struct kvm_book3e_206_tlb_entry *gtlbe, stlbe; |
| 548 | int tlbsel = tlbsel_of(index); |
| 549 | int esel = esel_of(index); |
| 550 | |
| 551 | gtlbe = get_entry(vcpu_e500, tlbsel, esel); |
| 552 | |
| 553 | switch (tlbsel) { |
| 554 | case 0: |
| 555 | priv = &vcpu_e500->gtlb_priv[tlbsel][esel]; |
| 556 | |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 557 | /* Triggers after clear_tlb_privs or on initial mapping */ |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 558 | if (!(priv->ref.flags & E500_TLB_VALID)) { |
| 559 | kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe); |
| 560 | } else { |
| 561 | kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K, |
| 562 | &priv->ref, eaddr, &stlbe); |
| 563 | write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0); |
| 564 | } |
| 565 | break; |
| 566 | |
| 567 | case 1: { |
| 568 | gfn_t gfn = gpaddr >> PAGE_SHIFT; |
| 569 | kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe, |
| 570 | esel); |
| 571 | break; |
| 572 | } |
| 573 | |
| 574 | default: |
| 575 | BUG(); |
| 576 | break; |
| 577 | } |
| 578 | } |
| 579 | |
| 580 | /************* MMU Notifiers *************/ |
| 581 | |
| 582 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) |
| 583 | { |
| 584 | trace_kvm_unmap_hva(hva); |
| 585 | |
| 586 | /* |
| 587 | * Flush all shadow tlb entries everywhere. This is slow, but |
| 588 | * we are 100% sure that we catch the to be unmapped page |
| 589 | */ |
| 590 | kvm_flush_remote_tlbs(kvm); |
| 591 | |
| 592 | return 0; |
| 593 | } |
| 594 | |
| 595 | int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end) |
| 596 | { |
| 597 | /* kvm_unmap_hva flushes everything anyways */ |
| 598 | kvm_unmap_hva(kvm, start); |
| 599 | |
| 600 | return 0; |
| 601 | } |
| 602 | |
| 603 | int kvm_age_hva(struct kvm *kvm, unsigned long hva) |
| 604 | { |
| 605 | /* XXX could be more clever ;) */ |
| 606 | return 0; |
| 607 | } |
| 608 | |
| 609 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) |
| 610 | { |
| 611 | /* XXX could be more clever ;) */ |
| 612 | return 0; |
| 613 | } |
| 614 | |
| 615 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) |
| 616 | { |
| 617 | /* The page will get remapped properly on its next fault */ |
| 618 | kvm_unmap_hva(kvm, hva); |
| 619 | } |
| 620 | |
| 621 | /*****************************************/ |
| 622 | |
| 623 | int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500) |
| 624 | { |
| 625 | host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY; |
| 626 | host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY; |
| 627 | |
| 628 | /* |
| 629 | * This should never happen on real e500 hardware, but is |
| 630 | * architecturally possible -- e.g. in some weird nested |
| 631 | * virtualization case. |
| 632 | */ |
| 633 | if (host_tlb_params[0].entries == 0 || |
| 634 | host_tlb_params[1].entries == 0) { |
| 635 | pr_err("%s: need to know host tlb size\n", __func__); |
| 636 | return -ENODEV; |
| 637 | } |
| 638 | |
| 639 | host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >> |
| 640 | TLBnCFG_ASSOC_SHIFT; |
| 641 | host_tlb_params[1].ways = host_tlb_params[1].entries; |
| 642 | |
| 643 | if (!is_power_of_2(host_tlb_params[0].entries) || |
| 644 | !is_power_of_2(host_tlb_params[0].ways) || |
| 645 | host_tlb_params[0].entries < host_tlb_params[0].ways || |
| 646 | host_tlb_params[0].ways == 0) { |
| 647 | pr_err("%s: bad tlb0 host config: %u entries %u ways\n", |
| 648 | __func__, host_tlb_params[0].entries, |
| 649 | host_tlb_params[0].ways); |
| 650 | return -ENODEV; |
| 651 | } |
| 652 | |
| 653 | host_tlb_params[0].sets = |
| 654 | host_tlb_params[0].entries / host_tlb_params[0].ways; |
| 655 | host_tlb_params[1].sets = 1; |
| 656 | |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 657 | vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) * |
| 658 | host_tlb_params[1].entries, |
| 659 | GFP_KERNEL); |
| 660 | if (!vcpu_e500->h2g_tlb1_rmap) |
Scott Wood | 4d2be6f | 2013-03-06 16:02:49 +0000 | [diff] [blame] | 661 | return -EINVAL; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 662 | |
| 663 | return 0; |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 664 | } |
| 665 | |
| 666 | void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500) |
| 667 | { |
| 668 | kfree(vcpu_e500->h2g_tlb1_rmap); |
Alexander Graf | b71c9e2 | 2013-01-11 15:22:45 +0100 | [diff] [blame] | 669 | } |