Linux Audio

Check our new training course

Loading...
   1/*
   2 * Copyright (C) 2008-2011 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 *
   9 * Description:
  10 * This file is based on arch/powerpc/kvm/44x_tlb.c,
  11 * by Hollis Blanchard <hollisb@us.ibm.com>.
  12 *
  13 * This program is free software; you can redistribute it and/or modify
  14 * it under the terms of the GNU General Public License, version 2, as
  15 * published by the Free Software Foundation.
  16 */
  17
  18#include <linux/kernel.h>
  19#include <linux/types.h>
  20#include <linux/slab.h>
  21#include <linux/string.h>
  22#include <linux/kvm.h>
  23#include <linux/kvm_host.h>
  24#include <linux/highmem.h>
  25#include <linux/log2.h>
  26#include <linux/uaccess.h>
  27#include <linux/sched.h>
  28#include <linux/rwsem.h>
  29#include <linux/vmalloc.h>
  30#include <linux/hugetlb.h>
  31#include <asm/kvm_ppc.h>
  32
  33#include "e500.h"
  34#include "trace.h"
  35#include "timing.h"
  36
  37#define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
  38
  39static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
  40
  41static inline unsigned int gtlb0_get_next_victim(
  42		struct kvmppc_vcpu_e500 *vcpu_e500)
  43{
  44	unsigned int victim;
  45
  46	victim = vcpu_e500->gtlb_nv[0]++;
  47	if (unlikely(vcpu_e500->gtlb_nv[0] >= vcpu_e500->gtlb_params[0].ways))
  48		vcpu_e500->gtlb_nv[0] = 0;
  49
  50	return victim;
  51}
  52
  53static inline unsigned int tlb1_max_shadow_size(void)
  54{
  55	/* reserve one entry for magic page */
  56	return host_tlb_params[1].entries - tlbcam_index - 1;
  57}
  58
  59static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
  60{
  61	return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
  62}
  63
  64static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
  65{
  66	/* Mask off reserved bits. */
  67	mas3 &= MAS3_ATTRIB_MASK;
  68
  69#ifndef CONFIG_KVM_BOOKE_HV
  70	if (!usermode) {
  71		/* Guest is in supervisor mode,
  72		 * so we need to translate guest
  73		 * supervisor permissions into user permissions. */
  74		mas3 &= ~E500_TLB_USER_PERM_MASK;
  75		mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
  76	}
  77	mas3 |= E500_TLB_SUPER_PERM_MASK;
  78#endif
  79	return mas3;
  80}
  81
  82static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
  83{
  84#ifdef CONFIG_SMP
  85	return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
  86#else
  87	return mas2 & MAS2_ATTRIB_MASK;
  88#endif
  89}
  90
  91/*
  92 * writing shadow tlb entry to host TLB
  93 */
  94static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
  95				     uint32_t mas0)
  96{
  97	unsigned long flags;
  98
  99	local_irq_save(flags);
 100	mtspr(SPRN_MAS0, mas0);
 101	mtspr(SPRN_MAS1, stlbe->mas1);
 102	mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
 103	mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
 104	mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
 105#ifdef CONFIG_KVM_BOOKE_HV
 106	mtspr(SPRN_MAS8, stlbe->mas8);
 107#endif
 108	asm volatile("isync; tlbwe" : : : "memory");
 109
 110#ifdef CONFIG_KVM_BOOKE_HV
 111	/* Must clear mas8 for other host tlbwe's */
 112	mtspr(SPRN_MAS8, 0);
 113	isync();
 114#endif
 115	local_irq_restore(flags);
 116
 117	trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
 118	                              stlbe->mas2, stlbe->mas7_3);
 119}
 120
 121/*
 122 * Acquire a mas0 with victim hint, as if we just took a TLB miss.
 123 *
 124 * We don't care about the address we're searching for, other than that it's
 125 * in the right set and is not present in the TLB.  Using a zero PID and a
 126 * userspace address means we don't have to set and then restore MAS5, or
 127 * calculate a proper MAS6 value.
 128 */
 129static u32 get_host_mas0(unsigned long eaddr)
 130{
 131	unsigned long flags;
 132	u32 mas0;
 133
 134	local_irq_save(flags);
 135	mtspr(SPRN_MAS6, 0);
 136	asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
 137	mas0 = mfspr(SPRN_MAS0);
 138	local_irq_restore(flags);
 139
 140	return mas0;
 141}
 142
 143/* sesel is for tlb1 only */
 144static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
 145		int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
 146{
 147	u32 mas0;
 148
 149	if (tlbsel == 0) {
 150		mas0 = get_host_mas0(stlbe->mas2);
 151		__write_host_tlbe(stlbe, mas0);
 152	} else {
 153		__write_host_tlbe(stlbe,
 154				  MAS0_TLBSEL(1) |
 155				  MAS0_ESEL(to_htlb1_esel(sesel)));
 156	}
 157}
 158
 159#ifdef CONFIG_KVM_E500V2
 160void kvmppc_map_magic(struct kvm_vcpu *vcpu)
 161{
 162	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 163	struct kvm_book3e_206_tlb_entry magic;
 164	ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
 165	unsigned int stid;
 166	pfn_t pfn;
 167
 168	pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
 169	get_page(pfn_to_page(pfn));
 170
 171	preempt_disable();
 172	stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
 173
 174	magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
 175		     MAS1_TSIZE(BOOK3E_PAGESZ_4K);
 176	magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
 177	magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
 178		       MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
 179	magic.mas8 = 0;
 180
 181	__write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
 182	preempt_enable();
 183}
 184#endif
 185
 186static void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500,
 187				int tlbsel, int esel)
 188{
 189	struct kvm_book3e_206_tlb_entry *gtlbe =
 190		get_entry(vcpu_e500, tlbsel, esel);
 191
 192	if (tlbsel == 1 &&
 193	    vcpu_e500->gtlb_priv[1][esel].ref.flags & E500_TLB_BITMAP) {
 194		u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
 195		int hw_tlb_indx;
 196		unsigned long flags;
 197
 198		local_irq_save(flags);
 199		while (tmp) {
 200			hw_tlb_indx = __ilog2_u64(tmp & -tmp);
 201			mtspr(SPRN_MAS0,
 202			      MAS0_TLBSEL(1) |
 203			      MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
 204			mtspr(SPRN_MAS1, 0);
 205			asm volatile("tlbwe");
 206			vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
 207			tmp &= tmp - 1;
 208		}
 209		mb();
 210		vcpu_e500->g2h_tlb1_map[esel] = 0;
 211		vcpu_e500->gtlb_priv[1][esel].ref.flags &= ~E500_TLB_BITMAP;
 212		local_irq_restore(flags);
 213
 214		return;
 215	}
 216
 217	/* Guest tlbe is backed by at most one host tlbe per shadow pid. */
 218	kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
 219}
 220
 221static int tlb0_set_base(gva_t addr, int sets, int ways)
 222{
 223	int set_base;
 224
 225	set_base = (addr >> PAGE_SHIFT) & (sets - 1);
 226	set_base *= ways;
 227
 228	return set_base;
 229}
 230
 231static int gtlb0_set_base(struct kvmppc_vcpu_e500 *vcpu_e500, gva_t addr)
 232{
 233	return tlb0_set_base(addr, vcpu_e500->gtlb_params[0].sets,
 234			     vcpu_e500->gtlb_params[0].ways);
 235}
 236
 237static unsigned int get_tlb_esel(struct kvm_vcpu *vcpu, int tlbsel)
 238{
 239	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 240	int esel = get_tlb_esel_bit(vcpu);
 241
 242	if (tlbsel == 0) {
 243		esel &= vcpu_e500->gtlb_params[0].ways - 1;
 244		esel += gtlb0_set_base(vcpu_e500, vcpu->arch.shared->mas2);
 245	} else {
 246		esel &= vcpu_e500->gtlb_params[tlbsel].entries - 1;
 247	}
 248
 249	return esel;
 250}
 251
 252/* Search the guest TLB for a matching entry. */
 253static int kvmppc_e500_tlb_index(struct kvmppc_vcpu_e500 *vcpu_e500,
 254		gva_t eaddr, int tlbsel, unsigned int pid, int as)
 255{
 256	int size = vcpu_e500->gtlb_params[tlbsel].entries;
 257	unsigned int set_base, offset;
 258	int i;
 259
 260	if (tlbsel == 0) {
 261		set_base = gtlb0_set_base(vcpu_e500, eaddr);
 262		size = vcpu_e500->gtlb_params[0].ways;
 263	} else {
 264		if (eaddr < vcpu_e500->tlb1_min_eaddr ||
 265				eaddr > vcpu_e500->tlb1_max_eaddr)
 266			return -1;
 267		set_base = 0;
 268	}
 269
 270	offset = vcpu_e500->gtlb_offset[tlbsel];
 271
 272	for (i = 0; i < size; i++) {
 273		struct kvm_book3e_206_tlb_entry *tlbe =
 274			&vcpu_e500->gtlb_arch[offset + set_base + i];
 275		unsigned int tid;
 276
 277		if (eaddr < get_tlb_eaddr(tlbe))
 278			continue;
 279
 280		if (eaddr > get_tlb_end(tlbe))
 281			continue;
 282
 283		tid = get_tlb_tid(tlbe);
 284		if (tid && (tid != pid))
 285			continue;
 286
 287		if (!get_tlb_v(tlbe))
 288			continue;
 289
 290		if (get_tlb_ts(tlbe) != as && as != -1)
 291			continue;
 292
 293		return set_base + i;
 294	}
 295
 296	return -1;
 297}
 298
 299static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
 300					 struct kvm_book3e_206_tlb_entry *gtlbe,
 301					 pfn_t pfn)
 302{
 303	ref->pfn = pfn;
 304	ref->flags = E500_TLB_VALID;
 305
 306	if (tlbe_is_writable(gtlbe))
 307		ref->flags |= E500_TLB_DIRTY;
 308}
 309
 310static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
 311{
 312	if (ref->flags & E500_TLB_VALID) {
 313		if (ref->flags & E500_TLB_DIRTY)
 314			kvm_release_pfn_dirty(ref->pfn);
 315		else
 316			kvm_release_pfn_clean(ref->pfn);
 317
 318		ref->flags = 0;
 319	}
 320}
 321
 322static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
 323{
 324	if (vcpu_e500->g2h_tlb1_map)
 325		memset(vcpu_e500->g2h_tlb1_map,
 326		       sizeof(u64) * vcpu_e500->gtlb_params[1].entries, 0);
 327	if (vcpu_e500->h2g_tlb1_rmap)
 328		memset(vcpu_e500->h2g_tlb1_rmap,
 329		       sizeof(unsigned int) * host_tlb_params[1].entries, 0);
 330}
 331
 332static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
 333{
 334	int tlbsel = 0;
 335	int i;
 336
 337	for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
 338		struct tlbe_ref *ref =
 339			&vcpu_e500->gtlb_priv[tlbsel][i].ref;
 340		kvmppc_e500_ref_release(ref);
 341	}
 342}
 343
 344static void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500)
 345{
 346	int stlbsel = 1;
 347	int i;
 348
 349	kvmppc_e500_tlbil_all(vcpu_e500);
 350
 351	for (i = 0; i < host_tlb_params[stlbsel].entries; i++) {
 352		struct tlbe_ref *ref =
 353			&vcpu_e500->tlb_refs[stlbsel][i];
 354		kvmppc_e500_ref_release(ref);
 355	}
 356
 357	clear_tlb_privs(vcpu_e500);
 358}
 359
 360static inline void kvmppc_e500_deliver_tlb_miss(struct kvm_vcpu *vcpu,
 361		unsigned int eaddr, int as)
 362{
 363	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 364	unsigned int victim, tsized;
 365	int tlbsel;
 366
 367	/* since we only have two TLBs, only lower bit is used. */
 368	tlbsel = (vcpu->arch.shared->mas4 >> 28) & 0x1;
 369	victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
 370	tsized = (vcpu->arch.shared->mas4 >> 7) & 0x1f;
 371
 372	vcpu->arch.shared->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(victim)
 373		| MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
 374	vcpu->arch.shared->mas1 = MAS1_VALID | (as ? MAS1_TS : 0)
 375		| MAS1_TID(get_tlbmiss_tid(vcpu))
 376		| MAS1_TSIZE(tsized);
 377	vcpu->arch.shared->mas2 = (eaddr & MAS2_EPN)
 378		| (vcpu->arch.shared->mas4 & MAS2_ATTRIB_MASK);
 379	vcpu->arch.shared->mas7_3 &= MAS3_U0 | MAS3_U1 | MAS3_U2 | MAS3_U3;
 380	vcpu->arch.shared->mas6 = (vcpu->arch.shared->mas6 & MAS6_SPID1)
 381		| (get_cur_pid(vcpu) << 16)
 382		| (as ? MAS6_SAS : 0);
 383}
 384
 385/* TID must be supplied by the caller */
 386static inline void kvmppc_e500_setup_stlbe(
 387	struct kvm_vcpu *vcpu,
 388	struct kvm_book3e_206_tlb_entry *gtlbe,
 389	int tsize, struct tlbe_ref *ref, u64 gvaddr,
 390	struct kvm_book3e_206_tlb_entry *stlbe)
 391{
 392	pfn_t pfn = ref->pfn;
 393	u32 pr = vcpu->arch.shared->msr & MSR_PR;
 394
 395	BUG_ON(!(ref->flags & E500_TLB_VALID));
 396
 397	/* Force IPROT=0 for all guest mappings. */
 398	stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
 399	stlbe->mas2 = (gvaddr & MAS2_EPN) |
 400		      e500_shadow_mas2_attrib(gtlbe->mas2, pr);
 401	stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
 402			e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
 403
 404#ifdef CONFIG_KVM_BOOKE_HV
 405	stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid;
 406#endif
 407}
 408
 409static inline void kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
 410	u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
 411	int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
 412	struct tlbe_ref *ref)
 413{
 414	struct kvm_memory_slot *slot;
 415	unsigned long pfn, hva;
 416	int pfnmap = 0;
 417	int tsize = BOOK3E_PAGESZ_4K;
 418
 419	/*
 420	 * Translate guest physical to true physical, acquiring
 421	 * a page reference if it is normal, non-reserved memory.
 422	 *
 423	 * gfn_to_memslot() must succeed because otherwise we wouldn't
 424	 * have gotten this far.  Eventually we should just pass the slot
 425	 * pointer through from the first lookup.
 426	 */
 427	slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
 428	hva = gfn_to_hva_memslot(slot, gfn);
 429
 430	if (tlbsel == 1) {
 431		struct vm_area_struct *vma;
 432		down_read(&current->mm->mmap_sem);
 433
 434		vma = find_vma(current->mm, hva);
 435		if (vma && hva >= vma->vm_start &&
 436		    (vma->vm_flags & VM_PFNMAP)) {
 437			/*
 438			 * This VMA is a physically contiguous region (e.g.
 439			 * /dev/mem) that bypasses normal Linux page
 440			 * management.  Find the overlap between the
 441			 * vma and the memslot.
 442			 */
 443
 444			unsigned long start, end;
 445			unsigned long slot_start, slot_end;
 446
 447			pfnmap = 1;
 448
 449			start = vma->vm_pgoff;
 450			end = start +
 451			      ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
 452
 453			pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
 454
 455			slot_start = pfn - (gfn - slot->base_gfn);
 456			slot_end = slot_start + slot->npages;
 457
 458			if (start < slot_start)
 459				start = slot_start;
 460			if (end > slot_end)
 461				end = slot_end;
 462
 463			tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
 464				MAS1_TSIZE_SHIFT;
 465
 466			/*
 467			 * e500 doesn't implement the lowest tsize bit,
 468			 * or 1K pages.
 469			 */
 470			tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
 471
 472			/*
 473			 * Now find the largest tsize (up to what the guest
 474			 * requested) that will cover gfn, stay within the
 475			 * range, and for which gfn and pfn are mutually
 476			 * aligned.
 477			 */
 478
 479			for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
 480				unsigned long gfn_start, gfn_end, tsize_pages;
 481				tsize_pages = 1 << (tsize - 2);
 482
 483				gfn_start = gfn & ~(tsize_pages - 1);
 484				gfn_end = gfn_start + tsize_pages;
 485
 486				if (gfn_start + pfn - gfn < start)
 487					continue;
 488				if (gfn_end + pfn - gfn > end)
 489					continue;
 490				if ((gfn & (tsize_pages - 1)) !=
 491				    (pfn & (tsize_pages - 1)))
 492					continue;
 493
 494				gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
 495				pfn &= ~(tsize_pages - 1);
 496				break;
 497			}
 498		} else if (vma && hva >= vma->vm_start &&
 499			   (vma->vm_flags & VM_HUGETLB)) {
 500			unsigned long psize = vma_kernel_pagesize(vma);
 501
 502			tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
 503				MAS1_TSIZE_SHIFT;
 504
 505			/*
 506			 * Take the largest page size that satisfies both host
 507			 * and guest mapping
 508			 */
 509			tsize = min(__ilog2(psize) - 10, tsize);
 510
 511			/*
 512			 * e500 doesn't implement the lowest tsize bit,
 513			 * or 1K pages.
 514			 */
 515			tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
 516		}
 517
 518		up_read(&current->mm->mmap_sem);
 519	}
 520
 521	if (likely(!pfnmap)) {
 522		unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
 523		pfn = gfn_to_pfn_memslot(vcpu_e500->vcpu.kvm, slot, gfn);
 524		if (is_error_pfn(pfn)) {
 525			printk(KERN_ERR "Couldn't get real page for gfn %lx!\n",
 526					(long)gfn);
 527			kvm_release_pfn_clean(pfn);
 528			return;
 529		}
 530
 531		/* Align guest and physical address to page map boundaries */
 532		pfn &= ~(tsize_pages - 1);
 533		gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
 534	}
 535
 536	/* Drop old ref and setup new one. */
 537	kvmppc_e500_ref_release(ref);
 538	kvmppc_e500_ref_setup(ref, gtlbe, pfn);
 539
 540	kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
 541				ref, gvaddr, stlbe);
 542}
 543
 544/* XXX only map the one-one case, for now use TLB0 */
 545static void kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500,
 546				 int esel,
 547				 struct kvm_book3e_206_tlb_entry *stlbe)
 548{
 549	struct kvm_book3e_206_tlb_entry *gtlbe;
 550	struct tlbe_ref *ref;
 551
 552	gtlbe = get_entry(vcpu_e500, 0, esel);
 553	ref = &vcpu_e500->gtlb_priv[0][esel].ref;
 554
 555	kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
 556			get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
 557			gtlbe, 0, stlbe, ref);
 558}
 559
 560/* Caller must ensure that the specified guest TLB entry is safe to insert into
 561 * the shadow TLB. */
 562/* XXX for both one-one and one-to-many , for now use TLB1 */
 563static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
 564		u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
 565		struct kvm_book3e_206_tlb_entry *stlbe, int esel)
 566{
 567	struct tlbe_ref *ref;
 568	unsigned int victim;
 569
 570	victim = vcpu_e500->host_tlb1_nv++;
 571
 572	if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
 573		vcpu_e500->host_tlb1_nv = 0;
 574
 575	ref = &vcpu_e500->tlb_refs[1][victim];
 576	kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe, ref);
 577
 578	vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << victim;
 579	vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
 580	if (vcpu_e500->h2g_tlb1_rmap[victim]) {
 581		unsigned int idx = vcpu_e500->h2g_tlb1_rmap[victim];
 582		vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << victim);
 583	}
 584	vcpu_e500->h2g_tlb1_rmap[victim] = esel;
 585
 586	return victim;
 587}
 588
 589static void kvmppc_recalc_tlb1map_range(struct kvmppc_vcpu_e500 *vcpu_e500)
 590{
 591	int size = vcpu_e500->gtlb_params[1].entries;
 592	unsigned int offset;
 593	gva_t eaddr;
 594	int i;
 595
 596	vcpu_e500->tlb1_min_eaddr = ~0UL;
 597	vcpu_e500->tlb1_max_eaddr = 0;
 598	offset = vcpu_e500->gtlb_offset[1];
 599
 600	for (i = 0; i < size; i++) {
 601		struct kvm_book3e_206_tlb_entry *tlbe =
 602			&vcpu_e500->gtlb_arch[offset + i];
 603
 604		if (!get_tlb_v(tlbe))
 605			continue;
 606
 607		eaddr = get_tlb_eaddr(tlbe);
 608		vcpu_e500->tlb1_min_eaddr =
 609				min(vcpu_e500->tlb1_min_eaddr, eaddr);
 610
 611		eaddr = get_tlb_end(tlbe);
 612		vcpu_e500->tlb1_max_eaddr =
 613				max(vcpu_e500->tlb1_max_eaddr, eaddr);
 614	}
 615}
 616
 617static int kvmppc_need_recalc_tlb1map_range(struct kvmppc_vcpu_e500 *vcpu_e500,
 618				struct kvm_book3e_206_tlb_entry *gtlbe)
 619{
 620	unsigned long start, end, size;
 621
 622	size = get_tlb_bytes(gtlbe);
 623	start = get_tlb_eaddr(gtlbe) & ~(size - 1);
 624	end = start + size - 1;
 625
 626	return vcpu_e500->tlb1_min_eaddr == start ||
 627			vcpu_e500->tlb1_max_eaddr == end;
 628}
 629
 630/* This function is supposed to be called for a adding a new valid tlb entry */
 631static void kvmppc_set_tlb1map_range(struct kvm_vcpu *vcpu,
 632				struct kvm_book3e_206_tlb_entry *gtlbe)
 633{
 634	unsigned long start, end, size;
 635	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 636
 637	if (!get_tlb_v(gtlbe))
 638		return;
 639
 640	size = get_tlb_bytes(gtlbe);
 641	start = get_tlb_eaddr(gtlbe) & ~(size - 1);
 642	end = start + size - 1;
 643
 644	vcpu_e500->tlb1_min_eaddr = min(vcpu_e500->tlb1_min_eaddr, start);
 645	vcpu_e500->tlb1_max_eaddr = max(vcpu_e500->tlb1_max_eaddr, end);
 646}
 647
 648static inline int kvmppc_e500_gtlbe_invalidate(
 649				struct kvmppc_vcpu_e500 *vcpu_e500,
 650				int tlbsel, int esel)
 651{
 652	struct kvm_book3e_206_tlb_entry *gtlbe =
 653		get_entry(vcpu_e500, tlbsel, esel);
 654
 655	if (unlikely(get_tlb_iprot(gtlbe)))
 656		return -1;
 657
 658	if (tlbsel == 1 && kvmppc_need_recalc_tlb1map_range(vcpu_e500, gtlbe))
 659		kvmppc_recalc_tlb1map_range(vcpu_e500);
 660
 661	gtlbe->mas1 = 0;
 662
 663	return 0;
 664}
 665
 666int kvmppc_e500_emul_mt_mmucsr0(struct kvmppc_vcpu_e500 *vcpu_e500, ulong value)
 667{
 668	int esel;
 669
 670	if (value & MMUCSR0_TLB0FI)
 671		for (esel = 0; esel < vcpu_e500->gtlb_params[0].entries; esel++)
 672			kvmppc_e500_gtlbe_invalidate(vcpu_e500, 0, esel);
 673	if (value & MMUCSR0_TLB1FI)
 674		for (esel = 0; esel < vcpu_e500->gtlb_params[1].entries; esel++)
 675			kvmppc_e500_gtlbe_invalidate(vcpu_e500, 1, esel);
 676
 677	/* Invalidate all vcpu id mappings */
 678	kvmppc_e500_tlbil_all(vcpu_e500);
 679
 680	return EMULATE_DONE;
 681}
 682
 683int kvmppc_e500_emul_tlbivax(struct kvm_vcpu *vcpu, int ra, int rb)
 684{
 685	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 686	unsigned int ia;
 687	int esel, tlbsel;
 688	gva_t ea;
 689
 690	ea = ((ra) ? kvmppc_get_gpr(vcpu, ra) : 0) + kvmppc_get_gpr(vcpu, rb);
 691
 692	ia = (ea >> 2) & 0x1;
 693
 694	/* since we only have two TLBs, only lower bit is used. */
 695	tlbsel = (ea >> 3) & 0x1;
 696
 697	if (ia) {
 698		/* invalidate all entries */
 699		for (esel = 0; esel < vcpu_e500->gtlb_params[tlbsel].entries;
 700		     esel++)
 701			kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
 702	} else {
 703		ea &= 0xfffff000;
 704		esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel,
 705				get_cur_pid(vcpu), -1);
 706		if (esel >= 0)
 707			kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
 708	}
 709
 710	/* Invalidate all vcpu id mappings */
 711	kvmppc_e500_tlbil_all(vcpu_e500);
 712
 713	return EMULATE_DONE;
 714}
 715
 716static void tlbilx_all(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
 717		       int pid, int rt)
 718{
 719	struct kvm_book3e_206_tlb_entry *tlbe;
 720	int tid, esel;
 721
 722	/* invalidate all entries */
 723	for (esel = 0; esel < vcpu_e500->gtlb_params[tlbsel].entries; esel++) {
 724		tlbe = get_entry(vcpu_e500, tlbsel, esel);
 725		tid = get_tlb_tid(tlbe);
 726		if (rt == 0 || tid == pid) {
 727			inval_gtlbe_on_host(vcpu_e500, tlbsel, esel);
 728			kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
 729		}
 730	}
 731}
 732
 733static void tlbilx_one(struct kvmppc_vcpu_e500 *vcpu_e500, int pid,
 734		       int ra, int rb)
 735{
 736	int tlbsel, esel;
 737	gva_t ea;
 738
 739	ea = kvmppc_get_gpr(&vcpu_e500->vcpu, rb);
 740	if (ra)
 741		ea += kvmppc_get_gpr(&vcpu_e500->vcpu, ra);
 742
 743	for (tlbsel = 0; tlbsel < 2; tlbsel++) {
 744		esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel, pid, -1);
 745		if (esel >= 0) {
 746			inval_gtlbe_on_host(vcpu_e500, tlbsel, esel);
 747			kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
 748			break;
 749		}
 750	}
 751}
 752
 753int kvmppc_e500_emul_tlbilx(struct kvm_vcpu *vcpu, int rt, int ra, int rb)
 754{
 755	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 756	int pid = get_cur_spid(vcpu);
 757
 758	if (rt == 0 || rt == 1) {
 759		tlbilx_all(vcpu_e500, 0, pid, rt);
 760		tlbilx_all(vcpu_e500, 1, pid, rt);
 761	} else if (rt == 3) {
 762		tlbilx_one(vcpu_e500, pid, ra, rb);
 763	}
 764
 765	return EMULATE_DONE;
 766}
 767
 768int kvmppc_e500_emul_tlbre(struct kvm_vcpu *vcpu)
 769{
 770	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 771	int tlbsel, esel;
 772	struct kvm_book3e_206_tlb_entry *gtlbe;
 773
 774	tlbsel = get_tlb_tlbsel(vcpu);
 775	esel = get_tlb_esel(vcpu, tlbsel);
 776
 777	gtlbe = get_entry(vcpu_e500, tlbsel, esel);
 778	vcpu->arch.shared->mas0 &= ~MAS0_NV(~0);
 779	vcpu->arch.shared->mas0 |= MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
 780	vcpu->arch.shared->mas1 = gtlbe->mas1;
 781	vcpu->arch.shared->mas2 = gtlbe->mas2;
 782	vcpu->arch.shared->mas7_3 = gtlbe->mas7_3;
 783
 784	return EMULATE_DONE;
 785}
 786
 787int kvmppc_e500_emul_tlbsx(struct kvm_vcpu *vcpu, int rb)
 788{
 789	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 790	int as = !!get_cur_sas(vcpu);
 791	unsigned int pid = get_cur_spid(vcpu);
 792	int esel, tlbsel;
 793	struct kvm_book3e_206_tlb_entry *gtlbe = NULL;
 794	gva_t ea;
 795
 796	ea = kvmppc_get_gpr(vcpu, rb);
 797
 798	for (tlbsel = 0; tlbsel < 2; tlbsel++) {
 799		esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel, pid, as);
 800		if (esel >= 0) {
 801			gtlbe = get_entry(vcpu_e500, tlbsel, esel);
 802			break;
 803		}
 804	}
 805
 806	if (gtlbe) {
 807		esel &= vcpu_e500->gtlb_params[tlbsel].ways - 1;
 808
 809		vcpu->arch.shared->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(esel)
 810			| MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
 811		vcpu->arch.shared->mas1 = gtlbe->mas1;
 812		vcpu->arch.shared->mas2 = gtlbe->mas2;
 813		vcpu->arch.shared->mas7_3 = gtlbe->mas7_3;
 814	} else {
 815		int victim;
 816
 817		/* since we only have two TLBs, only lower bit is used. */
 818		tlbsel = vcpu->arch.shared->mas4 >> 28 & 0x1;
 819		victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
 820
 821		vcpu->arch.shared->mas0 = MAS0_TLBSEL(tlbsel)
 822			| MAS0_ESEL(victim)
 823			| MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
 824		vcpu->arch.shared->mas1 =
 825			  (vcpu->arch.shared->mas6 & MAS6_SPID0)
 826			| (vcpu->arch.shared->mas6 & (MAS6_SAS ? MAS1_TS : 0))
 827			| (vcpu->arch.shared->mas4 & MAS4_TSIZED(~0));
 828		vcpu->arch.shared->mas2 &= MAS2_EPN;
 829		vcpu->arch.shared->mas2 |= vcpu->arch.shared->mas4 &
 830					   MAS2_ATTRIB_MASK;
 831		vcpu->arch.shared->mas7_3 &= MAS3_U0 | MAS3_U1 |
 832					     MAS3_U2 | MAS3_U3;
 833	}
 834
 835	kvmppc_set_exit_type(vcpu, EMULATED_TLBSX_EXITS);
 836	return EMULATE_DONE;
 837}
 838
 839/* sesel is for tlb1 only */
 840static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
 841			struct kvm_book3e_206_tlb_entry *gtlbe,
 842			struct kvm_book3e_206_tlb_entry *stlbe,
 843			int stlbsel, int sesel)
 844{
 845	int stid;
 846
 847	preempt_disable();
 848	stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
 849
 850	stlbe->mas1 |= MAS1_TID(stid);
 851	write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
 852	preempt_enable();
 853}
 854
 855int kvmppc_e500_emul_tlbwe(struct kvm_vcpu *vcpu)
 856{
 857	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 858	struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
 859	int tlbsel, esel, stlbsel, sesel;
 860	int recal = 0;
 861
 862	tlbsel = get_tlb_tlbsel(vcpu);
 863	esel = get_tlb_esel(vcpu, tlbsel);
 864
 865	gtlbe = get_entry(vcpu_e500, tlbsel, esel);
 866
 867	if (get_tlb_v(gtlbe)) {
 868		inval_gtlbe_on_host(vcpu_e500, tlbsel, esel);
 869		if ((tlbsel == 1) &&
 870			kvmppc_need_recalc_tlb1map_range(vcpu_e500, gtlbe))
 871			recal = 1;
 872	}
 873
 874	gtlbe->mas1 = vcpu->arch.shared->mas1;
 875	gtlbe->mas2 = vcpu->arch.shared->mas2;
 876	gtlbe->mas7_3 = vcpu->arch.shared->mas7_3;
 877
 878	trace_kvm_booke206_gtlb_write(vcpu->arch.shared->mas0, gtlbe->mas1,
 879	                              gtlbe->mas2, gtlbe->mas7_3);
 880
 881	if (tlbsel == 1) {
 882		/*
 883		 * If a valid tlb1 entry is overwritten then recalculate the
 884		 * min/max TLB1 map address range otherwise no need to look
 885		 * in tlb1 array.
 886		 */
 887		if (recal)
 888			kvmppc_recalc_tlb1map_range(vcpu_e500);
 889		else
 890			kvmppc_set_tlb1map_range(vcpu, gtlbe);
 891	}
 892
 893	/* Invalidate shadow mappings for the about-to-be-clobbered TLBE. */
 894	if (tlbe_is_host_safe(vcpu, gtlbe)) {
 895		u64 eaddr;
 896		u64 raddr;
 897
 898		switch (tlbsel) {
 899		case 0:
 900			/* TLB0 */
 901			gtlbe->mas1 &= ~MAS1_TSIZE(~0);
 902			gtlbe->mas1 |= MAS1_TSIZE(BOOK3E_PAGESZ_4K);
 903
 904			stlbsel = 0;
 905			kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
 906			sesel = 0; /* unused */
 907
 908			break;
 909
 910		case 1:
 911			/* TLB1 */
 912			eaddr = get_tlb_eaddr(gtlbe);
 913			raddr = get_tlb_raddr(gtlbe);
 914
 915			/* Create a 4KB mapping on the host.
 916			 * If the guest wanted a large page,
 917			 * only the first 4KB is mapped here and the rest
 918			 * are mapped on the fly. */
 919			stlbsel = 1;
 920			sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr,
 921				    raddr >> PAGE_SHIFT, gtlbe, &stlbe, esel);
 922			break;
 923
 924		default:
 925			BUG();
 926		}
 927
 928		write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
 929	}
 930
 931	kvmppc_set_exit_type(vcpu, EMULATED_TLBWE_EXITS);
 932	return EMULATE_DONE;
 933}
 934
 935static int kvmppc_e500_tlb_search(struct kvm_vcpu *vcpu,
 936				  gva_t eaddr, unsigned int pid, int as)
 937{
 938	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
 939	int esel, tlbsel;
 940
 941	for (tlbsel = 0; tlbsel < 2; tlbsel++) {
 942		esel = kvmppc_e500_tlb_index(vcpu_e500, eaddr, tlbsel, pid, as);
 943		if (esel >= 0)
 944			return index_of(tlbsel, esel);
 945	}
 946
 947	return -1;
 948}
 949
 950/* 'linear_address' is actually an encoding of AS|PID|EADDR . */
 951int kvmppc_core_vcpu_translate(struct kvm_vcpu *vcpu,
 952                               struct kvm_translation *tr)
 953{
 954	int index;
 955	gva_t eaddr;
 956	u8 pid;
 957	u8 as;
 958
 959	eaddr = tr->linear_address;
 960	pid = (tr->linear_address >> 32) & 0xff;
 961	as = (tr->linear_address >> 40) & 0x1;
 962
 963	index = kvmppc_e500_tlb_search(vcpu, eaddr, pid, as);
 964	if (index < 0) {
 965		tr->valid = 0;
 966		return 0;
 967	}
 968
 969	tr->physical_address = kvmppc_mmu_xlate(vcpu, index, eaddr);
 970	/* XXX what does "writeable" and "usermode" even mean? */
 971	tr->valid = 1;
 972
 973	return 0;
 974}
 975
 976
 977int kvmppc_mmu_itlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
 978{
 979	unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
 980
 981	return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
 982}
 983
 984int kvmppc_mmu_dtlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
 985{
 986	unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
 987
 988	return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
 989}
 990
 991void kvmppc_mmu_itlb_miss(struct kvm_vcpu *vcpu)
 992{
 993	unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
 994
 995	kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.pc, as);
 996}
 997
 998void kvmppc_mmu_dtlb_miss(struct kvm_vcpu *vcpu)
 999{
1000	unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
1001
1002	kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.fault_dear, as);
1003}
1004
1005gpa_t kvmppc_mmu_xlate(struct kvm_vcpu *vcpu, unsigned int index,
1006			gva_t eaddr)
1007{
1008	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1009	struct kvm_book3e_206_tlb_entry *gtlbe;
1010	u64 pgmask;
1011
1012	gtlbe = get_entry(vcpu_e500, tlbsel_of(index), esel_of(index));
1013	pgmask = get_tlb_bytes(gtlbe) - 1;
1014
1015	return get_tlb_raddr(gtlbe) | (eaddr & pgmask);
1016}
1017
1018void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
1019{
1020}
1021
1022void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
1023			unsigned int index)
1024{
1025	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1026	struct tlbe_priv *priv;
1027	struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
1028	int tlbsel = tlbsel_of(index);
1029	int esel = esel_of(index);
1030	int stlbsel, sesel;
1031
1032	gtlbe = get_entry(vcpu_e500, tlbsel, esel);
1033
1034	switch (tlbsel) {
1035	case 0:
1036		stlbsel = 0;
1037		sesel = 0; /* unused */
1038		priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
1039
1040		kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
1041					&priv->ref, eaddr, &stlbe);
1042		break;
1043
1044	case 1: {
1045		gfn_t gfn = gpaddr >> PAGE_SHIFT;
1046
1047		stlbsel = 1;
1048		sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn,
1049					     gtlbe, &stlbe, esel);
1050		break;
1051	}
1052
1053	default:
1054		BUG();
1055		break;
1056	}
1057
1058	write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
1059}
1060
1061static void free_gtlb(struct kvmppc_vcpu_e500 *vcpu_e500)
1062{
1063	int i;
1064
1065	clear_tlb1_bitmap(vcpu_e500);
1066	kfree(vcpu_e500->g2h_tlb1_map);
1067
1068	clear_tlb_refs(vcpu_e500);
1069	kfree(vcpu_e500->gtlb_priv[0]);
1070	kfree(vcpu_e500->gtlb_priv[1]);
1071
1072	if (vcpu_e500->shared_tlb_pages) {
1073		vfree((void *)(round_down((uintptr_t)vcpu_e500->gtlb_arch,
1074					  PAGE_SIZE)));
1075
1076		for (i = 0; i < vcpu_e500->num_shared_tlb_pages; i++) {
1077			set_page_dirty_lock(vcpu_e500->shared_tlb_pages[i]);
1078			put_page(vcpu_e500->shared_tlb_pages[i]);
1079		}
1080
1081		vcpu_e500->num_shared_tlb_pages = 0;
1082		vcpu_e500->shared_tlb_pages = NULL;
1083	} else {
1084		kfree(vcpu_e500->gtlb_arch);
1085	}
1086
1087	vcpu_e500->gtlb_arch = NULL;
1088}
1089
1090void kvmppc_get_sregs_e500_tlb(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1091{
1092	sregs->u.e.mas0 = vcpu->arch.shared->mas0;
1093	sregs->u.e.mas1 = vcpu->arch.shared->mas1;
1094	sregs->u.e.mas2 = vcpu->arch.shared->mas2;
1095	sregs->u.e.mas7_3 = vcpu->arch.shared->mas7_3;
1096	sregs->u.e.mas4 = vcpu->arch.shared->mas4;
1097	sregs->u.e.mas6 = vcpu->arch.shared->mas6;
1098
1099	sregs->u.e.mmucfg = vcpu->arch.mmucfg;
1100	sregs->u.e.tlbcfg[0] = vcpu->arch.tlbcfg[0];
1101	sregs->u.e.tlbcfg[1] = vcpu->arch.tlbcfg[1];
1102	sregs->u.e.tlbcfg[2] = 0;
1103	sregs->u.e.tlbcfg[3] = 0;
1104}
1105
1106int kvmppc_set_sregs_e500_tlb(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1107{
1108	if (sregs->u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1109		vcpu->arch.shared->mas0 = sregs->u.e.mas0;
1110		vcpu->arch.shared->mas1 = sregs->u.e.mas1;
1111		vcpu->arch.shared->mas2 = sregs->u.e.mas2;
1112		vcpu->arch.shared->mas7_3 = sregs->u.e.mas7_3;
1113		vcpu->arch.shared->mas4 = sregs->u.e.mas4;
1114		vcpu->arch.shared->mas6 = sregs->u.e.mas6;
1115	}
1116
1117	return 0;
1118}
1119
1120int kvm_vcpu_ioctl_config_tlb(struct kvm_vcpu *vcpu,
1121			      struct kvm_config_tlb *cfg)
1122{
1123	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1124	struct kvm_book3e_206_tlb_params params;
1125	char *virt;
1126	struct page **pages;
1127	struct tlbe_priv *privs[2] = {};
1128	u64 *g2h_bitmap = NULL;
1129	size_t array_len;
1130	u32 sets;
1131	int num_pages, ret, i;
1132
1133	if (cfg->mmu_type != KVM_MMU_FSL_BOOKE_NOHV)
1134		return -EINVAL;
1135
1136	if (copy_from_user(&params, (void __user *)(uintptr_t)cfg->params,
1137			   sizeof(params)))
1138		return -EFAULT;
1139
1140	if (params.tlb_sizes[1] > 64)
1141		return -EINVAL;
1142	if (params.tlb_ways[1] != params.tlb_sizes[1])
1143		return -EINVAL;
1144	if (params.tlb_sizes[2] != 0 || params.tlb_sizes[3] != 0)
1145		return -EINVAL;
1146	if (params.tlb_ways[2] != 0 || params.tlb_ways[3] != 0)
1147		return -EINVAL;
1148
1149	if (!is_power_of_2(params.tlb_ways[0]))
1150		return -EINVAL;
1151
1152	sets = params.tlb_sizes[0] >> ilog2(params.tlb_ways[0]);
1153	if (!is_power_of_2(sets))
1154		return -EINVAL;
1155
1156	array_len = params.tlb_sizes[0] + params.tlb_sizes[1];
1157	array_len *= sizeof(struct kvm_book3e_206_tlb_entry);
1158
1159	if (cfg->array_len < array_len)
1160		return -EINVAL;
1161
1162	num_pages = DIV_ROUND_UP(cfg->array + array_len - 1, PAGE_SIZE) -
1163		    cfg->array / PAGE_SIZE;
1164	pages = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
1165	if (!pages)
1166		return -ENOMEM;
1167
1168	ret = get_user_pages_fast(cfg->array, num_pages, 1, pages);
1169	if (ret < 0)
1170		goto err_pages;
1171
1172	if (ret != num_pages) {
1173		num_pages = ret;
1174		ret = -EFAULT;
1175		goto err_put_page;
1176	}
1177
1178	virt = vmap(pages, num_pages, VM_MAP, PAGE_KERNEL);
1179	if (!virt)
1180		goto err_put_page;
1181
1182	privs[0] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[0],
1183			   GFP_KERNEL);
1184	privs[1] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[1],
1185			   GFP_KERNEL);
1186
1187	if (!privs[0] || !privs[1])
1188		goto err_put_page;
1189
1190	g2h_bitmap = kzalloc(sizeof(u64) * params.tlb_sizes[1],
1191	                     GFP_KERNEL);
1192	if (!g2h_bitmap)
1193		goto err_put_page;
1194
1195	free_gtlb(vcpu_e500);
1196
1197	vcpu_e500->gtlb_priv[0] = privs[0];
1198	vcpu_e500->gtlb_priv[1] = privs[1];
1199	vcpu_e500->g2h_tlb1_map = g2h_bitmap;
1200
1201	vcpu_e500->gtlb_arch = (struct kvm_book3e_206_tlb_entry *)
1202		(virt + (cfg->array & (PAGE_SIZE - 1)));
1203
1204	vcpu_e500->gtlb_params[0].entries = params.tlb_sizes[0];
1205	vcpu_e500->gtlb_params[1].entries = params.tlb_sizes[1];
1206
1207	vcpu_e500->gtlb_offset[0] = 0;
1208	vcpu_e500->gtlb_offset[1] = params.tlb_sizes[0];
1209
1210	vcpu->arch.mmucfg = mfspr(SPRN_MMUCFG) & ~MMUCFG_LPIDSIZE;
1211
1212	vcpu->arch.tlbcfg[0] &= ~(TLBnCFG_N_ENTRY | TLBnCFG_ASSOC);
1213	if (params.tlb_sizes[0] <= 2048)
1214		vcpu->arch.tlbcfg[0] |= params.tlb_sizes[0];
1215	vcpu->arch.tlbcfg[0] |= params.tlb_ways[0] << TLBnCFG_ASSOC_SHIFT;
1216
1217	vcpu->arch.tlbcfg[1] &= ~(TLBnCFG_N_ENTRY | TLBnCFG_ASSOC);
1218	vcpu->arch.tlbcfg[1] |= params.tlb_sizes[1];
1219	vcpu->arch.tlbcfg[1] |= params.tlb_ways[1] << TLBnCFG_ASSOC_SHIFT;
1220
1221	vcpu_e500->shared_tlb_pages = pages;
1222	vcpu_e500->num_shared_tlb_pages = num_pages;
1223
1224	vcpu_e500->gtlb_params[0].ways = params.tlb_ways[0];
1225	vcpu_e500->gtlb_params[0].sets = sets;
1226
1227	vcpu_e500->gtlb_params[1].ways = params.tlb_sizes[1];
1228	vcpu_e500->gtlb_params[1].sets = 1;
1229
1230	kvmppc_recalc_tlb1map_range(vcpu_e500);
1231	return 0;
1232
1233err_put_page:
1234	kfree(privs[0]);
1235	kfree(privs[1]);
1236
1237	for (i = 0; i < num_pages; i++)
1238		put_page(pages[i]);
1239
1240err_pages:
1241	kfree(pages);
1242	return ret;
1243}
1244
1245int kvm_vcpu_ioctl_dirty_tlb(struct kvm_vcpu *vcpu,
1246			     struct kvm_dirty_tlb *dirty)
1247{
1248	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1249	kvmppc_recalc_tlb1map_range(vcpu_e500);
1250	clear_tlb_refs(vcpu_e500);
1251	return 0;
1252}
1253
1254int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
1255{
1256	struct kvm_vcpu *vcpu = &vcpu_e500->vcpu;
1257	int entry_size = sizeof(struct kvm_book3e_206_tlb_entry);
1258	int entries = KVM_E500_TLB0_SIZE + KVM_E500_TLB1_SIZE;
1259
1260	host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
1261	host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
1262
1263	/*
1264	 * This should never happen on real e500 hardware, but is
1265	 * architecturally possible -- e.g. in some weird nested
1266	 * virtualization case.
1267	 */
1268	if (host_tlb_params[0].entries == 0 ||
1269	    host_tlb_params[1].entries == 0) {
1270		pr_err("%s: need to know host tlb size\n", __func__);
1271		return -ENODEV;
1272	}
1273
1274	host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
1275				  TLBnCFG_ASSOC_SHIFT;
1276	host_tlb_params[1].ways = host_tlb_params[1].entries;
1277
1278	if (!is_power_of_2(host_tlb_params[0].entries) ||
1279	    !is_power_of_2(host_tlb_params[0].ways) ||
1280	    host_tlb_params[0].entries < host_tlb_params[0].ways ||
1281	    host_tlb_params[0].ways == 0) {
1282		pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
1283		       __func__, host_tlb_params[0].entries,
1284		       host_tlb_params[0].ways);
1285		return -ENODEV;
1286	}
1287
1288	host_tlb_params[0].sets =
1289		host_tlb_params[0].entries / host_tlb_params[0].ways;
1290	host_tlb_params[1].sets = 1;
1291
1292	vcpu_e500->gtlb_params[0].entries = KVM_E500_TLB0_SIZE;
1293	vcpu_e500->gtlb_params[1].entries = KVM_E500_TLB1_SIZE;
1294
1295	vcpu_e500->gtlb_params[0].ways = KVM_E500_TLB0_WAY_NUM;
1296	vcpu_e500->gtlb_params[0].sets =
1297		KVM_E500_TLB0_SIZE / KVM_E500_TLB0_WAY_NUM;
1298
1299	vcpu_e500->gtlb_params[1].ways = KVM_E500_TLB1_SIZE;
1300	vcpu_e500->gtlb_params[1].sets = 1;
1301
1302	vcpu_e500->gtlb_arch = kmalloc(entries * entry_size, GFP_KERNEL);
1303	if (!vcpu_e500->gtlb_arch)
1304		return -ENOMEM;
1305
1306	vcpu_e500->gtlb_offset[0] = 0;
1307	vcpu_e500->gtlb_offset[1] = KVM_E500_TLB0_SIZE;
1308
1309	vcpu_e500->tlb_refs[0] =
1310		kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[0].entries,
1311			GFP_KERNEL);
1312	if (!vcpu_e500->tlb_refs[0])
1313		goto err;
1314
1315	vcpu_e500->tlb_refs[1] =
1316		kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[1].entries,
1317			GFP_KERNEL);
1318	if (!vcpu_e500->tlb_refs[1])
1319		goto err;
1320
1321	vcpu_e500->gtlb_priv[0] = kzalloc(sizeof(struct tlbe_ref) *
1322					  vcpu_e500->gtlb_params[0].entries,
1323					  GFP_KERNEL);
1324	if (!vcpu_e500->gtlb_priv[0])
1325		goto err;
1326
1327	vcpu_e500->gtlb_priv[1] = kzalloc(sizeof(struct tlbe_ref) *
1328					  vcpu_e500->gtlb_params[1].entries,
1329					  GFP_KERNEL);
1330	if (!vcpu_e500->gtlb_priv[1])
1331		goto err;
1332
1333	vcpu_e500->g2h_tlb1_map = kzalloc(sizeof(unsigned int) *
1334					  vcpu_e500->gtlb_params[1].entries,
1335					  GFP_KERNEL);
1336	if (!vcpu_e500->g2h_tlb1_map)
1337		goto err;
1338
1339	vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
1340					   host_tlb_params[1].entries,
1341					   GFP_KERNEL);
1342	if (!vcpu_e500->h2g_tlb1_rmap)
1343		goto err;
1344
1345	/* Init TLB configuration register */
1346	vcpu->arch.tlbcfg[0] = mfspr(SPRN_TLB0CFG) &
1347			     ~(TLBnCFG_N_ENTRY | TLBnCFG_ASSOC);
1348	vcpu->arch.tlbcfg[0] |= vcpu_e500->gtlb_params[0].entries;
1349	vcpu->arch.tlbcfg[0] |=
1350		vcpu_e500->gtlb_params[0].ways << TLBnCFG_ASSOC_SHIFT;
1351
1352	vcpu->arch.tlbcfg[1] = mfspr(SPRN_TLB1CFG) &
1353			     ~(TLBnCFG_N_ENTRY | TLBnCFG_ASSOC);
1354	vcpu->arch.tlbcfg[1] |= vcpu_e500->gtlb_params[1].entries;
1355	vcpu->arch.tlbcfg[1] |=
1356		vcpu_e500->gtlb_params[1].ways << TLBnCFG_ASSOC_SHIFT;
1357
1358	kvmppc_recalc_tlb1map_range(vcpu_e500);
1359	return 0;
1360
1361err:
1362	free_gtlb(vcpu_e500);
1363	kfree(vcpu_e500->tlb_refs[0]);
1364	kfree(vcpu_e500->tlb_refs[1]);
1365	return -1;
1366}
1367
1368void kvmppc_e500_tlb_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
1369{
1370	free_gtlb(vcpu_e500);
1371	kfree(vcpu_e500->h2g_tlb1_rmap);
1372	kfree(vcpu_e500->tlb_refs[0]);
1373	kfree(vcpu_e500->tlb_refs[1]);
1374}