2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_host.h>
22 #include <linux/highmem.h>
23 #include <linux/gfp.h>
24 #include <linux/slab.h>
25 #include <linux/hugetlb.h>
26 #include <linux/vmalloc.h>
27 #include <linux/srcu.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/file.h>
31 #include <asm/tlbflush.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/kvm_book3s.h>
34 #include <asm/mmu-hash64.h>
35 #include <asm/hvcall.h>
36 #include <asm/synch.h>
37 #include <asm/ppc-opcode.h>
38 #include <asm/cputable.h>
40 #include "book3s_hv_cma.h"
42 /* POWER7 has 10-bit LPIDs, PPC970 has 6-bit LPIDs */
43 #define MAX_LPID_970 63
45 /* Power architecture requires HPT is at least 256kB */
46 #define PPC_MIN_HPT_ORDER 18
48 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
49 long pte_index, unsigned long pteh,
50 unsigned long ptel, unsigned long *pte_idx_ret);
51 static void kvmppc_rmap_reset(struct kvm *kvm);
53 long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
56 struct revmap_entry *rev;
57 struct page *page = NULL;
58 long order = KVM_DEFAULT_HPT_ORDER;
62 if (order < PPC_MIN_HPT_ORDER)
63 order = PPC_MIN_HPT_ORDER;
66 kvm->arch.hpt_cma_alloc = 0;
68 * try first to allocate it from the kernel page allocator.
69 * We keep the CMA reserved for failed allocation.
71 hpt = __get_free_pages(GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT |
72 __GFP_NOWARN, order - PAGE_SHIFT);
74 /* Next try to allocate from the preallocated pool */
76 VM_BUG_ON(order < KVM_CMA_CHUNK_ORDER);
77 page = kvm_alloc_hpt(1 << (order - PAGE_SHIFT));
79 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
80 kvm->arch.hpt_cma_alloc = 1;
85 /* Lastly try successively smaller sizes from the page allocator */
86 while (!hpt && order > PPC_MIN_HPT_ORDER) {
87 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT|
88 __GFP_NOWARN, order - PAGE_SHIFT);
96 kvm->arch.hpt_virt = hpt;
97 kvm->arch.hpt_order = order;
98 /* HPTEs are 2**4 bytes long */
99 kvm->arch.hpt_npte = 1ul << (order - 4);
100 /* 128 (2**7) bytes in each HPTEG */
101 kvm->arch.hpt_mask = (1ul << (order - 7)) - 1;
103 /* Allocate reverse map array */
104 rev = vmalloc(sizeof(struct revmap_entry) * kvm->arch.hpt_npte);
106 pr_err("kvmppc_alloc_hpt: Couldn't alloc reverse map array\n");
109 kvm->arch.revmap = rev;
110 kvm->arch.sdr1 = __pa(hpt) | (order - 18);
112 pr_info("KVM guest htab at %lx (order %ld), LPID %x\n",
113 hpt, order, kvm->arch.lpid);
116 *htab_orderp = order;
120 if (kvm->arch.hpt_cma_alloc)
121 kvm_release_hpt(page, 1 << (order - PAGE_SHIFT));
123 free_pages(hpt, order - PAGE_SHIFT);
127 long kvmppc_alloc_reset_hpt(struct kvm *kvm, u32 *htab_orderp)
132 mutex_lock(&kvm->lock);
133 if (kvm->arch.rma_setup_done) {
134 kvm->arch.rma_setup_done = 0;
135 /* order rma_setup_done vs. vcpus_running */
137 if (atomic_read(&kvm->arch.vcpus_running)) {
138 kvm->arch.rma_setup_done = 1;
142 if (kvm->arch.hpt_virt) {
143 order = kvm->arch.hpt_order;
144 /* Set the entire HPT to 0, i.e. invalid HPTEs */
145 memset((void *)kvm->arch.hpt_virt, 0, 1ul << order);
147 * Reset all the reverse-mapping chains for all memslots
149 kvmppc_rmap_reset(kvm);
150 /* Ensure that each vcpu will flush its TLB on next entry. */
151 cpumask_setall(&kvm->arch.need_tlb_flush);
152 *htab_orderp = order;
155 err = kvmppc_alloc_hpt(kvm, htab_orderp);
156 order = *htab_orderp;
159 mutex_unlock(&kvm->lock);
163 void kvmppc_free_hpt(struct kvm *kvm)
165 kvmppc_free_lpid(kvm->arch.lpid);
166 vfree(kvm->arch.revmap);
167 if (kvm->arch.hpt_cma_alloc)
168 kvm_release_hpt(virt_to_page(kvm->arch.hpt_virt),
169 1 << (kvm->arch.hpt_order - PAGE_SHIFT));
171 free_pages(kvm->arch.hpt_virt,
172 kvm->arch.hpt_order - PAGE_SHIFT);
175 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
176 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
178 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
181 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
182 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
184 return (pgsize == 0x10000) ? 0x1000 : 0;
187 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
188 unsigned long porder)
191 unsigned long npages;
192 unsigned long hp_v, hp_r;
193 unsigned long addr, hash;
195 unsigned long hp0, hp1;
196 unsigned long idx_ret;
198 struct kvm *kvm = vcpu->kvm;
200 psize = 1ul << porder;
201 npages = memslot->npages >> (porder - PAGE_SHIFT);
203 /* VRMA can't be > 1TB */
204 if (npages > 1ul << (40 - porder))
205 npages = 1ul << (40 - porder);
206 /* Can't use more than 1 HPTE per HPTEG */
207 if (npages > kvm->arch.hpt_mask + 1)
208 npages = kvm->arch.hpt_mask + 1;
210 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
211 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
212 hp1 = hpte1_pgsize_encoding(psize) |
213 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
215 for (i = 0; i < npages; ++i) {
217 /* can't use hpt_hash since va > 64 bits */
218 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & kvm->arch.hpt_mask;
220 * We assume that the hash table is empty and no
221 * vcpus are using it at this stage. Since we create
222 * at most one HPTE per HPTEG, we just assume entry 7
223 * is available and use it.
225 hash = (hash << 3) + 7;
226 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
228 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
230 if (ret != H_SUCCESS) {
231 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
238 int kvmppc_mmu_hv_init(void)
240 unsigned long host_lpid, rsvd_lpid;
242 if (!cpu_has_feature(CPU_FTR_HVMODE))
245 /* POWER7 has 10-bit LPIDs, PPC970 and e500mc have 6-bit LPIDs */
246 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
247 host_lpid = mfspr(SPRN_LPID); /* POWER7 */
248 rsvd_lpid = LPID_RSVD;
250 host_lpid = 0; /* PPC970 */
251 rsvd_lpid = MAX_LPID_970;
254 kvmppc_init_lpid(rsvd_lpid + 1);
256 kvmppc_claim_lpid(host_lpid);
257 /* rsvd_lpid is reserved for use in partition switching */
258 kvmppc_claim_lpid(rsvd_lpid);
263 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
265 unsigned long msr = vcpu->arch.intr_msr;
267 /* If transactional, change to suspend mode on IRQ delivery */
268 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
271 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
272 kvmppc_set_msr(vcpu, msr);
276 * This is called to get a reference to a guest page if there isn't
277 * one already in the memslot->arch.slot_phys[] array.
279 static long kvmppc_get_guest_page(struct kvm *kvm, unsigned long gfn,
280 struct kvm_memory_slot *memslot,
285 struct page *page, *hpage, *pages[1];
286 unsigned long s, pgsize;
287 unsigned long *physp;
288 unsigned int is_io, got, pgorder;
289 struct vm_area_struct *vma;
290 unsigned long pfn, i, npages;
292 physp = memslot->arch.slot_phys;
295 if (physp[gfn - memslot->base_gfn])
303 start = gfn_to_hva_memslot(memslot, gfn);
305 /* Instantiate and get the page we want access to */
306 np = get_user_pages_fast(start, 1, 1, pages);
308 /* Look up the vma for the page */
309 down_read(¤t->mm->mmap_sem);
310 vma = find_vma(current->mm, start);
311 if (!vma || vma->vm_start > start ||
312 start + psize > vma->vm_end ||
313 !(vma->vm_flags & VM_PFNMAP))
315 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
316 pfn = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
317 /* check alignment of pfn vs. requested page size */
318 if (psize > PAGE_SIZE && (pfn & ((psize >> PAGE_SHIFT) - 1)))
320 up_read(¤t->mm->mmap_sem);
324 got = KVMPPC_GOT_PAGE;
326 /* See if this is a large page */
328 if (PageHuge(page)) {
329 hpage = compound_head(page);
330 s <<= compound_order(hpage);
331 /* Get the whole large page if slot alignment is ok */
332 if (s > psize && slot_is_aligned(memslot, s) &&
333 !(memslot->userspace_addr & (s - 1))) {
343 pfn = page_to_pfn(page);
346 npages = pgsize >> PAGE_SHIFT;
347 pgorder = __ilog2(npages);
348 physp += (gfn - memslot->base_gfn) & ~(npages - 1);
349 spin_lock(&kvm->arch.slot_phys_lock);
350 for (i = 0; i < npages; ++i) {
352 physp[i] = ((pfn + i) << PAGE_SHIFT) +
353 got + is_io + pgorder;
357 spin_unlock(&kvm->arch.slot_phys_lock);
366 up_read(¤t->mm->mmap_sem);
370 long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
371 long pte_index, unsigned long pteh,
372 unsigned long ptel, unsigned long *pte_idx_ret)
374 unsigned long psize, gpa, gfn;
375 struct kvm_memory_slot *memslot;
378 if (kvm->arch.using_mmu_notifiers)
381 psize = hpte_page_size(pteh, ptel);
385 pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
387 /* Find the memslot (if any) for this address */
388 gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
389 gfn = gpa >> PAGE_SHIFT;
390 memslot = gfn_to_memslot(kvm, gfn);
391 if (memslot && !(memslot->flags & KVM_MEMSLOT_INVALID)) {
392 if (!slot_is_aligned(memslot, psize))
394 if (kvmppc_get_guest_page(kvm, gfn, memslot, psize) < 0)
399 /* Protect linux PTE lookup from page table destruction */
400 rcu_read_lock_sched(); /* this disables preemption too */
401 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
402 current->mm->pgd, false, pte_idx_ret);
403 rcu_read_unlock_sched();
404 if (ret == H_TOO_HARD) {
405 /* this can't happen */
406 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
407 ret = H_RESOURCE; /* or something */
414 * We come here on a H_ENTER call from the guest when we are not
415 * using mmu notifiers and we don't have the requested page pinned
418 long kvmppc_virtmode_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
419 long pte_index, unsigned long pteh,
422 return kvmppc_virtmode_do_h_enter(vcpu->kvm, flags, pte_index,
423 pteh, ptel, &vcpu->arch.gpr[4]);
426 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
432 for (i = 0; i < vcpu->arch.slb_nr; i++) {
433 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
436 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
441 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
442 return &vcpu->arch.slb[i];
447 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
450 unsigned long ra_mask;
452 ra_mask = hpte_page_size(v, r) - 1;
453 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
456 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
457 struct kvmppc_pte *gpte, bool data, bool iswrite)
459 struct kvm *kvm = vcpu->kvm;
460 struct kvmppc_slb *slbe;
462 unsigned long pp, key;
464 unsigned long *hptep;
466 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
470 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
475 /* real mode access */
476 slb_v = vcpu->kvm->arch.vrma_slb_v;
480 /* Find the HPTE in the hash table */
481 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
482 HPTE_V_VALID | HPTE_V_ABSENT);
487 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
488 v = hptep[0] & ~HPTE_V_HVLOCK;
489 gr = kvm->arch.revmap[index].guest_rpte;
491 /* Unlock the HPTE */
492 asm volatile("lwsync" : : : "memory");
497 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
499 /* Get PP bits and key for permission check */
500 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
501 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
504 /* Calculate permissions */
505 gpte->may_read = hpte_read_permission(pp, key);
506 gpte->may_write = hpte_write_permission(pp, key);
507 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
509 /* Storage key permission check for POWER7 */
510 if (data && virtmode && cpu_has_feature(CPU_FTR_ARCH_206)) {
511 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
518 /* Get the guest physical address */
519 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
524 * Quick test for whether an instruction is a load or a store.
525 * If the instruction is a load or a store, then this will indicate
526 * which it is, at least on server processors. (Embedded processors
527 * have some external PID instructions that don't follow the rule
528 * embodied here.) If the instruction isn't a load or store, then
529 * this doesn't return anything useful.
531 static int instruction_is_store(unsigned int instr)
536 if ((instr & 0xfc000000) == 0x7c000000)
537 mask = 0x100; /* major opcode 31 */
538 return (instr & mask) != 0;
541 static int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
542 unsigned long gpa, gva_t ea, int is_store)
546 unsigned long srr0 = kvmppc_get_pc(vcpu);
548 /* We try to load the last instruction. We don't let
549 * emulate_instruction do it as it doesn't check what
551 * If we fail, we just return to the guest and try executing it again.
553 if (vcpu->arch.last_inst == KVM_INST_FETCH_FAILED) {
554 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
555 if (ret != EMULATE_DONE || last_inst == KVM_INST_FETCH_FAILED)
557 vcpu->arch.last_inst = last_inst;
561 * WARNING: We do not know for sure whether the instruction we just
562 * read from memory is the same that caused the fault in the first
563 * place. If the instruction we read is neither an load or a store,
564 * then it can't access memory, so we don't need to worry about
565 * enforcing access permissions. So, assuming it is a load or
566 * store, we just check that its direction (load or store) is
567 * consistent with the original fault, since that's what we
568 * checked the access permissions against. If there is a mismatch
569 * we just return and retry the instruction.
572 if (instruction_is_store(kvmppc_get_last_inst(vcpu)) != !!is_store)
576 * Emulated accesses are emulated by looking at the hash for
577 * translation once, then performing the access later. The
578 * translation could be invalidated in the meantime in which
579 * point performing the subsequent memory access on the old
580 * physical address could possibly be a security hole for the
581 * guest (but not the host).
583 * This is less of an issue for MMIO stores since they aren't
584 * globally visible. It could be an issue for MMIO loads to
585 * a certain extent but we'll ignore it for now.
588 vcpu->arch.paddr_accessed = gpa;
589 vcpu->arch.vaddr_accessed = ea;
590 return kvmppc_emulate_mmio(run, vcpu);
593 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
594 unsigned long ea, unsigned long dsisr)
596 struct kvm *kvm = vcpu->kvm;
597 unsigned long *hptep, hpte[3], r;
598 unsigned long mmu_seq, psize, pte_size;
599 unsigned long gpa, gfn, hva, pfn;
600 struct kvm_memory_slot *memslot;
602 struct revmap_entry *rev;
603 struct page *page, *pages[1];
604 long index, ret, npages;
606 unsigned int writing, write_ok;
607 struct vm_area_struct *vma;
608 unsigned long rcbits;
611 * Real-mode code has already searched the HPT and found the
612 * entry we're interested in. Lock the entry and check that
613 * it hasn't changed. If it has, just return and re-execute the
616 if (ea != vcpu->arch.pgfault_addr)
618 index = vcpu->arch.pgfault_index;
619 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
620 rev = &kvm->arch.revmap[index];
622 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
624 hpte[0] = hptep[0] & ~HPTE_V_HVLOCK;
626 hpte[2] = r = rev->guest_rpte;
627 asm volatile("lwsync" : : : "memory");
631 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
632 hpte[1] != vcpu->arch.pgfault_hpte[1])
635 /* Translate the logical address and get the page */
636 psize = hpte_page_size(hpte[0], r);
637 gpa = (r & HPTE_R_RPN & ~(psize - 1)) | (ea & (psize - 1));
638 gfn = gpa >> PAGE_SHIFT;
639 memslot = gfn_to_memslot(kvm, gfn);
641 /* No memslot means it's an emulated MMIO region */
642 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
643 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
644 dsisr & DSISR_ISSTORE);
646 if (!kvm->arch.using_mmu_notifiers)
647 return -EFAULT; /* should never get here */
649 /* used to check for invalidations in progress */
650 mmu_seq = kvm->mmu_notifier_seq;
656 pte_size = PAGE_SIZE;
657 writing = (dsisr & DSISR_ISSTORE) != 0;
658 /* If writing != 0, then the HPTE must allow writing, if we get here */
660 hva = gfn_to_hva_memslot(memslot, gfn);
661 npages = get_user_pages_fast(hva, 1, writing, pages);
663 /* Check if it's an I/O mapping */
664 down_read(¤t->mm->mmap_sem);
665 vma = find_vma(current->mm, hva);
666 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
667 (vma->vm_flags & VM_PFNMAP)) {
668 pfn = vma->vm_pgoff +
669 ((hva - vma->vm_start) >> PAGE_SHIFT);
671 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
672 write_ok = vma->vm_flags & VM_WRITE;
674 up_read(¤t->mm->mmap_sem);
679 pfn = page_to_pfn(page);
680 if (PageHuge(page)) {
681 page = compound_head(page);
682 pte_size <<= compound_order(page);
684 /* if the guest wants write access, see if that is OK */
685 if (!writing && hpte_is_writable(r)) {
686 unsigned int hugepage_shift;
690 * We need to protect against page table destruction
691 * while looking up and updating the pte.
693 rcu_read_lock_sched();
694 ptep = find_linux_pte_or_hugepte(current->mm->pgd,
695 hva, &hugepage_shift);
697 pte = kvmppc_read_update_linux_pte(ptep, 1,
702 rcu_read_unlock_sched();
707 if (psize > pte_size)
710 /* Check WIMG vs. the actual page we're accessing */
711 if (!hpte_cache_flags_ok(r, is_io)) {
715 * Allow guest to map emulated device memory as
716 * uncacheable, but actually make it cacheable.
718 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
722 * Set the HPTE to point to pfn.
723 * Since the pfn is at PAGE_SIZE granularity, make sure we
724 * don't mask out lower-order bits if psize < PAGE_SIZE.
726 if (psize < PAGE_SIZE)
728 r = (r & ~(HPTE_R_PP0 - psize)) | ((pfn << PAGE_SHIFT) & ~(psize - 1));
729 if (hpte_is_writable(r) && !write_ok)
730 r = hpte_make_readonly(r);
733 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
735 if ((hptep[0] & ~HPTE_V_HVLOCK) != hpte[0] || hptep[1] != hpte[1] ||
736 rev->guest_rpte != hpte[2])
737 /* HPTE has been changed under us; let the guest retry */
739 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
741 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
744 /* Check if we might have been invalidated; let the guest retry if so */
746 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
751 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
752 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
753 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
755 if (hptep[0] & HPTE_V_VALID) {
756 /* HPTE was previously valid, so we need to invalidate it */
758 hptep[0] |= HPTE_V_ABSENT;
759 kvmppc_invalidate_hpte(kvm, hptep, index);
760 /* don't lose previous R and C bits */
761 r |= hptep[1] & (HPTE_R_R | HPTE_R_C);
763 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
769 asm volatile("ptesync" : : : "memory");
771 if (page && hpte_is_writable(r))
777 * We drop pages[0] here, not page because page might
778 * have been set to the head page of a compound, but
779 * we have to drop the reference on the correct tail
780 * page to match the get inside gup()
787 hptep[0] &= ~HPTE_V_HVLOCK;
792 static void kvmppc_rmap_reset(struct kvm *kvm)
794 struct kvm_memslots *slots;
795 struct kvm_memory_slot *memslot;
798 srcu_idx = srcu_read_lock(&kvm->srcu);
799 slots = kvm->memslots;
800 kvm_for_each_memslot(memslot, slots) {
802 * This assumes it is acceptable to lose reference and
803 * change bits across a reset.
805 memset(memslot->arch.rmap, 0,
806 memslot->npages * sizeof(*memslot->arch.rmap));
808 srcu_read_unlock(&kvm->srcu, srcu_idx);
811 static int kvm_handle_hva_range(struct kvm *kvm,
814 int (*handler)(struct kvm *kvm,
815 unsigned long *rmapp,
820 struct kvm_memslots *slots;
821 struct kvm_memory_slot *memslot;
823 slots = kvm_memslots(kvm);
824 kvm_for_each_memslot(memslot, slots) {
825 unsigned long hva_start, hva_end;
828 hva_start = max(start, memslot->userspace_addr);
829 hva_end = min(end, memslot->userspace_addr +
830 (memslot->npages << PAGE_SHIFT));
831 if (hva_start >= hva_end)
834 * {gfn(page) | page intersects with [hva_start, hva_end)} =
835 * {gfn, gfn+1, ..., gfn_end-1}.
837 gfn = hva_to_gfn_memslot(hva_start, memslot);
838 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
840 for (; gfn < gfn_end; ++gfn) {
841 gfn_t gfn_offset = gfn - memslot->base_gfn;
843 ret = handler(kvm, &memslot->arch.rmap[gfn_offset], gfn);
851 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
852 int (*handler)(struct kvm *kvm, unsigned long *rmapp,
855 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
858 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
861 struct revmap_entry *rev = kvm->arch.revmap;
862 unsigned long h, i, j;
863 unsigned long *hptep;
864 unsigned long ptel, psize, rcbits;
868 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
874 * To avoid an ABBA deadlock with the HPTE lock bit,
875 * we can't spin on the HPTE lock while holding the
878 i = *rmapp & KVMPPC_RMAP_INDEX;
879 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
880 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
881 /* unlock rmap before spinning on the HPTE lock */
883 while (hptep[0] & HPTE_V_HVLOCK)
889 /* chain is now empty */
890 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
892 /* remove i from chain */
896 rev[i].forw = rev[i].back = i;
897 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
900 /* Now check and modify the HPTE */
901 ptel = rev[i].guest_rpte;
902 psize = hpte_page_size(hptep[0], ptel);
903 if ((hptep[0] & HPTE_V_VALID) &&
904 hpte_rpn(ptel, psize) == gfn) {
905 if (kvm->arch.using_mmu_notifiers)
906 hptep[0] |= HPTE_V_ABSENT;
907 kvmppc_invalidate_hpte(kvm, hptep, i);
908 /* Harvest R and C */
909 rcbits = hptep[1] & (HPTE_R_R | HPTE_R_C);
910 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
911 if (rcbits & ~rev[i].guest_rpte) {
912 rev[i].guest_rpte = ptel | rcbits;
913 note_hpte_modification(kvm, &rev[i]);
917 hptep[0] &= ~HPTE_V_HVLOCK;
922 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
924 if (kvm->arch.using_mmu_notifiers)
925 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
929 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
931 if (kvm->arch.using_mmu_notifiers)
932 kvm_handle_hva_range(kvm, start, end, kvm_unmap_rmapp);
936 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
937 struct kvm_memory_slot *memslot)
939 unsigned long *rmapp;
943 rmapp = memslot->arch.rmap;
944 gfn = memslot->base_gfn;
945 for (n = memslot->npages; n; --n) {
947 * Testing the present bit without locking is OK because
948 * the memslot has been marked invalid already, and hence
949 * no new HPTEs referencing this page can be created,
950 * thus the present bit can't go from 0 to 1.
952 if (*rmapp & KVMPPC_RMAP_PRESENT)
953 kvm_unmap_rmapp(kvm, rmapp, gfn);
959 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
962 struct revmap_entry *rev = kvm->arch.revmap;
963 unsigned long head, i, j;
964 unsigned long *hptep;
969 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
970 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
973 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
978 i = head = *rmapp & KVMPPC_RMAP_INDEX;
980 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
983 /* If this HPTE isn't referenced, ignore it */
984 if (!(hptep[1] & HPTE_R_R))
987 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
988 /* unlock rmap before spinning on the HPTE lock */
990 while (hptep[0] & HPTE_V_HVLOCK)
995 /* Now check and modify the HPTE */
996 if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_R)) {
997 kvmppc_clear_ref_hpte(kvm, hptep, i);
998 if (!(rev[i].guest_rpte & HPTE_R_R)) {
999 rev[i].guest_rpte |= HPTE_R_R;
1000 note_hpte_modification(kvm, &rev[i]);
1004 hptep[0] &= ~HPTE_V_HVLOCK;
1005 } while ((i = j) != head);
1011 int kvm_age_hva_hv(struct kvm *kvm, unsigned long hva)
1013 if (!kvm->arch.using_mmu_notifiers)
1015 return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
1018 static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
1021 struct revmap_entry *rev = kvm->arch.revmap;
1022 unsigned long head, i, j;
1026 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1030 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1033 if (*rmapp & KVMPPC_RMAP_PRESENT) {
1034 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1036 hp = (unsigned long *)(kvm->arch.hpt_virt + (i << 4));
1038 if (hp[1] & HPTE_R_R)
1040 } while ((i = j) != head);
1049 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
1051 if (!kvm->arch.using_mmu_notifiers)
1053 return kvm_handle_hva(kvm, hva, kvm_test_age_rmapp);
1056 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1058 if (!kvm->arch.using_mmu_notifiers)
1060 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
1063 static int kvm_test_clear_dirty(struct kvm *kvm, unsigned long *rmapp)
1065 struct revmap_entry *rev = kvm->arch.revmap;
1066 unsigned long head, i, j;
1067 unsigned long *hptep;
1072 if (*rmapp & KVMPPC_RMAP_CHANGED) {
1073 *rmapp &= ~KVMPPC_RMAP_CHANGED;
1076 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1081 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1083 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
1086 if (!(hptep[1] & HPTE_R_C))
1089 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1090 /* unlock rmap before spinning on the HPTE lock */
1092 while (hptep[0] & HPTE_V_HVLOCK)
1097 /* Now check and modify the HPTE */
1098 if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_C)) {
1099 /* need to make it temporarily absent to clear C */
1100 hptep[0] |= HPTE_V_ABSENT;
1101 kvmppc_invalidate_hpte(kvm, hptep, i);
1102 hptep[1] &= ~HPTE_R_C;
1104 hptep[0] = (hptep[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
1105 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1106 rev[i].guest_rpte |= HPTE_R_C;
1107 note_hpte_modification(kvm, &rev[i]);
1111 hptep[0] &= ~HPTE_V_HVLOCK;
1112 } while ((i = j) != head);
1118 static void harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1119 struct kvm_memory_slot *memslot,
1124 if (!vpa->dirty || !vpa->pinned_addr)
1126 gfn = vpa->gpa >> PAGE_SHIFT;
1127 if (gfn < memslot->base_gfn ||
1128 gfn >= memslot->base_gfn + memslot->npages)
1133 __set_bit_le(gfn - memslot->base_gfn, map);
1136 long kvmppc_hv_get_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot,
1140 unsigned long *rmapp;
1141 struct kvm_vcpu *vcpu;
1144 rmapp = memslot->arch.rmap;
1145 for (i = 0; i < memslot->npages; ++i) {
1146 if (kvm_test_clear_dirty(kvm, rmapp) && map)
1147 __set_bit_le(i, map);
1151 /* Harvest dirty bits from VPA and DTL updates */
1152 /* Note: we never modify the SLB shadow buffer areas */
1153 kvm_for_each_vcpu(i, vcpu, kvm) {
1154 spin_lock(&vcpu->arch.vpa_update_lock);
1155 harvest_vpa_dirty(&vcpu->arch.vpa, memslot, map);
1156 harvest_vpa_dirty(&vcpu->arch.dtl, memslot, map);
1157 spin_unlock(&vcpu->arch.vpa_update_lock);
1163 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1164 unsigned long *nb_ret)
1166 struct kvm_memory_slot *memslot;
1167 unsigned long gfn = gpa >> PAGE_SHIFT;
1168 struct page *page, *pages[1];
1170 unsigned long hva, offset;
1172 unsigned long *physp;
1175 srcu_idx = srcu_read_lock(&kvm->srcu);
1176 memslot = gfn_to_memslot(kvm, gfn);
1177 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1179 if (!kvm->arch.using_mmu_notifiers) {
1180 physp = memslot->arch.slot_phys;
1183 physp += gfn - memslot->base_gfn;
1186 if (kvmppc_get_guest_page(kvm, gfn, memslot,
1191 page = pfn_to_page(pa >> PAGE_SHIFT);
1194 hva = gfn_to_hva_memslot(memslot, gfn);
1195 npages = get_user_pages_fast(hva, 1, 1, pages);
1200 srcu_read_unlock(&kvm->srcu, srcu_idx);
1202 offset = gpa & (PAGE_SIZE - 1);
1204 *nb_ret = PAGE_SIZE - offset;
1205 return page_address(page) + offset;
1208 srcu_read_unlock(&kvm->srcu, srcu_idx);
1212 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1215 struct page *page = virt_to_page(va);
1216 struct kvm_memory_slot *memslot;
1218 unsigned long *rmap;
1223 if (!dirty || !kvm->arch.using_mmu_notifiers)
1226 /* We need to mark this page dirty in the rmap chain */
1227 gfn = gpa >> PAGE_SHIFT;
1228 srcu_idx = srcu_read_lock(&kvm->srcu);
1229 memslot = gfn_to_memslot(kvm, gfn);
1231 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
1233 *rmap |= KVMPPC_RMAP_CHANGED;
1236 srcu_read_unlock(&kvm->srcu, srcu_idx);
1240 * Functions for reading and writing the hash table via reads and
1241 * writes on a file descriptor.
1243 * Reads return the guest view of the hash table, which has to be
1244 * pieced together from the real hash table and the guest_rpte
1245 * values in the revmap array.
1247 * On writes, each HPTE written is considered in turn, and if it
1248 * is valid, it is written to the HPT as if an H_ENTER with the
1249 * exact flag set was done. When the invalid count is non-zero
1250 * in the header written to the stream, the kernel will make
1251 * sure that that many HPTEs are invalid, and invalidate them
1255 struct kvm_htab_ctx {
1256 unsigned long index;
1257 unsigned long flags;
1262 #define HPTE_SIZE (2 * sizeof(unsigned long))
1265 * Returns 1 if this HPT entry has been modified or has pending
1268 static int hpte_dirty(struct revmap_entry *revp, unsigned long *hptp)
1270 unsigned long rcbits_unset;
1272 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1275 /* Also need to consider changes in reference and changed bits */
1276 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1277 if ((hptp[0] & HPTE_V_VALID) && (hptp[1] & rcbits_unset))
1283 static long record_hpte(unsigned long flags, unsigned long *hptp,
1284 unsigned long *hpte, struct revmap_entry *revp,
1285 int want_valid, int first_pass)
1288 unsigned long rcbits_unset;
1292 /* Unmodified entries are uninteresting except on the first pass */
1293 dirty = hpte_dirty(revp, hptp);
1294 if (!first_pass && !dirty)
1298 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1300 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1301 !(hptp[0] & HPTE_V_BOLTED))
1304 if (valid != want_valid)
1308 if (valid || dirty) {
1309 /* lock the HPTE so it's stable and read it */
1311 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1315 /* re-evaluate valid and dirty from synchronized HPTE value */
1316 valid = !!(v & HPTE_V_VALID);
1317 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1319 /* Harvest R and C into guest view if necessary */
1320 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1321 if (valid && (rcbits_unset & hptp[1])) {
1322 revp->guest_rpte |= (hptp[1] & (HPTE_R_R | HPTE_R_C)) |
1327 if (v & HPTE_V_ABSENT) {
1328 v &= ~HPTE_V_ABSENT;
1332 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1335 r = revp->guest_rpte;
1336 /* only clear modified if this is the right sort of entry */
1337 if (valid == want_valid && dirty) {
1338 r &= ~HPTE_GR_MODIFIED;
1339 revp->guest_rpte = r;
1341 asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
1342 hptp[0] &= ~HPTE_V_HVLOCK;
1344 if (!(valid == want_valid && (first_pass || dirty)))
1352 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1353 size_t count, loff_t *ppos)
1355 struct kvm_htab_ctx *ctx = file->private_data;
1356 struct kvm *kvm = ctx->kvm;
1357 struct kvm_get_htab_header hdr;
1358 unsigned long *hptp;
1359 struct revmap_entry *revp;
1360 unsigned long i, nb, nw;
1361 unsigned long __user *lbuf;
1362 struct kvm_get_htab_header __user *hptr;
1363 unsigned long flags;
1365 unsigned long hpte[2];
1367 if (!access_ok(VERIFY_WRITE, buf, count))
1370 first_pass = ctx->first_pass;
1374 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1375 revp = kvm->arch.revmap + i;
1376 lbuf = (unsigned long __user *)buf;
1379 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1380 /* Initialize header */
1381 hptr = (struct kvm_get_htab_header __user *)buf;
1386 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1388 /* Skip uninteresting entries, i.e. clean on not-first pass */
1390 while (i < kvm->arch.hpt_npte &&
1391 !hpte_dirty(revp, hptp)) {
1399 /* Grab a series of valid entries */
1400 while (i < kvm->arch.hpt_npte &&
1401 hdr.n_valid < 0xffff &&
1402 nb + HPTE_SIZE < count &&
1403 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1404 /* valid entry, write it out */
1406 if (__put_user(hpte[0], lbuf) ||
1407 __put_user(hpte[1], lbuf + 1))
1415 /* Now skip invalid entries while we can */
1416 while (i < kvm->arch.hpt_npte &&
1417 hdr.n_invalid < 0xffff &&
1418 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1419 /* found an invalid entry */
1426 if (hdr.n_valid || hdr.n_invalid) {
1427 /* write back the header */
1428 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1431 buf = (char __user *)lbuf;
1436 /* Check if we've wrapped around the hash table */
1437 if (i >= kvm->arch.hpt_npte) {
1439 ctx->first_pass = 0;
1449 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1450 size_t count, loff_t *ppos)
1452 struct kvm_htab_ctx *ctx = file->private_data;
1453 struct kvm *kvm = ctx->kvm;
1454 struct kvm_get_htab_header hdr;
1457 unsigned long __user *lbuf;
1458 unsigned long *hptp;
1459 unsigned long tmp[2];
1464 if (!access_ok(VERIFY_READ, buf, count))
1467 /* lock out vcpus from running while we're doing this */
1468 mutex_lock(&kvm->lock);
1469 rma_setup = kvm->arch.rma_setup_done;
1471 kvm->arch.rma_setup_done = 0; /* temporarily */
1472 /* order rma_setup_done vs. vcpus_running */
1474 if (atomic_read(&kvm->arch.vcpus_running)) {
1475 kvm->arch.rma_setup_done = 1;
1476 mutex_unlock(&kvm->lock);
1482 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1484 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1488 if (nb + hdr.n_valid * HPTE_SIZE > count)
1496 if (i >= kvm->arch.hpt_npte ||
1497 i + hdr.n_valid + hdr.n_invalid > kvm->arch.hpt_npte)
1500 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1501 lbuf = (unsigned long __user *)buf;
1502 for (j = 0; j < hdr.n_valid; ++j) {
1504 if (__get_user(v, lbuf) || __get_user(r, lbuf + 1))
1507 if (!(v & HPTE_V_VALID))
1512 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT))
1513 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1515 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1517 if (ret != H_SUCCESS) {
1518 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1519 "r=%lx\n", ret, i, v, r);
1522 if (!rma_setup && is_vrma_hpte(v)) {
1523 unsigned long psize = hpte_page_size(v, r);
1524 unsigned long senc = slb_pgsize_encoding(psize);
1527 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1528 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1529 lpcr = senc << (LPCR_VRMASD_SH - 4);
1530 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
1537 for (j = 0; j < hdr.n_invalid; ++j) {
1538 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT))
1539 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1547 /* Order HPTE updates vs. rma_setup_done */
1549 kvm->arch.rma_setup_done = rma_setup;
1550 mutex_unlock(&kvm->lock);
1557 static int kvm_htab_release(struct inode *inode, struct file *filp)
1559 struct kvm_htab_ctx *ctx = filp->private_data;
1561 filp->private_data = NULL;
1562 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1563 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1564 kvm_put_kvm(ctx->kvm);
1569 static const struct file_operations kvm_htab_fops = {
1570 .read = kvm_htab_read,
1571 .write = kvm_htab_write,
1572 .llseek = default_llseek,
1573 .release = kvm_htab_release,
1576 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1579 struct kvm_htab_ctx *ctx;
1582 /* reject flags we don't recognize */
1583 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1585 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1590 ctx->index = ghf->start_index;
1591 ctx->flags = ghf->flags;
1592 ctx->first_pass = 1;
1594 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1595 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1601 if (rwflag == O_RDONLY) {
1602 mutex_lock(&kvm->slots_lock);
1603 atomic_inc(&kvm->arch.hpte_mod_interest);
1604 /* make sure kvmppc_do_h_enter etc. see the increment */
1605 synchronize_srcu_expedited(&kvm->srcu);
1606 mutex_unlock(&kvm->slots_lock);
1612 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
1614 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
1616 if (cpu_has_feature(CPU_FTR_ARCH_206))
1617 vcpu->arch.slb_nr = 32; /* POWER7 */
1619 vcpu->arch.slb_nr = 64;
1621 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
1622 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
1624 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;