1.. SPDX-License-Identifier: GPL-2.0 2 3================= 4KVM Lock Overview 5================= 6 71. Acquisition Orders 8--------------------- 9 10The acquisition orders for mutexes are as follows: 11 12- kvm->lock is taken outside vcpu->mutex 13 14- kvm->lock is taken outside kvm->slots_lock and kvm->irq_lock 15 16- kvm->slots_lock is taken outside kvm->irq_lock, though acquiring 17 them together is quite rare. 18 19- kvm->mn_active_invalidate_count ensures that pairs of 20 invalidate_range_start() and invalidate_range_end() callbacks 21 use the same memslots array. kvm->slots_lock and kvm->slots_arch_lock 22 are taken on the waiting side in install_new_memslots, so MMU notifiers 23 must not take either kvm->slots_lock or kvm->slots_arch_lock. 24 25For SRCU: 26 27- ``synchronize_srcu(&kvm->srcu)`` is called inside critical sections 28 for kvm->lock, vcpu->mutex and kvm->slots_lock. These locks _cannot_ 29 be taken inside a kvm->srcu read-side critical section; that is, the 30 following is broken:: 31 32 srcu_read_lock(&kvm->srcu); 33 mutex_lock(&kvm->slots_lock); 34 35- kvm->slots_arch_lock instead is released before the call to 36 ``synchronize_srcu()``. It _can_ therefore be taken inside a 37 kvm->srcu read-side critical section, for example while processing 38 a vmexit. 39 40On x86: 41 42- vcpu->mutex is taken outside kvm->arch.hyperv.hv_lock and kvm->arch.xen.xen_lock 43 44- kvm->arch.mmu_lock is an rwlock. kvm->arch.tdp_mmu_pages_lock and 45 kvm->arch.mmu_unsync_pages_lock are taken inside kvm->arch.mmu_lock, and 46 cannot be taken without already holding kvm->arch.mmu_lock (typically with 47 ``read_lock`` for the TDP MMU, thus the need for additional spinlocks). 48 49Everything else is a leaf: no other lock is taken inside the critical 50sections. 51 522. Exception 53------------ 54 55Fast page fault: 56 57Fast page fault is the fast path which fixes the guest page fault out of 58the mmu-lock on x86. Currently, the page fault can be fast in one of the 59following two cases: 60 611. Access Tracking: The SPTE is not present, but it is marked for access 62 tracking. That means we need to restore the saved R/X bits. This is 63 described in more detail later below. 64 652. Write-Protection: The SPTE is present and the fault is caused by 66 write-protect. That means we just need to change the W bit of the spte. 67 68What we use to avoid all the race is the Host-writable bit and MMU-writable bit 69on the spte: 70 71- Host-writable means the gfn is writable in the host kernel page tables and in 72 its KVM memslot. 73- MMU-writable means the gfn is writable in the guest's mmu and it is not 74 write-protected by shadow page write-protection. 75 76On fast page fault path, we will use cmpxchg to atomically set the spte W 77bit if spte.HOST_WRITEABLE = 1 and spte.WRITE_PROTECT = 1, to restore the saved 78R/X bits if for an access-traced spte, or both. This is safe because whenever 79changing these bits can be detected by cmpxchg. 80 81But we need carefully check these cases: 82 831) The mapping from gfn to pfn 84 85The mapping from gfn to pfn may be changed since we can only ensure the pfn 86is not changed during cmpxchg. This is a ABA problem, for example, below case 87will happen: 88 89+------------------------------------------------------------------------+ 90| At the beginning:: | 91| | 92| gpte = gfn1 | 93| gfn1 is mapped to pfn1 on host | 94| spte is the shadow page table entry corresponding with gpte and | 95| spte = pfn1 | 96+------------------------------------------------------------------------+ 97| On fast page fault path: | 98+------------------------------------+-----------------------------------+ 99| CPU 0: | CPU 1: | 100+------------------------------------+-----------------------------------+ 101| :: | | 102| | | 103| old_spte = *spte; | | 104+------------------------------------+-----------------------------------+ 105| | pfn1 is swapped out:: | 106| | | 107| | spte = 0; | 108| | | 109| | pfn1 is re-alloced for gfn2. | 110| | | 111| | gpte is changed to point to | 112| | gfn2 by the guest:: | 113| | | 114| | spte = pfn1; | 115+------------------------------------+-----------------------------------+ 116| :: | 117| | 118| if (cmpxchg(spte, old_spte, old_spte+W) | 119| mark_page_dirty(vcpu->kvm, gfn1) | 120| OOPS!!! | 121+------------------------------------------------------------------------+ 122 123We dirty-log for gfn1, that means gfn2 is lost in dirty-bitmap. 124 125For direct sp, we can easily avoid it since the spte of direct sp is fixed 126to gfn. For indirect sp, we disabled fast page fault for simplicity. 127 128A solution for indirect sp could be to pin the gfn, for example via 129kvm_vcpu_gfn_to_pfn_atomic, before the cmpxchg. After the pinning: 130 131- We have held the refcount of pfn that means the pfn can not be freed and 132 be reused for another gfn. 133- The pfn is writable and therefore it cannot be shared between different gfns 134 by KSM. 135 136Then, we can ensure the dirty bitmaps is correctly set for a gfn. 137 1382) Dirty bit tracking 139 140In the origin code, the spte can be fast updated (non-atomically) if the 141spte is read-only and the Accessed bit has already been set since the 142Accessed bit and Dirty bit can not be lost. 143 144But it is not true after fast page fault since the spte can be marked 145writable between reading spte and updating spte. Like below case: 146 147+------------------------------------------------------------------------+ 148| At the beginning:: | 149| | 150| spte.W = 0 | 151| spte.Accessed = 1 | 152+------------------------------------+-----------------------------------+ 153| CPU 0: | CPU 1: | 154+------------------------------------+-----------------------------------+ 155| In mmu_spte_clear_track_bits():: | | 156| | | 157| old_spte = *spte; | | 158| | | 159| | | 160| /* 'if' condition is satisfied. */| | 161| if (old_spte.Accessed == 1 && | | 162| old_spte.W == 0) | | 163| spte = 0ull; | | 164+------------------------------------+-----------------------------------+ 165| | on fast page fault path:: | 166| | | 167| | spte.W = 1 | 168| | | 169| | memory write on the spte:: | 170| | | 171| | spte.Dirty = 1 | 172+------------------------------------+-----------------------------------+ 173| :: | | 174| | | 175| else | | 176| old_spte = xchg(spte, 0ull) | | 177| if (old_spte.Accessed == 1) | | 178| kvm_set_pfn_accessed(spte.pfn);| | 179| if (old_spte.Dirty == 1) | | 180| kvm_set_pfn_dirty(spte.pfn); | | 181| OOPS!!! | | 182+------------------------------------+-----------------------------------+ 183 184The Dirty bit is lost in this case. 185 186In order to avoid this kind of issue, we always treat the spte as "volatile" 187if it can be updated out of mmu-lock, see spte_has_volatile_bits(), it means, 188the spte is always atomically updated in this case. 189 1903) flush tlbs due to spte updated 191 192If the spte is updated from writable to readonly, we should flush all TLBs, 193otherwise rmap_write_protect will find a read-only spte, even though the 194writable spte might be cached on a CPU's TLB. 195 196As mentioned before, the spte can be updated to writable out of mmu-lock on 197fast page fault path, in order to easily audit the path, we see if TLBs need 198be flushed caused by this reason in mmu_spte_update() since this is a common 199function to update spte (present -> present). 200 201Since the spte is "volatile" if it can be updated out of mmu-lock, we always 202atomically update the spte, the race caused by fast page fault can be avoided, 203See the comments in spte_has_volatile_bits() and mmu_spte_update(). 204 205Lockless Access Tracking: 206 207This is used for Intel CPUs that are using EPT but do not support the EPT A/D 208bits. In this case, PTEs are tagged as A/D disabled (using ignored bits), and 209when the KVM MMU notifier is called to track accesses to a page (via 210kvm_mmu_notifier_clear_flush_young), it marks the PTE not-present in hardware 211by clearing the RWX bits in the PTE and storing the original R & X bits in more 212unused/ignored bits. When the VM tries to access the page later on, a fault is 213generated and the fast page fault mechanism described above is used to 214atomically restore the PTE to a Present state. The W bit is not saved when the 215PTE is marked for access tracking and during restoration to the Present state, 216the W bit is set depending on whether or not it was a write access. If it 217wasn't, then the W bit will remain clear until a write access happens, at which 218time it will be set using the Dirty tracking mechanism described above. 219 2203. Reference 221------------ 222 223``kvm_lock`` 224^^^^^^^^^^^^ 225 226:Type: mutex 227:Arch: any 228:Protects: - vm_list 229 230``kvm_count_lock`` 231^^^^^^^^^^^^^^^^^^ 232 233:Type: raw_spinlock_t 234:Arch: any 235:Protects: - hardware virtualization enable/disable 236:Comment: 'raw' because hardware enabling/disabling must be atomic /wrt 237 migration. 238 239``kvm->mn_invalidate_lock`` 240^^^^^^^^^^^^^^^^^^^^^^^^^^^ 241 242:Type: spinlock_t 243:Arch: any 244:Protects: mn_active_invalidate_count, mn_memslots_update_rcuwait 245 246``kvm_arch::tsc_write_lock`` 247^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 248 249:Type: raw_spinlock_t 250:Arch: x86 251:Protects: - kvm_arch::{last_tsc_write,last_tsc_nsec,last_tsc_offset} 252 - tsc offset in vmcb 253:Comment: 'raw' because updating the tsc offsets must not be preempted. 254 255``kvm->mmu_lock`` 256^^^^^^^^^^^^^^^^^ 257:Type: spinlock_t or rwlock_t 258:Arch: any 259:Protects: -shadow page/shadow tlb entry 260:Comment: it is a spinlock since it is used in mmu notifier. 261 262``kvm->srcu`` 263^^^^^^^^^^^^^ 264:Type: srcu lock 265:Arch: any 266:Protects: - kvm->memslots 267 - kvm->buses 268:Comment: The srcu read lock must be held while accessing memslots (e.g. 269 when using gfn_to_* functions) and while accessing in-kernel 270 MMIO/PIO address->device structure mapping (kvm->buses). 271 The srcu index can be stored in kvm_vcpu->srcu_idx per vcpu 272 if it is needed by multiple functions. 273 274``kvm->slots_arch_lock`` 275^^^^^^^^^^^^^^^^^^^^^^^^ 276:Type: mutex 277:Arch: any (only needed on x86 though) 278:Protects: any arch-specific fields of memslots that have to be modified 279 in a ``kvm->srcu`` read-side critical section. 280:Comment: must be held before reading the pointer to the current memslots, 281 until after all changes to the memslots are complete 282 283``wakeup_vcpus_on_cpu_lock`` 284^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 285:Type: spinlock_t 286:Arch: x86 287:Protects: wakeup_vcpus_on_cpu 288:Comment: This is a per-CPU lock and it is used for VT-d posted-interrupts. 289 When VT-d posted-interrupts is supported and the VM has assigned 290 devices, we put the blocked vCPU on the list blocked_vcpu_on_cpu 291 protected by blocked_vcpu_on_cpu_lock, when VT-d hardware issues 292 wakeup notification event since external interrupts from the 293 assigned devices happens, we will find the vCPU on the list to 294 wakeup. 295