1 // SPDX-License-Identifier: GPL-2.0 2 3 #ifndef __KVM_X86_MMU_TDP_ITER_H 4 #define __KVM_X86_MMU_TDP_ITER_H 5 6 #include <linux/kvm_host.h> 7 8 #include "mmu.h" 9 #include "spte.h" 10 11 /* 12 * TDP MMU SPTEs are RCU protected to allow paging structures (non-leaf SPTEs) 13 * to be zapped while holding mmu_lock for read, and to allow TLB flushes to be 14 * batched without having to collect the list of zapped SPs. Flows that can 15 * remove SPs must service pending TLB flushes prior to dropping RCU protection. 16 */ 17 static inline u64 kvm_tdp_mmu_read_spte(tdp_ptep_t sptep) 18 { 19 return READ_ONCE(*rcu_dereference(sptep)); 20 } 21 22 static inline u64 kvm_tdp_mmu_write_spte_atomic(tdp_ptep_t sptep, u64 new_spte) 23 { 24 return xchg(rcu_dereference(sptep), new_spte); 25 } 26 27 static inline void __kvm_tdp_mmu_write_spte(tdp_ptep_t sptep, u64 new_spte) 28 { 29 WRITE_ONCE(*rcu_dereference(sptep), new_spte); 30 } 31 32 static inline u64 kvm_tdp_mmu_write_spte(tdp_ptep_t sptep, u64 old_spte, 33 u64 new_spte, int level) 34 { 35 /* 36 * Atomically write the SPTE if it is a shadow-present, leaf SPTE with 37 * volatile bits, i.e. has bits that can be set outside of mmu_lock. 38 * The Writable bit can be set by KVM's fast page fault handler, and 39 * Accessed and Dirty bits can be set by the CPU. 40 * 41 * Note, non-leaf SPTEs do have Accessed bits and those bits are 42 * technically volatile, but KVM doesn't consume the Accessed bit of 43 * non-leaf SPTEs, i.e. KVM doesn't care if it clobbers the bit. This 44 * logic needs to be reassessed if KVM were to use non-leaf Accessed 45 * bits, e.g. to skip stepping down into child SPTEs when aging SPTEs. 46 */ 47 if (is_shadow_present_pte(old_spte) && is_last_spte(old_spte, level) && 48 spte_has_volatile_bits(old_spte)) 49 return kvm_tdp_mmu_write_spte_atomic(sptep, new_spte); 50 51 __kvm_tdp_mmu_write_spte(sptep, new_spte); 52 return old_spte; 53 } 54 55 /* 56 * A TDP iterator performs a pre-order walk over a TDP paging structure. 57 */ 58 struct tdp_iter { 59 /* 60 * The iterator will traverse the paging structure towards the mapping 61 * for this GFN. 62 */ 63 gfn_t next_last_level_gfn; 64 /* 65 * The next_last_level_gfn at the time when the thread last 66 * yielded. Only yielding when the next_last_level_gfn != 67 * yielded_gfn helps ensure forward progress. 68 */ 69 gfn_t yielded_gfn; 70 /* Pointers to the page tables traversed to reach the current SPTE */ 71 tdp_ptep_t pt_path[PT64_ROOT_MAX_LEVEL]; 72 /* A pointer to the current SPTE */ 73 tdp_ptep_t sptep; 74 /* The lowest GFN mapped by the current SPTE */ 75 gfn_t gfn; 76 /* The level of the root page given to the iterator */ 77 int root_level; 78 /* The lowest level the iterator should traverse to */ 79 int min_level; 80 /* The iterator's current level within the paging structure */ 81 int level; 82 /* The address space ID, i.e. SMM vs. regular. */ 83 int as_id; 84 /* A snapshot of the value at sptep */ 85 u64 old_spte; 86 /* 87 * Whether the iterator has a valid state. This will be false if the 88 * iterator walks off the end of the paging structure. 89 */ 90 bool valid; 91 /* 92 * True if KVM dropped mmu_lock and yielded in the middle of a walk, in 93 * which case tdp_iter_next() needs to restart the walk at the root 94 * level instead of advancing to the next entry. 95 */ 96 bool yielded; 97 }; 98 99 /* 100 * Iterates over every SPTE mapping the GFN range [start, end) in a 101 * preorder traversal. 102 */ 103 #define for_each_tdp_pte_min_level(iter, root, min_level, start, end) \ 104 for (tdp_iter_start(&iter, root, min_level, start); \ 105 iter.valid && iter.gfn < end; \ 106 tdp_iter_next(&iter)) 107 108 #define for_each_tdp_pte(iter, root, start, end) \ 109 for_each_tdp_pte_min_level(iter, root, PG_LEVEL_4K, start, end) 110 111 tdp_ptep_t spte_to_child_pt(u64 pte, int level); 112 113 void tdp_iter_start(struct tdp_iter *iter, struct kvm_mmu_page *root, 114 int min_level, gfn_t next_last_level_gfn); 115 void tdp_iter_next(struct tdp_iter *iter); 116 void tdp_iter_restart(struct tdp_iter *iter); 117 118 #endif /* __KVM_X86_MMU_TDP_ITER_H */ 119