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 */
 17static inline u64 kvm_tdp_mmu_read_spte(tdp_ptep_t sptep)
 18{
 19	return READ_ONCE(*rcu_dereference(sptep));
 20}
 21
 22static inline u64 kvm_tdp_mmu_write_spte_atomic(tdp_ptep_t sptep, u64 new_spte)
 23{
 24	KVM_MMU_WARN_ON(is_ept_ve_possible(new_spte));
 25	return xchg(rcu_dereference(sptep), new_spte);
 26}
 27
 28static inline void __kvm_tdp_mmu_write_spte(tdp_ptep_t sptep, u64 new_spte)
 29{
 30	KVM_MMU_WARN_ON(is_ept_ve_possible(new_spte));
 31	WRITE_ONCE(*rcu_dereference(sptep), new_spte);
 32}
 33
 34/*
 35 * SPTEs must be modified atomically if they are shadow-present, leaf
 36 * SPTEs, and have volatile bits, i.e. has bits that can be set outside
 37 * of mmu_lock.  The Writable bit can be set by KVM's fast page fault
 38 * handler, and Accessed and Dirty bits can be set by the CPU.
 39 *
 40 * Note, non-leaf SPTEs do have Accessed bits and those bits are
 41 * technically volatile, but KVM doesn't consume the Accessed bit of
 42 * non-leaf SPTEs, i.e. KVM doesn't care if it clobbers the bit.  This
 43 * logic needs to be reassessed if KVM were to use non-leaf Accessed
 44 * bits, e.g. to skip stepping down into child SPTEs when aging SPTEs.
 45 */
 46static inline bool kvm_tdp_mmu_spte_need_atomic_write(u64 old_spte, int level)
 47{
 48	return is_shadow_present_pte(old_spte) &&
 49	       is_last_spte(old_spte, level) &&
 50	       spte_has_volatile_bits(old_spte);
 51}
 52
 53static inline u64 kvm_tdp_mmu_write_spte(tdp_ptep_t sptep, u64 old_spte,
 54					 u64 new_spte, int level)
 55{
 56	if (kvm_tdp_mmu_spte_need_atomic_write(old_spte, level))
 57		return kvm_tdp_mmu_write_spte_atomic(sptep, new_spte);
 58
 59	__kvm_tdp_mmu_write_spte(sptep, new_spte);
 60	return old_spte;
 61}
 62
 63static inline u64 tdp_mmu_clear_spte_bits(tdp_ptep_t sptep, u64 old_spte,
 64					  u64 mask, int level)
 65{
 66	atomic64_t *sptep_atomic;
 67
 68	if (kvm_tdp_mmu_spte_need_atomic_write(old_spte, level)) {
 69		sptep_atomic = (atomic64_t *)rcu_dereference(sptep);
 70		return (u64)atomic64_fetch_and(~mask, sptep_atomic);
 71	}
 72
 73	__kvm_tdp_mmu_write_spte(sptep, old_spte & ~mask);
 74	return old_spte;
 75}
 76
 77/*
 78 * A TDP iterator performs a pre-order walk over a TDP paging structure.
 79 */
 80struct tdp_iter {
 81	/*
 82	 * The iterator will traverse the paging structure towards the mapping
 83	 * for this GFN.
 84	 */
 85	gfn_t next_last_level_gfn;
 86	/*
 87	 * The next_last_level_gfn at the time when the thread last
 88	 * yielded. Only yielding when the next_last_level_gfn !=
 89	 * yielded_gfn helps ensure forward progress.
 90	 */
 91	gfn_t yielded_gfn;
 92	/* Pointers to the page tables traversed to reach the current SPTE */
 93	tdp_ptep_t pt_path[PT64_ROOT_MAX_LEVEL];
 94	/* A pointer to the current SPTE */
 95	tdp_ptep_t sptep;
 96	/* The lowest GFN mapped by the current SPTE */
 97	gfn_t gfn;
 98	/* The level of the root page given to the iterator */
 99	int root_level;
100	/* The lowest level the iterator should traverse to */
101	int min_level;
102	/* The iterator's current level within the paging structure */
103	int level;
104	/* The address space ID, i.e. SMM vs. regular. */
105	int as_id;
106	/* A snapshot of the value at sptep */
107	u64 old_spte;
108	/*
109	 * Whether the iterator has a valid state. This will be false if the
110	 * iterator walks off the end of the paging structure.
111	 */
112	bool valid;
113	/*
114	 * True if KVM dropped mmu_lock and yielded in the middle of a walk, in
115	 * which case tdp_iter_next() needs to restart the walk at the root
116	 * level instead of advancing to the next entry.
117	 */
118	bool yielded;
119};
120
121/*
122 * Iterates over every SPTE mapping the GFN range [start, end) in a
123 * preorder traversal.
124 */
125#define for_each_tdp_pte_min_level(iter, root, min_level, start, end) \
126	for (tdp_iter_start(&iter, root, min_level, start); \
127	     iter.valid && iter.gfn < end;		     \
128	     tdp_iter_next(&iter))
129
130#define for_each_tdp_pte(iter, root, start, end) \
131	for_each_tdp_pte_min_level(iter, root, PG_LEVEL_4K, start, end)
132
133tdp_ptep_t spte_to_child_pt(u64 pte, int level);
134
135void tdp_iter_start(struct tdp_iter *iter, struct kvm_mmu_page *root,
136		    int min_level, gfn_t next_last_level_gfn);
137void tdp_iter_next(struct tdp_iter *iter);
138void tdp_iter_restart(struct tdp_iter *iter);
139
140#endif /* __KVM_X86_MMU_TDP_ITER_H */