Skip to content

Commit 06ce2c6

Browse files
paulusmackavikivity
paulusmack
authored and
avikivity
committed
KVM: PPC: Maintain a doubly-linked list of guest HPTEs for each gfn
This expands the reverse mapping array to contain two links for each HPTE which are used to link together HPTEs that correspond to the same guest logical page. Each circular list of HPTEs is pointed to by the rmap array entry for the guest logical page, pointed to by the relevant memslot. Links are 32-bit HPT entry indexes rather than full 64-bit pointers, to save space. We use 3 of the remaining 32 bits in the rmap array entries as a lock bit, a referenced bit and a present bit (the present bit is needed since HPTE index 0 is valid). The bit lock for the rmap chain nests inside the HPTE lock bit. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Alexander Graf <agraf@suse.de> Signed-off-by: Avi Kivity <avi@redhat.com>
1 parent 9d0ef5e commit 06ce2c6

File tree

3 files changed

+117
-2
lines changed

3 files changed

+117
-2
lines changed

‎arch/powerpc/include/asm/kvm_book3s_64.h

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -113,6 +113,11 @@ static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
113113
return 0; /* error */
114114
}
115115

116+
static inline unsigned long hpte_rpn(unsigned long ptel, unsigned long psize)
117+
{
118+
return ((ptel & HPTE_R_RPN) & ~(psize - 1)) >> PAGE_SHIFT;
119+
}
120+
116121
static inline int hpte_cache_flags_ok(unsigned long ptel, unsigned long io_type)
117122
{
118123
unsigned int wimg = ptel & HPTE_R_WIMG;
@@ -139,6 +144,19 @@ static inline unsigned long hpte_cache_bits(unsigned long pte_val)
139144
#endif
140145
}
141146

147+
static inline void lock_rmap(unsigned long *rmap)
148+
{
149+
do {
150+
while (test_bit(KVMPPC_RMAP_LOCK_BIT, rmap))
151+
cpu_relax();
152+
} while (test_and_set_bit_lock(KVMPPC_RMAP_LOCK_BIT, rmap));
153+
}
154+
155+
static inline void unlock_rmap(unsigned long *rmap)
156+
{
157+
__clear_bit_unlock(KVMPPC_RMAP_LOCK_BIT, rmap);
158+
}
159+
142160
static inline bool slot_is_aligned(struct kvm_memory_slot *memslot,
143161
unsigned long pagesize)
144162
{

‎arch/powerpc/include/asm/kvm_host.h

Lines changed: 16 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -170,12 +170,27 @@ struct kvmppc_rma_info {
170170
/*
171171
* The reverse mapping array has one entry for each HPTE,
172172
* which stores the guest's view of the second word of the HPTE
173-
* (including the guest physical address of the mapping).
173+
* (including the guest physical address of the mapping),
174+
* plus forward and backward pointers in a doubly-linked ring
175+
* of HPTEs that map the same host page. The pointers in this
176+
* ring are 32-bit HPTE indexes, to save space.
174177
*/
175178
struct revmap_entry {
176179
unsigned long guest_rpte;
180+
unsigned int forw, back;
177181
};
178182

183+
/*
184+
* We use the top bit of each memslot->rmap entry as a lock bit,
185+
* and bit 32 as a present flag. The bottom 32 bits are the
186+
* index in the guest HPT of a HPTE that points to the page.
187+
*/
188+
#define KVMPPC_RMAP_LOCK_BIT 63
189+
#define KVMPPC_RMAP_REF_BIT 33
190+
#define KVMPPC_RMAP_REFERENCED (1ul << KVMPPC_RMAP_REF_BIT)
191+
#define KVMPPC_RMAP_PRESENT 0x100000000ul
192+
#define KVMPPC_RMAP_INDEX 0xfffffffful
193+
179194
/* Low-order bits in kvm->arch.slot_phys[][] */
180195
#define KVMPPC_PAGE_ORDER_MASK 0x1f
181196
#define KVMPPC_PAGE_NO_CACHE HPTE_R_I /* 0x20 */

‎arch/powerpc/kvm/book3s_hv_rm_mmu.c

Lines changed: 83 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -54,6 +54,70 @@ static void *real_vmalloc_addr(void *x)
5454
return __va(addr);
5555
}
5656

57+
/*
58+
* Add this HPTE into the chain for the real page.
59+
* Must be called with the chain locked; it unlocks the chain.
60+
*/
61+
static void kvmppc_add_revmap_chain(struct kvm *kvm, struct revmap_entry *rev,
62+
unsigned long *rmap, long pte_index, int realmode)
63+
{
64+
struct revmap_entry *head, *tail;
65+
unsigned long i;
66+
67+
if (*rmap & KVMPPC_RMAP_PRESENT) {
68+
i = *rmap & KVMPPC_RMAP_INDEX;
69+
head = &kvm->arch.revmap[i];
70+
if (realmode)
71+
head = real_vmalloc_addr(head);
72+
tail = &kvm->arch.revmap[head->back];
73+
if (realmode)
74+
tail = real_vmalloc_addr(tail);
75+
rev->forw = i;
76+
rev->back = head->back;
77+
tail->forw = pte_index;
78+
head->back = pte_index;
79+
} else {
80+
rev->forw = rev->back = pte_index;
81+
i = pte_index;
82+
}
83+
smp_wmb();
84+
*rmap = i | KVMPPC_RMAP_REFERENCED | KVMPPC_RMAP_PRESENT; /* unlock */
85+
}
86+
87+
/* Remove this HPTE from the chain for a real page */
88+
static void remove_revmap_chain(struct kvm *kvm, long pte_index,
89+
unsigned long hpte_v)
90+
{
91+
struct revmap_entry *rev, *next, *prev;
92+
unsigned long gfn, ptel, head;
93+
struct kvm_memory_slot *memslot;
94+
unsigned long *rmap;
95+
96+
rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
97+
ptel = rev->guest_rpte;
98+
gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
99+
memslot = builtin_gfn_to_memslot(kvm, gfn);
100+
if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
101+
return;
102+
103+
rmap = real_vmalloc_addr(&memslot->rmap[gfn - memslot->base_gfn]);
104+
lock_rmap(rmap);
105+
106+
head = *rmap & KVMPPC_RMAP_INDEX;
107+
next = real_vmalloc_addr(&kvm->arch.revmap[rev->forw]);
108+
prev = real_vmalloc_addr(&kvm->arch.revmap[rev->back]);
109+
next->back = rev->back;
110+
prev->forw = rev->forw;
111+
if (head == pte_index) {
112+
head = rev->forw;
113+
if (head == pte_index)
114+
*rmap &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
115+
else
116+
*rmap = (*rmap & ~KVMPPC_RMAP_INDEX) | head;
117+
}
118+
unlock_rmap(rmap);
119+
}
120+
57121
long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
58122
long pte_index, unsigned long pteh, unsigned long ptel)
59123
{
@@ -66,6 +130,7 @@ long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
66130
struct kvm_memory_slot *memslot;
67131
unsigned long *physp, pte_size;
68132
unsigned long is_io;
133+
unsigned long *rmap;
69134
bool realmode = vcpu->arch.vcore->vcore_state == VCORE_RUNNING;
70135

71136
psize = hpte_page_size(pteh, ptel);
@@ -83,6 +148,7 @@ long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
83148
if (!slot_is_aligned(memslot, psize))
84149
return H_PARAMETER;
85150
slot_fn = gfn - memslot->base_gfn;
151+
rmap = &memslot->rmap[slot_fn];
86152

87153
physp = kvm->arch.slot_phys[memslot->id];
88154
if (!physp)
@@ -164,13 +230,25 @@ long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
164230
}
165231

166232
/* Save away the guest's idea of the second HPTE dword */
167-
rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
233+
rev = &kvm->arch.revmap[pte_index];
234+
if (realmode)
235+
rev = real_vmalloc_addr(rev);
168236
if (rev)
169237
rev->guest_rpte = g_ptel;
238+
239+
/* Link HPTE into reverse-map chain */
240+
if (realmode)
241+
rmap = real_vmalloc_addr(rmap);
242+
lock_rmap(rmap);
243+
kvmppc_add_revmap_chain(kvm, rev, rmap, pte_index, realmode);
244+
170245
hpte[1] = ptel;
246+
247+
/* Write the first HPTE dword, unlocking the HPTE and making it valid */
171248
eieio();
172249
hpte[0] = pteh;
173250
asm volatile("ptesync" : : : "memory");
251+
174252
vcpu->arch.gpr[4] = pte_index;
175253
return H_SUCCESS;
176254
}
@@ -220,6 +298,8 @@ long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,
220298
vcpu->arch.gpr[4] = v = hpte[0] & ~HPTE_V_HVLOCK;
221299
vcpu->arch.gpr[5] = r = hpte[1];
222300
rb = compute_tlbie_rb(v, r, pte_index);
301+
remove_revmap_chain(kvm, pte_index, v);
302+
smp_wmb();
223303
hpte[0] = 0;
224304
if (!(flags & H_LOCAL)) {
225305
while(!try_lock_tlbie(&kvm->arch.tlbie_lock))
@@ -293,6 +373,8 @@ long kvmppc_h_bulk_remove(struct kvm_vcpu *vcpu)
293373
flags |= (hp[1] >> 5) & 0x0c;
294374
args[i * 2] = ((0x80 | flags) << 56) + pte_index;
295375
tlbrb[n_inval++] = compute_tlbie_rb(hp[0], hp[1], pte_index);
376+
remove_revmap_chain(kvm, pte_index, hp[0]);
377+
smp_wmb();
296378
hp[0] = 0;
297379
}
298380
if (n_inval == 0)

0 commit comments

Comments
 (0)