2 * PowerPC Radix MMU mulation helpers for QEMU.
4 * Copyright (c) 2016 Suraj Jitindar Singh, IBM Corporation
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
22 #include "exec/exec-all.h"
23 #include "qemu/error-report.h"
24 #include "sysemu/kvm.h"
28 #include "mmu-radix64.h"
29 #include "mmu-book3s-v3.h"
31 static bool ppc_radix64_get_fully_qualified_addr(const CPUPPCState
*env
,
33 uint64_t *lpid
, uint64_t *pid
)
35 /* When EA(2:11) are nonzero, raise a segment interrupt */
36 if (eaddr
& ~R_EADDR_VALID_MASK
) {
40 if (FIELD_EX64(env
->msr
, MSR
, HV
)) { /* MSR[HV] -> Hypervisor/bare metal */
41 switch (eaddr
& R_EADDR_QUADRANT
) {
42 case R_EADDR_QUADRANT0
:
44 *pid
= env
->spr
[SPR_BOOKS_PID
];
46 case R_EADDR_QUADRANT1
:
47 *lpid
= env
->spr
[SPR_LPIDR
];
48 *pid
= env
->spr
[SPR_BOOKS_PID
];
50 case R_EADDR_QUADRANT2
:
51 *lpid
= env
->spr
[SPR_LPIDR
];
54 case R_EADDR_QUADRANT3
:
59 g_assert_not_reached();
61 } else { /* !MSR[HV] -> Guest */
62 switch (eaddr
& R_EADDR_QUADRANT
) {
63 case R_EADDR_QUADRANT0
: /* Guest application */
64 *lpid
= env
->spr
[SPR_LPIDR
];
65 *pid
= env
->spr
[SPR_BOOKS_PID
];
67 case R_EADDR_QUADRANT1
: /* Illegal */
68 case R_EADDR_QUADRANT2
:
70 case R_EADDR_QUADRANT3
: /* Guest OS */
71 *lpid
= env
->spr
[SPR_LPIDR
];
72 *pid
= 0; /* pid set to 0 -> addresses guest operating system */
75 g_assert_not_reached();
82 static void ppc_radix64_raise_segi(PowerPCCPU
*cpu
, MMUAccessType access_type
,
85 CPUState
*cs
= CPU(cpu
);
86 CPUPPCState
*env
= &cpu
->env
;
88 switch (access_type
) {
90 /* Instruction Segment Interrupt */
91 cs
->exception_index
= POWERPC_EXCP_ISEG
;
95 /* Data Segment Interrupt */
96 cs
->exception_index
= POWERPC_EXCP_DSEG
;
97 env
->spr
[SPR_DAR
] = eaddr
;
100 g_assert_not_reached();
105 static inline const char *access_str(MMUAccessType access_type
)
107 return access_type
== MMU_DATA_LOAD
? "reading" :
108 (access_type
== MMU_DATA_STORE
? "writing" : "execute");
111 static void ppc_radix64_raise_si(PowerPCCPU
*cpu
, MMUAccessType access_type
,
112 vaddr eaddr
, uint32_t cause
)
114 CPUState
*cs
= CPU(cpu
);
115 CPUPPCState
*env
= &cpu
->env
;
117 qemu_log_mask(CPU_LOG_MMU
, "%s for %s @0x%"VADDR_PRIx
" cause %08x\n",
118 __func__
, access_str(access_type
),
121 switch (access_type
) {
123 /* Instruction Storage Interrupt */
124 cs
->exception_index
= POWERPC_EXCP_ISI
;
125 env
->error_code
= cause
;
128 cause
|= DSISR_ISSTORE
;
131 /* Data Storage Interrupt */
132 cs
->exception_index
= POWERPC_EXCP_DSI
;
133 env
->spr
[SPR_DSISR
] = cause
;
134 env
->spr
[SPR_DAR
] = eaddr
;
138 g_assert_not_reached();
142 static void ppc_radix64_raise_hsi(PowerPCCPU
*cpu
, MMUAccessType access_type
,
143 vaddr eaddr
, hwaddr g_raddr
, uint32_t cause
)
145 CPUState
*cs
= CPU(cpu
);
146 CPUPPCState
*env
= &cpu
->env
;
148 qemu_log_mask(CPU_LOG_MMU
, "%s for %s @0x%"VADDR_PRIx
" 0x%"
149 HWADDR_PRIx
" cause %08x\n",
150 __func__
, access_str(access_type
),
151 eaddr
, g_raddr
, cause
);
153 switch (access_type
) {
155 /* H Instruction Storage Interrupt */
156 cs
->exception_index
= POWERPC_EXCP_HISI
;
157 env
->spr
[SPR_ASDR
] = g_raddr
;
158 env
->error_code
= cause
;
161 cause
|= DSISR_ISSTORE
;
164 /* H Data Storage Interrupt */
165 cs
->exception_index
= POWERPC_EXCP_HDSI
;
166 env
->spr
[SPR_HDSISR
] = cause
;
167 env
->spr
[SPR_HDAR
] = eaddr
;
168 env
->spr
[SPR_ASDR
] = g_raddr
;
172 g_assert_not_reached();
176 static bool ppc_radix64_check_prot(PowerPCCPU
*cpu
, MMUAccessType access_type
,
177 uint64_t pte
, int *fault_cause
, int *prot
,
178 int mmu_idx
, bool partition_scoped
)
180 CPUPPCState
*env
= &cpu
->env
;
183 /* Check Page Attributes (pte58:59) */
184 if ((pte
& R_PTE_ATT
) == R_PTE_ATT_NI_IO
&& access_type
== MMU_INST_FETCH
) {
186 * Radix PTE entries with the non-idempotent I/O attribute are treated
189 *fault_cause
|= SRR1_NOEXEC_GUARD
;
193 /* Determine permissions allowed by Encoded Access Authority */
194 if (!partition_scoped
&& (pte
& R_PTE_EAA_PRIV
) &&
195 FIELD_EX64(env
->msr
, MSR
, PR
)) {
197 } else if (mmuidx_pr(mmu_idx
) || (pte
& R_PTE_EAA_PRIV
) ||
199 *prot
= ppc_radix64_get_prot_eaa(pte
);
200 } else { /* !MSR_PR && !(pte & R_PTE_EAA_PRIV) && !partition_scoped */
201 *prot
= ppc_radix64_get_prot_eaa(pte
);
202 *prot
&= ppc_radix64_get_prot_amr(cpu
); /* Least combined permissions */
205 /* Check if requested access type is allowed */
206 need_prot
= prot_for_access_type(access_type
);
207 if (need_prot
& ~*prot
) { /* Page Protected for that Access */
208 *fault_cause
|= access_type
== MMU_INST_FETCH
? SRR1_NOEXEC_GUARD
:
216 static void ppc_radix64_set_rc(PowerPCCPU
*cpu
, MMUAccessType access_type
,
217 uint64_t pte
, hwaddr pte_addr
, int *prot
)
219 CPUState
*cs
= CPU(cpu
);
222 npte
= pte
| R_PTE_R
; /* Always set reference bit */
224 if (access_type
== MMU_DATA_STORE
) { /* Store/Write */
225 npte
|= R_PTE_C
; /* Set change bit */
228 * Treat the page as read-only for now, so that a later write
229 * will pass through this function again to set the C bit.
231 *prot
&= ~PAGE_WRITE
;
234 if (pte
^ npte
) { /* If pte has changed then write it back */
235 stq_phys(cs
->as
, pte_addr
, npte
);
239 static bool ppc_radix64_is_valid_level(int level
, int psize
, uint64_t nls
)
244 * Check if this is a valid level, according to POWER9 and POWER10
245 * Processor User's Manuals, sections 4.10.4.1 and 5.10.6.1, respectively:
246 * Supported Radix Tree Configurations and Resulting Page Sizes.
248 * Note: these checks are specific to POWER9 and POWER10 CPUs. Any future
249 * CPUs that supports a different Radix MMU configuration will need their
250 * own implementation.
253 case 0: /* Root Page Dir */
254 ret
= psize
== 52 && nls
== 13;
261 ret
= nls
== 9 || nls
== 5;
267 if (unlikely(!ret
)) {
268 qemu_log_mask(LOG_GUEST_ERROR
, "invalid radix configuration: "
269 "level %d size %d nls %"PRIu64
"\n",
275 static int ppc_radix64_next_level(AddressSpace
*as
, vaddr eaddr
,
276 uint64_t *pte_addr
, uint64_t *nls
,
277 int *psize
, uint64_t *pte
, int *fault_cause
)
279 uint64_t index
, mask
, nlb
, pde
;
281 /* Read page <directory/table> entry from guest address space */
282 pde
= ldq_phys(as
, *pte_addr
);
283 if (!(pde
& R_PTE_VALID
)) { /* Invalid Entry */
284 *fault_cause
|= DSISR_NOPTE
;
290 if (!(pde
& R_PTE_LEAF
)) { /* Prepare for next iteration */
291 *nls
= pde
& R_PDE_NLS
;
292 index
= eaddr
>> (*psize
- *nls
); /* Shift */
293 index
&= ((1UL << *nls
) - 1); /* Mask */
294 nlb
= pde
& R_PDE_NLB
;
295 mask
= MAKE_64BIT_MASK(0, *nls
+ 3);
298 qemu_log_mask(LOG_GUEST_ERROR
,
299 "%s: misaligned page dir/table base: 0x"TARGET_FMT_lx
300 " page dir size: 0x"TARGET_FMT_lx
"\n",
301 __func__
, nlb
, mask
+ 1);
304 *pte_addr
= nlb
+ index
* sizeof(pde
);
309 static int ppc_radix64_walk_tree(AddressSpace
*as
, vaddr eaddr
,
310 uint64_t base_addr
, uint64_t nls
,
311 hwaddr
*raddr
, int *psize
, uint64_t *pte
,
312 int *fault_cause
, hwaddr
*pte_addr
)
314 uint64_t index
, pde
, rpn
, mask
;
317 index
= eaddr
>> (*psize
- nls
); /* Shift */
318 index
&= ((1UL << nls
) - 1); /* Mask */
319 mask
= MAKE_64BIT_MASK(0, nls
+ 3);
321 if (base_addr
& mask
) {
322 qemu_log_mask(LOG_GUEST_ERROR
,
323 "%s: misaligned page dir base: 0x"TARGET_FMT_lx
324 " page dir size: 0x"TARGET_FMT_lx
"\n",
325 __func__
, base_addr
, mask
+ 1);
328 *pte_addr
= base_addr
+ index
* sizeof(pde
);
333 if (!ppc_radix64_is_valid_level(level
++, *psize
, nls
)) {
334 *fault_cause
|= DSISR_R_BADCONFIG
;
338 ret
= ppc_radix64_next_level(as
, eaddr
, pte_addr
, &nls
, psize
, &pde
,
343 } while (!(pde
& R_PTE_LEAF
));
346 rpn
= pde
& R_PTE_RPN
;
347 mask
= (1UL << *psize
) - 1;
349 /* Or high bits of rpn and low bits to ea to form whole real addr */
350 *raddr
= (rpn
& ~mask
) | (eaddr
& mask
);
354 static bool validate_pate(PowerPCCPU
*cpu
, uint64_t lpid
, ppc_v3_pate_t
*pate
)
356 CPUPPCState
*env
= &cpu
->env
;
358 if (!(pate
->dw0
& PATE0_HR
)) {
361 if (lpid
== 0 && !FIELD_EX64(env
->msr
, MSR
, HV
)) {
364 if ((pate
->dw0
& PATE1_R_PRTS
) < 5) {
367 /* More checks ... */
371 static int ppc_radix64_partition_scoped_xlate(PowerPCCPU
*cpu
,
372 MMUAccessType access_type
,
373 vaddr eaddr
, hwaddr g_raddr
,
375 hwaddr
*h_raddr
, int *h_prot
,
376 int *h_page_size
, bool pde_addr
,
377 int mmu_idx
, bool guest_visible
)
383 qemu_log_mask(CPU_LOG_MMU
, "%s for %s @0x%"VADDR_PRIx
384 " mmu_idx %u 0x%"HWADDR_PRIx
"\n",
385 __func__
, access_str(access_type
),
386 eaddr
, mmu_idx
, g_raddr
);
388 *h_page_size
= PRTBE_R_GET_RTS(pate
.dw0
);
389 /* No valid pte or access denied due to protection */
390 if (ppc_radix64_walk_tree(CPU(cpu
)->as
, g_raddr
, pate
.dw0
& PRTBE_R_RPDB
,
391 pate
.dw0
& PRTBE_R_RPDS
, h_raddr
, h_page_size
,
392 &pte
, &fault_cause
, &pte_addr
) ||
393 ppc_radix64_check_prot(cpu
, access_type
, pte
,
394 &fault_cause
, h_prot
, mmu_idx
, true)) {
395 if (pde_addr
) { /* address being translated was that of a guest pde */
396 fault_cause
|= DSISR_PRTABLE_FAULT
;
399 ppc_radix64_raise_hsi(cpu
, access_type
, eaddr
, g_raddr
, fault_cause
);
405 ppc_radix64_set_rc(cpu
, access_type
, pte
, pte_addr
, h_prot
);
412 * The spapr vhc has a flat partition scope provided by qemu memory when
415 * When running a nested guest, the addressing is 2-level radix on top of the
416 * vhc memory, so it works practically identically to the bare metal 2-level
417 * radix. So that code is selected directly. A cleaner and more flexible nested
418 * hypervisor implementation would allow the vhc to provide a ->nested_xlate()
419 * function but that is not required for the moment.
421 static bool vhyp_flat_addressing(PowerPCCPU
*cpu
)
424 return !vhyp_cpu_in_nested(cpu
);
429 static int ppc_radix64_process_scoped_xlate(PowerPCCPU
*cpu
,
430 MMUAccessType access_type
,
431 vaddr eaddr
, uint64_t pid
,
432 ppc_v3_pate_t pate
, hwaddr
*g_raddr
,
433 int *g_prot
, int *g_page_size
,
434 int mmu_idx
, bool guest_visible
)
436 CPUState
*cs
= CPU(cpu
);
437 CPUPPCState
*env
= &cpu
->env
;
438 uint64_t offset
, size
, prtb
, prtbe_addr
, prtbe0
, base_addr
, nls
, index
, pte
;
439 int fault_cause
= 0, h_page_size
, h_prot
;
440 hwaddr h_raddr
, pte_addr
;
443 qemu_log_mask(CPU_LOG_MMU
, "%s for %s @0x%"VADDR_PRIx
444 " mmu_idx %u pid %"PRIu64
"\n",
445 __func__
, access_str(access_type
),
446 eaddr
, mmu_idx
, pid
);
448 prtb
= (pate
.dw1
& PATE1_R_PRTB
);
449 size
= 1ULL << ((pate
.dw1
& PATE1_R_PRTS
) + 12);
450 if (prtb
& (size
- 1)) {
451 /* Process Table not properly aligned */
453 ppc_radix64_raise_si(cpu
, access_type
, eaddr
, DSISR_R_BADCONFIG
);
458 /* Index Process Table by PID to Find Corresponding Process Table Entry */
459 offset
= pid
* sizeof(struct prtb_entry
);
460 if (offset
>= size
) {
461 /* offset exceeds size of the process table */
463 ppc_radix64_raise_si(cpu
, access_type
, eaddr
, DSISR_NOPTE
);
467 prtbe_addr
= prtb
+ offset
;
469 if (vhyp_flat_addressing(cpu
)) {
470 prtbe0
= ldq_phys(cs
->as
, prtbe_addr
);
473 * Process table addresses are subject to partition-scoped
476 * On a Radix host, the partition-scoped page table for LPID=0
477 * is only used to translate the effective addresses of the
478 * process table entries.
480 ret
= ppc_radix64_partition_scoped_xlate(cpu
, 0, eaddr
, prtbe_addr
,
481 pate
, &h_raddr
, &h_prot
,
483 /* mmu_idx is 5 because we're translating from hypervisor scope */
488 prtbe0
= ldq_phys(cs
->as
, h_raddr
);
491 /* Walk Radix Tree from Process Table Entry to Convert EA to RA */
492 *g_page_size
= PRTBE_R_GET_RTS(prtbe0
);
493 base_addr
= prtbe0
& PRTBE_R_RPDB
;
494 nls
= prtbe0
& PRTBE_R_RPDS
;
495 if (FIELD_EX64(env
->msr
, MSR
, HV
) || vhyp_flat_addressing(cpu
)) {
497 * Can treat process table addresses as real addresses
499 ret
= ppc_radix64_walk_tree(cs
->as
, eaddr
& R_EADDR_MASK
, base_addr
,
500 nls
, g_raddr
, g_page_size
, &pte
,
501 &fault_cause
, &pte_addr
);
505 ppc_radix64_raise_si(cpu
, access_type
, eaddr
, fault_cause
);
513 index
= (eaddr
& R_EADDR_MASK
) >> (*g_page_size
- nls
); /* Shift */
514 index
&= ((1UL << nls
) - 1); /* Mask */
515 pte_addr
= base_addr
+ (index
* sizeof(pte
));
518 * Each process table address is subject to a partition-scoped
522 ret
= ppc_radix64_partition_scoped_xlate(cpu
, 0, eaddr
, pte_addr
,
523 pate
, &h_raddr
, &h_prot
,
525 /* mmu_idx is 5 because we're translating from hypervisor scope */
531 if (!ppc_radix64_is_valid_level(level
++, *g_page_size
, nls
)) {
532 fault_cause
|= DSISR_R_BADCONFIG
;
535 ret
= ppc_radix64_next_level(cs
->as
, eaddr
& R_EADDR_MASK
,
536 &h_raddr
, &nls
, g_page_size
,
543 ppc_radix64_raise_si(cpu
, access_type
, eaddr
, fault_cause
);
548 } while (!(pte
& R_PTE_LEAF
));
550 rpn
= pte
& R_PTE_RPN
;
551 mask
= (1UL << *g_page_size
) - 1;
553 /* Or high bits of rpn and low bits to ea to form whole real addr */
554 *g_raddr
= (rpn
& ~mask
) | (eaddr
& mask
);
557 if (ppc_radix64_check_prot(cpu
, access_type
, pte
, &fault_cause
,
558 g_prot
, mmu_idx
, false)) {
559 /* Access denied due to protection */
561 ppc_radix64_raise_si(cpu
, access_type
, eaddr
, fault_cause
);
567 ppc_radix64_set_rc(cpu
, access_type
, pte
, pte_addr
, g_prot
);
574 * Radix tree translation is a 2 steps translation process:
576 * 1. Process-scoped translation: Guest Eff Addr -> Guest Real Addr
577 * 2. Partition-scoped translation: Guest Real Addr -> Host Real Addr
580 * +-------------+----------------+---------------+
581 * | | HV = 0 | HV = 1 |
582 * +-------------+----------------+---------------+
583 * | Relocation | Partition | No |
584 * | = Off | Scoped | Translation |
585 * Relocation +-------------+----------------+---------------+
586 * | Relocation | Partition & | Process |
587 * | = On | Process Scoped | Scoped |
588 * +-------------+----------------+---------------+
590 static bool ppc_radix64_xlate_impl(PowerPCCPU
*cpu
, vaddr eaddr
,
591 MMUAccessType access_type
, hwaddr
*raddr
,
592 int *psizep
, int *protp
, int mmu_idx
,
595 CPUPPCState
*env
= &cpu
->env
;
602 assert(!(mmuidx_hv(mmu_idx
) && cpu
->vhyp
));
604 relocation
= !mmuidx_real(mmu_idx
);
606 /* HV or virtual hypervisor Real Mode Access */
607 if (!relocation
&& (mmuidx_hv(mmu_idx
) || vhyp_flat_addressing(cpu
))) {
608 /* In real mode top 4 effective addr bits (mostly) ignored */
609 *raddr
= eaddr
& 0x0FFFFFFFFFFFFFFFULL
;
611 /* In HV mode, add HRMOR if top EA bit is clear */
612 if (mmuidx_hv(mmu_idx
) || !env
->has_hv_mode
) {
613 if (!(eaddr
>> 63)) {
614 *raddr
|= env
->spr
[SPR_HRMOR
];
617 *protp
= PAGE_READ
| PAGE_WRITE
| PAGE_EXEC
;
618 *psizep
= TARGET_PAGE_BITS
;
623 * Check UPRT (we avoid the check in real mode to deal with
624 * transitional states during kexec.
626 if (guest_visible
&& !ppc64_use_proc_tbl(cpu
)) {
627 qemu_log_mask(LOG_GUEST_ERROR
,
628 "LPCR:UPRT not set in radix mode ! LPCR="
629 TARGET_FMT_lx
"\n", env
->spr
[SPR_LPCR
]);
632 /* Virtual Mode Access - get the fully qualified address */
633 if (!ppc_radix64_get_fully_qualified_addr(&cpu
->env
, eaddr
, &lpid
, &pid
)) {
635 ppc_radix64_raise_segi(cpu
, access_type
, eaddr
);
640 /* Get Partition Table */
642 PPCVirtualHypervisorClass
*vhc
;
643 vhc
= PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu
->vhyp
);
644 if (!vhc
->get_pate(cpu
->vhyp
, cpu
, lpid
, &pate
)) {
646 ppc_radix64_raise_hsi(cpu
, access_type
, eaddr
, eaddr
,
652 if (!ppc64_v3_get_pate(cpu
, lpid
, &pate
)) {
654 ppc_radix64_raise_hsi(cpu
, access_type
, eaddr
, eaddr
,
659 if (!validate_pate(cpu
, lpid
, &pate
)) {
661 ppc_radix64_raise_hsi(cpu
, access_type
, eaddr
, eaddr
,
669 *protp
= PAGE_READ
| PAGE_WRITE
| PAGE_EXEC
;
672 * Perform process-scoped translation if relocation enabled.
674 * - Translates an effective address to a host real address in
675 * quadrants 0 and 3 when HV=1.
677 * - Translates an effective address to a guest real address.
680 int ret
= ppc_radix64_process_scoped_xlate(cpu
, access_type
, eaddr
, pid
,
681 pate
, &g_raddr
, &prot
,
682 &psize
, mmu_idx
, guest_visible
);
686 *psizep
= MIN(*psizep
, psize
);
689 g_raddr
= eaddr
& R_EADDR_MASK
;
692 if (vhyp_flat_addressing(cpu
)) {
696 * Perform partition-scoped translation if !HV or HV access to
697 * quadrants 1 or 2. Translates a guest real address to a host
700 if (lpid
|| !mmuidx_hv(mmu_idx
)) {
703 ret
= ppc_radix64_partition_scoped_xlate(cpu
, access_type
, eaddr
,
704 g_raddr
, pate
, raddr
,
705 &prot
, &psize
, false,
706 mmu_idx
, guest_visible
);
710 *psizep
= MIN(*psizep
, psize
);
720 bool ppc_radix64_xlate(PowerPCCPU
*cpu
, vaddr eaddr
, MMUAccessType access_type
,
721 hwaddr
*raddrp
, int *psizep
, int *protp
, int mmu_idx
,
724 bool ret
= ppc_radix64_xlate_impl(cpu
, eaddr
, access_type
, raddrp
,
725 psizep
, protp
, mmu_idx
, guest_visible
);
727 qemu_log_mask(CPU_LOG_MMU
, "%s for %s @0x%"VADDR_PRIx
728 " mmu_idx %u (prot %c%c%c) -> 0x%"HWADDR_PRIx
"\n",
729 __func__
, access_str(access_type
),
731 *protp
& PAGE_READ
? 'r' : '-',
732 *protp
& PAGE_WRITE
? 'w' : '-',
733 *protp
& PAGE_EXEC
? 'x' : '-',