3 * Copyright (C) 2001 Todd Inglett, IBM Corporation
5 * pSeries LPAR support.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 /* Enables debugging of low-level hash table routines - careful! */
25 #include <linux/kernel.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/console.h>
28 #include <linux/export.h>
29 #include <asm/processor.h>
32 #include <asm/pgtable.h>
33 #include <asm/machdep.h>
34 #include <asm/mmu_context.h>
35 #include <asm/iommu.h>
36 #include <asm/tlbflush.h>
39 #include <asm/cputable.h>
42 #include <asm/trace.h>
43 #include <asm/firmware.h>
45 #include "plpar_wrappers.h"
50 EXPORT_SYMBOL(plpar_hcall
);
51 EXPORT_SYMBOL(plpar_hcall9
);
52 EXPORT_SYMBOL(plpar_hcall_norets
);
54 extern void pSeries_find_serial_port(void);
56 void vpa_init(int cpu
)
58 int hwcpu
= get_hard_smp_processor_id(cpu
);
61 struct paca_struct
*pp
;
62 struct dtl_entry
*dtl
;
64 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
65 lppaca_of(cpu
).vmxregs_in_use
= 1;
67 addr
= __pa(&lppaca_of(cpu
));
68 ret
= register_vpa(hwcpu
, addr
);
71 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
72 "%lx failed with %ld\n", cpu
, hwcpu
, addr
, ret
);
76 * PAPR says this feature is SLB-Buffer but firmware never
77 * reports that. All SPLPAR support SLB shadow buffer.
79 addr
= __pa(&slb_shadow
[cpu
]);
80 if (firmware_has_feature(FW_FEATURE_SPLPAR
)) {
81 ret
= register_slb_shadow(hwcpu
, addr
);
83 pr_err("WARNING: SLB shadow buffer registration for "
84 "cpu %d (hw %d) of area %lx failed with %ld\n",
85 cpu
, hwcpu
, addr
, ret
);
89 * Register dispatch trace log, if one has been allocated.
92 dtl
= pp
->dispatch_log
;
96 lppaca_of(cpu
).dtl_idx
= 0;
98 /* hypervisor reads buffer length from this field */
99 dtl
->enqueue_to_dispatch_time
= DISPATCH_LOG_BYTES
;
100 ret
= register_dtl(hwcpu
, __pa(dtl
));
102 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
103 "failed with %ld\n", smp_processor_id(),
105 lppaca_of(cpu
).dtl_enable_mask
= 2;
109 static long pSeries_lpar_hpte_insert(unsigned long hpte_group
,
110 unsigned long vpn
, unsigned long pa
,
111 unsigned long rflags
, unsigned long vflags
,
112 int psize
, int apsize
, int ssize
)
114 unsigned long lpar_rc
;
117 unsigned long hpte_v
, hpte_r
;
119 if (!(vflags
& HPTE_V_BOLTED
))
120 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
121 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
122 hpte_group
, vpn
, pa
, rflags
, vflags
, psize
);
124 hpte_v
= hpte_encode_v(vpn
, psize
, apsize
, ssize
) | vflags
| HPTE_V_VALID
;
125 hpte_r
= hpte_encode_r(pa
, psize
, apsize
) | rflags
;
127 if (!(vflags
& HPTE_V_BOLTED
))
128 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v
, hpte_r
);
130 /* Now fill in the actual HPTE */
131 /* Set CEC cookie to 0 */
133 /* I-cache Invalidate = 0 */
134 /* I-cache synchronize = 0 */
138 /* Make pHyp happy */
139 if ((rflags
& _PAGE_NO_CACHE
) & !(rflags
& _PAGE_WRITETHRU
))
140 hpte_r
&= ~_PAGE_COHERENT
;
141 if (firmware_has_feature(FW_FEATURE_XCMO
) && !(hpte_r
& HPTE_R_N
))
142 flags
|= H_COALESCE_CAND
;
144 lpar_rc
= plpar_pte_enter(flags
, hpte_group
, hpte_v
, hpte_r
, &slot
);
145 if (unlikely(lpar_rc
== H_PTEG_FULL
)) {
146 if (!(vflags
& HPTE_V_BOLTED
))
152 * Since we try and ioremap PHBs we don't own, the pte insert
153 * will fail. However we must catch the failure in hash_page
154 * or we will loop forever, so return -2 in this case.
156 if (unlikely(lpar_rc
!= H_SUCCESS
)) {
157 if (!(vflags
& HPTE_V_BOLTED
))
158 pr_devel(" lpar err %ld\n", lpar_rc
);
161 if (!(vflags
& HPTE_V_BOLTED
))
162 pr_devel(" -> slot: %lu\n", slot
& 7);
164 /* Because of iSeries, we have to pass down the secondary
165 * bucket bit here as well
167 return (slot
& 7) | (!!(vflags
& HPTE_V_SECONDARY
) << 3);
170 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock
);
172 static long pSeries_lpar_hpte_remove(unsigned long hpte_group
)
174 unsigned long slot_offset
;
175 unsigned long lpar_rc
;
177 unsigned long dummy1
, dummy2
;
179 /* pick a random slot to start at */
180 slot_offset
= mftb() & 0x7;
182 for (i
= 0; i
< HPTES_PER_GROUP
; i
++) {
184 /* don't remove a bolted entry */
185 lpar_rc
= plpar_pte_remove(H_ANDCOND
, hpte_group
+ slot_offset
,
186 (0x1UL
<< 4), &dummy1
, &dummy2
);
187 if (lpar_rc
== H_SUCCESS
)
191 * The test for adjunct partition is performed before the
192 * ANDCOND test. H_RESOURCE may be returned, so we need to
193 * check for that as well.
195 BUG_ON(lpar_rc
!= H_NOT_FOUND
&& lpar_rc
!= H_RESOURCE
);
204 static void pSeries_lpar_hptab_clear(void)
206 unsigned long size_bytes
= 1UL << ppc64_pft_size
;
207 unsigned long hpte_count
= size_bytes
>> 4;
215 /* Read in batches of 4,
216 * invalidate only valid entries not in the VRMA
217 * hpte_count will be a multiple of 4
219 for (i
= 0; i
< hpte_count
; i
+= 4) {
220 lpar_rc
= plpar_pte_read_4_raw(0, i
, (void *)ptes
);
221 if (lpar_rc
!= H_SUCCESS
)
223 for (j
= 0; j
< 4; j
++){
224 if ((ptes
[j
].pteh
& HPTE_V_VRMA_MASK
) ==
227 if (ptes
[j
].pteh
& HPTE_V_VALID
)
228 plpar_pte_remove_raw(0, i
+ j
, 0,
229 &(ptes
[j
].pteh
), &(ptes
[j
].ptel
));
235 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
236 * the low 3 bits of flags happen to line up. So no transform is needed.
237 * We can probably optimize here and assume the high bits of newpp are
238 * already zero. For now I am paranoid.
240 static long pSeries_lpar_hpte_updatepp(unsigned long slot
,
243 int psize
, int ssize
, int local
)
245 unsigned long lpar_rc
;
246 unsigned long flags
= (newpp
& 7) | H_AVPN
;
247 unsigned long want_v
;
249 want_v
= hpte_encode_avpn(vpn
, psize
, ssize
);
251 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
252 want_v
, slot
, flags
, psize
);
254 lpar_rc
= plpar_pte_protect(flags
, slot
, want_v
);
256 if (lpar_rc
== H_NOT_FOUND
) {
257 pr_devel("not found !\n");
263 BUG_ON(lpar_rc
!= H_SUCCESS
);
268 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot
)
270 unsigned long dword0
;
271 unsigned long lpar_rc
;
272 unsigned long dummy_word1
;
275 /* Read 1 pte at a time */
276 /* Do not need RPN to logical page translation */
277 /* No cross CEC PFT access */
280 lpar_rc
= plpar_pte_read(flags
, slot
, &dword0
, &dummy_word1
);
282 BUG_ON(lpar_rc
!= H_SUCCESS
);
287 static long pSeries_lpar_hpte_find(unsigned long vpn
, int psize
, int ssize
)
292 unsigned long want_v
, hpte_v
;
294 hash
= hpt_hash(vpn
, mmu_psize_defs
[psize
].shift
, ssize
);
295 want_v
= hpte_encode_avpn(vpn
, psize
, ssize
);
297 /* Bolted entries are always in the primary group */
298 slot
= (hash
& htab_hash_mask
) * HPTES_PER_GROUP
;
299 for (i
= 0; i
< HPTES_PER_GROUP
; i
++) {
300 hpte_v
= pSeries_lpar_hpte_getword0(slot
);
302 if (HPTE_V_COMPARE(hpte_v
, want_v
) && (hpte_v
& HPTE_V_VALID
))
311 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp
,
313 int psize
, int ssize
)
316 unsigned long lpar_rc
, slot
, vsid
, flags
;
318 vsid
= get_kernel_vsid(ea
, ssize
);
319 vpn
= hpt_vpn(ea
, vsid
, ssize
);
321 slot
= pSeries_lpar_hpte_find(vpn
, psize
, ssize
);
325 lpar_rc
= plpar_pte_protect(flags
, slot
, 0);
327 BUG_ON(lpar_rc
!= H_SUCCESS
);
330 static void pSeries_lpar_hpte_invalidate(unsigned long slot
, unsigned long vpn
,
331 int psize
, int ssize
, int local
)
333 unsigned long want_v
;
334 unsigned long lpar_rc
;
335 unsigned long dummy1
, dummy2
;
337 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
338 slot
, vpn
, psize
, local
);
340 want_v
= hpte_encode_avpn(vpn
, psize
, ssize
);
341 lpar_rc
= plpar_pte_remove(H_AVPN
, slot
, want_v
, &dummy1
, &dummy2
);
342 if (lpar_rc
== H_NOT_FOUND
)
345 BUG_ON(lpar_rc
!= H_SUCCESS
);
348 static void pSeries_lpar_hpte_removebolted(unsigned long ea
,
349 int psize
, int ssize
)
352 unsigned long slot
, vsid
;
354 vsid
= get_kernel_vsid(ea
, ssize
);
355 vpn
= hpt_vpn(ea
, vsid
, ssize
);
357 slot
= pSeries_lpar_hpte_find(vpn
, psize
, ssize
);
360 pSeries_lpar_hpte_invalidate(slot
, vpn
, psize
, ssize
, 0);
363 /* Flag bits for H_BULK_REMOVE */
364 #define HBR_REQUEST 0x4000000000000000UL
365 #define HBR_RESPONSE 0x8000000000000000UL
366 #define HBR_END 0xc000000000000000UL
367 #define HBR_AVPN 0x0200000000000000UL
368 #define HBR_ANDCOND 0x0100000000000000UL
371 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
374 static void pSeries_lpar_flush_hash_range(unsigned long number
, int local
)
377 unsigned long i
, pix
, rc
;
378 unsigned long flags
= 0;
379 struct ppc64_tlb_batch
*batch
= &__get_cpu_var(ppc64_tlb_batch
);
380 int lock_tlbie
= !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE
);
381 unsigned long param
[9];
382 unsigned long hash
, index
, shift
, hidx
, slot
;
387 spin_lock_irqsave(&pSeries_lpar_tlbie_lock
, flags
);
389 psize
= batch
->psize
;
390 ssize
= batch
->ssize
;
392 for (i
= 0; i
< number
; i
++) {
395 pte_iterate_hashed_subpages(pte
, psize
, vpn
, index
, shift
) {
396 hash
= hpt_hash(vpn
, shift
, ssize
);
397 hidx
= __rpte_to_hidx(pte
, index
);
398 if (hidx
& _PTEIDX_SECONDARY
)
400 slot
= (hash
& htab_hash_mask
) * HPTES_PER_GROUP
;
401 slot
+= hidx
& _PTEIDX_GROUP_IX
;
402 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE
)) {
403 pSeries_lpar_hpte_invalidate(slot
, vpn
, psize
,
406 param
[pix
] = HBR_REQUEST
| HBR_AVPN
| slot
;
407 param
[pix
+1] = hpte_encode_avpn(vpn
, psize
,
411 rc
= plpar_hcall9(H_BULK_REMOVE
, param
,
412 param
[0], param
[1], param
[2],
413 param
[3], param
[4], param
[5],
415 BUG_ON(rc
!= H_SUCCESS
);
419 } pte_iterate_hashed_end();
422 param
[pix
] = HBR_END
;
423 rc
= plpar_hcall9(H_BULK_REMOVE
, param
, param
[0], param
[1],
424 param
[2], param
[3], param
[4], param
[5],
426 BUG_ON(rc
!= H_SUCCESS
);
430 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock
, flags
);
433 static int __init
disable_bulk_remove(char *str
)
435 if (strcmp(str
, "off") == 0 &&
436 firmware_has_feature(FW_FEATURE_BULK_REMOVE
)) {
437 printk(KERN_INFO
"Disabling BULK_REMOVE firmware feature");
438 powerpc_firmware_features
&= ~FW_FEATURE_BULK_REMOVE
;
443 __setup("bulk_remove=", disable_bulk_remove
);
445 void __init
hpte_init_lpar(void)
447 ppc_md
.hpte_invalidate
= pSeries_lpar_hpte_invalidate
;
448 ppc_md
.hpte_updatepp
= pSeries_lpar_hpte_updatepp
;
449 ppc_md
.hpte_updateboltedpp
= pSeries_lpar_hpte_updateboltedpp
;
450 ppc_md
.hpte_insert
= pSeries_lpar_hpte_insert
;
451 ppc_md
.hpte_remove
= pSeries_lpar_hpte_remove
;
452 ppc_md
.hpte_removebolted
= pSeries_lpar_hpte_removebolted
;
453 ppc_md
.flush_hash_range
= pSeries_lpar_flush_hash_range
;
454 ppc_md
.hpte_clear_all
= pSeries_lpar_hptab_clear
;
457 #ifdef CONFIG_PPC_SMLPAR
458 #define CMO_FREE_HINT_DEFAULT 1
459 static int cmo_free_hint_flag
= CMO_FREE_HINT_DEFAULT
;
461 static int __init
cmo_free_hint(char *str
)
464 parm
= strstrip(str
);
466 if (strcasecmp(parm
, "no") == 0 || strcasecmp(parm
, "off") == 0) {
467 printk(KERN_INFO
"cmo_free_hint: CMO free page hinting is not active.\n");
468 cmo_free_hint_flag
= 0;
472 cmo_free_hint_flag
= 1;
473 printk(KERN_INFO
"cmo_free_hint: CMO free page hinting is active.\n");
475 if (strcasecmp(parm
, "yes") == 0 || strcasecmp(parm
, "on") == 0)
481 __setup("cmo_free_hint=", cmo_free_hint
);
483 static void pSeries_set_page_state(struct page
*page
, int order
,
487 unsigned long cmo_page_sz
, addr
;
489 cmo_page_sz
= cmo_get_page_size();
490 addr
= __pa((unsigned long)page_address(page
));
492 for (i
= 0; i
< (1 << order
); i
++, addr
+= PAGE_SIZE
) {
493 for (j
= 0; j
< PAGE_SIZE
; j
+= cmo_page_sz
)
494 plpar_hcall_norets(H_PAGE_INIT
, state
, addr
+ j
, 0);
498 void arch_free_page(struct page
*page
, int order
)
500 if (!cmo_free_hint_flag
|| !firmware_has_feature(FW_FEATURE_CMO
))
503 pSeries_set_page_state(page
, order
, H_PAGE_SET_UNUSED
);
505 EXPORT_SYMBOL(arch_free_page
);
509 #ifdef CONFIG_TRACEPOINTS
511 * We optimise our hcall path by placing hcall_tracepoint_refcount
512 * directly in the TOC so we can check if the hcall tracepoints are
513 * enabled via a single load.
516 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
517 extern long hcall_tracepoint_refcount
;
520 * Since the tracing code might execute hcalls we need to guard against
521 * recursion. One example of this are spinlocks calling H_YIELD on
522 * shared processor partitions.
524 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth
);
526 void hcall_tracepoint_regfunc(void)
528 hcall_tracepoint_refcount
++;
531 void hcall_tracepoint_unregfunc(void)
533 hcall_tracepoint_refcount
--;
536 void __trace_hcall_entry(unsigned long opcode
, unsigned long *args
)
542 * We cannot call tracepoints inside RCU idle regions which
543 * means we must not trace H_CEDE.
545 if (opcode
== H_CEDE
)
548 local_irq_save(flags
);
550 depth
= &__get_cpu_var(hcall_trace_depth
);
557 trace_hcall_entry(opcode
, args
);
561 local_irq_restore(flags
);
564 void __trace_hcall_exit(long opcode
, unsigned long retval
,
565 unsigned long *retbuf
)
570 if (opcode
== H_CEDE
)
573 local_irq_save(flags
);
575 depth
= &__get_cpu_var(hcall_trace_depth
);
581 trace_hcall_exit(opcode
, retval
, retbuf
);
586 local_irq_restore(flags
);
592 * H_GET_MPP hcall returns info in 7 parms
594 int h_get_mpp(struct hvcall_mpp_data
*mpp_data
)
597 unsigned long retbuf
[PLPAR_HCALL9_BUFSIZE
];
599 rc
= plpar_hcall9(H_GET_MPP
, retbuf
);
601 mpp_data
->entitled_mem
= retbuf
[0];
602 mpp_data
->mapped_mem
= retbuf
[1];
604 mpp_data
->group_num
= (retbuf
[2] >> 2 * 8) & 0xffff;
605 mpp_data
->pool_num
= retbuf
[2] & 0xffff;
607 mpp_data
->mem_weight
= (retbuf
[3] >> 7 * 8) & 0xff;
608 mpp_data
->unallocated_mem_weight
= (retbuf
[3] >> 6 * 8) & 0xff;
609 mpp_data
->unallocated_entitlement
= retbuf
[3] & 0xffffffffffff;
611 mpp_data
->pool_size
= retbuf
[4];
612 mpp_data
->loan_request
= retbuf
[5];
613 mpp_data
->backing_mem
= retbuf
[6];
617 EXPORT_SYMBOL(h_get_mpp
);
619 int h_get_mpp_x(struct hvcall_mpp_x_data
*mpp_x_data
)
622 unsigned long retbuf
[PLPAR_HCALL9_BUFSIZE
] = { 0 };
624 rc
= plpar_hcall9(H_GET_MPP_X
, retbuf
);
626 mpp_x_data
->coalesced_bytes
= retbuf
[0];
627 mpp_x_data
->pool_coalesced_bytes
= retbuf
[1];
628 mpp_x_data
->pool_purr_cycles
= retbuf
[2];
629 mpp_x_data
->pool_spurr_cycles
= retbuf
[3];