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 <asm/processor.h>
31 #include <asm/pgtable.h>
32 #include <asm/machdep.h>
33 #include <asm/abs_addr.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>
44 #include "plpar_wrappers.h"
49 EXPORT_SYMBOL(plpar_hcall);
50 EXPORT_SYMBOL(plpar_hcall9);
51 EXPORT_SYMBOL(plpar_hcall_norets);
53 extern void pSeries_find_serial_port(void);
56 static int vtermno; /* virtual terminal# for udbg */
58 #define __ALIGNED__ __attribute__((__aligned__(sizeof(long))))
59 static void udbg_hvsi_putc(char c)
61 /* packet's seqno isn't used anyways */
62 uint8_t packet[] __ALIGNED__ = { 0xff, 5, 0, 0, c };
69 rc = plpar_put_term_char(vtermno, sizeof(packet), packet);
70 } while (rc == H_BUSY);
73 static long hvsi_udbg_buf_len;
74 static uint8_t hvsi_udbg_buf[256];
76 static int udbg_hvsi_getc_poll(void)
81 if (hvsi_udbg_buf_len == 0) {
82 rc = plpar_get_term_char(vtermno, &hvsi_udbg_buf_len, hvsi_udbg_buf);
83 if (rc != H_SUCCESS || hvsi_udbg_buf[0] != 0xff) {
84 /* bad read or non-data packet */
85 hvsi_udbg_buf_len = 0;
87 /* remove the packet header */
88 for (i = 4; i < hvsi_udbg_buf_len; i++)
89 hvsi_udbg_buf[i-4] = hvsi_udbg_buf[i];
90 hvsi_udbg_buf_len -= 4;
94 if (hvsi_udbg_buf_len <= 0 || hvsi_udbg_buf_len > 256) {
96 hvsi_udbg_buf_len = 0;
100 ch = hvsi_udbg_buf[0];
101 /* shift remaining data down */
102 for (i = 1; i < hvsi_udbg_buf_len; i++) {
103 hvsi_udbg_buf[i-1] = hvsi_udbg_buf[i];
110 static int udbg_hvsi_getc(void)
114 ch = udbg_hvsi_getc_poll();
116 /* This shouldn't be needed...but... */
117 volatile unsigned long delay;
118 for (delay=0; delay < 2000000; delay++)
126 static void udbg_putcLP(char c)
136 rc = plpar_put_term_char(vtermno, 1, buf);
137 } while(rc == H_BUSY);
140 /* Buffered chars getc */
141 static long inbuflen;
142 static long inbuf[2]; /* must be 2 longs */
144 static int udbg_getc_pollLP(void)
146 /* The interface is tricky because it may return up to 16 chars.
147 * We save them statically for future calls to udbg_getc().
149 char ch, *buf = (char *)inbuf;
153 /* get some more chars. */
155 rc = plpar_get_term_char(vtermno, &inbuflen, buf);
157 inbuflen = 0; /* otherwise inbuflen is garbage */
159 if (inbuflen <= 0 || inbuflen > 16) {
160 /* Catch error case as well as other oddities (corruption) */
165 for (i = 1; i < inbuflen; i++) /* shuffle them down. */
171 static int udbg_getcLP(void)
175 ch = udbg_getc_pollLP();
177 /* This shouldn't be needed...but... */
178 volatile unsigned long delay;
179 for (delay=0; delay < 2000000; delay++)
187 /* call this from early_init() for a working debug console on
188 * vterm capable LPAR machines
190 void __init udbg_init_debug_lpar(void)
193 udbg_putc = udbg_putcLP;
194 udbg_getc = udbg_getcLP;
195 udbg_getc_poll = udbg_getc_pollLP;
197 register_early_udbg_console();
200 /* returns 0 if couldn't find or use /chosen/stdout as console */
201 void __init find_udbg_vterm(void)
203 struct device_node *stdout_node;
207 /* find the boot console from /chosen/stdout */
210 name = of_get_property(of_chosen, "linux,stdout-path", NULL);
213 stdout_node = of_find_node_by_path(name);
216 name = of_get_property(stdout_node, "name", NULL);
218 printk(KERN_WARNING "stdout node missing 'name' property!\n");
222 /* Check if it's a virtual terminal */
223 if (strncmp(name, "vty", 3) != 0)
225 termno = of_get_property(stdout_node, "reg", NULL);
230 if (of_device_is_compatible(stdout_node, "hvterm1")) {
231 udbg_putc = udbg_putcLP;
232 udbg_getc = udbg_getcLP;
233 udbg_getc_poll = udbg_getc_pollLP;
234 add_preferred_console("hvc", termno[0] & 0xff, NULL);
235 } else if (of_device_is_compatible(stdout_node, "hvterm-protocol")) {
237 udbg_putc = udbg_hvsi_putc;
238 udbg_getc = udbg_hvsi_getc;
239 udbg_getc_poll = udbg_hvsi_getc_poll;
240 add_preferred_console("hvsi", termno[0] & 0xff, NULL);
243 of_node_put(stdout_node);
246 void vpa_init(int cpu)
248 int hwcpu = get_hard_smp_processor_id(cpu);
251 struct paca_struct *pp;
252 struct dtl_entry *dtl;
254 if (cpu_has_feature(CPU_FTR_ALTIVEC))
255 lppaca_of(cpu).vmxregs_in_use = 1;
257 addr = __pa(&lppaca_of(cpu));
258 ret = register_vpa(hwcpu, addr);
261 printk(KERN_ERR "WARNING: vpa_init: VPA registration for "
262 "cpu %d (hw %d) of area %lx returns %ld\n",
263 cpu, hwcpu, addr, ret);
267 * PAPR says this feature is SLB-Buffer but firmware never
268 * reports that. All SPLPAR support SLB shadow buffer.
270 addr = __pa(&slb_shadow[cpu]);
271 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
272 ret = register_slb_shadow(hwcpu, addr);
275 "WARNING: vpa_init: SLB shadow buffer "
276 "registration for cpu %d (hw %d) of area %lx "
277 "returns %ld\n", cpu, hwcpu, addr, ret);
281 * Register dispatch trace log, if one has been allocated.
284 dtl = pp->dispatch_log;
288 lppaca_of(cpu).dtl_idx = 0;
290 /* hypervisor reads buffer length from this field */
291 dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
292 ret = register_dtl(hwcpu, __pa(dtl));
294 pr_warn("DTL registration failed for cpu %d (%ld)\n",
296 lppaca_of(cpu).dtl_enable_mask = 2;
300 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
301 unsigned long va, unsigned long pa,
302 unsigned long rflags, unsigned long vflags,
303 int psize, int ssize)
305 unsigned long lpar_rc;
308 unsigned long hpte_v, hpte_r;
310 if (!(vflags & HPTE_V_BOLTED))
311 pr_devel("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
312 "rflags=%lx, vflags=%lx, psize=%d)\n",
313 hpte_group, va, pa, rflags, vflags, psize);
315 hpte_v = hpte_encode_v(va, psize, ssize) | vflags | HPTE_V_VALID;
316 hpte_r = hpte_encode_r(pa, psize) | rflags;
318 if (!(vflags & HPTE_V_BOLTED))
319 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
321 /* Now fill in the actual HPTE */
322 /* Set CEC cookie to 0 */
324 /* I-cache Invalidate = 0 */
325 /* I-cache synchronize = 0 */
329 /* Make pHyp happy */
330 if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
331 hpte_r &= ~_PAGE_COHERENT;
332 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
333 flags |= H_COALESCE_CAND;
335 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
336 if (unlikely(lpar_rc == H_PTEG_FULL)) {
337 if (!(vflags & HPTE_V_BOLTED))
343 * Since we try and ioremap PHBs we don't own, the pte insert
344 * will fail. However we must catch the failure in hash_page
345 * or we will loop forever, so return -2 in this case.
347 if (unlikely(lpar_rc != H_SUCCESS)) {
348 if (!(vflags & HPTE_V_BOLTED))
349 pr_devel(" lpar err %lu\n", lpar_rc);
352 if (!(vflags & HPTE_V_BOLTED))
353 pr_devel(" -> slot: %lu\n", slot & 7);
355 /* Because of iSeries, we have to pass down the secondary
356 * bucket bit here as well
358 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
361 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
363 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
365 unsigned long slot_offset;
366 unsigned long lpar_rc;
368 unsigned long dummy1, dummy2;
370 /* pick a random slot to start at */
371 slot_offset = mftb() & 0x7;
373 for (i = 0; i < HPTES_PER_GROUP; i++) {
375 /* don't remove a bolted entry */
376 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
377 (0x1UL << 4), &dummy1, &dummy2);
378 if (lpar_rc == H_SUCCESS)
380 BUG_ON(lpar_rc != H_NOT_FOUND);
389 static void pSeries_lpar_hptab_clear(void)
391 unsigned long size_bytes = 1UL << ppc64_pft_size;
392 unsigned long hpte_count = size_bytes >> 4;
400 /* Read in batches of 4,
401 * invalidate only valid entries not in the VRMA
402 * hpte_count will be a multiple of 4
404 for (i = 0; i < hpte_count; i += 4) {
405 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
406 if (lpar_rc != H_SUCCESS)
408 for (j = 0; j < 4; j++){
409 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
412 if (ptes[j].pteh & HPTE_V_VALID)
413 plpar_pte_remove_raw(0, i + j, 0,
414 &(ptes[j].pteh), &(ptes[j].ptel));
420 * This computes the AVPN and B fields of the first dword of a HPTE,
421 * for use when we want to match an existing PTE. The bottom 7 bits
422 * of the returned value are zero.
424 static inline unsigned long hpte_encode_avpn(unsigned long va, int psize,
429 v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
430 v <<= HPTE_V_AVPN_SHIFT;
431 v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
436 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
437 * the low 3 bits of flags happen to line up. So no transform is needed.
438 * We can probably optimize here and assume the high bits of newpp are
439 * already zero. For now I am paranoid.
441 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
444 int psize, int ssize, int local)
446 unsigned long lpar_rc;
447 unsigned long flags = (newpp & 7) | H_AVPN;
448 unsigned long want_v;
450 want_v = hpte_encode_avpn(va, psize, ssize);
452 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
453 want_v, slot, flags, psize);
455 lpar_rc = plpar_pte_protect(flags, slot, want_v);
457 if (lpar_rc == H_NOT_FOUND) {
458 pr_devel("not found !\n");
464 BUG_ON(lpar_rc != H_SUCCESS);
469 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
471 unsigned long dword0;
472 unsigned long lpar_rc;
473 unsigned long dummy_word1;
476 /* Read 1 pte at a time */
477 /* Do not need RPN to logical page translation */
478 /* No cross CEC PFT access */
481 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
483 BUG_ON(lpar_rc != H_SUCCESS);
488 static long pSeries_lpar_hpte_find(unsigned long va, int psize, int ssize)
493 unsigned long want_v, hpte_v;
495 hash = hpt_hash(va, mmu_psize_defs[psize].shift, ssize);
496 want_v = hpte_encode_avpn(va, psize, ssize);
498 /* Bolted entries are always in the primary group */
499 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
500 for (i = 0; i < HPTES_PER_GROUP; i++) {
501 hpte_v = pSeries_lpar_hpte_getword0(slot);
503 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
512 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
514 int psize, int ssize)
516 unsigned long lpar_rc, slot, vsid, va, flags;
518 vsid = get_kernel_vsid(ea, ssize);
519 va = hpt_va(ea, vsid, ssize);
521 slot = pSeries_lpar_hpte_find(va, psize, ssize);
525 lpar_rc = plpar_pte_protect(flags, slot, 0);
527 BUG_ON(lpar_rc != H_SUCCESS);
530 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
531 int psize, int ssize, int local)
533 unsigned long want_v;
534 unsigned long lpar_rc;
535 unsigned long dummy1, dummy2;
537 pr_devel(" inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
538 slot, va, psize, local);
540 want_v = hpte_encode_avpn(va, psize, ssize);
541 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
542 if (lpar_rc == H_NOT_FOUND)
545 BUG_ON(lpar_rc != H_SUCCESS);
548 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
549 int psize, int ssize)
551 unsigned long slot, vsid, va;
553 vsid = get_kernel_vsid(ea, ssize);
554 va = hpt_va(ea, vsid, ssize);
556 slot = pSeries_lpar_hpte_find(va, psize, ssize);
559 pSeries_lpar_hpte_invalidate(slot, va, psize, ssize, 0);
562 /* Flag bits for H_BULK_REMOVE */
563 #define HBR_REQUEST 0x4000000000000000UL
564 #define HBR_RESPONSE 0x8000000000000000UL
565 #define HBR_END 0xc000000000000000UL
566 #define HBR_AVPN 0x0200000000000000UL
567 #define HBR_ANDCOND 0x0100000000000000UL
570 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
573 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
575 unsigned long i, pix, rc;
576 unsigned long flags = 0;
577 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
578 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
579 unsigned long param[9];
581 unsigned long hash, index, shift, hidx, slot;
586 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
588 psize = batch->psize;
589 ssize = batch->ssize;
591 for (i = 0; i < number; i++) {
592 va = batch->vaddr[i];
594 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
595 hash = hpt_hash(va, shift, ssize);
596 hidx = __rpte_to_hidx(pte, index);
597 if (hidx & _PTEIDX_SECONDARY)
599 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
600 slot += hidx & _PTEIDX_GROUP_IX;
601 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
602 pSeries_lpar_hpte_invalidate(slot, va, psize,
605 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
606 param[pix+1] = hpte_encode_avpn(va, psize,
610 rc = plpar_hcall9(H_BULK_REMOVE, param,
611 param[0], param[1], param[2],
612 param[3], param[4], param[5],
614 BUG_ON(rc != H_SUCCESS);
618 } pte_iterate_hashed_end();
621 param[pix] = HBR_END;
622 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
623 param[2], param[3], param[4], param[5],
625 BUG_ON(rc != H_SUCCESS);
629 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
632 static int __init disable_bulk_remove(char *str)
634 if (strcmp(str, "off") == 0 &&
635 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
636 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
637 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
642 __setup("bulk_remove=", disable_bulk_remove);
644 void __init hpte_init_lpar(void)
646 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
647 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
648 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
649 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
650 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
651 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
652 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
653 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
656 #ifdef CONFIG_PPC_SMLPAR
657 #define CMO_FREE_HINT_DEFAULT 1
658 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
660 static int __init cmo_free_hint(char *str)
663 parm = strstrip(str);
665 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
666 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
667 cmo_free_hint_flag = 0;
671 cmo_free_hint_flag = 1;
672 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
674 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
680 __setup("cmo_free_hint=", cmo_free_hint);
682 static void pSeries_set_page_state(struct page *page, int order,
686 unsigned long cmo_page_sz, addr;
688 cmo_page_sz = cmo_get_page_size();
689 addr = __pa((unsigned long)page_address(page));
691 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
692 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
693 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
697 void arch_free_page(struct page *page, int order)
699 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
702 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
704 EXPORT_SYMBOL(arch_free_page);
708 #ifdef CONFIG_TRACEPOINTS
710 * We optimise our hcall path by placing hcall_tracepoint_refcount
711 * directly in the TOC so we can check if the hcall tracepoints are
712 * enabled via a single load.
715 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
716 extern long hcall_tracepoint_refcount;
719 * Since the tracing code might execute hcalls we need to guard against
720 * recursion. One example of this are spinlocks calling H_YIELD on
721 * shared processor partitions.
723 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
725 void hcall_tracepoint_regfunc(void)
727 hcall_tracepoint_refcount++;
730 void hcall_tracepoint_unregfunc(void)
732 hcall_tracepoint_refcount--;
735 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
740 local_irq_save(flags);
742 depth = &__get_cpu_var(hcall_trace_depth);
748 trace_hcall_entry(opcode, args);
752 local_irq_restore(flags);
755 void __trace_hcall_exit(long opcode, unsigned long retval,
756 unsigned long *retbuf)
761 local_irq_save(flags);
763 depth = &__get_cpu_var(hcall_trace_depth);
769 trace_hcall_exit(opcode, retval, retbuf);
773 local_irq_restore(flags);
779 * H_GET_MPP hcall returns info in 7 parms
781 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
784 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
786 rc = plpar_hcall9(H_GET_MPP, retbuf);
788 mpp_data->entitled_mem = retbuf[0];
789 mpp_data->mapped_mem = retbuf[1];
791 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
792 mpp_data->pool_num = retbuf[2] & 0xffff;
794 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
795 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
796 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
798 mpp_data->pool_size = retbuf[4];
799 mpp_data->loan_request = retbuf[5];
800 mpp_data->backing_mem = retbuf[6];
804 EXPORT_SYMBOL(h_get_mpp);
806 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
809 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
811 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
813 mpp_x_data->coalesced_bytes = retbuf[0];
814 mpp_x_data->pool_coalesced_bytes = retbuf[1];
815 mpp_x_data->pool_purr_cycles = retbuf[2];
816 mpp_x_data->pool_spurr_cycles = retbuf[3];