1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Copyright (C) 2001 Todd Inglett, IBM Corporation
6 * pSeries LPAR support.
9 /* Enables debugging of low-level hash table routines - careful! */
11 #define pr_fmt(fmt) "lpar: " fmt
13 #include <linux/kernel.h>
14 #include <linux/dma-mapping.h>
15 #include <linux/console.h>
16 #include <linux/export.h>
17 #include <linux/jump_label.h>
18 #include <linux/delay.h>
19 #include <linux/stop_machine.h>
20 #include <linux/spinlock.h>
21 #include <linux/cpuhotplug.h>
22 #include <linux/workqueue.h>
23 #include <linux/proc_fs.h>
24 #include <asm/processor.h>
27 #include <asm/pgtable.h>
28 #include <asm/machdep.h>
29 #include <asm/mmu_context.h>
30 #include <asm/iommu.h>
33 #include <asm/cputable.h>
36 #include <asm/trace.h>
37 #include <asm/firmware.h>
38 #include <asm/plpar_wrappers.h>
39 #include <asm/kexec.h>
40 #include <asm/fadump.h>
41 #include <asm/asm-prototypes.h>
42 #include <asm/debugfs.h>
46 /* Flag bits for H_BULK_REMOVE */
47 #define HBR_REQUEST 0x4000000000000000UL
48 #define HBR_RESPONSE 0x8000000000000000UL
49 #define HBR_END 0xc000000000000000UL
50 #define HBR_AVPN 0x0200000000000000UL
51 #define HBR_ANDCOND 0x0100000000000000UL
55 EXPORT_SYMBOL(plpar_hcall);
56 EXPORT_SYMBOL(plpar_hcall9);
57 EXPORT_SYMBOL(plpar_hcall_norets);
60 * H_BLOCK_REMOVE supported block size for this page size in segment who's base
61 * page size is that page size.
63 * The first index is the segment base page size, the second one is the actual
66 static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
69 * Due to the involved complexity, and that the current hypervisor is only
70 * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
71 * buffer size to 8 size block.
73 #define HBLKRM_SUPPORTED_BLOCK_SIZE 8
75 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
76 static u8 dtl_mask = DTL_LOG_PREEMPT;
81 void alloc_dtl_buffers(unsigned long *time_limit)
84 struct paca_struct *pp;
85 struct dtl_entry *dtl;
87 for_each_possible_cpu(cpu) {
91 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
93 pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
95 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
96 pr_warn("Stolen time statistics will be unreliable\n");
102 pp->dispatch_log = dtl;
103 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
106 if (time_limit && time_after(jiffies, *time_limit)) {
108 *time_limit = jiffies + HZ;
113 void register_dtl_buffer(int cpu)
116 struct paca_struct *pp;
117 struct dtl_entry *dtl;
118 int hwcpu = get_hard_smp_processor_id(cpu);
121 dtl = pp->dispatch_log;
122 if (dtl && dtl_mask) {
125 lppaca_of(cpu).dtl_idx = 0;
127 /* hypervisor reads buffer length from this field */
128 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
129 ret = register_dtl(hwcpu, __pa(dtl));
131 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
134 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
138 #ifdef CONFIG_PPC_SPLPAR
140 struct delayed_work work;
144 struct vcpu_dispatch_data {
155 int numa_remote_disp;
160 * This represents the number of cpus in the hypervisor. Since there is no
161 * architected way to discover the number of processors in the host, we
162 * provision for dealing with NR_CPUS. This is currently 2048 by default, and
163 * is sufficient for our purposes. This will need to be tweaked if
164 * CONFIG_NR_CPUS is changed.
166 #define NR_CPUS_H NR_CPUS
168 DEFINE_RWLOCK(dtl_access_lock);
169 static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
170 static DEFINE_PER_CPU(u64, dtl_entry_ridx);
171 static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
172 static enum cpuhp_state dtl_worker_state;
173 static DEFINE_MUTEX(dtl_enable_mutex);
174 static int vcpudispatch_stats_on __read_mostly;
175 static int vcpudispatch_stats_freq = 50;
176 static __be32 *vcpu_associativity, *pcpu_associativity;
179 static void free_dtl_buffers(unsigned long *time_limit)
181 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
183 struct paca_struct *pp;
185 for_each_possible_cpu(cpu) {
187 if (!pp->dispatch_log)
189 kmem_cache_free(dtl_cache, pp->dispatch_log);
191 pp->dispatch_log = 0;
192 pp->dispatch_log_end = 0;
195 if (time_limit && time_after(jiffies, *time_limit)) {
197 *time_limit = jiffies + HZ;
203 static int init_cpu_associativity(void)
205 vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
206 VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
207 pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
208 VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
210 if (!vcpu_associativity || !pcpu_associativity) {
211 pr_err("error allocating memory for associativity information\n");
218 static void destroy_cpu_associativity(void)
220 kfree(vcpu_associativity);
221 kfree(pcpu_associativity);
222 vcpu_associativity = pcpu_associativity = 0;
225 static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
230 assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
232 rc = hcall_vphn(cpu, flag, &assoc[0]);
240 static __be32 *get_pcpu_associativity(int cpu)
242 return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
245 static __be32 *get_vcpu_associativity(int cpu)
247 return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
250 static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
252 __be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
254 if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
257 last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
258 cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
260 if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
263 return cpu_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
266 static int cpu_home_node_dispatch_distance(int disp_cpu)
268 __be32 *disp_cpu_assoc, *vcpu_assoc;
269 int vcpu_id = smp_processor_id();
271 if (disp_cpu >= NR_CPUS_H) {
272 pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
273 disp_cpu, NR_CPUS_H);
277 disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
278 vcpu_assoc = get_vcpu_associativity(vcpu_id);
280 if (!disp_cpu_assoc || !vcpu_assoc)
283 return cpu_distance(disp_cpu_assoc, vcpu_assoc);
286 static void update_vcpu_disp_stat(int disp_cpu)
288 struct vcpu_dispatch_data *disp;
291 disp = this_cpu_ptr(&vcpu_disp_data);
292 if (disp->last_disp_cpu == -1) {
293 disp->last_disp_cpu = disp_cpu;
299 if (disp->last_disp_cpu == disp_cpu ||
300 (cpu_first_thread_sibling(disp->last_disp_cpu) ==
301 cpu_first_thread_sibling(disp_cpu)))
302 disp->same_cpu_disp++;
304 distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
307 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
312 disp->same_chip_disp++;
315 disp->diff_chip_disp++;
318 disp->far_chip_disp++;
321 pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
330 distance = cpu_home_node_dispatch_distance(disp_cpu);
332 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
337 disp->numa_home_disp++;
340 disp->numa_remote_disp++;
343 disp->numa_far_disp++;
346 pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
353 disp->last_disp_cpu = disp_cpu;
356 static void process_dtl_buffer(struct work_struct *work)
358 struct dtl_entry dtle;
359 u64 i = __this_cpu_read(dtl_entry_ridx);
360 struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
361 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
362 struct lppaca *vpa = local_paca->lppaca_ptr;
363 struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
365 if (!local_paca->dispatch_log)
368 /* if we have been migrated away, we cancel ourself */
369 if (d->cpu != smp_processor_id()) {
370 pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
375 if (i == be64_to_cpu(vpa->dtl_idx))
378 while (i < be64_to_cpu(vpa->dtl_idx)) {
381 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
382 /* buffer has overflowed */
383 pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
385 be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
386 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
387 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
390 update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
394 dtl = local_paca->dispatch_log;
397 __this_cpu_write(dtl_entry_ridx, i);
400 schedule_delayed_work_on(d->cpu, to_delayed_work(work),
401 HZ / vcpudispatch_stats_freq);
404 static int dtl_worker_online(unsigned int cpu)
406 struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
408 memset(d, 0, sizeof(*d));
409 INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
412 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
413 per_cpu(dtl_entry_ridx, cpu) = 0;
414 register_dtl_buffer(cpu);
416 per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
419 schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
423 static int dtl_worker_offline(unsigned int cpu)
425 struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
427 cancel_delayed_work_sync(&d->work);
429 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
430 unregister_dtl(get_hard_smp_processor_id(cpu));
436 static void set_global_dtl_mask(u8 mask)
441 for_each_present_cpu(cpu)
442 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
445 static void reset_global_dtl_mask(void)
449 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
450 dtl_mask = DTL_LOG_PREEMPT;
454 for_each_present_cpu(cpu)
455 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
458 static int dtl_worker_enable(unsigned long *time_limit)
462 if (!write_trylock(&dtl_access_lock)) {
467 set_global_dtl_mask(DTL_LOG_ALL);
469 /* Setup dtl buffers and register those */
470 alloc_dtl_buffers(time_limit);
472 state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
473 dtl_worker_online, dtl_worker_offline);
475 pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
476 free_dtl_buffers(time_limit);
477 reset_global_dtl_mask();
478 write_unlock(&dtl_access_lock);
482 dtl_worker_state = state;
488 static void dtl_worker_disable(unsigned long *time_limit)
490 cpuhp_remove_state(dtl_worker_state);
491 free_dtl_buffers(time_limit);
492 reset_global_dtl_mask();
493 write_unlock(&dtl_access_lock);
496 static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
497 size_t count, loff_t *ppos)
499 unsigned long time_limit = jiffies + HZ;
500 struct vcpu_dispatch_data *disp;
507 if (copy_from_user(buf, p, count))
511 rc = kstrtoint(buf, 0, &cmd);
512 if (rc || cmd < 0 || cmd > 1) {
513 pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
514 return rc ? rc : -EINVAL;
517 mutex_lock(&dtl_enable_mutex);
519 if ((cmd == 0 && !vcpudispatch_stats_on) ||
520 (cmd == 1 && vcpudispatch_stats_on))
524 rc = init_cpu_associativity();
528 for_each_possible_cpu(cpu) {
529 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
530 memset(disp, 0, sizeof(*disp));
531 disp->last_disp_cpu = -1;
534 rc = dtl_worker_enable(&time_limit);
536 destroy_cpu_associativity();
540 dtl_worker_disable(&time_limit);
541 destroy_cpu_associativity();
544 vcpudispatch_stats_on = cmd;
547 mutex_unlock(&dtl_enable_mutex);
553 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
556 struct vcpu_dispatch_data *disp;
558 if (!vcpudispatch_stats_on) {
559 seq_puts(p, "off\n");
563 for_each_online_cpu(cpu) {
564 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
565 seq_printf(p, "cpu%d", cpu);
566 seq_put_decimal_ull(p, " ", disp->total_disp);
567 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
568 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
569 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
570 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
571 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
572 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
573 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
580 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
582 return single_open(file, vcpudispatch_stats_display, NULL);
585 static const struct file_operations vcpudispatch_stats_proc_ops = {
586 .open = vcpudispatch_stats_open,
588 .write = vcpudispatch_stats_write,
590 .release = single_release,
593 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
594 const char __user *p, size_t count, loff_t *ppos)
602 if (copy_from_user(buf, p, count))
606 rc = kstrtoint(buf, 0, &freq);
607 if (rc || freq < 1 || freq > HZ) {
608 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
610 return rc ? rc : -EINVAL;
613 vcpudispatch_stats_freq = freq;
618 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
620 seq_printf(p, "%d\n", vcpudispatch_stats_freq);
624 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
626 return single_open(file, vcpudispatch_stats_freq_display, NULL);
629 static const struct file_operations vcpudispatch_stats_freq_proc_ops = {
630 .open = vcpudispatch_stats_freq_open,
632 .write = vcpudispatch_stats_freq_write,
634 .release = single_release,
637 static int __init vcpudispatch_stats_procfs_init(void)
639 if (!lppaca_shared_proc(get_lppaca()))
642 if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
643 &vcpudispatch_stats_proc_ops))
644 pr_err("vcpudispatch_stats: error creating procfs file\n");
645 else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
646 &vcpudispatch_stats_freq_proc_ops))
647 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
652 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
653 #endif /* CONFIG_PPC_SPLPAR */
655 void vpa_init(int cpu)
657 int hwcpu = get_hard_smp_processor_id(cpu);
662 * The spec says it "may be problematic" if CPU x registers the VPA of
663 * CPU y. We should never do that, but wail if we ever do.
665 WARN_ON(cpu != smp_processor_id());
667 if (cpu_has_feature(CPU_FTR_ALTIVEC))
668 lppaca_of(cpu).vmxregs_in_use = 1;
670 if (cpu_has_feature(CPU_FTR_ARCH_207S))
671 lppaca_of(cpu).ebb_regs_in_use = 1;
673 addr = __pa(&lppaca_of(cpu));
674 ret = register_vpa(hwcpu, addr);
677 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
678 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
682 #ifdef CONFIG_PPC_BOOK3S_64
684 * PAPR says this feature is SLB-Buffer but firmware never
685 * reports that. All SPLPAR support SLB shadow buffer.
687 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
688 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
689 ret = register_slb_shadow(hwcpu, addr);
691 pr_err("WARNING: SLB shadow buffer registration for "
692 "cpu %d (hw %d) of area %lx failed with %ld\n",
693 cpu, hwcpu, addr, ret);
695 #endif /* CONFIG_PPC_BOOK3S_64 */
698 * Register dispatch trace log, if one has been allocated.
700 register_dtl_buffer(cpu);
703 #ifdef CONFIG_PPC_BOOK3S_64
705 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
706 unsigned long vpn, unsigned long pa,
707 unsigned long rflags, unsigned long vflags,
708 int psize, int apsize, int ssize)
710 unsigned long lpar_rc;
713 unsigned long hpte_v, hpte_r;
715 if (!(vflags & HPTE_V_BOLTED))
716 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
717 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
718 hpte_group, vpn, pa, rflags, vflags, psize);
720 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
721 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
723 if (!(vflags & HPTE_V_BOLTED))
724 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
726 /* Now fill in the actual HPTE */
727 /* Set CEC cookie to 0 */
729 /* I-cache Invalidate = 0 */
730 /* I-cache synchronize = 0 */
734 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
735 flags |= H_COALESCE_CAND;
737 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
738 if (unlikely(lpar_rc == H_PTEG_FULL)) {
739 pr_devel("Hash table group is full\n");
744 * Since we try and ioremap PHBs we don't own, the pte insert
745 * will fail. However we must catch the failure in hash_page
746 * or we will loop forever, so return -2 in this case.
748 if (unlikely(lpar_rc != H_SUCCESS)) {
749 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
752 if (!(vflags & HPTE_V_BOLTED))
753 pr_devel(" -> slot: %lu\n", slot & 7);
755 /* Because of iSeries, we have to pass down the secondary
756 * bucket bit here as well
758 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
761 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
763 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
765 unsigned long slot_offset;
766 unsigned long lpar_rc;
768 unsigned long dummy1, dummy2;
770 /* pick a random slot to start at */
771 slot_offset = mftb() & 0x7;
773 for (i = 0; i < HPTES_PER_GROUP; i++) {
775 /* don't remove a bolted entry */
776 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
777 HPTE_V_BOLTED, &dummy1, &dummy2);
778 if (lpar_rc == H_SUCCESS)
782 * The test for adjunct partition is performed before the
783 * ANDCOND test. H_RESOURCE may be returned, so we need to
784 * check for that as well.
786 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
795 static void manual_hpte_clear_all(void)
797 unsigned long size_bytes = 1UL << ppc64_pft_size;
798 unsigned long hpte_count = size_bytes >> 4;
806 /* Read in batches of 4,
807 * invalidate only valid entries not in the VRMA
808 * hpte_count will be a multiple of 4
810 for (i = 0; i < hpte_count; i += 4) {
811 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
812 if (lpar_rc != H_SUCCESS) {
813 pr_info("Failed to read hash page table at %ld err %ld\n",
817 for (j = 0; j < 4; j++){
818 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
821 if (ptes[j].pteh & HPTE_V_VALID)
822 plpar_pte_remove_raw(0, i + j, 0,
823 &(ptes[j].pteh), &(ptes[j].ptel));
828 static int hcall_hpte_clear_all(void)
833 rc = plpar_hcall_norets(H_CLEAR_HPT);
834 } while (rc == H_CONTINUE);
839 static void pseries_hpte_clear_all(void)
843 rc = hcall_hpte_clear_all();
845 manual_hpte_clear_all();
847 #ifdef __LITTLE_ENDIAN__
849 * Reset exceptions to big endian.
851 * FIXME this is a hack for kexec, we need to reset the exception
852 * endian before starting the new kernel and this is a convenient place
855 * This is also called on boot when a fadump happens. In that case we
856 * must not change the exception endian mode.
858 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
859 pseries_big_endian_exceptions();
864 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
865 * the low 3 bits of flags happen to line up. So no transform is needed.
866 * We can probably optimize here and assume the high bits of newpp are
867 * already zero. For now I am paranoid.
869 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
872 int psize, int apsize,
873 int ssize, unsigned long inv_flags)
875 unsigned long lpar_rc;
877 unsigned long want_v;
879 want_v = hpte_encode_avpn(vpn, psize, ssize);
881 flags = (newpp & 7) | H_AVPN;
882 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
883 /* Move pp0 into bit 8 (IBM 55) */
884 flags |= (newpp & HPTE_R_PP0) >> 55;
886 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
887 want_v, slot, flags, psize);
889 lpar_rc = plpar_pte_protect(flags, slot, want_v);
891 if (lpar_rc == H_NOT_FOUND) {
892 pr_devel("not found !\n");
898 BUG_ON(lpar_rc != H_SUCCESS);
903 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
912 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
914 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
915 if (lpar_rc != H_SUCCESS) {
916 pr_info("Failed to read hash page table at %ld err %ld\n",
917 hpte_group, lpar_rc);
921 for (j = 0; j < 4; j++) {
922 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
923 (ptes[j].pteh & HPTE_V_VALID))
931 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
935 unsigned long want_v;
936 unsigned long hpte_group;
938 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
939 want_v = hpte_encode_avpn(vpn, psize, ssize);
941 /* Bolted entries are always in the primary group */
942 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
943 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
946 return hpte_group + slot;
949 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
951 int psize, int ssize)
954 unsigned long lpar_rc, slot, vsid, flags;
956 vsid = get_kernel_vsid(ea, ssize);
957 vpn = hpt_vpn(ea, vsid, ssize);
959 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
963 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
964 /* Move pp0 into bit 8 (IBM 55) */
965 flags |= (newpp & HPTE_R_PP0) >> 55;
967 lpar_rc = plpar_pte_protect(flags, slot, 0);
969 BUG_ON(lpar_rc != H_SUCCESS);
972 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
973 int psize, int apsize,
974 int ssize, int local)
976 unsigned long want_v;
977 unsigned long lpar_rc;
978 unsigned long dummy1, dummy2;
980 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
981 slot, vpn, psize, local);
983 want_v = hpte_encode_avpn(vpn, psize, ssize);
984 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
985 if (lpar_rc == H_NOT_FOUND)
988 BUG_ON(lpar_rc != H_SUCCESS);
993 * As defined in the PAPR's section 14.5.4.1.8
994 * The control mask doesn't include the returned reference and change bit from
997 #define HBLKR_AVPN 0x0100000000000000UL
998 #define HBLKR_CTRL_MASK 0xf800000000000000UL
999 #define HBLKR_CTRL_SUCCESS 0x8000000000000000UL
1000 #define HBLKR_CTRL_ERRNOTFOUND 0x8800000000000000UL
1001 #define HBLKR_CTRL_ERRBUSY 0xa000000000000000UL
1004 * Returned true if we are supporting this block size for the specified segment
1005 * base page size and actual page size.
1007 * Currently, we only support 8 size block.
1009 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1011 return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1015 * H_BLOCK_REMOVE caller.
1016 * @idx should point to the latest @param entry set with a PTEX.
1017 * If PTE cannot be processed because another CPUs has already locked that
1018 * group, those entries are put back in @param starting at index 1.
1019 * If entries has to be retried and @retry_busy is set to true, these entries
1020 * are retried until success. If @retry_busy is set to false, the returned
1021 * is the number of entries yet to process.
1023 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1026 unsigned long i, rc, new_idx;
1027 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1030 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1035 if (idx > PLPAR_HCALL9_BUFSIZE) {
1036 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1037 idx = PLPAR_HCALL9_BUFSIZE;
1038 } else if (idx < PLPAR_HCALL9_BUFSIZE)
1039 param[idx] = HBR_END;
1041 rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1043 param[1], param[2], param[3], param[4], /* TS0-7 */
1044 param[5], param[6], param[7], param[8]);
1045 if (rc == H_SUCCESS)
1048 BUG_ON(rc != H_PARTIAL);
1050 /* Check that the unprocessed entries were 'not found' or 'busy' */
1051 for (i = 0; i < idx-1; i++) {
1052 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1054 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1055 param[++new_idx] = param[i+1];
1059 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1060 && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1064 * If there were entries found busy, retry these entries if requested,
1065 * of if all the entries have to be retried.
1067 if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1075 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1077 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1078 * to make sure that we avoid bouncing the hypervisor tlbie lock.
1080 #define PPC64_HUGE_HPTE_BATCH 12
1082 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1083 int count, int psize, int ssize)
1085 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1086 unsigned long shift, current_vpgb, vpgb;
1089 shift = mmu_psize_defs[psize].shift;
1091 for (i = 0; i < count; i++) {
1093 * Shifting 3 bits more on the right to get a
1094 * 8 pages aligned virtual addresse.
1096 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1097 if (!pix || vpgb != current_vpgb) {
1099 * Need to start a new 8 pages block, flush
1100 * the current one if needed.
1103 (void)call_block_remove(pix, param, true);
1104 current_vpgb = vpgb;
1105 param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1109 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1110 if (pix == PLPAR_HCALL9_BUFSIZE) {
1111 pix = call_block_remove(pix, param, false);
1113 * pix = 0 means that all the entries were
1114 * removed, we can start a new block.
1115 * Otherwise, this means that there are entries
1116 * to retry, and pix points to latest one, so
1117 * we should increment it and try to continue
1125 (void)call_block_remove(pix, param, true);
1128 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1129 int count, int psize, int ssize)
1131 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1132 int i = 0, pix = 0, rc;
1134 for (i = 0; i < count; i++) {
1136 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1137 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1140 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1141 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1144 rc = plpar_hcall9(H_BULK_REMOVE, param,
1145 param[0], param[1], param[2],
1146 param[3], param[4], param[5],
1147 param[6], param[7]);
1148 BUG_ON(rc != H_SUCCESS);
1154 param[pix] = HBR_END;
1155 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1156 param[2], param[3], param[4], param[5],
1157 param[6], param[7]);
1158 BUG_ON(rc != H_SUCCESS);
1162 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1164 int count, int psize,
1167 unsigned long flags = 0;
1168 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1171 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1173 /* Assuming THP size is 16M */
1174 if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1175 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1177 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1180 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1183 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1185 unsigned char *hpte_slot_array,
1186 int psize, int ssize, int local)
1189 unsigned long s_addr = addr;
1190 unsigned int max_hpte_count, valid;
1191 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1192 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1193 unsigned long shift, hidx, vpn = 0, hash, slot;
1195 shift = mmu_psize_defs[psize].shift;
1196 max_hpte_count = 1U << (PMD_SHIFT - shift);
1198 for (i = 0; i < max_hpte_count; i++) {
1199 valid = hpte_valid(hpte_slot_array, i);
1202 hidx = hpte_hash_index(hpte_slot_array, i);
1205 addr = s_addr + (i * (1ul << shift));
1206 vpn = hpt_vpn(addr, vsid, ssize);
1207 hash = hpt_hash(vpn, shift, ssize);
1208 if (hidx & _PTEIDX_SECONDARY)
1211 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1212 slot += hidx & _PTEIDX_GROUP_IX;
1214 slot_array[index] = slot;
1215 vpn_array[index] = vpn;
1216 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1218 * Now do a bluk invalidate
1220 __pSeries_lpar_hugepage_invalidate(slot_array,
1222 PPC64_HUGE_HPTE_BATCH,
1229 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1230 index, psize, ssize);
1233 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1235 unsigned char *hpte_slot_array,
1236 int psize, int ssize, int local)
1238 WARN(1, "%s called without THP support\n", __func__);
1242 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1243 int psize, int ssize)
1246 unsigned long slot, vsid;
1248 vsid = get_kernel_vsid(ea, ssize);
1249 vpn = hpt_vpn(ea, vsid, ssize);
1251 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1256 * lpar doesn't use the passed actual page size
1258 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1263 static inline unsigned long compute_slot(real_pte_t pte,
1265 unsigned long index,
1266 unsigned long shift,
1269 unsigned long slot, hash, hidx;
1271 hash = hpt_hash(vpn, shift, ssize);
1272 hidx = __rpte_to_hidx(pte, index);
1273 if (hidx & _PTEIDX_SECONDARY)
1275 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1276 slot += hidx & _PTEIDX_GROUP_IX;
1281 * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1282 * "all within the same naturally aligned 8 page virtual address block".
1284 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1285 unsigned long *param)
1288 unsigned long i, pix = 0;
1289 unsigned long index, shift, slot, current_vpgb, vpgb;
1293 psize = batch->psize;
1294 ssize = batch->ssize;
1296 for (i = 0; i < number; i++) {
1297 vpn = batch->vpn[i];
1298 pte = batch->pte[i];
1299 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1301 * Shifting 3 bits more on the right to get a
1302 * 8 pages aligned virtual addresse.
1304 vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1305 if (!pix || vpgb != current_vpgb) {
1307 * Need to start a new 8 pages block, flush
1308 * the current one if needed.
1311 (void)call_block_remove(pix, param,
1313 current_vpgb = vpgb;
1314 param[0] = hpte_encode_avpn(vpn, psize,
1319 slot = compute_slot(pte, vpn, index, shift, ssize);
1320 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1322 if (pix == PLPAR_HCALL9_BUFSIZE) {
1323 pix = call_block_remove(pix, param, false);
1325 * pix = 0 means that all the entries were
1326 * removed, we can start a new block.
1327 * Otherwise, this means that there are entries
1328 * to retry, and pix points to latest one, so
1329 * we should increment it and try to continue
1335 } pte_iterate_hashed_end();
1339 (void)call_block_remove(pix, param, true);
1343 * TLB Block Invalidate Characteristics
1345 * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1346 * is able to process for each couple segment base page size, actual page size.
1348 * The ibm,get-system-parameter properties is returning a buffer with the
1351 * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1353 * TLB Block Invalidate Specifiers:
1354 * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1355 * [ 1 byte Number of page sizes (N) that are supported for the specified
1356 * TLB invalidate block size ]
1357 * [ 1 byte Encoded segment base page size and actual page size
1358 * MSB=0 means 4k segment base page size and actual page size
1359 * MSB=1 the penc value in mmu_psize_def ]
1362 * Next TLB Block Invalidate Specifiers...
1366 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1367 unsigned int block_size)
1369 if (block_size > hblkrm_size[bpsize][psize])
1370 hblkrm_size[bpsize][psize] = block_size;
1374 * Decode the Encoded segment base page size and actual page size.
1376 * - bit 7 is the L bit
1377 * - bits 0-5 are the penc value
1378 * If the L bit is 0, this means 4K segment base page size and actual page size
1379 * otherwise the penc value should be read.
1381 #define HBLKRM_L_MASK 0x80
1382 #define HBLKRM_PENC_MASK 0x3f
1383 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1384 unsigned int block_size)
1386 unsigned int bpsize, psize;
1388 /* First, check the L bit, if not set, this means 4K */
1389 if ((lp & HBLKRM_L_MASK) == 0) {
1390 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1394 lp &= HBLKRM_PENC_MASK;
1395 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1396 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1398 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1399 if (def->penc[psize] == lp) {
1400 set_hblkrm_bloc_size(bpsize, psize, block_size);
1407 #define SPLPAR_TLB_BIC_TOKEN 50
1410 * The size of the TLB Block Invalidate Characteristics is variable. But at the
1411 * maximum it will be the number of possible page sizes *2 + 10 bytes.
1412 * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1413 * (128 bytes) for the buffer to get plenty of space.
1415 #define SPLPAR_TLB_BIC_MAXLENGTH 128
1417 void __init pseries_lpar_read_hblkrm_characteristics(void)
1419 unsigned char local_buffer[SPLPAR_TLB_BIC_MAXLENGTH];
1420 int call_status, len, idx, bpsize;
1422 if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1425 spin_lock(&rtas_data_buf_lock);
1426 memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
1427 call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
1429 SPLPAR_TLB_BIC_TOKEN,
1430 __pa(rtas_data_buf),
1431 RTAS_DATA_BUF_SIZE);
1432 memcpy(local_buffer, rtas_data_buf, SPLPAR_TLB_BIC_MAXLENGTH);
1433 local_buffer[SPLPAR_TLB_BIC_MAXLENGTH - 1] = '\0';
1434 spin_unlock(&rtas_data_buf_lock);
1436 if (call_status != 0) {
1437 pr_warn("%s %s Error calling get-system-parameter (0x%x)\n",
1438 __FILE__, __func__, call_status);
1443 * The first two (2) bytes of the data in the buffer are the length of
1444 * the returned data, not counting these first two (2) bytes.
1446 len = be16_to_cpu(*((u16 *)local_buffer)) + 2;
1447 if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1448 pr_warn("%s too large returned buffer %d", __func__, len);
1454 u8 block_shift = local_buffer[idx++];
1456 unsigned int npsize;
1461 block_size = 1 << block_shift;
1463 for (npsize = local_buffer[idx++];
1464 npsize > 0 && idx < len; npsize--)
1465 check_lp_set_hblkrm((unsigned int) local_buffer[idx++],
1469 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1470 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1471 if (hblkrm_size[bpsize][idx])
1472 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1473 bpsize, idx, hblkrm_size[bpsize][idx]);
1477 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1480 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1483 unsigned long i, pix, rc;
1484 unsigned long flags = 0;
1485 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1486 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1487 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1488 unsigned long index, shift, slot;
1493 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1495 if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1496 do_block_remove(number, batch, param);
1500 psize = batch->psize;
1501 ssize = batch->ssize;
1503 for (i = 0; i < number; i++) {
1504 vpn = batch->vpn[i];
1505 pte = batch->pte[i];
1506 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1507 slot = compute_slot(pte, vpn, index, shift, ssize);
1508 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1510 * lpar doesn't use the passed actual page size
1512 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1515 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1516 param[pix+1] = hpte_encode_avpn(vpn, psize,
1520 rc = plpar_hcall9(H_BULK_REMOVE, param,
1521 param[0], param[1], param[2],
1522 param[3], param[4], param[5],
1523 param[6], param[7]);
1524 BUG_ON(rc != H_SUCCESS);
1528 } pte_iterate_hashed_end();
1531 param[pix] = HBR_END;
1532 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1533 param[2], param[3], param[4], param[5],
1534 param[6], param[7]);
1535 BUG_ON(rc != H_SUCCESS);
1540 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1543 static int __init disable_bulk_remove(char *str)
1545 if (strcmp(str, "off") == 0 &&
1546 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1547 pr_info("Disabling BULK_REMOVE firmware feature");
1548 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1553 __setup("bulk_remove=", disable_bulk_remove);
1555 #define HPT_RESIZE_TIMEOUT 10000 /* ms */
1557 struct hpt_resize_state {
1558 unsigned long shift;
1562 static int pseries_lpar_resize_hpt_commit(void *data)
1564 struct hpt_resize_state *state = data;
1566 state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1567 if (state->commit_rc != H_SUCCESS)
1570 /* Hypervisor has transitioned the HTAB, update our globals */
1571 ppc64_pft_size = state->shift;
1572 htab_size_bytes = 1UL << ppc64_pft_size;
1573 htab_hash_mask = (htab_size_bytes >> 7) - 1;
1579 * Must be called in process context. The caller must hold the
1582 static int pseries_lpar_resize_hpt(unsigned long shift)
1584 struct hpt_resize_state state = {
1586 .commit_rc = H_FUNCTION,
1588 unsigned int delay, total_delay = 0;
1594 if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1597 pr_info("Attempting to resize HPT to shift %lu\n", shift);
1601 rc = plpar_resize_hpt_prepare(0, shift);
1602 while (H_IS_LONG_BUSY(rc)) {
1603 delay = get_longbusy_msecs(rc);
1604 total_delay += delay;
1605 if (total_delay > HPT_RESIZE_TIMEOUT) {
1606 /* prepare with shift==0 cancels an in-progress resize */
1607 rc = plpar_resize_hpt_prepare(0, 0);
1608 if (rc != H_SUCCESS)
1609 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1614 rc = plpar_resize_hpt_prepare(0, shift);
1623 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1626 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1629 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1635 rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1641 switch (state.commit_rc) {
1646 pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1652 pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1653 shift, (long long) ktime_ms_delta(t1, t0),
1654 (long long) ktime_ms_delta(t2, t1));
1659 static int pseries_lpar_register_process_table(unsigned long base,
1660 unsigned long page_size, unsigned long table_size)
1663 unsigned long flags = 0;
1666 flags |= PROC_TABLE_NEW;
1667 if (radix_enabled())
1668 flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
1670 flags |= PROC_TABLE_HPT_SLB;
1672 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
1673 page_size, table_size);
1674 if (!H_IS_LONG_BUSY(rc))
1676 mdelay(get_longbusy_msecs(rc));
1678 if (rc != H_SUCCESS) {
1679 pr_err("Failed to register process table (rc=%ld)\n", rc);
1685 void __init hpte_init_pseries(void)
1687 mmu_hash_ops.hpte_invalidate = pSeries_lpar_hpte_invalidate;
1688 mmu_hash_ops.hpte_updatepp = pSeries_lpar_hpte_updatepp;
1689 mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1690 mmu_hash_ops.hpte_insert = pSeries_lpar_hpte_insert;
1691 mmu_hash_ops.hpte_remove = pSeries_lpar_hpte_remove;
1692 mmu_hash_ops.hpte_removebolted = pSeries_lpar_hpte_removebolted;
1693 mmu_hash_ops.flush_hash_range = pSeries_lpar_flush_hash_range;
1694 mmu_hash_ops.hpte_clear_all = pseries_hpte_clear_all;
1695 mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1697 if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1698 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1701 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1702 * to inform the hypervisor that we wish to use the HPT.
1704 if (cpu_has_feature(CPU_FTR_ARCH_300))
1705 pseries_lpar_register_process_table(0, 0, 0);
1708 void radix_init_pseries(void)
1710 pr_info("Using radix MMU under hypervisor\n");
1712 pseries_lpar_register_process_table(__pa(process_tb),
1713 0, PRTB_SIZE_SHIFT - 12);
1716 #ifdef CONFIG_PPC_SMLPAR
1717 #define CMO_FREE_HINT_DEFAULT 1
1718 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1720 static int __init cmo_free_hint(char *str)
1723 parm = strstrip(str);
1725 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1726 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1727 cmo_free_hint_flag = 0;
1731 cmo_free_hint_flag = 1;
1732 pr_info("%s: CMO free page hinting is active.\n", __func__);
1734 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1740 __setup("cmo_free_hint=", cmo_free_hint);
1742 static void pSeries_set_page_state(struct page *page, int order,
1743 unsigned long state)
1746 unsigned long cmo_page_sz, addr;
1748 cmo_page_sz = cmo_get_page_size();
1749 addr = __pa((unsigned long)page_address(page));
1751 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1752 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1753 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1757 void arch_free_page(struct page *page, int order)
1759 if (radix_enabled())
1761 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1764 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1766 EXPORT_SYMBOL(arch_free_page);
1768 #endif /* CONFIG_PPC_SMLPAR */
1769 #endif /* CONFIG_PPC_BOOK3S_64 */
1771 #ifdef CONFIG_TRACEPOINTS
1772 #ifdef CONFIG_JUMP_LABEL
1773 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1775 int hcall_tracepoint_regfunc(void)
1777 static_key_slow_inc(&hcall_tracepoint_key);
1781 void hcall_tracepoint_unregfunc(void)
1783 static_key_slow_dec(&hcall_tracepoint_key);
1787 * We optimise our hcall path by placing hcall_tracepoint_refcount
1788 * directly in the TOC so we can check if the hcall tracepoints are
1789 * enabled via a single load.
1792 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1793 extern long hcall_tracepoint_refcount;
1795 int hcall_tracepoint_regfunc(void)
1797 hcall_tracepoint_refcount++;
1801 void hcall_tracepoint_unregfunc(void)
1803 hcall_tracepoint_refcount--;
1808 * Since the tracing code might execute hcalls we need to guard against
1809 * recursion. One example of this are spinlocks calling H_YIELD on
1810 * shared processor partitions.
1812 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1815 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1817 unsigned long flags;
1818 unsigned int *depth;
1821 * We cannot call tracepoints inside RCU idle regions which
1822 * means we must not trace H_CEDE.
1824 if (opcode == H_CEDE)
1827 local_irq_save(flags);
1829 depth = this_cpu_ptr(&hcall_trace_depth);
1836 trace_hcall_entry(opcode, args);
1840 local_irq_restore(flags);
1843 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1845 unsigned long flags;
1846 unsigned int *depth;
1848 if (opcode == H_CEDE)
1851 local_irq_save(flags);
1853 depth = this_cpu_ptr(&hcall_trace_depth);
1859 trace_hcall_exit(opcode, retval, retbuf);
1864 local_irq_restore(flags);
1870 * H_GET_MPP hcall returns info in 7 parms
1872 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1875 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1877 rc = plpar_hcall9(H_GET_MPP, retbuf);
1879 mpp_data->entitled_mem = retbuf[0];
1880 mpp_data->mapped_mem = retbuf[1];
1882 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1883 mpp_data->pool_num = retbuf[2] & 0xffff;
1885 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1886 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1887 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1889 mpp_data->pool_size = retbuf[4];
1890 mpp_data->loan_request = retbuf[5];
1891 mpp_data->backing_mem = retbuf[6];
1895 EXPORT_SYMBOL(h_get_mpp);
1897 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1900 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1902 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1904 mpp_x_data->coalesced_bytes = retbuf[0];
1905 mpp_x_data->pool_coalesced_bytes = retbuf[1];
1906 mpp_x_data->pool_purr_cycles = retbuf[2];
1907 mpp_x_data->pool_spurr_cycles = retbuf[3];
1912 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
1914 unsigned long protovsid;
1915 unsigned long va_bits = VA_BITS;
1916 unsigned long modinv, vsid_modulus;
1917 unsigned long max_mod_inv, tmp_modinv;
1919 if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1922 if (ssize == MMU_SEGSIZE_256M) {
1923 modinv = VSID_MULINV_256M;
1924 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1926 modinv = VSID_MULINV_1T;
1927 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1931 * vsid outside our range.
1933 if (vsid >= vsid_modulus)
1937 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1938 * and vsid = (protovsid * x) % vsid_modulus, then we say:
1939 * protovsid = (vsid * modinv) % vsid_modulus
1942 /* Check if (vsid * modinv) overflow (63 bits) */
1943 max_mod_inv = 0x7fffffffffffffffull / vsid;
1944 if (modinv < max_mod_inv)
1945 return (vsid * modinv) % vsid_modulus;
1947 tmp_modinv = modinv/max_mod_inv;
1948 modinv %= max_mod_inv;
1950 protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1951 protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1956 static int __init reserve_vrma_context_id(void)
1958 unsigned long protovsid;
1961 * Reserve context ids which map to reserved virtual addresses. For now
1962 * we only reserve the context id which maps to the VRMA VSID. We ignore
1963 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1964 * enable adjunct support via the "ibm,client-architecture-support"
1967 protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1968 hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1971 machine_device_initcall(pseries, reserve_vrma_context_id);
1973 #ifdef CONFIG_DEBUG_FS
1974 /* debugfs file interface for vpa data */
1975 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1978 int cpu = (long)filp->private_data;
1979 struct lppaca *lppaca = &lppaca_of(cpu);
1981 return simple_read_from_buffer(buf, len, pos, lppaca,
1982 sizeof(struct lppaca));
1985 static const struct file_operations vpa_fops = {
1986 .open = simple_open,
1987 .read = vpa_file_read,
1988 .llseek = default_llseek,
1991 static int __init vpa_debugfs_init(void)
1995 static struct dentry *vpa_dir;
1997 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2000 vpa_dir = debugfs_create_dir("vpa", powerpc_debugfs_root);
2002 pr_warn("%s: can't create vpa root dir\n", __func__);
2006 /* set up the per-cpu vpa file*/
2007 for_each_possible_cpu(i) {
2010 sprintf(name, "cpu-%ld", i);
2012 d = debugfs_create_file(name, 0400, vpa_dir, (void *)i,
2015 pr_warn("%s: can't create per-cpu vpa file\n",
2023 machine_arch_initcall(pseries, vpa_debugfs_init);
2024 #endif /* CONFIG_DEBUG_FS */