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();
526 destroy_cpu_associativity();
530 for_each_possible_cpu(cpu) {
531 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
532 memset(disp, 0, sizeof(*disp));
533 disp->last_disp_cpu = -1;
536 rc = dtl_worker_enable(&time_limit);
538 destroy_cpu_associativity();
542 dtl_worker_disable(&time_limit);
543 destroy_cpu_associativity();
546 vcpudispatch_stats_on = cmd;
549 mutex_unlock(&dtl_enable_mutex);
555 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
558 struct vcpu_dispatch_data *disp;
560 if (!vcpudispatch_stats_on) {
561 seq_puts(p, "off\n");
565 for_each_online_cpu(cpu) {
566 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
567 seq_printf(p, "cpu%d", cpu);
568 seq_put_decimal_ull(p, " ", disp->total_disp);
569 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
570 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
571 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
572 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
573 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
574 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
575 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
582 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
584 return single_open(file, vcpudispatch_stats_display, NULL);
587 static const struct file_operations vcpudispatch_stats_proc_ops = {
588 .open = vcpudispatch_stats_open,
590 .write = vcpudispatch_stats_write,
592 .release = single_release,
595 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
596 const char __user *p, size_t count, loff_t *ppos)
604 if (copy_from_user(buf, p, count))
608 rc = kstrtoint(buf, 0, &freq);
609 if (rc || freq < 1 || freq > HZ) {
610 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
612 return rc ? rc : -EINVAL;
615 vcpudispatch_stats_freq = freq;
620 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
622 seq_printf(p, "%d\n", vcpudispatch_stats_freq);
626 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
628 return single_open(file, vcpudispatch_stats_freq_display, NULL);
631 static const struct file_operations vcpudispatch_stats_freq_proc_ops = {
632 .open = vcpudispatch_stats_freq_open,
634 .write = vcpudispatch_stats_freq_write,
636 .release = single_release,
639 static int __init vcpudispatch_stats_procfs_init(void)
641 if (!lppaca_shared_proc(get_lppaca()))
644 if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
645 &vcpudispatch_stats_proc_ops))
646 pr_err("vcpudispatch_stats: error creating procfs file\n");
647 else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
648 &vcpudispatch_stats_freq_proc_ops))
649 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
654 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
655 #endif /* CONFIG_PPC_SPLPAR */
657 void vpa_init(int cpu)
659 int hwcpu = get_hard_smp_processor_id(cpu);
664 * The spec says it "may be problematic" if CPU x registers the VPA of
665 * CPU y. We should never do that, but wail if we ever do.
667 WARN_ON(cpu != smp_processor_id());
669 if (cpu_has_feature(CPU_FTR_ALTIVEC))
670 lppaca_of(cpu).vmxregs_in_use = 1;
672 if (cpu_has_feature(CPU_FTR_ARCH_207S))
673 lppaca_of(cpu).ebb_regs_in_use = 1;
675 addr = __pa(&lppaca_of(cpu));
676 ret = register_vpa(hwcpu, addr);
679 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
680 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
684 #ifdef CONFIG_PPC_BOOK3S_64
686 * PAPR says this feature is SLB-Buffer but firmware never
687 * reports that. All SPLPAR support SLB shadow buffer.
689 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
690 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
691 ret = register_slb_shadow(hwcpu, addr);
693 pr_err("WARNING: SLB shadow buffer registration for "
694 "cpu %d (hw %d) of area %lx failed with %ld\n",
695 cpu, hwcpu, addr, ret);
697 #endif /* CONFIG_PPC_BOOK3S_64 */
700 * Register dispatch trace log, if one has been allocated.
702 register_dtl_buffer(cpu);
705 #ifdef CONFIG_PPC_BOOK3S_64
707 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
708 unsigned long vpn, unsigned long pa,
709 unsigned long rflags, unsigned long vflags,
710 int psize, int apsize, int ssize)
712 unsigned long lpar_rc;
715 unsigned long hpte_v, hpte_r;
717 if (!(vflags & HPTE_V_BOLTED))
718 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
719 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
720 hpte_group, vpn, pa, rflags, vflags, psize);
722 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
723 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
725 if (!(vflags & HPTE_V_BOLTED))
726 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
728 /* Now fill in the actual HPTE */
729 /* Set CEC cookie to 0 */
731 /* I-cache Invalidate = 0 */
732 /* I-cache synchronize = 0 */
736 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
737 flags |= H_COALESCE_CAND;
739 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
740 if (unlikely(lpar_rc == H_PTEG_FULL)) {
741 pr_devel("Hash table group is full\n");
746 * Since we try and ioremap PHBs we don't own, the pte insert
747 * will fail. However we must catch the failure in hash_page
748 * or we will loop forever, so return -2 in this case.
750 if (unlikely(lpar_rc != H_SUCCESS)) {
751 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
754 if (!(vflags & HPTE_V_BOLTED))
755 pr_devel(" -> slot: %lu\n", slot & 7);
757 /* Because of iSeries, we have to pass down the secondary
758 * bucket bit here as well
760 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
763 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
765 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
767 unsigned long slot_offset;
768 unsigned long lpar_rc;
770 unsigned long dummy1, dummy2;
772 /* pick a random slot to start at */
773 slot_offset = mftb() & 0x7;
775 for (i = 0; i < HPTES_PER_GROUP; i++) {
777 /* don't remove a bolted entry */
778 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
779 (0x1UL << 4), &dummy1, &dummy2);
780 if (lpar_rc == H_SUCCESS)
784 * The test for adjunct partition is performed before the
785 * ANDCOND test. H_RESOURCE may be returned, so we need to
786 * check for that as well.
788 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
797 static void manual_hpte_clear_all(void)
799 unsigned long size_bytes = 1UL << ppc64_pft_size;
800 unsigned long hpte_count = size_bytes >> 4;
808 /* Read in batches of 4,
809 * invalidate only valid entries not in the VRMA
810 * hpte_count will be a multiple of 4
812 for (i = 0; i < hpte_count; i += 4) {
813 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
814 if (lpar_rc != H_SUCCESS) {
815 pr_info("Failed to read hash page table at %ld err %ld\n",
819 for (j = 0; j < 4; j++){
820 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
823 if (ptes[j].pteh & HPTE_V_VALID)
824 plpar_pte_remove_raw(0, i + j, 0,
825 &(ptes[j].pteh), &(ptes[j].ptel));
830 static int hcall_hpte_clear_all(void)
835 rc = plpar_hcall_norets(H_CLEAR_HPT);
836 } while (rc == H_CONTINUE);
841 static void pseries_hpte_clear_all(void)
845 rc = hcall_hpte_clear_all();
847 manual_hpte_clear_all();
849 #ifdef __LITTLE_ENDIAN__
851 * Reset exceptions to big endian.
853 * FIXME this is a hack for kexec, we need to reset the exception
854 * endian before starting the new kernel and this is a convenient place
857 * This is also called on boot when a fadump happens. In that case we
858 * must not change the exception endian mode.
860 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
861 pseries_big_endian_exceptions();
866 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
867 * the low 3 bits of flags happen to line up. So no transform is needed.
868 * We can probably optimize here and assume the high bits of newpp are
869 * already zero. For now I am paranoid.
871 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
874 int psize, int apsize,
875 int ssize, unsigned long inv_flags)
877 unsigned long lpar_rc;
879 unsigned long want_v;
881 want_v = hpte_encode_avpn(vpn, psize, ssize);
883 flags = (newpp & 7) | H_AVPN;
884 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
885 /* Move pp0 into bit 8 (IBM 55) */
886 flags |= (newpp & HPTE_R_PP0) >> 55;
888 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
889 want_v, slot, flags, psize);
891 lpar_rc = plpar_pte_protect(flags, slot, want_v);
893 if (lpar_rc == H_NOT_FOUND) {
894 pr_devel("not found !\n");
900 BUG_ON(lpar_rc != H_SUCCESS);
905 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
914 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
916 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
917 if (lpar_rc != H_SUCCESS) {
918 pr_info("Failed to read hash page table at %ld err %ld\n",
919 hpte_group, lpar_rc);
923 for (j = 0; j < 4; j++) {
924 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
925 (ptes[j].pteh & HPTE_V_VALID))
933 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
937 unsigned long want_v;
938 unsigned long hpte_group;
940 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
941 want_v = hpte_encode_avpn(vpn, psize, ssize);
943 /* Bolted entries are always in the primary group */
944 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
945 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
948 return hpte_group + slot;
951 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
953 int psize, int ssize)
956 unsigned long lpar_rc, slot, vsid, flags;
958 vsid = get_kernel_vsid(ea, ssize);
959 vpn = hpt_vpn(ea, vsid, ssize);
961 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
965 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
966 /* Move pp0 into bit 8 (IBM 55) */
967 flags |= (newpp & HPTE_R_PP0) >> 55;
969 lpar_rc = plpar_pte_protect(flags, slot, 0);
971 BUG_ON(lpar_rc != H_SUCCESS);
974 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
975 int psize, int apsize,
976 int ssize, int local)
978 unsigned long want_v;
979 unsigned long lpar_rc;
980 unsigned long dummy1, dummy2;
982 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
983 slot, vpn, psize, local);
985 want_v = hpte_encode_avpn(vpn, psize, ssize);
986 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
987 if (lpar_rc == H_NOT_FOUND)
990 BUG_ON(lpar_rc != H_SUCCESS);
995 * As defined in the PAPR's section 14.5.4.1.8
996 * The control mask doesn't include the returned reference and change bit from
999 #define HBLKR_AVPN 0x0100000000000000UL
1000 #define HBLKR_CTRL_MASK 0xf800000000000000UL
1001 #define HBLKR_CTRL_SUCCESS 0x8000000000000000UL
1002 #define HBLKR_CTRL_ERRNOTFOUND 0x8800000000000000UL
1003 #define HBLKR_CTRL_ERRBUSY 0xa000000000000000UL
1006 * Returned true if we are supporting this block size for the specified segment
1007 * base page size and actual page size.
1009 * Currently, we only support 8 size block.
1011 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1013 return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1017 * H_BLOCK_REMOVE caller.
1018 * @idx should point to the latest @param entry set with a PTEX.
1019 * If PTE cannot be processed because another CPUs has already locked that
1020 * group, those entries are put back in @param starting at index 1.
1021 * If entries has to be retried and @retry_busy is set to true, these entries
1022 * are retried until success. If @retry_busy is set to false, the returned
1023 * is the number of entries yet to process.
1025 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1028 unsigned long i, rc, new_idx;
1029 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1032 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1037 if (idx > PLPAR_HCALL9_BUFSIZE) {
1038 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1039 idx = PLPAR_HCALL9_BUFSIZE;
1040 } else if (idx < PLPAR_HCALL9_BUFSIZE)
1041 param[idx] = HBR_END;
1043 rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1045 param[1], param[2], param[3], param[4], /* TS0-7 */
1046 param[5], param[6], param[7], param[8]);
1047 if (rc == H_SUCCESS)
1050 BUG_ON(rc != H_PARTIAL);
1052 /* Check that the unprocessed entries were 'not found' or 'busy' */
1053 for (i = 0; i < idx-1; i++) {
1054 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1056 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1057 param[++new_idx] = param[i+1];
1061 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1062 && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1066 * If there were entries found busy, retry these entries if requested,
1067 * of if all the entries have to be retried.
1069 if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1077 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1079 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1080 * to make sure that we avoid bouncing the hypervisor tlbie lock.
1082 #define PPC64_HUGE_HPTE_BATCH 12
1084 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1085 int count, int psize, int ssize)
1087 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1088 unsigned long shift, current_vpgb, vpgb;
1091 shift = mmu_psize_defs[psize].shift;
1093 for (i = 0; i < count; i++) {
1095 * Shifting 3 bits more on the right to get a
1096 * 8 pages aligned virtual addresse.
1098 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1099 if (!pix || vpgb != current_vpgb) {
1101 * Need to start a new 8 pages block, flush
1102 * the current one if needed.
1105 (void)call_block_remove(pix, param, true);
1106 current_vpgb = vpgb;
1107 param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1111 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1112 if (pix == PLPAR_HCALL9_BUFSIZE) {
1113 pix = call_block_remove(pix, param, false);
1115 * pix = 0 means that all the entries were
1116 * removed, we can start a new block.
1117 * Otherwise, this means that there are entries
1118 * to retry, and pix points to latest one, so
1119 * we should increment it and try to continue
1127 (void)call_block_remove(pix, param, true);
1130 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1131 int count, int psize, int ssize)
1133 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1134 int i = 0, pix = 0, rc;
1136 for (i = 0; i < count; i++) {
1138 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1139 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1142 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1143 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1146 rc = plpar_hcall9(H_BULK_REMOVE, param,
1147 param[0], param[1], param[2],
1148 param[3], param[4], param[5],
1149 param[6], param[7]);
1150 BUG_ON(rc != H_SUCCESS);
1156 param[pix] = HBR_END;
1157 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1158 param[2], param[3], param[4], param[5],
1159 param[6], param[7]);
1160 BUG_ON(rc != H_SUCCESS);
1164 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1166 int count, int psize,
1169 unsigned long flags = 0;
1170 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1173 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1175 /* Assuming THP size is 16M */
1176 if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1177 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1179 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1182 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1185 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1187 unsigned char *hpte_slot_array,
1188 int psize, int ssize, int local)
1191 unsigned long s_addr = addr;
1192 unsigned int max_hpte_count, valid;
1193 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1194 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1195 unsigned long shift, hidx, vpn = 0, hash, slot;
1197 shift = mmu_psize_defs[psize].shift;
1198 max_hpte_count = 1U << (PMD_SHIFT - shift);
1200 for (i = 0; i < max_hpte_count; i++) {
1201 valid = hpte_valid(hpte_slot_array, i);
1204 hidx = hpte_hash_index(hpte_slot_array, i);
1207 addr = s_addr + (i * (1ul << shift));
1208 vpn = hpt_vpn(addr, vsid, ssize);
1209 hash = hpt_hash(vpn, shift, ssize);
1210 if (hidx & _PTEIDX_SECONDARY)
1213 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1214 slot += hidx & _PTEIDX_GROUP_IX;
1216 slot_array[index] = slot;
1217 vpn_array[index] = vpn;
1218 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1220 * Now do a bluk invalidate
1222 __pSeries_lpar_hugepage_invalidate(slot_array,
1224 PPC64_HUGE_HPTE_BATCH,
1231 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1232 index, psize, ssize);
1235 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1237 unsigned char *hpte_slot_array,
1238 int psize, int ssize, int local)
1240 WARN(1, "%s called without THP support\n", __func__);
1244 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1245 int psize, int ssize)
1248 unsigned long slot, vsid;
1250 vsid = get_kernel_vsid(ea, ssize);
1251 vpn = hpt_vpn(ea, vsid, ssize);
1253 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1258 * lpar doesn't use the passed actual page size
1260 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1265 static inline unsigned long compute_slot(real_pte_t pte,
1267 unsigned long index,
1268 unsigned long shift,
1271 unsigned long slot, hash, hidx;
1273 hash = hpt_hash(vpn, shift, ssize);
1274 hidx = __rpte_to_hidx(pte, index);
1275 if (hidx & _PTEIDX_SECONDARY)
1277 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1278 slot += hidx & _PTEIDX_GROUP_IX;
1283 * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1284 * "all within the same naturally aligned 8 page virtual address block".
1286 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1287 unsigned long *param)
1290 unsigned long i, pix = 0;
1291 unsigned long index, shift, slot, current_vpgb, vpgb;
1295 psize = batch->psize;
1296 ssize = batch->ssize;
1298 for (i = 0; i < number; i++) {
1299 vpn = batch->vpn[i];
1300 pte = batch->pte[i];
1301 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1303 * Shifting 3 bits more on the right to get a
1304 * 8 pages aligned virtual addresse.
1306 vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1307 if (!pix || vpgb != current_vpgb) {
1309 * Need to start a new 8 pages block, flush
1310 * the current one if needed.
1313 (void)call_block_remove(pix, param,
1315 current_vpgb = vpgb;
1316 param[0] = hpte_encode_avpn(vpn, psize,
1321 slot = compute_slot(pte, vpn, index, shift, ssize);
1322 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1324 if (pix == PLPAR_HCALL9_BUFSIZE) {
1325 pix = call_block_remove(pix, param, false);
1327 * pix = 0 means that all the entries were
1328 * removed, we can start a new block.
1329 * Otherwise, this means that there are entries
1330 * to retry, and pix points to latest one, so
1331 * we should increment it and try to continue
1337 } pte_iterate_hashed_end();
1341 (void)call_block_remove(pix, param, true);
1345 * TLB Block Invalidate Characteristics
1347 * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1348 * is able to process for each couple segment base page size, actual page size.
1350 * The ibm,get-system-parameter properties is returning a buffer with the
1353 * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1355 * TLB Block Invalidate Specifiers:
1356 * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1357 * [ 1 byte Number of page sizes (N) that are supported for the specified
1358 * TLB invalidate block size ]
1359 * [ 1 byte Encoded segment base page size and actual page size
1360 * MSB=0 means 4k segment base page size and actual page size
1361 * MSB=1 the penc value in mmu_psize_def ]
1364 * Next TLB Block Invalidate Specifiers...
1368 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1369 unsigned int block_size)
1371 if (block_size > hblkrm_size[bpsize][psize])
1372 hblkrm_size[bpsize][psize] = block_size;
1376 * Decode the Encoded segment base page size and actual page size.
1378 * - bit 7 is the L bit
1379 * - bits 0-5 are the penc value
1380 * If the L bit is 0, this means 4K segment base page size and actual page size
1381 * otherwise the penc value should be read.
1383 #define HBLKRM_L_MASK 0x80
1384 #define HBLKRM_PENC_MASK 0x3f
1385 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1386 unsigned int block_size)
1388 unsigned int bpsize, psize;
1390 /* First, check the L bit, if not set, this means 4K */
1391 if ((lp & HBLKRM_L_MASK) == 0) {
1392 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1396 lp &= HBLKRM_PENC_MASK;
1397 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1398 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1400 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1401 if (def->penc[psize] == lp) {
1402 set_hblkrm_bloc_size(bpsize, psize, block_size);
1409 #define SPLPAR_TLB_BIC_TOKEN 50
1412 * The size of the TLB Block Invalidate Characteristics is variable. But at the
1413 * maximum it will be the number of possible page sizes *2 + 10 bytes.
1414 * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1415 * (128 bytes) for the buffer to get plenty of space.
1417 #define SPLPAR_TLB_BIC_MAXLENGTH 128
1419 void __init pseries_lpar_read_hblkrm_characteristics(void)
1421 const s32 token = rtas_token("ibm,get-system-parameter");
1422 unsigned char local_buffer[SPLPAR_TLB_BIC_MAXLENGTH];
1423 int call_status, len, idx, bpsize;
1425 if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1429 spin_lock(&rtas_data_buf_lock);
1430 memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
1431 call_status = rtas_call(token, 3, 1, NULL, SPLPAR_TLB_BIC_TOKEN,
1432 __pa(rtas_data_buf), RTAS_DATA_BUF_SIZE);
1433 memcpy(local_buffer, rtas_data_buf, SPLPAR_TLB_BIC_MAXLENGTH);
1434 local_buffer[SPLPAR_TLB_BIC_MAXLENGTH - 1] = '\0';
1435 spin_unlock(&rtas_data_buf_lock);
1436 } while (rtas_busy_delay(call_status));
1438 if (call_status != 0) {
1439 pr_warn("%s %s Error calling get-system-parameter (0x%x)\n",
1440 __FILE__, __func__, call_status);
1445 * The first two (2) bytes of the data in the buffer are the length of
1446 * the returned data, not counting these first two (2) bytes.
1448 len = be16_to_cpu(*((u16 *)local_buffer)) + 2;
1449 if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1450 pr_warn("%s too large returned buffer %d", __func__, len);
1456 u8 block_shift = local_buffer[idx++];
1458 unsigned int npsize;
1463 block_size = 1 << block_shift;
1465 for (npsize = local_buffer[idx++];
1466 npsize > 0 && idx < len; npsize--)
1467 check_lp_set_hblkrm((unsigned int) local_buffer[idx++],
1471 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1472 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1473 if (hblkrm_size[bpsize][idx])
1474 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1475 bpsize, idx, hblkrm_size[bpsize][idx]);
1479 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1482 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1485 unsigned long i, pix, rc;
1486 unsigned long flags = 0;
1487 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1488 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1489 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1490 unsigned long index, shift, slot;
1495 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1497 if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1498 do_block_remove(number, batch, param);
1502 psize = batch->psize;
1503 ssize = batch->ssize;
1505 for (i = 0; i < number; i++) {
1506 vpn = batch->vpn[i];
1507 pte = batch->pte[i];
1508 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1509 slot = compute_slot(pte, vpn, index, shift, ssize);
1510 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1512 * lpar doesn't use the passed actual page size
1514 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1517 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1518 param[pix+1] = hpte_encode_avpn(vpn, psize,
1522 rc = plpar_hcall9(H_BULK_REMOVE, param,
1523 param[0], param[1], param[2],
1524 param[3], param[4], param[5],
1525 param[6], param[7]);
1526 BUG_ON(rc != H_SUCCESS);
1530 } pte_iterate_hashed_end();
1533 param[pix] = HBR_END;
1534 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1535 param[2], param[3], param[4], param[5],
1536 param[6], param[7]);
1537 BUG_ON(rc != H_SUCCESS);
1542 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1545 static int __init disable_bulk_remove(char *str)
1547 if (strcmp(str, "off") == 0 &&
1548 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1549 pr_info("Disabling BULK_REMOVE firmware feature");
1550 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1555 __setup("bulk_remove=", disable_bulk_remove);
1557 #define HPT_RESIZE_TIMEOUT 10000 /* ms */
1559 struct hpt_resize_state {
1560 unsigned long shift;
1564 static int pseries_lpar_resize_hpt_commit(void *data)
1566 struct hpt_resize_state *state = data;
1568 state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1569 if (state->commit_rc != H_SUCCESS)
1572 /* Hypervisor has transitioned the HTAB, update our globals */
1573 ppc64_pft_size = state->shift;
1574 htab_size_bytes = 1UL << ppc64_pft_size;
1575 htab_hash_mask = (htab_size_bytes >> 7) - 1;
1581 * Must be called in process context. The caller must hold the
1584 static int pseries_lpar_resize_hpt(unsigned long shift)
1586 struct hpt_resize_state state = {
1588 .commit_rc = H_FUNCTION,
1590 unsigned int delay, total_delay = 0;
1596 if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1599 pr_info("Attempting to resize HPT to shift %lu\n", shift);
1603 rc = plpar_resize_hpt_prepare(0, shift);
1604 while (H_IS_LONG_BUSY(rc)) {
1605 delay = get_longbusy_msecs(rc);
1606 total_delay += delay;
1607 if (total_delay > HPT_RESIZE_TIMEOUT) {
1608 /* prepare with shift==0 cancels an in-progress resize */
1609 rc = plpar_resize_hpt_prepare(0, 0);
1610 if (rc != H_SUCCESS)
1611 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1616 rc = plpar_resize_hpt_prepare(0, shift);
1625 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1628 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1631 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1637 rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1643 switch (state.commit_rc) {
1648 pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1654 pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1655 shift, (long long) ktime_ms_delta(t1, t0),
1656 (long long) ktime_ms_delta(t2, t1));
1661 static int pseries_lpar_register_process_table(unsigned long base,
1662 unsigned long page_size, unsigned long table_size)
1665 unsigned long flags = 0;
1668 flags |= PROC_TABLE_NEW;
1669 if (radix_enabled())
1670 flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
1672 flags |= PROC_TABLE_HPT_SLB;
1674 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
1675 page_size, table_size);
1676 if (!H_IS_LONG_BUSY(rc))
1678 mdelay(get_longbusy_msecs(rc));
1680 if (rc != H_SUCCESS) {
1681 pr_err("Failed to register process table (rc=%ld)\n", rc);
1687 void __init hpte_init_pseries(void)
1689 mmu_hash_ops.hpte_invalidate = pSeries_lpar_hpte_invalidate;
1690 mmu_hash_ops.hpte_updatepp = pSeries_lpar_hpte_updatepp;
1691 mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1692 mmu_hash_ops.hpte_insert = pSeries_lpar_hpte_insert;
1693 mmu_hash_ops.hpte_remove = pSeries_lpar_hpte_remove;
1694 mmu_hash_ops.hpte_removebolted = pSeries_lpar_hpte_removebolted;
1695 mmu_hash_ops.flush_hash_range = pSeries_lpar_flush_hash_range;
1696 mmu_hash_ops.hpte_clear_all = pseries_hpte_clear_all;
1697 mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1699 if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1700 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1703 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1704 * to inform the hypervisor that we wish to use the HPT.
1706 if (cpu_has_feature(CPU_FTR_ARCH_300))
1707 pseries_lpar_register_process_table(0, 0, 0);
1710 void radix_init_pseries(void)
1712 pr_info("Using radix MMU under hypervisor\n");
1714 pseries_lpar_register_process_table(__pa(process_tb),
1715 0, PRTB_SIZE_SHIFT - 12);
1718 #ifdef CONFIG_PPC_SMLPAR
1719 #define CMO_FREE_HINT_DEFAULT 1
1720 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1722 static int __init cmo_free_hint(char *str)
1725 parm = strstrip(str);
1727 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1728 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1729 cmo_free_hint_flag = 0;
1733 cmo_free_hint_flag = 1;
1734 pr_info("%s: CMO free page hinting is active.\n", __func__);
1736 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1742 __setup("cmo_free_hint=", cmo_free_hint);
1744 static void pSeries_set_page_state(struct page *page, int order,
1745 unsigned long state)
1748 unsigned long cmo_page_sz, addr;
1750 cmo_page_sz = cmo_get_page_size();
1751 addr = __pa((unsigned long)page_address(page));
1753 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1754 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1755 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1759 void arch_free_page(struct page *page, int order)
1761 if (radix_enabled())
1763 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1766 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1768 EXPORT_SYMBOL(arch_free_page);
1770 #endif /* CONFIG_PPC_SMLPAR */
1771 #endif /* CONFIG_PPC_BOOK3S_64 */
1773 #ifdef CONFIG_TRACEPOINTS
1774 #ifdef CONFIG_JUMP_LABEL
1775 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1777 int hcall_tracepoint_regfunc(void)
1779 static_key_slow_inc(&hcall_tracepoint_key);
1783 void hcall_tracepoint_unregfunc(void)
1785 static_key_slow_dec(&hcall_tracepoint_key);
1789 * We optimise our hcall path by placing hcall_tracepoint_refcount
1790 * directly in the TOC so we can check if the hcall tracepoints are
1791 * enabled via a single load.
1794 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1795 extern long hcall_tracepoint_refcount;
1797 int hcall_tracepoint_regfunc(void)
1799 hcall_tracepoint_refcount++;
1803 void hcall_tracepoint_unregfunc(void)
1805 hcall_tracepoint_refcount--;
1810 * Since the tracing code might execute hcalls we need to guard against
1811 * recursion. One example of this are spinlocks calling H_YIELD on
1812 * shared processor partitions.
1814 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1817 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1819 unsigned long flags;
1820 unsigned int *depth;
1823 * We cannot call tracepoints inside RCU idle regions which
1824 * means we must not trace H_CEDE.
1826 if (opcode == H_CEDE)
1829 local_irq_save(flags);
1831 depth = this_cpu_ptr(&hcall_trace_depth);
1838 trace_hcall_entry(opcode, args);
1842 local_irq_restore(flags);
1845 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1847 unsigned long flags;
1848 unsigned int *depth;
1850 if (opcode == H_CEDE)
1853 local_irq_save(flags);
1855 depth = this_cpu_ptr(&hcall_trace_depth);
1861 trace_hcall_exit(opcode, retval, retbuf);
1866 local_irq_restore(flags);
1872 * H_GET_MPP hcall returns info in 7 parms
1874 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1877 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1879 rc = plpar_hcall9(H_GET_MPP, retbuf);
1881 mpp_data->entitled_mem = retbuf[0];
1882 mpp_data->mapped_mem = retbuf[1];
1884 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1885 mpp_data->pool_num = retbuf[2] & 0xffff;
1887 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1888 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1889 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1891 mpp_data->pool_size = retbuf[4];
1892 mpp_data->loan_request = retbuf[5];
1893 mpp_data->backing_mem = retbuf[6];
1897 EXPORT_SYMBOL(h_get_mpp);
1899 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1902 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1904 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1906 mpp_x_data->coalesced_bytes = retbuf[0];
1907 mpp_x_data->pool_coalesced_bytes = retbuf[1];
1908 mpp_x_data->pool_purr_cycles = retbuf[2];
1909 mpp_x_data->pool_spurr_cycles = retbuf[3];
1914 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
1916 unsigned long protovsid;
1917 unsigned long va_bits = VA_BITS;
1918 unsigned long modinv, vsid_modulus;
1919 unsigned long max_mod_inv, tmp_modinv;
1921 if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1924 if (ssize == MMU_SEGSIZE_256M) {
1925 modinv = VSID_MULINV_256M;
1926 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1928 modinv = VSID_MULINV_1T;
1929 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1933 * vsid outside our range.
1935 if (vsid >= vsid_modulus)
1939 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1940 * and vsid = (protovsid * x) % vsid_modulus, then we say:
1941 * protovsid = (vsid * modinv) % vsid_modulus
1944 /* Check if (vsid * modinv) overflow (63 bits) */
1945 max_mod_inv = 0x7fffffffffffffffull / vsid;
1946 if (modinv < max_mod_inv)
1947 return (vsid * modinv) % vsid_modulus;
1949 tmp_modinv = modinv/max_mod_inv;
1950 modinv %= max_mod_inv;
1952 protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1953 protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1958 static int __init reserve_vrma_context_id(void)
1960 unsigned long protovsid;
1963 * Reserve context ids which map to reserved virtual addresses. For now
1964 * we only reserve the context id which maps to the VRMA VSID. We ignore
1965 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1966 * enable adjunct support via the "ibm,client-architecture-support"
1969 protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1970 hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1973 machine_device_initcall(pseries, reserve_vrma_context_id);
1975 #ifdef CONFIG_DEBUG_FS
1976 /* debugfs file interface for vpa data */
1977 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1980 int cpu = (long)filp->private_data;
1981 struct lppaca *lppaca = &lppaca_of(cpu);
1983 return simple_read_from_buffer(buf, len, pos, lppaca,
1984 sizeof(struct lppaca));
1987 static const struct file_operations vpa_fops = {
1988 .open = simple_open,
1989 .read = vpa_file_read,
1990 .llseek = default_llseek,
1993 static int __init vpa_debugfs_init(void)
1997 struct dentry *vpa_dir;
1999 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2002 vpa_dir = debugfs_create_dir("vpa", powerpc_debugfs_root);
2004 pr_warn("%s: can't create vpa root dir\n", __func__);
2008 /* set up the per-cpu vpa file*/
2009 for_each_possible_cpu(i) {
2012 sprintf(name, "cpu-%ld", i);
2014 d = debugfs_create_file(name, 0400, vpa_dir, (void *)i,
2017 pr_warn("%s: can't create per-cpu vpa file\n",
2025 machine_arch_initcall(pseries, vpa_debugfs_init);
2026 #endif /* CONFIG_DEBUG_FS */