1 /*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License, version 2, as
6 * published by the Free Software Foundation.
7 */
8
9 #include <linux/cpu.h>
10 #include <linux/kvm_host.h>
11 #include <linux/preempt.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/init.h>
16 #include <linux/memblock.h>
17 #include <linux/sizes.h>
18 #include <linux/cma.h>
19 #include <linux/bitops.h>
20
21 #include <asm/cputable.h>
22 #include <asm/kvm_ppc.h>
23 #include <asm/kvm_book3s.h>
24 #include <asm/archrandom.h>
25 #include <asm/xics.h>
26 #include <asm/dbell.h>
27 #include <asm/cputhreads.h>
28
29 #define KVM_CMA_CHUNK_ORDER 18
30
31 /*
32 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
33 * should be power of 2.
34 */
35 #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
36 /*
37 * By default we reserve 5% of memory for hash pagetable allocation.
38 */
39 static unsigned long kvm_cma_resv_ratio = 5;
40
41 static struct cma *kvm_cma;
42
early_parse_kvm_cma_resv(char * p)43 static int __init early_parse_kvm_cma_resv(char *p)
44 {
45 pr_debug("%s(%s)\n", __func__, p);
46 if (!p)
47 return -EINVAL;
48 return kstrtoul(p, 0, &kvm_cma_resv_ratio);
49 }
50 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
51
kvm_alloc_hpt(unsigned long nr_pages)52 struct page *kvm_alloc_hpt(unsigned long nr_pages)
53 {
54 VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT);
55
56 return cma_alloc(kvm_cma, nr_pages, order_base_2(HPT_ALIGN_PAGES));
57 }
58 EXPORT_SYMBOL_GPL(kvm_alloc_hpt);
59
kvm_release_hpt(struct page * page,unsigned long nr_pages)60 void kvm_release_hpt(struct page *page, unsigned long nr_pages)
61 {
62 cma_release(kvm_cma, page, nr_pages);
63 }
64 EXPORT_SYMBOL_GPL(kvm_release_hpt);
65
66 /**
67 * kvm_cma_reserve() - reserve area for kvm hash pagetable
68 *
69 * This function reserves memory from early allocator. It should be
70 * called by arch specific code once the memblock allocator
71 * has been activated and all other subsystems have already allocated/reserved
72 * memory.
73 */
kvm_cma_reserve(void)74 void __init kvm_cma_reserve(void)
75 {
76 unsigned long align_size;
77 struct memblock_region *reg;
78 phys_addr_t selected_size = 0;
79
80 /*
81 * We need CMA reservation only when we are in HV mode
82 */
83 if (!cpu_has_feature(CPU_FTR_HVMODE))
84 return;
85 /*
86 * We cannot use memblock_phys_mem_size() here, because
87 * memblock_analyze() has not been called yet.
88 */
89 for_each_memblock(memory, reg)
90 selected_size += memblock_region_memory_end_pfn(reg) -
91 memblock_region_memory_base_pfn(reg);
92
93 selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
94 if (selected_size) {
95 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
96 (unsigned long)selected_size / SZ_1M);
97 align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
98 cma_declare_contiguous(0, selected_size, 0, align_size,
99 KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, &kvm_cma);
100 }
101 }
102
103 /*
104 * Real-mode H_CONFER implementation.
105 * We check if we are the only vcpu out of this virtual core
106 * still running in the guest and not ceded. If so, we pop up
107 * to the virtual-mode implementation; if not, just return to
108 * the guest.
109 */
kvmppc_rm_h_confer(struct kvm_vcpu * vcpu,int target,unsigned int yield_count)110 long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target,
111 unsigned int yield_count)
112 {
113 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
114 int ptid = local_paca->kvm_hstate.ptid;
115 int threads_running;
116 int threads_ceded;
117 int threads_conferring;
118 u64 stop = get_tb() + 10 * tb_ticks_per_usec;
119 int rv = H_SUCCESS; /* => don't yield */
120
121 set_bit(ptid, &vc->conferring_threads);
122 while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) {
123 threads_running = VCORE_ENTRY_MAP(vc);
124 threads_ceded = vc->napping_threads;
125 threads_conferring = vc->conferring_threads;
126 if ((threads_ceded | threads_conferring) == threads_running) {
127 rv = H_TOO_HARD; /* => do yield */
128 break;
129 }
130 }
131 clear_bit(ptid, &vc->conferring_threads);
132 return rv;
133 }
134
135 /*
136 * When running HV mode KVM we need to block certain operations while KVM VMs
137 * exist in the system. We use a counter of VMs to track this.
138 *
139 * One of the operations we need to block is onlining of secondaries, so we
140 * protect hv_vm_count with get/put_online_cpus().
141 */
142 static atomic_t hv_vm_count;
143
kvm_hv_vm_activated(void)144 void kvm_hv_vm_activated(void)
145 {
146 get_online_cpus();
147 atomic_inc(&hv_vm_count);
148 put_online_cpus();
149 }
150 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated);
151
kvm_hv_vm_deactivated(void)152 void kvm_hv_vm_deactivated(void)
153 {
154 get_online_cpus();
155 atomic_dec(&hv_vm_count);
156 put_online_cpus();
157 }
158 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated);
159
kvm_hv_mode_active(void)160 bool kvm_hv_mode_active(void)
161 {
162 return atomic_read(&hv_vm_count) != 0;
163 }
164
165 extern int hcall_real_table[], hcall_real_table_end[];
166
kvmppc_hcall_impl_hv_realmode(unsigned long cmd)167 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd)
168 {
169 cmd /= 4;
170 if (cmd < hcall_real_table_end - hcall_real_table &&
171 hcall_real_table[cmd])
172 return 1;
173
174 return 0;
175 }
176 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode);
177
kvmppc_hwrng_present(void)178 int kvmppc_hwrng_present(void)
179 {
180 return powernv_hwrng_present();
181 }
182 EXPORT_SYMBOL_GPL(kvmppc_hwrng_present);
183
kvmppc_h_random(struct kvm_vcpu * vcpu)184 long kvmppc_h_random(struct kvm_vcpu *vcpu)
185 {
186 if (powernv_get_random_real_mode(&vcpu->arch.gpr[4]))
187 return H_SUCCESS;
188
189 return H_HARDWARE;
190 }
191
rm_writeb(unsigned long paddr,u8 val)192 static inline void rm_writeb(unsigned long paddr, u8 val)
193 {
194 __asm__ __volatile__("stbcix %0,0,%1"
195 : : "r" (val), "r" (paddr) : "memory");
196 }
197
198 /*
199 * Send an interrupt or message to another CPU.
200 * This can only be called in real mode.
201 * The caller needs to include any barrier needed to order writes
202 * to memory vs. the IPI/message.
203 */
kvmhv_rm_send_ipi(int cpu)204 void kvmhv_rm_send_ipi(int cpu)
205 {
206 unsigned long xics_phys;
207
208 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
209 if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
210 cpu_first_thread_sibling(cpu) ==
211 cpu_first_thread_sibling(raw_smp_processor_id())) {
212 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
213 msg |= cpu_thread_in_core(cpu);
214 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
215 return;
216 }
217
218 /* Else poke the target with an IPI */
219 xics_phys = paca[cpu].kvm_hstate.xics_phys;
220 rm_writeb(xics_phys + XICS_MFRR, IPI_PRIORITY);
221 }
222
223 /*
224 * The following functions are called from the assembly code
225 * in book3s_hv_rmhandlers.S.
226 */
kvmhv_interrupt_vcore(struct kvmppc_vcore * vc,int active)227 static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active)
228 {
229 int cpu = vc->pcpu;
230
231 /* Order setting of exit map vs. msgsnd/IPI */
232 smp_mb();
233 for (; active; active >>= 1, ++cpu)
234 if (active & 1)
235 kvmhv_rm_send_ipi(cpu);
236 }
237
kvmhv_commence_exit(int trap)238 void kvmhv_commence_exit(int trap)
239 {
240 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
241 int ptid = local_paca->kvm_hstate.ptid;
242 struct kvm_split_mode *sip = local_paca->kvm_hstate.kvm_split_mode;
243 int me, ee, i;
244
245 /* Set our bit in the threads-exiting-guest map in the 0xff00
246 bits of vcore->entry_exit_map */
247 me = 0x100 << ptid;
248 do {
249 ee = vc->entry_exit_map;
250 } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee);
251
252 /* Are we the first here? */
253 if ((ee >> 8) != 0)
254 return;
255
256 /*
257 * Trigger the other threads in this vcore to exit the guest.
258 * If this is a hypervisor decrementer interrupt then they
259 * will be already on their way out of the guest.
260 */
261 if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER)
262 kvmhv_interrupt_vcore(vc, ee & ~(1 << ptid));
263
264 /*
265 * If we are doing dynamic micro-threading, interrupt the other
266 * subcores to pull them out of their guests too.
267 */
268 if (!sip)
269 return;
270
271 for (i = 0; i < MAX_SUBCORES; ++i) {
272 vc = sip->master_vcs[i];
273 if (!vc)
274 break;
275 do {
276 ee = vc->entry_exit_map;
277 /* Already asked to exit? */
278 if ((ee >> 8) != 0)
279 break;
280 } while (cmpxchg(&vc->entry_exit_map, ee,
281 ee | VCORE_EXIT_REQ) != ee);
282 if ((ee >> 8) == 0)
283 kvmhv_interrupt_vcore(vc, ee);
284 }
285 }
286