xref: /openbmc/linux/arch/powerpc/kvm/book3s_xive.h (revision 3a35093a)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright 2017 Benjamin Herrenschmidt, IBM Corporation
4  */
5 
6 #ifndef _KVM_PPC_BOOK3S_XIVE_H
7 #define _KVM_PPC_BOOK3S_XIVE_H
8 
9 #ifdef CONFIG_KVM_XICS
10 #include "book3s_xics.h"
11 
12 /*
13  * The XIVE Interrupt source numbers are within the range 0 to
14  * KVMPPC_XICS_NR_IRQS.
15  */
16 #define KVMPPC_XIVE_FIRST_IRQ	0
17 #define KVMPPC_XIVE_NR_IRQS	KVMPPC_XICS_NR_IRQS
18 
19 /*
20  * State for one guest irq source.
21  *
22  * For each guest source we allocate a HW interrupt in the XIVE
23  * which we use for all SW triggers. It will be unused for
24  * pass-through but it's easier to keep around as the same
25  * guest interrupt can alternatively be emulated or pass-through
26  * if a physical device is hot unplugged and replaced with an
27  * emulated one.
28  *
29  * This state structure is very similar to the XICS one with
30  * additional XIVE specific tracking.
31  */
32 struct kvmppc_xive_irq_state {
33 	bool valid;			/* Interrupt entry is valid */
34 
35 	u32 number;			/* Guest IRQ number */
36 	u32 ipi_number;			/* XIVE IPI HW number */
37 	struct xive_irq_data ipi_data;	/* XIVE IPI associated data */
38 	u32 pt_number;			/* XIVE Pass-through number if any */
39 	struct xive_irq_data *pt_data;	/* XIVE Pass-through associated data */
40 
41 	/* Targetting as set by guest */
42 	u8 guest_priority;		/* Guest set priority */
43 	u8 saved_priority;		/* Saved priority when masking */
44 
45 	/* Actual targetting */
46 	u32 act_server;			/* Actual server */
47 	u8 act_priority;		/* Actual priority */
48 
49 	/* Various state bits */
50 	bool in_eoi;			/* Synchronize with H_EOI */
51 	bool old_p;			/* P bit state when masking */
52 	bool old_q;			/* Q bit state when masking */
53 	bool lsi;			/* level-sensitive interrupt */
54 	bool asserted;			/* Only for emulated LSI: current state */
55 
56 	/* Saved for migration state */
57 	bool in_queue;
58 	bool saved_p;
59 	bool saved_q;
60 	u8 saved_scan_prio;
61 
62 	/* Xive native */
63 	u32 eisn;			/* Guest Effective IRQ number */
64 };
65 
66 /* Select the "right" interrupt (IPI vs. passthrough) */
67 static inline void kvmppc_xive_select_irq(struct kvmppc_xive_irq_state *state,
68 					  u32 *out_hw_irq,
69 					  struct xive_irq_data **out_xd)
70 {
71 	if (state->pt_number) {
72 		if (out_hw_irq)
73 			*out_hw_irq = state->pt_number;
74 		if (out_xd)
75 			*out_xd = state->pt_data;
76 	} else {
77 		if (out_hw_irq)
78 			*out_hw_irq = state->ipi_number;
79 		if (out_xd)
80 			*out_xd = &state->ipi_data;
81 	}
82 }
83 
84 /*
85  * This corresponds to an "ICS" in XICS terminology, we use it
86  * as a mean to break up source information into multiple structures.
87  */
88 struct kvmppc_xive_src_block {
89 	arch_spinlock_t lock;
90 	u16 id;
91 	struct kvmppc_xive_irq_state irq_state[KVMPPC_XICS_IRQ_PER_ICS];
92 };
93 
94 struct kvmppc_xive;
95 
96 struct kvmppc_xive_ops {
97 	int (*reset_mapped)(struct kvm *kvm, unsigned long guest_irq);
98 };
99 
100 struct kvmppc_xive {
101 	struct kvm *kvm;
102 	struct kvm_device *dev;
103 	struct dentry *dentry;
104 
105 	/* VP block associated with the VM */
106 	u32	vp_base;
107 
108 	/* Blocks of sources */
109 	struct kvmppc_xive_src_block *src_blocks[KVMPPC_XICS_MAX_ICS_ID + 1];
110 	u32	max_sbid;
111 
112 	/*
113 	 * For state save, we lazily scan the queues on the first interrupt
114 	 * being migrated. We don't have a clean way to reset that flags
115 	 * so we keep track of the number of valid sources and how many of
116 	 * them were migrated so we can reset when all of them have been
117 	 * processed.
118 	 */
119 	u32	src_count;
120 	u32	saved_src_count;
121 
122 	/*
123 	 * Some irqs are delayed on restore until the source is created,
124 	 * keep track here of how many of them
125 	 */
126 	u32	delayed_irqs;
127 
128 	/* Which queues (priorities) are in use by the guest */
129 	u8	qmap;
130 
131 	/* Queue orders */
132 	u32	q_order;
133 	u32	q_page_order;
134 
135 	/* Flags */
136 	u8	single_escalation;
137 
138 	/* Number of entries in the VP block */
139 	u32	nr_servers;
140 
141 	struct kvmppc_xive_ops *ops;
142 	struct address_space   *mapping;
143 	struct mutex mapping_lock;
144 	struct mutex lock;
145 };
146 
147 #define KVMPPC_XIVE_Q_COUNT	8
148 
149 struct kvmppc_xive_vcpu {
150 	struct kvmppc_xive	*xive;
151 	struct kvm_vcpu		*vcpu;
152 	bool			valid;
153 
154 	/* Server number. This is the HW CPU ID from a guest perspective */
155 	u32			server_num;
156 
157 	/*
158 	 * HW VP corresponding to this VCPU. This is the base of the VP
159 	 * block plus the server number.
160 	 */
161 	u32			vp_id;
162 	u32			vp_chip_id;
163 	u32			vp_cam;
164 
165 	/* IPI used for sending ... IPIs */
166 	u32			vp_ipi;
167 	struct xive_irq_data	vp_ipi_data;
168 
169 	/* Local emulation state */
170 	uint8_t			cppr;	/* guest CPPR */
171 	uint8_t			hw_cppr;/* Hardware CPPR */
172 	uint8_t			mfrr;
173 	uint8_t			pending;
174 
175 	/* Each VP has 8 queues though we only provision some */
176 	struct xive_q		queues[KVMPPC_XIVE_Q_COUNT];
177 	u32			esc_virq[KVMPPC_XIVE_Q_COUNT];
178 	char			*esc_virq_names[KVMPPC_XIVE_Q_COUNT];
179 
180 	/* Stash a delayed irq on restore from migration (see set_icp) */
181 	u32			delayed_irq;
182 
183 	/* Stats */
184 	u64			stat_rm_h_xirr;
185 	u64			stat_rm_h_ipoll;
186 	u64			stat_rm_h_cppr;
187 	u64			stat_rm_h_eoi;
188 	u64			stat_rm_h_ipi;
189 	u64			stat_vm_h_xirr;
190 	u64			stat_vm_h_ipoll;
191 	u64			stat_vm_h_cppr;
192 	u64			stat_vm_h_eoi;
193 	u64			stat_vm_h_ipi;
194 };
195 
196 static inline struct kvm_vcpu *kvmppc_xive_find_server(struct kvm *kvm, u32 nr)
197 {
198 	struct kvm_vcpu *vcpu = NULL;
199 	int i;
200 
201 	kvm_for_each_vcpu(i, vcpu, kvm) {
202 		if (vcpu->arch.xive_vcpu && nr == vcpu->arch.xive_vcpu->server_num)
203 			return vcpu;
204 	}
205 	return NULL;
206 }
207 
208 static inline struct kvmppc_xive_src_block *kvmppc_xive_find_source(struct kvmppc_xive *xive,
209 		u32 irq, u16 *source)
210 {
211 	u32 bid = irq >> KVMPPC_XICS_ICS_SHIFT;
212 	u16 src = irq & KVMPPC_XICS_SRC_MASK;
213 
214 	if (source)
215 		*source = src;
216 	if (bid > KVMPPC_XICS_MAX_ICS_ID)
217 		return NULL;
218 	return xive->src_blocks[bid];
219 }
220 
221 /*
222  * When the XIVE resources are allocated at the HW level, the VP
223  * structures describing the vCPUs of a guest are distributed among
224  * the chips to optimize the PowerBUS usage. For best performance, the
225  * guest vCPUs can be pinned to match the VP structure distribution.
226  *
227  * Currently, the VP identifiers are deduced from the vCPU id using
228  * the kvmppc_pack_vcpu_id() routine which is not incorrect but not
229  * optimal either. It VSMT is used, the result is not continuous and
230  * the constraints on HW resources described above can not be met.
231  */
232 static inline u32 kvmppc_xive_vp(struct kvmppc_xive *xive, u32 server)
233 {
234 	return xive->vp_base + kvmppc_pack_vcpu_id(xive->kvm, server);
235 }
236 
237 static inline bool kvmppc_xive_vp_in_use(struct kvm *kvm, u32 vp_id)
238 {
239 	struct kvm_vcpu *vcpu = NULL;
240 	int i;
241 
242 	kvm_for_each_vcpu(i, vcpu, kvm) {
243 		if (vcpu->arch.xive_vcpu && vp_id == vcpu->arch.xive_vcpu->vp_id)
244 			return true;
245 	}
246 	return false;
247 }
248 
249 /*
250  * Mapping between guest priorities and host priorities
251  * is as follow.
252  *
253  * Guest request for 0...6 are honored. Guest request for anything
254  * higher results in a priority of 6 being applied.
255  *
256  * Similar mapping is done for CPPR values
257  */
258 static inline u8 xive_prio_from_guest(u8 prio)
259 {
260 	if (prio == 0xff || prio < 6)
261 		return prio;
262 	return 6;
263 }
264 
265 static inline u8 xive_prio_to_guest(u8 prio)
266 {
267 	return prio;
268 }
269 
270 static inline u32 __xive_read_eq(__be32 *qpage, u32 msk, u32 *idx, u32 *toggle)
271 {
272 	u32 cur;
273 
274 	if (!qpage)
275 		return 0;
276 	cur = be32_to_cpup(qpage + *idx);
277 	if ((cur >> 31) == *toggle)
278 		return 0;
279 	*idx = (*idx + 1) & msk;
280 	if (*idx == 0)
281 		(*toggle) ^= 1;
282 	return cur & 0x7fffffff;
283 }
284 
285 extern unsigned long xive_rm_h_xirr(struct kvm_vcpu *vcpu);
286 extern unsigned long xive_rm_h_ipoll(struct kvm_vcpu *vcpu, unsigned long server);
287 extern int xive_rm_h_ipi(struct kvm_vcpu *vcpu, unsigned long server,
288 			 unsigned long mfrr);
289 extern int xive_rm_h_cppr(struct kvm_vcpu *vcpu, unsigned long cppr);
290 extern int xive_rm_h_eoi(struct kvm_vcpu *vcpu, unsigned long xirr);
291 
292 extern unsigned long (*__xive_vm_h_xirr)(struct kvm_vcpu *vcpu);
293 extern unsigned long (*__xive_vm_h_ipoll)(struct kvm_vcpu *vcpu, unsigned long server);
294 extern int (*__xive_vm_h_ipi)(struct kvm_vcpu *vcpu, unsigned long server,
295 			      unsigned long mfrr);
296 extern int (*__xive_vm_h_cppr)(struct kvm_vcpu *vcpu, unsigned long cppr);
297 extern int (*__xive_vm_h_eoi)(struct kvm_vcpu *vcpu, unsigned long xirr);
298 
299 /*
300  * Common Xive routines for XICS-over-XIVE and XIVE native
301  */
302 void kvmppc_xive_disable_vcpu_interrupts(struct kvm_vcpu *vcpu);
303 int kvmppc_xive_debug_show_queues(struct seq_file *m, struct kvm_vcpu *vcpu);
304 void kvmppc_xive_debug_show_sources(struct seq_file *m,
305 				    struct kvmppc_xive_src_block *sb);
306 struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
307 	struct kvmppc_xive *xive, int irq);
308 void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
309 int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio);
310 int kvmppc_xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio,
311 				  bool single_escalation);
312 struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type);
313 void xive_cleanup_single_escalation(struct kvm_vcpu *vcpu,
314 				    struct kvmppc_xive_vcpu *xc, int irq);
315 int kvmppc_xive_compute_vp_id(struct kvmppc_xive *xive, u32 cpu, u32 *vp);
316 int kvmppc_xive_set_nr_servers(struct kvmppc_xive *xive, u64 addr);
317 
318 #endif /* CONFIG_KVM_XICS */
319 #endif /* _KVM_PPC_BOOK3S_XICS_H */
320