1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling diagnose instructions
4  *
5  * Copyright IBM Corp. 2008, 2020
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  *               Christian Borntraeger <borntraeger@de.ibm.com>
9  */
10 
11 #include <linux/kvm.h>
12 #include <linux/kvm_host.h>
13 #include <asm/gmap.h>
14 #include <asm/virtio-ccw.h>
15 #include "kvm-s390.h"
16 #include "trace.h"
17 #include "trace-s390.h"
18 #include "gaccess.h"
19 
diag_release_pages(struct kvm_vcpu * vcpu)20 static int diag_release_pages(struct kvm_vcpu *vcpu)
21 {
22 	unsigned long start, end;
23 	unsigned long prefix  = kvm_s390_get_prefix(vcpu);
24 
25 	start = vcpu->run->s.regs.gprs[(vcpu->arch.sie_block->ipa & 0xf0) >> 4];
26 	end = vcpu->run->s.regs.gprs[vcpu->arch.sie_block->ipa & 0xf] + PAGE_SIZE;
27 	vcpu->stat.instruction_diagnose_10++;
28 
29 	if (start & ~PAGE_MASK || end & ~PAGE_MASK || start >= end
30 	    || start < 2 * PAGE_SIZE)
31 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
32 
33 	VCPU_EVENT(vcpu, 5, "diag release pages %lX %lX", start, end);
34 
35 	/*
36 	 * We checked for start >= end above, so lets check for the
37 	 * fast path (no prefix swap page involved)
38 	 */
39 	if (end <= prefix || start >= prefix + 2 * PAGE_SIZE) {
40 		gmap_discard(vcpu->arch.gmap, start, end);
41 	} else {
42 		/*
43 		 * This is slow path.  gmap_discard will check for start
44 		 * so lets split this into before prefix, prefix, after
45 		 * prefix and let gmap_discard make some of these calls
46 		 * NOPs.
47 		 */
48 		gmap_discard(vcpu->arch.gmap, start, prefix);
49 		if (start <= prefix)
50 			gmap_discard(vcpu->arch.gmap, 0, PAGE_SIZE);
51 		if (end > prefix + PAGE_SIZE)
52 			gmap_discard(vcpu->arch.gmap, PAGE_SIZE, 2 * PAGE_SIZE);
53 		gmap_discard(vcpu->arch.gmap, prefix + 2 * PAGE_SIZE, end);
54 	}
55 	return 0;
56 }
57 
__diag_page_ref_service(struct kvm_vcpu * vcpu)58 static int __diag_page_ref_service(struct kvm_vcpu *vcpu)
59 {
60 	struct prs_parm {
61 		u16 code;
62 		u16 subcode;
63 		u16 parm_len;
64 		u16 parm_version;
65 		u64 token_addr;
66 		u64 select_mask;
67 		u64 compare_mask;
68 		u64 zarch;
69 	};
70 	struct prs_parm parm;
71 	int rc;
72 	u16 rx = (vcpu->arch.sie_block->ipa & 0xf0) >> 4;
73 	u16 ry = (vcpu->arch.sie_block->ipa & 0x0f);
74 
75 	VCPU_EVENT(vcpu, 3, "diag page reference parameter block at 0x%llx",
76 		   vcpu->run->s.regs.gprs[rx]);
77 	vcpu->stat.instruction_diagnose_258++;
78 	if (vcpu->run->s.regs.gprs[rx] & 7)
79 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
80 	rc = read_guest(vcpu, vcpu->run->s.regs.gprs[rx], rx, &parm, sizeof(parm));
81 	if (rc)
82 		return kvm_s390_inject_prog_cond(vcpu, rc);
83 	if (parm.parm_version != 2 || parm.parm_len < 5 || parm.code != 0x258)
84 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
85 
86 	switch (parm.subcode) {
87 	case 0: /* TOKEN */
88 		VCPU_EVENT(vcpu, 3, "pageref token addr 0x%llx "
89 			   "select mask 0x%llx compare mask 0x%llx",
90 			   parm.token_addr, parm.select_mask, parm.compare_mask);
91 		if (vcpu->arch.pfault_token != KVM_S390_PFAULT_TOKEN_INVALID) {
92 			/*
93 			 * If the pagefault handshake is already activated,
94 			 * the token must not be changed.  We have to return
95 			 * decimal 8 instead, as mandated in SC24-6084.
96 			 */
97 			vcpu->run->s.regs.gprs[ry] = 8;
98 			return 0;
99 		}
100 
101 		if ((parm.compare_mask & parm.select_mask) != parm.compare_mask ||
102 		    parm.token_addr & 7 || parm.zarch != 0x8000000000000000ULL)
103 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
104 
105 		if (kvm_is_error_gpa(vcpu->kvm, parm.token_addr))
106 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
107 
108 		vcpu->arch.pfault_token = parm.token_addr;
109 		vcpu->arch.pfault_select = parm.select_mask;
110 		vcpu->arch.pfault_compare = parm.compare_mask;
111 		vcpu->run->s.regs.gprs[ry] = 0;
112 		rc = 0;
113 		break;
114 	case 1: /*
115 		 * CANCEL
116 		 * Specification allows to let already pending tokens survive
117 		 * the cancel, therefore to reduce code complexity, we assume
118 		 * all outstanding tokens are already pending.
119 		 */
120 		VCPU_EVENT(vcpu, 3, "pageref cancel addr 0x%llx", parm.token_addr);
121 		if (parm.token_addr || parm.select_mask ||
122 		    parm.compare_mask || parm.zarch)
123 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
124 
125 		vcpu->run->s.regs.gprs[ry] = 0;
126 		/*
127 		 * If the pfault handling was not established or is already
128 		 * canceled SC24-6084 requests to return decimal 4.
129 		 */
130 		if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
131 			vcpu->run->s.regs.gprs[ry] = 4;
132 		else
133 			vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
134 
135 		rc = 0;
136 		break;
137 	default:
138 		rc = -EOPNOTSUPP;
139 		break;
140 	}
141 
142 	return rc;
143 }
144 
__diag_time_slice_end(struct kvm_vcpu * vcpu)145 static int __diag_time_slice_end(struct kvm_vcpu *vcpu)
146 {
147 	VCPU_EVENT(vcpu, 5, "%s", "diag time slice end");
148 	vcpu->stat.instruction_diagnose_44++;
149 	kvm_vcpu_on_spin(vcpu, true);
150 	return 0;
151 }
152 
153 static int forward_cnt;
154 static unsigned long cur_slice;
155 
diag9c_forwarding_overrun(void)156 static int diag9c_forwarding_overrun(void)
157 {
158 	/* Reset the count on a new slice */
159 	if (time_after(jiffies, cur_slice)) {
160 		cur_slice = jiffies;
161 		forward_cnt = diag9c_forwarding_hz / HZ;
162 	}
163 	return forward_cnt-- <= 0 ? 1 : 0;
164 }
165 
__diag_time_slice_end_directed(struct kvm_vcpu * vcpu)166 static int __diag_time_slice_end_directed(struct kvm_vcpu *vcpu)
167 {
168 	struct kvm_vcpu *tcpu;
169 	int tid;
170 
171 	tid = vcpu->run->s.regs.gprs[(vcpu->arch.sie_block->ipa & 0xf0) >> 4];
172 	vcpu->stat.instruction_diagnose_9c++;
173 
174 	/* yield to self */
175 	if (tid == vcpu->vcpu_id)
176 		goto no_yield;
177 
178 	/* yield to invalid */
179 	tcpu = kvm_get_vcpu_by_id(vcpu->kvm, tid);
180 	if (!tcpu)
181 		goto no_yield;
182 
183 	/* target guest VCPU already running */
184 	if (READ_ONCE(tcpu->cpu) >= 0) {
185 		if (!diag9c_forwarding_hz || diag9c_forwarding_overrun())
186 			goto no_yield;
187 
188 		/* target host CPU already running */
189 		if (!vcpu_is_preempted(tcpu->cpu))
190 			goto no_yield;
191 		smp_yield_cpu(tcpu->cpu);
192 		VCPU_EVENT(vcpu, 5,
193 			   "diag time slice end directed to %d: yield forwarded",
194 			   tid);
195 		vcpu->stat.diag_9c_forward++;
196 		return 0;
197 	}
198 
199 	if (kvm_vcpu_yield_to(tcpu) <= 0)
200 		goto no_yield;
201 
202 	VCPU_EVENT(vcpu, 5, "diag time slice end directed to %d: done", tid);
203 	return 0;
204 no_yield:
205 	VCPU_EVENT(vcpu, 5, "diag time slice end directed to %d: ignored", tid);
206 	vcpu->stat.diag_9c_ignored++;
207 	return 0;
208 }
209 
__diag_ipl_functions(struct kvm_vcpu * vcpu)210 static int __diag_ipl_functions(struct kvm_vcpu *vcpu)
211 {
212 	unsigned int reg = vcpu->arch.sie_block->ipa & 0xf;
213 	unsigned long subcode = vcpu->run->s.regs.gprs[reg] & 0xffff;
214 
215 	VCPU_EVENT(vcpu, 3, "diag ipl functions, subcode %lx", subcode);
216 	vcpu->stat.instruction_diagnose_308++;
217 	switch (subcode) {
218 	case 3:
219 		vcpu->run->s390_reset_flags = KVM_S390_RESET_CLEAR;
220 		break;
221 	case 4:
222 		vcpu->run->s390_reset_flags = 0;
223 		break;
224 	default:
225 		return -EOPNOTSUPP;
226 	}
227 
228 	/*
229 	 * no need to check the return value of vcpu_stop as it can only have
230 	 * an error for protvirt, but protvirt means user cpu state
231 	 */
232 	if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
233 		kvm_s390_vcpu_stop(vcpu);
234 	vcpu->run->s390_reset_flags |= KVM_S390_RESET_SUBSYSTEM;
235 	vcpu->run->s390_reset_flags |= KVM_S390_RESET_IPL;
236 	vcpu->run->s390_reset_flags |= KVM_S390_RESET_CPU_INIT;
237 	vcpu->run->exit_reason = KVM_EXIT_S390_RESET;
238 	VCPU_EVENT(vcpu, 3, "requesting userspace resets %llx",
239 	  vcpu->run->s390_reset_flags);
240 	trace_kvm_s390_request_resets(vcpu->run->s390_reset_flags);
241 	return -EREMOTE;
242 }
243 
__diag_virtio_hypercall(struct kvm_vcpu * vcpu)244 static int __diag_virtio_hypercall(struct kvm_vcpu *vcpu)
245 {
246 	int ret;
247 
248 	vcpu->stat.instruction_diagnose_500++;
249 	/* No virtio-ccw notification? Get out quickly. */
250 	if (!vcpu->kvm->arch.css_support ||
251 	    (vcpu->run->s.regs.gprs[1] != KVM_S390_VIRTIO_CCW_NOTIFY))
252 		return -EOPNOTSUPP;
253 
254 	VCPU_EVENT(vcpu, 4, "diag 0x500 schid 0x%8.8x queue 0x%x cookie 0x%llx",
255 			    (u32) vcpu->run->s.regs.gprs[2],
256 			    (u32) vcpu->run->s.regs.gprs[3],
257 			    vcpu->run->s.regs.gprs[4]);
258 
259 	/*
260 	 * The layout is as follows:
261 	 * - gpr 2 contains the subchannel id (passed as addr)
262 	 * - gpr 3 contains the virtqueue index (passed as datamatch)
263 	 * - gpr 4 contains the index on the bus (optionally)
264 	 */
265 	ret = kvm_io_bus_write_cookie(vcpu, KVM_VIRTIO_CCW_NOTIFY_BUS,
266 				      vcpu->run->s.regs.gprs[2] & 0xffffffff,
267 				      8, &vcpu->run->s.regs.gprs[3],
268 				      vcpu->run->s.regs.gprs[4]);
269 
270 	/*
271 	 * Return cookie in gpr 2, but don't overwrite the register if the
272 	 * diagnose will be handled by userspace.
273 	 */
274 	if (ret != -EOPNOTSUPP)
275 		vcpu->run->s.regs.gprs[2] = ret;
276 	/* kvm_io_bus_write_cookie returns -EOPNOTSUPP if it found no match. */
277 	return ret < 0 ? ret : 0;
278 }
279 
kvm_s390_handle_diag(struct kvm_vcpu * vcpu)280 int kvm_s390_handle_diag(struct kvm_vcpu *vcpu)
281 {
282 	int code = kvm_s390_get_base_disp_rs(vcpu, NULL) & 0xffff;
283 
284 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
285 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
286 
287 	trace_kvm_s390_handle_diag(vcpu, code);
288 	switch (code) {
289 	case 0x10:
290 		return diag_release_pages(vcpu);
291 	case 0x44:
292 		return __diag_time_slice_end(vcpu);
293 	case 0x9c:
294 		return __diag_time_slice_end_directed(vcpu);
295 	case 0x258:
296 		return __diag_page_ref_service(vcpu);
297 	case 0x308:
298 		return __diag_ipl_functions(vcpu);
299 	case 0x500:
300 		return __diag_virtio_hypercall(vcpu);
301 	default:
302 		vcpu->stat.instruction_diagnose_other++;
303 		return -EOPNOTSUPP;
304 	}
305 }
306