| 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/gmap_helpers.h> |
| 15 | #include <asm/virtio-ccw.h> |
| 16 | #include "kvm-s390.h" |
| 17 | #include "trace.h" |
| 18 | #include "trace-s390.h" |
| 19 | #include "gaccess.h" |
| 20 | |
| 21 | static void do_discard_gfn_range(struct kvm_vcpu *vcpu, gfn_t gfn_start, gfn_t gfn_end) |
| 22 | { |
| 23 | struct kvm_memslot_iter iter; |
| 24 | struct kvm_memory_slot *slot; |
| 25 | struct kvm_memslots *slots; |
| 26 | unsigned long start, end; |
| 27 | |
| 28 | slots = kvm_vcpu_memslots(vcpu); |
| 29 | |
| 30 | kvm_for_each_memslot_in_gfn_range(&iter, slots, gfn_start, gfn_end) { |
| 31 | slot = iter.slot; |
| 32 | start = __gfn_to_hva_memslot(slot, max(gfn_start, slot->base_gfn)); |
| 33 | end = __gfn_to_hva_memslot(slot, min(gfn_end, slot->base_gfn + slot->npages)); |
| 34 | gmap_helper_discard(vcpu->kvm->mm, start, end); |
| 35 | } |
| 36 | } |
| 37 | |
| 38 | static int diag_release_pages(struct kvm_vcpu *vcpu) |
| 39 | { |
| 40 | unsigned long start, end; |
| 41 | unsigned long prefix = kvm_s390_get_prefix(vcpu); |
| 42 | |
| 43 | start = vcpu->run->s.regs.gprs[(vcpu->arch.sie_block->ipa & 0xf0) >> 4]; |
| 44 | end = vcpu->run->s.regs.gprs[vcpu->arch.sie_block->ipa & 0xf] + PAGE_SIZE; |
| 45 | vcpu->stat.instruction_diagnose_10++; |
| 46 | |
| 47 | if (start & ~PAGE_MASK || end & ~PAGE_MASK || start >= end |
| 48 | || start < 2 * PAGE_SIZE) |
| 49 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 50 | |
| 51 | VCPU_EVENT(vcpu, 5, "diag release pages %lX %lX" , start, end); |
| 52 | |
| 53 | mmap_read_lock(mm: vcpu->kvm->mm); |
| 54 | /* |
| 55 | * We checked for start >= end above, so lets check for the |
| 56 | * fast path (no prefix swap page involved) |
| 57 | */ |
| 58 | if (end <= prefix || start >= prefix + 2 * PAGE_SIZE) { |
| 59 | do_discard_gfn_range(vcpu, gfn_start: gpa_to_gfn(gpa: start), gfn_end: gpa_to_gfn(gpa: end)); |
| 60 | } else { |
| 61 | /* |
| 62 | * This is slow path. gmap_discard will check for start |
| 63 | * so lets split this into before prefix, prefix, after |
| 64 | * prefix and let gmap_discard make some of these calls |
| 65 | * NOPs. |
| 66 | */ |
| 67 | do_discard_gfn_range(vcpu, gfn_start: gpa_to_gfn(gpa: start), gfn_end: gpa_to_gfn(gpa: prefix)); |
| 68 | if (start <= prefix) |
| 69 | do_discard_gfn_range(vcpu, gfn_start: 0, gfn_end: 1); |
| 70 | if (end > prefix + PAGE_SIZE) |
| 71 | do_discard_gfn_range(vcpu, gfn_start: 1, gfn_end: 2); |
| 72 | do_discard_gfn_range(vcpu, gfn_start: gpa_to_gfn(gpa: prefix) + 2, gfn_end: gpa_to_gfn(gpa: end)); |
| 73 | } |
| 74 | mmap_read_unlock(mm: vcpu->kvm->mm); |
| 75 | return 0; |
| 76 | } |
| 77 | |
| 78 | static int __diag_page_ref_service(struct kvm_vcpu *vcpu) |
| 79 | { |
| 80 | struct prs_parm { |
| 81 | u16 code; |
| 82 | u16 subcode; |
| 83 | u16 parm_len; |
| 84 | u16 parm_version; |
| 85 | u64 token_addr; |
| 86 | u64 select_mask; |
| 87 | u64 compare_mask; |
| 88 | u64 zarch; |
| 89 | }; |
| 90 | struct prs_parm parm; |
| 91 | int rc; |
| 92 | u16 rx = (vcpu->arch.sie_block->ipa & 0xf0) >> 4; |
| 93 | u16 ry = (vcpu->arch.sie_block->ipa & 0x0f); |
| 94 | |
| 95 | VCPU_EVENT(vcpu, 3, "diag page reference parameter block at 0x%llx" , |
| 96 | vcpu->run->s.regs.gprs[rx]); |
| 97 | vcpu->stat.instruction_diagnose_258++; |
| 98 | if (vcpu->run->s.regs.gprs[rx] & 7) |
| 99 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 100 | rc = read_guest_real(vcpu, gra: vcpu->run->s.regs.gprs[rx], data: &parm, len: sizeof(parm)); |
| 101 | if (rc) |
| 102 | return kvm_s390_inject_prog_cond(vcpu, rc); |
| 103 | if (parm.parm_version != 2 || parm.parm_len < 5 || parm.code != 0x258) |
| 104 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 105 | |
| 106 | switch (parm.subcode) { |
| 107 | case 0: /* TOKEN */ |
| 108 | VCPU_EVENT(vcpu, 3, "pageref token addr 0x%llx " |
| 109 | "select mask 0x%llx compare mask 0x%llx" , |
| 110 | parm.token_addr, parm.select_mask, parm.compare_mask); |
| 111 | if (vcpu->arch.pfault_token != KVM_S390_PFAULT_TOKEN_INVALID) { |
| 112 | /* |
| 113 | * If the pagefault handshake is already activated, |
| 114 | * the token must not be changed. We have to return |
| 115 | * decimal 8 instead, as mandated in SC24-6084. |
| 116 | */ |
| 117 | vcpu->run->s.regs.gprs[ry] = 8; |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | if ((parm.compare_mask & parm.select_mask) != parm.compare_mask || |
| 122 | parm.token_addr & 7 || parm.zarch != 0x8000000000000000ULL) |
| 123 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 124 | |
| 125 | if (!kvm_is_gpa_in_memslot(vcpu->kvm, parm.token_addr)) |
| 126 | return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); |
| 127 | |
| 128 | vcpu->arch.pfault_token = parm.token_addr; |
| 129 | vcpu->arch.pfault_select = parm.select_mask; |
| 130 | vcpu->arch.pfault_compare = parm.compare_mask; |
| 131 | vcpu->run->s.regs.gprs[ry] = 0; |
| 132 | rc = 0; |
| 133 | break; |
| 134 | case 1: /* |
| 135 | * CANCEL |
| 136 | * Specification allows to let already pending tokens survive |
| 137 | * the cancel, therefore to reduce code complexity, we assume |
| 138 | * all outstanding tokens are already pending. |
| 139 | */ |
| 140 | VCPU_EVENT(vcpu, 3, "pageref cancel addr 0x%llx" , parm.token_addr); |
| 141 | if (parm.token_addr || parm.select_mask || |
| 142 | parm.compare_mask || parm.zarch) |
| 143 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 144 | |
| 145 | vcpu->run->s.regs.gprs[ry] = 0; |
| 146 | /* |
| 147 | * If the pfault handling was not established or is already |
| 148 | * canceled SC24-6084 requests to return decimal 4. |
| 149 | */ |
| 150 | if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) |
| 151 | vcpu->run->s.regs.gprs[ry] = 4; |
| 152 | else |
| 153 | vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; |
| 154 | |
| 155 | rc = 0; |
| 156 | break; |
| 157 | default: |
| 158 | rc = -EOPNOTSUPP; |
| 159 | break; |
| 160 | } |
| 161 | |
| 162 | return rc; |
| 163 | } |
| 164 | |
| 165 | static int __diag_time_slice_end(struct kvm_vcpu *vcpu) |
| 166 | { |
| 167 | VCPU_EVENT(vcpu, 5, "%s" , "diag time slice end" ); |
| 168 | vcpu->stat.instruction_diagnose_44++; |
| 169 | kvm_vcpu_on_spin(vcpu, yield_to_kernel_mode: true); |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | static int forward_cnt; |
| 174 | static unsigned long cur_slice; |
| 175 | |
| 176 | static int diag9c_forwarding_overrun(void) |
| 177 | { |
| 178 | /* Reset the count on a new slice */ |
| 179 | if (time_after(jiffies, cur_slice)) { |
| 180 | cur_slice = jiffies; |
| 181 | forward_cnt = diag9c_forwarding_hz / HZ; |
| 182 | } |
| 183 | return forward_cnt-- <= 0 ? 1 : 0; |
| 184 | } |
| 185 | |
| 186 | static int __diag_time_slice_end_directed(struct kvm_vcpu *vcpu) |
| 187 | { |
| 188 | struct kvm_vcpu *tcpu; |
| 189 | int tcpu_cpu; |
| 190 | int tid; |
| 191 | |
| 192 | tid = vcpu->run->s.regs.gprs[(vcpu->arch.sie_block->ipa & 0xf0) >> 4]; |
| 193 | vcpu->stat.instruction_diagnose_9c++; |
| 194 | |
| 195 | /* yield to self */ |
| 196 | if (tid == vcpu->vcpu_id) |
| 197 | goto no_yield; |
| 198 | |
| 199 | /* yield to invalid */ |
| 200 | tcpu = kvm_get_vcpu_by_id(kvm: vcpu->kvm, id: tid); |
| 201 | if (!tcpu) |
| 202 | goto no_yield; |
| 203 | |
| 204 | /* target guest VCPU already running */ |
| 205 | tcpu_cpu = READ_ONCE(tcpu->cpu); |
| 206 | if (tcpu_cpu >= 0) { |
| 207 | if (!diag9c_forwarding_hz || diag9c_forwarding_overrun()) |
| 208 | goto no_yield; |
| 209 | |
| 210 | /* target host CPU already running */ |
| 211 | if (!vcpu_is_preempted(cpu: tcpu_cpu)) |
| 212 | goto no_yield; |
| 213 | smp_yield_cpu(tcpu_cpu); |
| 214 | VCPU_EVENT(vcpu, 5, |
| 215 | "diag time slice end directed to %d: yield forwarded" , |
| 216 | tid); |
| 217 | vcpu->stat.diag_9c_forward++; |
| 218 | return 0; |
| 219 | } |
| 220 | |
| 221 | if (kvm_vcpu_yield_to(target: tcpu) <= 0) |
| 222 | goto no_yield; |
| 223 | |
| 224 | VCPU_EVENT(vcpu, 5, "diag time slice end directed to %d: done" , tid); |
| 225 | return 0; |
| 226 | no_yield: |
| 227 | VCPU_EVENT(vcpu, 5, "diag time slice end directed to %d: ignored" , tid); |
| 228 | vcpu->stat.diag_9c_ignored++; |
| 229 | return 0; |
| 230 | } |
| 231 | |
| 232 | static int __diag_ipl_functions(struct kvm_vcpu *vcpu) |
| 233 | { |
| 234 | unsigned int reg = vcpu->arch.sie_block->ipa & 0xf; |
| 235 | unsigned long subcode = vcpu->run->s.regs.gprs[reg] & 0xffff; |
| 236 | |
| 237 | VCPU_EVENT(vcpu, 3, "diag ipl functions, subcode %lx" , subcode); |
| 238 | vcpu->stat.instruction_diagnose_308++; |
| 239 | switch (subcode) { |
| 240 | case 3: |
| 241 | vcpu->run->s390_reset_flags = KVM_S390_RESET_CLEAR; |
| 242 | break; |
| 243 | case 4: |
| 244 | vcpu->run->s390_reset_flags = 0; |
| 245 | break; |
| 246 | default: |
| 247 | return -EOPNOTSUPP; |
| 248 | } |
| 249 | |
| 250 | /* |
| 251 | * no need to check the return value of vcpu_stop as it can only have |
| 252 | * an error for protvirt, but protvirt means user cpu state |
| 253 | */ |
| 254 | if (!kvm_s390_user_cpu_state_ctrl(kvm: vcpu->kvm)) |
| 255 | kvm_s390_vcpu_stop(vcpu); |
| 256 | vcpu->run->s390_reset_flags |= KVM_S390_RESET_SUBSYSTEM; |
| 257 | vcpu->run->s390_reset_flags |= KVM_S390_RESET_IPL; |
| 258 | vcpu->run->s390_reset_flags |= KVM_S390_RESET_CPU_INIT; |
| 259 | vcpu->run->exit_reason = KVM_EXIT_S390_RESET; |
| 260 | VCPU_EVENT(vcpu, 3, "requesting userspace resets %llx" , |
| 261 | vcpu->run->s390_reset_flags); |
| 262 | trace_kvm_s390_request_resets(resets: vcpu->run->s390_reset_flags); |
| 263 | return -EREMOTE; |
| 264 | } |
| 265 | |
| 266 | static int __diag_virtio_hypercall(struct kvm_vcpu *vcpu) |
| 267 | { |
| 268 | int ret; |
| 269 | |
| 270 | vcpu->stat.instruction_diagnose_500++; |
| 271 | /* No virtio-ccw notification? Get out quickly. */ |
| 272 | if (!vcpu->kvm->arch.css_support || |
| 273 | (vcpu->run->s.regs.gprs[1] != KVM_S390_VIRTIO_CCW_NOTIFY)) |
| 274 | return -EOPNOTSUPP; |
| 275 | |
| 276 | VCPU_EVENT(vcpu, 4, "diag 0x500 schid 0x%8.8x queue 0x%x cookie 0x%llx" , |
| 277 | (u32) vcpu->run->s.regs.gprs[2], |
| 278 | (u32) vcpu->run->s.regs.gprs[3], |
| 279 | vcpu->run->s.regs.gprs[4]); |
| 280 | |
| 281 | /* |
| 282 | * The layout is as follows: |
| 283 | * - gpr 2 contains the subchannel id (passed as addr) |
| 284 | * - gpr 3 contains the virtqueue index (passed as datamatch) |
| 285 | * - gpr 4 contains the index on the bus (optionally) |
| 286 | */ |
| 287 | ret = kvm_io_bus_write_cookie(vcpu, bus_idx: KVM_VIRTIO_CCW_NOTIFY_BUS, |
| 288 | addr: vcpu->run->s.regs.gprs[2] & 0xffffffff, |
| 289 | len: 8, val: &vcpu->run->s.regs.gprs[3], |
| 290 | cookie: vcpu->run->s.regs.gprs[4]); |
| 291 | |
| 292 | /* |
| 293 | * Return cookie in gpr 2, but don't overwrite the register if the |
| 294 | * diagnose will be handled by userspace. |
| 295 | */ |
| 296 | if (ret != -EOPNOTSUPP) |
| 297 | vcpu->run->s.regs.gprs[2] = ret; |
| 298 | /* kvm_io_bus_write_cookie returns -EOPNOTSUPP if it found no match. */ |
| 299 | return ret < 0 ? ret : 0; |
| 300 | } |
| 301 | |
| 302 | int kvm_s390_handle_diag(struct kvm_vcpu *vcpu) |
| 303 | { |
| 304 | int code = kvm_s390_get_base_disp_rs(vcpu, NULL) & 0xffff; |
| 305 | |
| 306 | if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) |
| 307 | return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); |
| 308 | |
| 309 | trace_kvm_s390_handle_diag(vcpu, code); |
| 310 | switch (code) { |
| 311 | case 0x10: |
| 312 | return diag_release_pages(vcpu); |
| 313 | case 0x44: |
| 314 | return __diag_time_slice_end(vcpu); |
| 315 | case 0x9c: |
| 316 | return __diag_time_slice_end_directed(vcpu); |
| 317 | case 0x258: |
| 318 | return __diag_page_ref_service(vcpu); |
| 319 | case 0x308: |
| 320 | return __diag_ipl_functions(vcpu); |
| 321 | case 0x500: |
| 322 | return __diag_virtio_hypercall(vcpu); |
| 323 | default: |
| 324 | vcpu->stat.instruction_diagnose_other++; |
| 325 | return -EOPNOTSUPP; |
| 326 | } |
| 327 | } |
| 328 | |