1 // SPDX-License-Identifier: GPL-2.0
3 * store hypervisor information instruction emulation functions.
5 * Copyright IBM Corp. 2016
6 * Author(s): Janosch Frank <frankja@linux.vnet.ibm.com>
8 #include <linux/errno.h>
9 #include <linux/pagemap.h>
10 #include <linux/vmalloc.h>
11 #include <linux/syscalls.h>
12 #include <linux/mutex.h>
13 #include <asm/asm-offsets.h>
16 #include <asm/sysinfo.h>
17 #include <asm/ebcdic.h>
18 #include <asm/facility.h>
19 #include <asm/sthyi.h>
22 #define DED_WEIGHT 0xffff
24 * CP and IFL as EBCDIC strings, SP/0x40 determines the end of string
25 * as they are justified with spaces.
27 #define CP 0xc3d7404040404040UL
28 #define IFL 0xc9c6d34040404040UL
32 HDR_STACK_INCM = 0x20,
57 u8 infhflg2; /* reserved */
58 u8 infhval1; /* reserved */
59 u8 infhval2; /* reserved */
84 u8 infmflg1; /* reserved */
85 u8 infmflg2; /* reserved */
87 u8 infmval2; /* reserved */
102 u8 infpflg2; /* reserved */
104 u8 infpval2; /* reserved */
136 struct lpar_cpu_inf {
142 * STHYI requires extensive locking in the higher hypervisors
143 * and is very computational/memory expensive. Therefore we
144 * cache the retrieved data whose valid period is 1s.
146 #define CACHE_VALID_JIFFIES HZ
153 static DEFINE_MUTEX(sthyi_mutex);
154 static struct sthyi_info sthyi_cache;
156 static inline u64 cpu_id(u8 ctidx, void *diag224_buf)
158 return *((u64 *)(diag224_buf + (ctidx + 1) * DIAG204_CPU_NAME_LEN));
162 * Scales the cpu capping from the lpar range to the one expected in
165 * diag204 reports a cap in hundredths of processor units.
166 * z/VM's range for one core is 0 - 0x10000.
168 static u32 scale_cap(u32 in)
170 return (0x10000 * in) / 100;
173 static void fill_hdr(struct sthyi_sctns *sctns)
175 sctns->hdr.infhdln = sizeof(sctns->hdr);
176 sctns->hdr.infmoff = sizeof(sctns->hdr);
177 sctns->hdr.infmlen = sizeof(sctns->mac);
178 sctns->hdr.infplen = sizeof(sctns->par);
179 sctns->hdr.infpoff = sctns->hdr.infhdln + sctns->hdr.infmlen;
180 sctns->hdr.infhtotl = sctns->hdr.infpoff + sctns->hdr.infplen;
183 static void fill_stsi_mac(struct sthyi_sctns *sctns,
184 struct sysinfo_1_1_1 *sysinfo)
186 sclp_ocf_cpc_name_copy(sctns->mac.infmname);
187 if (*(u64 *)sctns->mac.infmname != 0)
188 sctns->mac.infmval1 |= MAC_NAME_VLD;
190 if (stsi(sysinfo, 1, 1, 1))
193 memcpy(sctns->mac.infmtype, sysinfo->type, sizeof(sctns->mac.infmtype));
194 memcpy(sctns->mac.infmmanu, sysinfo->manufacturer, sizeof(sctns->mac.infmmanu));
195 memcpy(sctns->mac.infmpman, sysinfo->plant, sizeof(sctns->mac.infmpman));
196 memcpy(sctns->mac.infmseq, sysinfo->sequence, sizeof(sctns->mac.infmseq));
198 sctns->mac.infmval1 |= MAC_ID_VLD;
201 static void fill_stsi_par(struct sthyi_sctns *sctns,
202 struct sysinfo_2_2_2 *sysinfo)
204 if (stsi(sysinfo, 2, 2, 2))
207 sctns->par.infppnum = sysinfo->lpar_number;
208 memcpy(sctns->par.infppnam, sysinfo->name, sizeof(sctns->par.infppnam));
210 sctns->par.infpval1 |= PAR_ID_VLD;
213 static void fill_stsi(struct sthyi_sctns *sctns)
217 /* Errors are handled through the validity bits in the response. */
218 sysinfo = (void *)__get_free_page(GFP_KERNEL);
222 fill_stsi_mac(sctns, sysinfo);
223 fill_stsi_par(sctns, sysinfo);
225 free_pages((unsigned long)sysinfo, 0);
228 static void fill_diag_mac(struct sthyi_sctns *sctns,
229 struct diag204_x_phys_block *block,
234 for (i = 0; i < block->hdr.cpus; i++) {
235 switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
237 if (block->cpus[i].weight == DED_WEIGHT)
238 sctns->mac.infmdcps++;
240 sctns->mac.infmscps++;
243 if (block->cpus[i].weight == DED_WEIGHT)
244 sctns->mac.infmdifl++;
246 sctns->mac.infmsifl++;
250 sctns->mac.infmval1 |= MAC_CNT_VLD;
253 /* Returns a pointer to the the next partition block. */
254 static struct diag204_x_part_block *lpar_cpu_inf(struct lpar_cpu_inf *part_inf,
257 struct diag204_x_part_block *block)
259 int i, capped = 0, weight_cp = 0, weight_ifl = 0;
260 struct cpu_inf *cpu_inf;
262 for (i = 0; i < block->hdr.rcpus; i++) {
263 if (!(block->cpus[i].cflag & DIAG204_CPU_ONLINE))
266 switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
268 cpu_inf = &part_inf->cp;
269 if (block->cpus[i].cur_weight < DED_WEIGHT)
270 weight_cp |= block->cpus[i].cur_weight;
273 cpu_inf = &part_inf->ifl;
274 if (block->cpus[i].cur_weight < DED_WEIGHT)
275 weight_ifl |= block->cpus[i].cur_weight;
284 capped |= block->cpus[i].cflag & DIAG204_CPU_CAPPED;
285 cpu_inf->lpar_cap |= block->cpus[i].cpu_type_cap;
286 cpu_inf->lpar_grp_cap |= block->cpus[i].group_cpu_type_cap;
288 if (block->cpus[i].weight == DED_WEIGHT)
289 cpu_inf->cpu_num_ded += 1;
291 cpu_inf->cpu_num_shd += 1;
294 if (this_lpar && capped) {
295 part_inf->cp.lpar_weight = weight_cp;
296 part_inf->ifl.lpar_weight = weight_ifl;
298 part_inf->cp.all_weight += weight_cp;
299 part_inf->ifl.all_weight += weight_ifl;
300 return (struct diag204_x_part_block *)&block->cpus[i];
303 static void fill_diag(struct sthyi_sctns *sctns)
308 void *diag224_buf = NULL;
309 struct diag204_x_info_blk_hdr *ti_hdr;
310 struct diag204_x_part_block *part_block;
311 struct diag204_x_phys_block *phys_block;
312 struct lpar_cpu_inf lpar_inf = {};
314 /* Errors are handled through the validity bits in the response. */
315 pages = diag204((unsigned long)DIAG204_SUBC_RSI |
316 (unsigned long)DIAG204_INFO_EXT, 0, NULL);
320 diag204_buf = __vmalloc_node(array_size(pages, PAGE_SIZE),
321 PAGE_SIZE, GFP_KERNEL, NUMA_NO_NODE,
322 __builtin_return_address(0));
326 r = diag204((unsigned long)DIAG204_SUBC_STIB7 |
327 (unsigned long)DIAG204_INFO_EXT, pages, diag204_buf);
331 diag224_buf = (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
332 if (!diag224_buf || diag224(diag224_buf))
335 ti_hdr = diag204_buf;
336 part_block = diag204_buf + sizeof(*ti_hdr);
338 for (i = 0; i < ti_hdr->npar; i++) {
340 * For the calling lpar we also need to get the cpu
341 * caps and weights. The time information block header
342 * specifies the offset to the partition block of the
343 * caller lpar, so we know when we process its data.
345 this_lpar = (void *)part_block - diag204_buf == ti_hdr->this_part;
346 part_block = lpar_cpu_inf(&lpar_inf, this_lpar, diag224_buf,
350 phys_block = (struct diag204_x_phys_block *)part_block;
351 part_block = diag204_buf + ti_hdr->this_part;
352 if (part_block->hdr.mtid)
353 sctns->par.infpflg1 = PAR_MT_EN;
355 sctns->par.infpval1 |= PAR_GRP_VLD;
356 sctns->par.infplgcp = scale_cap(lpar_inf.cp.lpar_grp_cap);
357 sctns->par.infplgif = scale_cap(lpar_inf.ifl.lpar_grp_cap);
358 memcpy(sctns->par.infplgnm, part_block->hdr.hardware_group_name,
359 sizeof(sctns->par.infplgnm));
361 sctns->par.infpscps = lpar_inf.cp.cpu_num_shd;
362 sctns->par.infpdcps = lpar_inf.cp.cpu_num_ded;
363 sctns->par.infpsifl = lpar_inf.ifl.cpu_num_shd;
364 sctns->par.infpdifl = lpar_inf.ifl.cpu_num_ded;
365 sctns->par.infpval1 |= PAR_PCNT_VLD;
367 sctns->par.infpabcp = scale_cap(lpar_inf.cp.lpar_cap);
368 sctns->par.infpabif = scale_cap(lpar_inf.ifl.lpar_cap);
369 sctns->par.infpval1 |= PAR_ABS_VLD;
372 * Everything below needs global performance data to be
375 if (!(ti_hdr->flags & DIAG204_LPAR_PHYS_FLG)) {
376 sctns->hdr.infhflg1 |= HDR_PERF_UNAV;
380 fill_diag_mac(sctns, phys_block, diag224_buf);
382 if (lpar_inf.cp.lpar_weight) {
383 sctns->par.infpwbcp = sctns->mac.infmscps * 0x10000 *
384 lpar_inf.cp.lpar_weight / lpar_inf.cp.all_weight;
387 if (lpar_inf.ifl.lpar_weight) {
388 sctns->par.infpwbif = sctns->mac.infmsifl * 0x10000 *
389 lpar_inf.ifl.lpar_weight / lpar_inf.ifl.all_weight;
391 sctns->par.infpval1 |= PAR_WGHT_VLD;
394 free_page((unsigned long)diag224_buf);
398 static int sthyi(u64 vaddr, u64 *rc)
400 union register_pair r1 = { .even = 0, }; /* subcode */
401 union register_pair r2 = { .even = vaddr, };
405 ".insn rre,0xB2560000,%[r1],%[r2]\n"
408 : [cc] "=&d" (cc), [r2] "+&d" (r2.pair)
415 static int fill_dst(void *dst, u64 *rc)
417 struct sthyi_sctns *sctns = (struct sthyi_sctns *)dst;
420 * If the facility is on, we don't want to emulate the instruction.
421 * We ask the hypervisor to provide the data.
423 if (test_facility(74))
424 return sthyi((u64)dst, rc);
433 static int sthyi_init_cache(void)
435 if (sthyi_cache.info)
437 sthyi_cache.info = (void *)get_zeroed_page(GFP_KERNEL);
438 if (!sthyi_cache.info)
440 sthyi_cache.end = jiffies - 1; /* expired */
444 static int sthyi_update_cache(u64 *rc)
448 memset(sthyi_cache.info, 0, PAGE_SIZE);
449 r = fill_dst(sthyi_cache.info, rc);
452 sthyi_cache.end = jiffies + CACHE_VALID_JIFFIES;
457 * sthyi_fill - Fill page with data returned by the STHYI instruction
459 * @dst: Pointer to zeroed page
460 * @rc: Pointer for storing the return code of the instruction
462 * Fills the destination with system information returned by the STHYI
463 * instruction. The data is generated by emulation or execution of STHYI,
464 * if available. The return value is either a negative error value or
465 * the condition code that would be returned, the rc parameter is the
466 * return code which is passed in register R2 + 1.
468 int sthyi_fill(void *dst, u64 *rc)
472 mutex_lock(&sthyi_mutex);
473 r = sthyi_init_cache();
477 if (time_is_before_jiffies(sthyi_cache.end)) {
479 r = sthyi_update_cache(rc);
484 memcpy(dst, sthyi_cache.info, PAGE_SIZE);
486 mutex_unlock(&sthyi_mutex);
489 EXPORT_SYMBOL_GPL(sthyi_fill);
491 SYSCALL_DEFINE4(s390_sthyi, unsigned long, function_code, void __user *, buffer,
492 u64 __user *, return_code, unsigned long, flags)
500 if (function_code != STHYI_FC_CP_IFL_CAP)
502 info = (void *)get_zeroed_page(GFP_KERNEL);
505 r = sthyi_fill(info, &sthyi_rc);
508 if (return_code && put_user(sthyi_rc, return_code)) {
512 if (copy_to_user(buffer, info, PAGE_SIZE))
515 free_page((unsigned long)info);