[linux-2.6-block.git] / arch / s390 / kernel / sthyi.c

// SPDX-License-Identifier: GPL-2.0
/*
 * store hypervisor information instruction emulation functions.
 *
 * Copyright IBM Corp. 2016
 * Author(s): Janosch Frank <frankja@linux.vnet.ibm.com>
 */
#include <linux/errno.h>
#include <linux/pagemap.h>
#include <linux/vmalloc.h>
#include <linux/syscalls.h>
#include <linux/mutex.h>
#include <asm/asm-offsets.h>
#include <asm/sclp.h>
#include <asm/diag.h>
#include <asm/sysinfo.h>
#include <asm/ebcdic.h>
#include <asm/facility.h>
#include <asm/sthyi.h>
#include "entry.h"

#define DED_WEIGHT 0xffff
/*
 * CP and IFL as EBCDIC strings, SP/0x40 determines the end of string
 * as they are justified with spaces.
 */
#define CP  0xc3d7404040404040UL
#define IFL 0xc9c6d34040404040UL

enum hdr_flags {
	HDR_NOT_LPAR   = 0x10,
	HDR_STACK_INCM = 0x20,
	HDR_STSI_UNAV  = 0x40,
	HDR_PERF_UNAV  = 0x80,
};

enum mac_validity {
	MAC_NAME_VLD = 0x20,
	MAC_ID_VLD   = 0x40,
	MAC_CNT_VLD  = 0x80,
};

enum par_flag {
	PAR_MT_EN = 0x80,
};

enum par_validity {
	PAR_GRP_VLD  = 0x08,
	PAR_ID_VLD   = 0x10,
	PAR_ABS_VLD  = 0x20,
	PAR_WGHT_VLD = 0x40,
	PAR_PCNT_VLD  = 0x80,
};

struct hdr_sctn {
	u8 infhflg1;
	u8 infhflg2; /* reserved */
	u8 infhval1; /* reserved */
	u8 infhval2; /* reserved */
	u8 reserved[3];
	u8 infhygct;
	u16 infhtotl;
	u16 infhdln;
	u16 infmoff;
	u16 infmlen;
	u16 infpoff;
	u16 infplen;
	u16 infhoff1;
	u16 infhlen1;
	u16 infgoff1;
	u16 infglen1;
	u16 infhoff2;
	u16 infhlen2;
	u16 infgoff2;
	u16 infglen2;
	u16 infhoff3;
	u16 infhlen3;
	u16 infgoff3;
	u16 infglen3;
	u8 reserved2[4];
} __packed;

struct mac_sctn {
	u8 infmflg1; /* reserved */
	u8 infmflg2; /* reserved */
	u8 infmval1;
	u8 infmval2; /* reserved */
	u16 infmscps;
	u16 infmdcps;
	u16 infmsifl;
	u16 infmdifl;
	char infmname[8];
	char infmtype[4];
	char infmmanu[16];
	char infmseq[16];
	char infmpman[4];
	u8 reserved[4];
} __packed;

struct par_sctn {
	u8 infpflg1;
	u8 infpflg2; /* reserved */
	u8 infpval1;
	u8 infpval2; /* reserved */
	u16 infppnum;
	u16 infpscps;
	u16 infpdcps;
	u16 infpsifl;
	u16 infpdifl;
	u16 reserved;
	char infppnam[8];
	u32 infpwbcp;
	u32 infpabcp;
	u32 infpwbif;
	u32 infpabif;
	char infplgnm[8];
	u32 infplgcp;
	u32 infplgif;
} __packed;

struct sthyi_sctns {
	struct hdr_sctn hdr;
	struct mac_sctn mac;
	struct par_sctn par;
} __packed;

struct cpu_inf {
	u64 lpar_cap;
	u64 lpar_grp_cap;
	u64 lpar_weight;
	u64 all_weight;
	int cpu_num_ded;
	int cpu_num_shd;
};

struct lpar_cpu_inf {
	struct cpu_inf cp;
	struct cpu_inf ifl;
};

/*
 * STHYI requires extensive locking in the higher hypervisors
 * and is very computational/memory expensive. Therefore we
 * cache the retrieved data whose valid period is 1s.
 */
#define CACHE_VALID_JIFFIES	HZ

struct sthyi_info {
	void *info;
	unsigned long end;
};

static DEFINE_MUTEX(sthyi_mutex);
static struct sthyi_info sthyi_cache;

static inline u64 cpu_id(u8 ctidx, void *diag224_buf)
{
	return *((u64 *)(diag224_buf + (ctidx + 1) * DIAG204_CPU_NAME_LEN));
}

/*
 * Scales the cpu capping from the lpar range to the one expected in
 * sthyi data.
 *
 * diag204 reports a cap in hundredths of processor units.
 * z/VM's range for one core is 0 - 0x10000.
 */
static u32 scale_cap(u32 in)
{
	return (0x10000 * in) / 100;
}

static void fill_hdr(struct sthyi_sctns *sctns)
{
	sctns->hdr.infhdln = sizeof(sctns->hdr);
	sctns->hdr.infmoff = sizeof(sctns->hdr);
	sctns->hdr.infmlen = sizeof(sctns->mac);
	sctns->hdr.infplen = sizeof(sctns->par);
	sctns->hdr.infpoff = sctns->hdr.infhdln + sctns->hdr.infmlen;
	sctns->hdr.infhtotl = sctns->hdr.infpoff + sctns->hdr.infplen;
}

static void fill_stsi_mac(struct sthyi_sctns *sctns,
			  struct sysinfo_1_1_1 *sysinfo)
{
	if (stsi(sysinfo, 1, 1, 1))
		return;

	sclp_ocf_cpc_name_copy(sctns->mac.infmname);

	memcpy(sctns->mac.infmtype, sysinfo->type, sizeof(sctns->mac.infmtype));
	memcpy(sctns->mac.infmmanu, sysinfo->manufacturer, sizeof(sctns->mac.infmmanu));
	memcpy(sctns->mac.infmpman, sysinfo->plant, sizeof(sctns->mac.infmpman));
	memcpy(sctns->mac.infmseq, sysinfo->sequence, sizeof(sctns->mac.infmseq));

	sctns->mac.infmval1 |= MAC_ID_VLD | MAC_NAME_VLD;
}

static void fill_stsi_par(struct sthyi_sctns *sctns,
			  struct sysinfo_2_2_2 *sysinfo)
{
	if (stsi(sysinfo, 2, 2, 2))
		return;

	sctns->par.infppnum = sysinfo->lpar_number;
	memcpy(sctns->par.infppnam, sysinfo->name, sizeof(sctns->par.infppnam));

	sctns->par.infpval1 |= PAR_ID_VLD;
}

static void fill_stsi(struct sthyi_sctns *sctns)
{
	void *sysinfo;

	/* Errors are handled through the validity bits in the response. */
	sysinfo = (void *)__get_free_page(GFP_KERNEL);
	if (!sysinfo)
		return;

	fill_stsi_mac(sctns, sysinfo);
	fill_stsi_par(sctns, sysinfo);

	free_pages((unsigned long)sysinfo, 0);
}

static void fill_diag_mac(struct sthyi_sctns *sctns,
			  struct diag204_x_phys_block *block,
			  void *diag224_buf)
{
	int i;

	for (i = 0; i < block->hdr.cpus; i++) {
		switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
		case CP:
			if (block->cpus[i].weight == DED_WEIGHT)
				sctns->mac.infmdcps++;
			else
				sctns->mac.infmscps++;
			break;
		case IFL:
			if (block->cpus[i].weight == DED_WEIGHT)
				sctns->mac.infmdifl++;
			else
				sctns->mac.infmsifl++;
			break;
		}
	}
	sctns->mac.infmval1 |= MAC_CNT_VLD;
}

/* Returns a pointer to the the next partition block. */
static struct diag204_x_part_block *lpar_cpu_inf(struct lpar_cpu_inf *part_inf,
						 bool this_lpar,
						 void *diag224_buf,
						 struct diag204_x_part_block *block)
{
	int i, capped = 0, weight_cp = 0, weight_ifl = 0;
	struct cpu_inf *cpu_inf;

	for (i = 0; i < block->hdr.rcpus; i++) {
		if (!(block->cpus[i].cflag & DIAG204_CPU_ONLINE))
			continue;

		switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
		case CP:
			cpu_inf = &part_inf->cp;
			if (block->cpus[i].cur_weight < DED_WEIGHT)
				weight_cp |= block->cpus[i].cur_weight;
			break;
		case IFL:
			cpu_inf = &part_inf->ifl;
			if (block->cpus[i].cur_weight < DED_WEIGHT)
				weight_ifl |= block->cpus[i].cur_weight;
			break;
		default:
			continue;
		}

		if (!this_lpar)
			continue;

		capped |= block->cpus[i].cflag & DIAG204_CPU_CAPPED;
		cpu_inf->lpar_cap |= block->cpus[i].cpu_type_cap;
		cpu_inf->lpar_grp_cap |= block->cpus[i].group_cpu_type_cap;

		if (block->cpus[i].weight == DED_WEIGHT)
			cpu_inf->cpu_num_ded += 1;
		else
			cpu_inf->cpu_num_shd += 1;
	}

	if (this_lpar && capped) {
		part_inf->cp.lpar_weight = weight_cp;
		part_inf->ifl.lpar_weight = weight_ifl;
	}
	part_inf->cp.all_weight += weight_cp;
	part_inf->ifl.all_weight += weight_ifl;
	return (struct diag204_x_part_block *)&block->cpus[i];
}

static void fill_diag(struct sthyi_sctns *sctns)
{
	int i, r, pages;
	bool this_lpar;
	void *diag204_buf;
	void *diag224_buf = NULL;
	struct diag204_x_info_blk_hdr *ti_hdr;
	struct diag204_x_part_block *part_block;
	struct diag204_x_phys_block *phys_block;
	struct lpar_cpu_inf lpar_inf = {};

	/* Errors are handled through the validity bits in the response. */
	pages = diag204((unsigned long)DIAG204_SUBC_RSI |
			(unsigned long)DIAG204_INFO_EXT, 0, NULL);
	if (pages <= 0)
		return;

	diag204_buf = vmalloc(array_size(pages, PAGE_SIZE));
	if (!diag204_buf)
		return;

	r = diag204((unsigned long)DIAG204_SUBC_STIB7 |
		    (unsigned long)DIAG204_INFO_EXT, pages, diag204_buf);
	if (r < 0)
		goto out;

	diag224_buf = (void *)__get_free_page(GFP_KERNEL | GFP_DMA);
	if (!diag224_buf || diag224(diag224_buf))
		goto out;

	ti_hdr = diag204_buf;
	part_block = diag204_buf + sizeof(*ti_hdr);

	for (i = 0; i < ti_hdr->npar; i++) {
		/*
		 * For the calling lpar we also need to get the cpu
		 * caps and weights. The time information block header
		 * specifies the offset to the partition block of the
		 * caller lpar, so we know when we process its data.
		 */
		this_lpar = (void *)part_block - diag204_buf == ti_hdr->this_part;
		part_block = lpar_cpu_inf(&lpar_inf, this_lpar, diag224_buf,
					  part_block);
	}

	phys_block = (struct diag204_x_phys_block *)part_block;
	part_block = diag204_buf + ti_hdr->this_part;
	if (part_block->hdr.mtid)
		sctns->par.infpflg1 = PAR_MT_EN;

	sctns->par.infpval1 |= PAR_GRP_VLD;
	sctns->par.infplgcp = scale_cap(lpar_inf.cp.lpar_grp_cap);
	sctns->par.infplgif = scale_cap(lpar_inf.ifl.lpar_grp_cap);
	memcpy(sctns->par.infplgnm, part_block->hdr.hardware_group_name,
	       sizeof(sctns->par.infplgnm));

	sctns->par.infpscps = lpar_inf.cp.cpu_num_shd;
	sctns->par.infpdcps = lpar_inf.cp.cpu_num_ded;
	sctns->par.infpsifl = lpar_inf.ifl.cpu_num_shd;
	sctns->par.infpdifl = lpar_inf.ifl.cpu_num_ded;
	sctns->par.infpval1 |= PAR_PCNT_VLD;

	sctns->par.infpabcp = scale_cap(lpar_inf.cp.lpar_cap);
	sctns->par.infpabif = scale_cap(lpar_inf.ifl.lpar_cap);
	sctns->par.infpval1 |= PAR_ABS_VLD;

	/*
	 * Everything below needs global performance data to be
	 * meaningful.
	 */
	if (!(ti_hdr->flags & DIAG204_LPAR_PHYS_FLG)) {
		sctns->hdr.infhflg1 |= HDR_PERF_UNAV;
		goto out;
	}

	fill_diag_mac(sctns, phys_block, diag224_buf);

	if (lpar_inf.cp.lpar_weight) {
		sctns->par.infpwbcp = sctns->mac.infmscps * 0x10000 *
			lpar_inf.cp.lpar_weight / lpar_inf.cp.all_weight;
	}

	if (lpar_inf.ifl.lpar_weight) {
		sctns->par.infpwbif = sctns->mac.infmsifl * 0x10000 *
			lpar_inf.ifl.lpar_weight / lpar_inf.ifl.all_weight;
	}
	sctns->par.infpval1 |= PAR_WGHT_VLD;

out:
	free_page((unsigned long)diag224_buf);
	vfree(diag204_buf);
}

static int sthyi(u64 vaddr, u64 *rc)
{
	register u64 code asm("0") = 0;
	register u64 addr asm("2") = vaddr;
	register u64 rcode asm("3");
	int cc;

	asm volatile(
		".insn   rre,0xB2560000,%[code],%[addr]\n"
		"ipm     %[cc]\n"
		"srl     %[cc],28\n"
		: [cc] "=d" (cc), "=d" (rcode)
		: [code] "d" (code), [addr] "a" (addr)
		: "memory", "cc");
	*rc = rcode;
	return cc;
}

static int fill_dst(void *dst, u64 *rc)
{
	struct sthyi_sctns *sctns = (struct sthyi_sctns *)dst;

	/*
	 * If the facility is on, we don't want to emulate the instruction.
	 * We ask the hypervisor to provide the data.
	 */
	if (test_facility(74))
		return sthyi((u64)dst, rc);

	fill_hdr(sctns);
	fill_stsi(sctns);
	fill_diag(sctns);
	*rc = 0;
	return 0;
}

static int sthyi_init_cache(void)
{
	if (sthyi_cache.info)
		return 0;
	sthyi_cache.info = (void *)get_zeroed_page(GFP_KERNEL);
	if (!sthyi_cache.info)
		return -ENOMEM;
	sthyi_cache.end = jiffies - 1; /* expired */
	return 0;
}

static int sthyi_update_cache(u64 *rc)
{
	int r;

	memset(sthyi_cache.info, 0, PAGE_SIZE);
	r = fill_dst(sthyi_cache.info, rc);
	if (r)
		return r;
	sthyi_cache.end = jiffies + CACHE_VALID_JIFFIES;
	return r;
}

/*
 * sthyi_fill - Fill page with data returned by the STHYI instruction
 *
 * @dst: Pointer to zeroed page
 * @rc:  Pointer for storing the return code of the instruction
 *
 * Fills the destination with system information returned by the STHYI
 * instruction. The data is generated by emulation or execution of STHYI,
 * if available. The return value is the condition code that would be
 * returned, the rc parameter is the return code which is passed in
 * register R2 + 1.
 */
int sthyi_fill(void *dst, u64 *rc)
{
	int r;

	mutex_lock(&sthyi_mutex);
	r = sthyi_init_cache();
	if (r)
		goto out;

	if (time_is_before_jiffies(sthyi_cache.end)) {
		/* cache expired */
		r = sthyi_update_cache(rc);
		if (r)
			goto out;
	}
	*rc = 0;
	memcpy(dst, sthyi_cache.info, PAGE_SIZE);
out:
	mutex_unlock(&sthyi_mutex);
	return r;
}
EXPORT_SYMBOL_GPL(sthyi_fill);

SYSCALL_DEFINE4(s390_sthyi, unsigned long, function_code, void __user *, buffer,
		u64 __user *, return_code, unsigned long, flags)
{
	u64 sthyi_rc;
	void *info;
	int r;

	if (flags)
		return -EINVAL;
	if (function_code != STHYI_FC_CP_IFL_CAP)
		return -EOPNOTSUPP;
	info = (void *)get_zeroed_page(GFP_KERNEL);
	if (!info)
		return -ENOMEM;
	r = sthyi_fill(info, &sthyi_rc);
	if (r < 0)
		goto out;
	if (return_code && put_user(sthyi_rc, return_code)) {
		r = -EFAULT;
		goto out;
	}
	if (copy_to_user(buffer, info, PAGE_SIZE))
		r = -EFAULT;
out:
	free_page((unsigned long)info);
	return r;
}
Commit	Line	Data
b1c0de0e	1	// SPDX-License-Identifier: GPL-2.0
95ca2cb5 JF	2	/*
	3	* store hypervisor information instruction emulation functions.
	4	*
95ca2cb5 JF	5	* Copyright IBM Corp. 2016
	6	* Author(s): Janosch Frank <frankja@linux.vnet.ibm.com>
	7	*/
95ca2cb5 JF	8	#include <linux/errno.h>
	9	#include <linux/pagemap.h>
	10	#include <linux/vmalloc.h>
3d8757b8 QH	11	#include <linux/syscalls.h>
3d8757b8 QH	12	#include <linux/mutex.h>
95ca2cb5 JF	13	#include <asm/asm-offsets.h>
	14	#include <asm/sclp.h>
	15	#include <asm/diag.h>
	16	#include <asm/sysinfo.h>
	17	#include <asm/ebcdic.h>
b7c92f1a	18	#include <asm/facility.h>
3d8757b8 QH	19	#include <asm/sthyi.h>
3d8757b8 QH	20	#include "entry.h"
95ca2cb5 JF	21
	22	#define DED_WEIGHT 0xffff
	23	/*
	24	* CP and IFL as EBCDIC strings, SP/0x40 determines the end of string
	25	* as they are justified with spaces.
	26	*/
	27	#define CP 0xc3d7404040404040UL
	28	#define IFL 0xc9c6d34040404040UL
	29
	30	enum hdr_flags {
	31	HDR_NOT_LPAR = 0x10,
	32	HDR_STACK_INCM = 0x20,
	33	HDR_STSI_UNAV = 0x40,
	34	HDR_PERF_UNAV = 0x80,
	35	};
	36
	37	enum mac_validity {
	38	MAC_NAME_VLD = 0x20,
	39	MAC_ID_VLD = 0x40,
	40	MAC_CNT_VLD = 0x80,
	41	};
	42
	43	enum par_flag {
	44	PAR_MT_EN = 0x80,
	45	};
	46
	47	enum par_validity {
	48	PAR_GRP_VLD = 0x08,
	49	PAR_ID_VLD = 0x10,
	50	PAR_ABS_VLD = 0x20,
	51	PAR_WGHT_VLD = 0x40,
	52	PAR_PCNT_VLD = 0x80,
	53	};
	54
	55	struct hdr_sctn {
	56	u8 infhflg1;
	57	u8 infhflg2; /* reserved */
	58	u8 infhval1; /* reserved */
	59	u8 infhval2; /* reserved */
	60	u8 reserved[3];
	61	u8 infhygct;
	62	u16 infhtotl;
	63	u16 infhdln;
	64	u16 infmoff;
	65	u16 infmlen;
	66	u16 infpoff;
	67	u16 infplen;
	68	u16 infhoff1;
	69	u16 infhlen1;
	70	u16 infgoff1;
	71	u16 infglen1;
	72	u16 infhoff2;
	73	u16 infhlen2;
	74	u16 infgoff2;
	75	u16 infglen2;
	76	u16 infhoff3;
	77	u16 infhlen3;
	78	u16 infgoff3;
	79	u16 infglen3;
	80	u8 reserved2[4];
	81	} __packed;
	82
	83	struct mac_sctn {
	84	u8 infmflg1; /* reserved */
85	u8 infmflg2; /* reserved */
86	u8 infmval1;
87	u8 infmval2; /* reserved */
88	u16 infmscps;
89	u16 infmdcps;
90	u16 infmsifl;
91	u16 infmdifl;
92	char infmname[8];
93	char infmtype[4];
94	char infmmanu[16];
95	char infmseq[16];
96	char infmpman[4];
97	u8 reserved[4];
98	} __packed;
99
100	struct par_sctn {
101	u8 infpflg1;
102	u8 infpflg2; /* reserved */
103	u8 infpval1;
104	u8 infpval2; /* reserved */
105	u16 infppnum;
106	u16 infpscps;
107	u16 infpdcps;
108	u16 infpsifl;
109	u16 infpdifl;
110	u16 reserved;
111	char infppnam[8];
112	u32 infpwbcp;
113	u32 infpabcp;
114	u32 infpwbif;
115	u32 infpabif;
116	char infplgnm[8];
117	u32 infplgcp;
118	u32 infplgif;
119	} __packed;
120
121	struct sthyi_sctns {
122	struct hdr_sctn hdr;
123	struct mac_sctn mac;
124	struct par_sctn par;
125	} __packed;
126
127	struct cpu_inf {
128	u64 lpar_cap;
129	u64 lpar_grp_cap;
130	u64 lpar_weight;
131	u64 all_weight;
132	int cpu_num_ded;
133	int cpu_num_shd;
134	};
135
136	struct lpar_cpu_inf {
137	struct cpu_inf cp;
138	struct cpu_inf ifl;
139	};
140
9fb6c9b3 QH	141	/*
	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.
	145	*/
	146	#define CACHE_VALID_JIFFIES HZ
	147
	148	struct sthyi_info {
	149	void *info;
	150	unsigned long end;
	151	};
	152
	153	static DEFINE_MUTEX(sthyi_mutex);
	154	static struct sthyi_info sthyi_cache;
	155
95ca2cb5 JF	156	static inline u64 cpu_id(u8 ctidx, void *diag224_buf)
	157	{
	158	return ((u64 )(diag224_buf + (ctidx + 1) * DIAG204_CPU_NAME_LEN));
	159	}
	160
	161	/*
	162	* Scales the cpu capping from the lpar range to the one expected in
	163	* sthyi data.
	164	*
	165	* diag204 reports a cap in hundredths of processor units.
	166	* z/VM's range for one core is 0 - 0x10000.
	167	*/
	168	static u32 scale_cap(u32 in)
	169	{
	170	return (0x10000 * in) / 100;
	171	}
	172
	173	static void fill_hdr(struct sthyi_sctns *sctns)
	174	{
	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;
	181	}
	182
	183	static void fill_stsi_mac(struct sthyi_sctns *sctns,
	184	struct sysinfo_1_1_1 *sysinfo)
	185	{
	186	if (stsi(sysinfo, 1, 1, 1))
	187	return;
	188
	189	sclp_ocf_cpc_name_copy(sctns->mac.infmname);
	190
	191	memcpy(sctns->mac.infmtype, sysinfo->type, sizeof(sctns->mac.infmtype));
	192	memcpy(sctns->mac.infmmanu, sysinfo->manufacturer, sizeof(sctns->mac.infmmanu));
	193	memcpy(sctns->mac.infmpman, sysinfo->plant, sizeof(sctns->mac.infmpman));
	194	memcpy(sctns->mac.infmseq, sysinfo->sequence, sizeof(sctns->mac.infmseq));
	195
	196	sctns->mac.infmval1 \|= MAC_ID_VLD \| MAC_NAME_VLD;
	197	}
	198
	199	static void fill_stsi_par(struct sthyi_sctns *sctns,
	200	struct sysinfo_2_2_2 *sysinfo)
	201	{
	202	if (stsi(sysinfo, 2, 2, 2))
	203	return;
	204
	205	sctns->par.infppnum = sysinfo->lpar_number;
	206	memcpy(sctns->par.infppnam, sysinfo->name, sizeof(sctns->par.infppnam));
	207
	208	sctns->par.infpval1 \|= PAR_ID_VLD;
	209	}
	210
	211	static void fill_stsi(struct sthyi_sctns *sctns)
	212	{
	213	void *sysinfo;
	214
	215	/* Errors are handled through the validity bits in the response. */
	216	sysinfo = (void *)__get_free_page(GFP_KERNEL);
	217	if (!sysinfo)
	218	return;
	219
220	fill_stsi_mac(sctns, sysinfo);
221	fill_stsi_par(sctns, sysinfo);
222
223	free_pages((unsigned long)sysinfo, 0);
224	}
225
226	static void fill_diag_mac(struct sthyi_sctns *sctns,
227	struct diag204_x_phys_block *block,
228	void *diag224_buf)
229	{
230	int i;
231
232	for (i = 0; i < block->hdr.cpus; i++) {
233	switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
234	case CP:
235	if (block->cpus[i].weight == DED_WEIGHT)
236	sctns->mac.infmdcps++;
237	else
238	sctns->mac.infmscps++;
239	break;
240	case IFL:
241	if (block->cpus[i].weight == DED_WEIGHT)
242	sctns->mac.infmdifl++;
243	else
244	sctns->mac.infmsifl++;
245	break;
246	}
247	}
248	sctns->mac.infmval1 \|= MAC_CNT_VLD;
249	}
250
251	/* Returns a pointer to the the next partition block. */
252	static struct diag204_x_part_block lpar_cpu_inf(struct lpar_cpu_inf part_inf,
253	bool this_lpar,
254	void *diag224_buf,
255	struct diag204_x_part_block *block)
256	{
257	int i, capped = 0, weight_cp = 0, weight_ifl = 0;
258	struct cpu_inf *cpu_inf;
259
260	for (i = 0; i < block->hdr.rcpus; i++) {
261	if (!(block->cpus[i].cflag & DIAG204_CPU_ONLINE))
262	continue;
263
264	switch (cpu_id(block->cpus[i].ctidx, diag224_buf)) {
265	case CP:
266	cpu_inf = &part_inf->cp;
267	if (block->cpus[i].cur_weight < DED_WEIGHT)
268	weight_cp \|= block->cpus[i].cur_weight;
269	break;
270	case IFL:
271	cpu_inf = &part_inf->ifl;
272	if (block->cpus[i].cur_weight < DED_WEIGHT)
273	weight_ifl \|= block->cpus[i].cur_weight;
274	break;
275	default:
276	continue;
277	}
278
279	if (!this_lpar)
280	continue;
281
282	capped \|= block->cpus[i].cflag & DIAG204_CPU_CAPPED;
283	cpu_inf->lpar_cap \|= block->cpus[i].cpu_type_cap;
284	cpu_inf->lpar_grp_cap \|= block->cpus[i].group_cpu_type_cap;
285
286	if (block->cpus[i].weight == DED_WEIGHT)
287	cpu_inf->cpu_num_ded += 1;
288	else
289	cpu_inf->cpu_num_shd += 1;
290	}
291
292	if (this_lpar && capped) {
293	part_inf->cp.lpar_weight = weight_cp;
294	part_inf->ifl.lpar_weight = weight_ifl;
295	}
296	part_inf->cp.all_weight += weight_cp;
297	part_inf->ifl.all_weight += weight_ifl;
298	return (struct diag204_x_part_block *)&block->cpus[i];
299	}
300
301	static void fill_diag(struct sthyi_sctns *sctns)
302	{
303	int i, r, pages;
304	bool this_lpar;
305	void *diag204_buf;
306	void *diag224_buf = NULL;
307	struct diag204_x_info_blk_hdr *ti_hdr;
308	struct diag204_x_part_block *part_block;
309	struct diag204_x_phys_block *phys_block;
310	struct lpar_cpu_inf lpar_inf = {};
311
312	/* Errors are handled through the validity bits in the response. */
313	pages = diag204((unsigned long)DIAG204_SUBC_RSI \|
314	(unsigned long)DIAG204_INFO_EXT, 0, NULL);
315	if (pages <= 0)
316	return;
317
42bc47b3	318	diag204_buf = vmalloc(array_size(pages, PAGE_SIZE));
95ca2cb5 JF	319	if (!diag204_buf)
	320	return;
	321
	322	r = diag204((unsigned long)DIAG204_SUBC_STIB7 \|
	323	(unsigned long)DIAG204_INFO_EXT, pages, diag204_buf);
	324	if (r < 0)
	325	goto out;
	326
45c7ee43	327	diag224_buf = (void *)__get_free_page(GFP_KERNEL \| GFP_DMA);
95ca2cb5 JF	328	if (!diag224_buf \|\| diag224(diag224_buf))
	329	goto out;
	330
	331	ti_hdr = diag204_buf;
	332	part_block = diag204_buf + sizeof(*ti_hdr);
	333
	334	for (i = 0; i < ti_hdr->npar; i++) {
	335	/*
	336	* For the calling lpar we also need to get the cpu
	337	* caps and weights. The time information block header
	338	* specifies the offset to the partition block of the
	339	* caller lpar, so we know when we process its data.
	340	*/
	341	this_lpar = (void *)part_block - diag204_buf == ti_hdr->this_part;
	342	part_block = lpar_cpu_inf(&lpar_inf, this_lpar, diag224_buf,
	343	part_block);
	344	}
	345
	346	phys_block = (struct diag204_x_phys_block *)part_block;
	347	part_block = diag204_buf + ti_hdr->this_part;
	348	if (part_block->hdr.mtid)
	349	sctns->par.infpflg1 = PAR_MT_EN;
	350
	351	sctns->par.infpval1 \|= PAR_GRP_VLD;
	352	sctns->par.infplgcp = scale_cap(lpar_inf.cp.lpar_grp_cap);
	353	sctns->par.infplgif = scale_cap(lpar_inf.ifl.lpar_grp_cap);
	354	memcpy(sctns->par.infplgnm, part_block->hdr.hardware_group_name,
	355	sizeof(sctns->par.infplgnm));
	356
	357	sctns->par.infpscps = lpar_inf.cp.cpu_num_shd;
	358	sctns->par.infpdcps = lpar_inf.cp.cpu_num_ded;
	359	sctns->par.infpsifl = lpar_inf.ifl.cpu_num_shd;
	360	sctns->par.infpdifl = lpar_inf.ifl.cpu_num_ded;
	361	sctns->par.infpval1 \|= PAR_PCNT_VLD;
	362
	363	sctns->par.infpabcp = scale_cap(lpar_inf.cp.lpar_cap);
	364	sctns->par.infpabif = scale_cap(lpar_inf.ifl.lpar_cap);
	365	sctns->par.infpval1 \|= PAR_ABS_VLD;
	366
	367	/*
	368	* Everything below needs global performance data to be
	369	* meaningful.
	370	*/
	371	if (!(ti_hdr->flags & DIAG204_LPAR_PHYS_FLG)) {
	372	sctns->hdr.infhflg1 \|= HDR_PERF_UNAV;
	373	goto out;
	374	}
	375
	376	fill_diag_mac(sctns, phys_block, diag224_buf);
	377
	378	if (lpar_inf.cp.lpar_weight) {
	379	sctns->par.infpwbcp = sctns->mac.infmscps * 0x10000 *
	380	lpar_inf.cp.lpar_weight / lpar_inf.cp.all_weight;
	381	}
	382
	383	if (lpar_inf.ifl.lpar_weight) {
	384	sctns->par.infpwbif = sctns->mac.infmsifl * 0x10000 *
	385	lpar_inf.ifl.lpar_weight / lpar_inf.ifl.all_weight;
	386	}
	387	sctns->par.infpval1 \|= PAR_WGHT_VLD;
	388
	389	out:
45c7ee43	390	free_page((unsigned long)diag224_buf);
95ca2cb5 JF	391	vfree(diag204_buf);
	392	}
	393
b7c92f1a	394	static int sthyi(u64 vaddr, u64 *rc)
95ca2cb5 JF	395	{
	396	register u64 code asm("0") = 0;
	397	register u64 addr asm("2") = vaddr;
b7c92f1a	398	register u64 rcode asm("3");
95ca2cb5 JF	399	int cc;
	400
	401	asm volatile(
	402	".insn rre,0xB2560000,%[code],%[addr]\n"
	403	"ipm %[cc]\n"
	404	"srl %[cc],28\n"
b7c92f1a	405	: [cc] "=d" (cc), "=d" (rcode)
95ca2cb5	406	: [code] "d" (code), [addr] "a" (addr)
b7c92f1a QH	407	: "memory", "cc");
b7c92f1a QH	408	*rc = rcode;
95ca2cb5 JF	409	return cc;
	410	}
	411
9fb6c9b3 QH	412	static int fill_dst(void dst, u64 rc)
	413	{
	414	struct sthyi_sctns sctns = (struct sthyi_sctns )dst;
	415
	416	/*
	417	* If the facility is on, we don't want to emulate the instruction.
	418	* We ask the hypervisor to provide the data.
	419	*/
	420	if (test_facility(74))
	421	return sthyi((u64)dst, rc);
	422
	423	fill_hdr(sctns);
	424	fill_stsi(sctns);
	425	fill_diag(sctns);
	426	*rc = 0;
	427	return 0;
	428	}
	429
	430	static int sthyi_init_cache(void)
	431	{
	432	if (sthyi_cache.info)
	433	return 0;
	434	sthyi_cache.info = (void *)get_zeroed_page(GFP_KERNEL);
	435	if (!sthyi_cache.info)
	436	return -ENOMEM;
	437	sthyi_cache.end = jiffies - 1; /* expired */
	438	return 0;
	439	}
	440
	441	static int sthyi_update_cache(u64 *rc)
	442	{
	443	int r;
	444
	445	memset(sthyi_cache.info, 0, PAGE_SIZE);
	446	r = fill_dst(sthyi_cache.info, rc);
	447	if (r)
	448	return r;
	449	sthyi_cache.end = jiffies + CACHE_VALID_JIFFIES;
	450	return r;
	451	}
	452
b7c92f1a QH	453	/*
	454	* sthyi_fill - Fill page with data returned by the STHYI instruction
	455	*
	456	* @dst: Pointer to zeroed page
	457	* @rc: Pointer for storing the return code of the instruction
	458	*
	459	* Fills the destination with system information returned by the STHYI
	460	* instruction. The data is generated by emulation or execution of STHYI,
	461	* if available. The return value is the condition code that would be
	462	* returned, the rc parameter is the return code which is passed in
	463	* register R2 + 1.
	464	*/
	465	int sthyi_fill(void dst, u64 rc)
95ca2cb5	466	{
9fb6c9b3	467	int r;
95ca2cb5	468
9fb6c9b3 QH	469	mutex_lock(&sthyi_mutex);
	470	r = sthyi_init_cache();
	471	if (r)
	472	goto out;
95ca2cb5	473
9fb6c9b3 QH	474	if (time_is_before_jiffies(sthyi_cache.end)) {
	475	/* cache expired */
	476	r = sthyi_update_cache(rc);
	477	if (r)
	478	goto out;
	479	}
b7c92f1a	480	*rc = 0;
9fb6c9b3 QH	481	memcpy(dst, sthyi_cache.info, PAGE_SIZE);
	482	out:
	483	mutex_unlock(&sthyi_mutex);
	484	return r;
95ca2cb5	485	}
b7c92f1a	486	EXPORT_SYMBOL_GPL(sthyi_fill);
3d8757b8 QH	487
	488	SYSCALL_DEFINE4(s390_sthyi, unsigned long, function_code, void __user *, buffer,
	489	u64 __user *, return_code, unsigned long, flags)
	490	{
	491	u64 sthyi_rc;
	492	void *info;
	493	int r;
	494
	495	if (flags)
	496	return -EINVAL;
	497	if (function_code != STHYI_FC_CP_IFL_CAP)
	498	return -EOPNOTSUPP;
	499	info = (void *)get_zeroed_page(GFP_KERNEL);
	500	if (!info)
	501	return -ENOMEM;
	502	r = sthyi_fill(info, &sthyi_rc);
	503	if (r < 0)
	504	goto out;
	505	if (return_code && put_user(sthyi_rc, return_code)) {
	506	r = -EFAULT;
	507	goto out;
	508	}
	509	if (copy_to_user(buffer, info, PAGE_SIZE))
	510	r = -EFAULT;
	511	out:
	512	free_page((unsigned long)info);
	513	return r;
	514	}