s390/diag324: Retrieve power readings via diag 0x324

Retrieve electrical power readings for resources in a computing
environment via diag 0x324. diag 0x324 stores the power readings in the
power information block (pib).

Provide power readings from pib via diag324 ioctl interface.  diag324
ioctl provides new pib to the user only if the threshold time has passed
since the last call. Otherwise, cache data is returned. When there are
no active readers, cleanup of pib buffer is performed.

Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>

arch/s390/include/asm/diag.h

@@ -38,6 +38,7 @@ enum diag_stat_enum {
 	DIAG_STAT_X308,
 	DIAG_STAT_X318,
 	DIAG_STAT_X320,
+	DIAG_STAT_X324,
 	DIAG_STAT_X49C,
 	DIAG_STAT_X500,
 	NR_DIAG_STAT

arch/s390/include/asm/sclp.h

@@ -89,6 +89,7 @@ struct sclp_info {
 	unsigned char has_gisaf : 1;
 	unsigned char has_diag318 : 1;
 	unsigned char has_diag320 : 1;
+	unsigned char has_diag324 : 1;
 	unsigned char has_sipl : 1;
 	unsigned char has_sipl_eckd : 1;
 	unsigned char has_dirq : 1;

arch/s390/include/uapi/asm/diag.h

@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * Diag ioctls and its associated structures definitions.
+ *
+ * Copyright IBM Corp. 2024
+ */
+
+#ifndef __S390_UAPI_ASM_DIAG_H
+#define __S390_UAPI_ASM_DIAG_H
+
+#include <linux/types.h>
+
+#define DIAG_MAGIC_STR 'D'
+
+struct diag324_pib {
+	__u64 address;
+	__u64 sequence;
+};
+
+/* Diag ioctl definitions */
+#define DIAG324_GET_PIBBUF	_IOWR(DIAG_MAGIC_STR, 0x77, struct diag324_pib)
+#define DIAG324_GET_PIBLEN	_IOR(DIAG_MAGIC_STR, 0x78, size_t)
+
+#endif /* __S390_UAPI_ASM_DIAG_H */

arch/s390/kernel/Makefile

@@ -44,6 +44,7 @@ obj-y	+= runtime_instr.o cache.o fpu.o dumpstack.o guarded_storage.o sthyi.o
 obj-y	+= entry.o reipl.o kdebugfs.o alternative.o
 obj-y	+= nospec-branch.o ipl_vmparm.o machine_kexec_reloc.o unwind_bc.o
 obj-y	+= smp.o text_amode31.o stacktrace.o abs_lowcore.o facility.o uv.o wti.o
+obj-y	+= diag/
 
 extra-y				+= vmlinux.lds
 

arch/s390/kernel/diag.c

@@ -53,6 +53,7 @@ static const struct diag_desc diag_map[NR_DIAG_STAT] = {
 	[DIAG_STAT_X308] = { .code = 0x308, .name = "List-Directed IPL" },
 	[DIAG_STAT_X318] = { .code = 0x318, .name = "CP Name and Version Codes" },
 	[DIAG_STAT_X320] = { .code = 0x320, .name = "Certificate Store" },
+	[DIAG_STAT_X324] = { .code = 0x324, .name = "Power Information Block" },
 	[DIAG_STAT_X49C] = { .code = 0x49c, .name = "Warning-Track Interruption" },
 	[DIAG_STAT_X500] = { .code = 0x500, .name = "Virtio Service" },
 };

arch/s390/kernel/diag/Makefile

@@ -1 +1 @@
-obj-y	:= diag_misc.o
+obj-y	:= diag_misc.o diag324.o

arch/s390/kernel/diag/diag324.c

@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Request power readings for resources in a computing environment via
+ * diag 0x324. diag 0x324 stores the power readings in the power information
+ * block (pib).
+ *
+ * Copyright IBM Corp. 2024
+ */
+
+#define pr_fmt(fmt)	"diag324: " fmt
+#include <linux/fs.h>
+#include <linux/gfp.h>
+#include <linux/ioctl.h>
+#include <linux/jiffies.h>
+#include <linux/kernel.h>
+#include <linux/ktime.h>
+#include <linux/string.h>
+#include <linux/slab.h>
+#include <linux/timer.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+#include <linux/vmalloc.h>
+
+#include <asm/diag.h>
+#include <asm/sclp.h>
+#include <asm/timex.h>
+#include <uapi/asm/diag.h>
+#include "diag_ioctl.h"
+
+enum subcode {
+	DIAG324_SUBC_0 = 0,
+	DIAG324_SUBC_1 = 1,
+	DIAG324_SUBC_2 = 2,
+};
+
+enum retcode {
+	DIAG324_RET_SUCCESS		= 0x0001,
+	DIAG324_RET_SUBC_NOTAVAIL	= 0x0103,
+	DIAG324_RET_INSUFFICIENT_SIZE	= 0x0104,
+	DIAG324_RET_READING_UNAVAILABLE	= 0x0105,
+};
+
+union diag324_response {
+	u64 response;
+	struct {
+		u64 installed	: 32;
+		u64		: 16;
+		u64 rc		: 16;
+	} sc0;
+	struct {
+		u64 format	: 16;
+		u64		: 16;
+		u64 pib_len	: 16;
+		u64 rc		: 16;
+	} sc1;
+	struct {
+		u64		: 48;
+		u64 rc		: 16;
+	} sc2;
+};
+
+union diag324_request {
+	u64 request;
+	struct {
+		u64		: 32;
+		u64 allocated	: 16;
+		u64		: 12;
+		u64 sc		: 4;
+	} sc2;
+};
+
+struct pib {
+	u32		: 8;
+	u32 num		: 8;
+	u32 len		: 16;
+	u32		: 24;
+	u32 hlen	: 8;
+	u64		: 64;
+	u64 intv;
+	u8  r[];
+} __packed;
+
+struct pibdata {
+	struct pib *pib;
+	ktime_t expire;
+	u64 sequence;
+	size_t len;
+	int rc;
+};
+
+static DEFINE_MUTEX(pibmutex);
+static struct pibdata pibdata;
+
+#define PIBWORK_DELAY (5 * NSEC_PER_SEC)
+
+static void pibwork_handler(struct work_struct *work);
+static DECLARE_DELAYED_WORK(pibwork, pibwork_handler);
+
+static unsigned long diag324(unsigned long subcode, void *addr)
+{
+	union register_pair rp = { .even = (unsigned long)addr };
+
+	diag_stat_inc(DIAG_STAT_X324);
+	asm volatile("diag	%[rp],%[subcode],0x324\n"
+		     : [rp] "+d" (rp.pair)
+		     : [subcode] "d" (subcode)
+		     : "memory");
+	return rp.odd;
+}
+
+static void pibwork_handler(struct work_struct *work)
+{
+	struct pibdata *data = &pibdata;
+	ktime_t timedout;
+
+	mutex_lock(&pibmutex);
+	timedout = ktime_add_ns(data->expire, PIBWORK_DELAY);
+	if (ktime_before(ktime_get(), timedout)) {
+		mod_delayed_work(system_wq, &pibwork, nsecs_to_jiffies(PIBWORK_DELAY));
+		goto out;
+	}
+	vfree(data->pib);
+	data->pib = NULL;
+out:
+	mutex_unlock(&pibmutex);
+}
+
+static void pib_update(struct pibdata *data)
+{
+	union diag324_request req = { .sc2.sc = DIAG324_SUBC_2, .sc2.allocated = data->len };
+	union diag324_response res;
+	int rc;
+
+	memset(data->pib, 0, data->len);
+	res.response = diag324(req.request, data->pib);
+	switch (res.sc2.rc) {
+	case DIAG324_RET_SUCCESS:
+		rc = 0;
+		break;
+	case DIAG324_RET_SUBC_NOTAVAIL:
+		rc = -ENOENT;
+		break;
+	case DIAG324_RET_INSUFFICIENT_SIZE:
+		rc = -EMSGSIZE;
+		break;
+	case DIAG324_RET_READING_UNAVAILABLE:
+		rc = -EBUSY;
+		break;
+	default:
+		rc = -EINVAL;
+	}
+	data->rc = rc;
+}
+
+long diag324_pibbuf(unsigned long arg)
+{
+	struct diag324_pib __user *udata = (struct diag324_pib __user *)arg;
+	struct pibdata *data = &pibdata;
+	static bool first = true;
+	u64 address;
+	int rc;
+
+	if (!data->len)
+		return -EOPNOTSUPP;
+	if (get_user(address, &udata->address))
+		return -EFAULT;
+	mutex_lock(&pibmutex);
+	rc = -ENOMEM;
+	if (!data->pib)
+		data->pib = vmalloc(data->len);
+	if (!data->pib)
+		goto out;
+	if (first || ktime_after(ktime_get(), data->expire)) {
+		pib_update(data);
+		data->sequence++;
+		data->expire = ktime_add_ns(ktime_get(), tod_to_ns(data->pib->intv));
+		mod_delayed_work(system_wq, &pibwork, nsecs_to_jiffies(PIBWORK_DELAY));
+		first = false;
+	}
+	rc = data->rc;
+	if (rc != 0 && rc != -EBUSY)
+		goto out;
+	rc = copy_to_user((void __user *)address, data->pib, data->pib->len);
+	rc |= put_user(data->sequence, &udata->sequence);
+	if (rc)
+		rc = -EFAULT;
+out:
+	mutex_unlock(&pibmutex);
+	return rc;
+}
+
+long diag324_piblen(unsigned long arg)
+{
+	struct pibdata *data = &pibdata;
+
+	if (!data->len)
+		return -EOPNOTSUPP;
+	if (put_user(data->len, (size_t __user *)arg))
+		return -EFAULT;
+	return 0;
+}
+
+static int __init diag324_init(void)
+{
+	union diag324_response res;
+	unsigned long installed;
+
+	if (!sclp.has_diag324)
+		return -EOPNOTSUPP;
+	res.response = diag324(DIAG324_SUBC_0, NULL);
+	if (res.sc0.rc != DIAG324_RET_SUCCESS)
+		return -EOPNOTSUPP;
+	installed = res.response;
+	if (!test_bit_inv(DIAG324_SUBC_1, &installed))
+		return -EOPNOTSUPP;
+	if (!test_bit_inv(DIAG324_SUBC_2, &installed))
+		return -EOPNOTSUPP;
+	res.response = diag324(DIAG324_SUBC_1, NULL);
+	if (res.sc1.rc != DIAG324_RET_SUCCESS)
+		return -EOPNOTSUPP;
+	pibdata.len = res.sc1.pib_len;
+	return 0;
+}
+device_initcall(diag324_init);

arch/s390/kernel/diag/diag_ioctl.h

@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _DIAG_IOCTL_H
+#define _DIAG_IOCTL_H
+
+#include <linux/types.h>
+
+long diag324_pibbuf(unsigned long arg);
+long diag324_piblen(unsigned long arg);
+
+#endif /* _DIAG_IOCTL_H */

arch/s390/kernel/diag/diag_misc.c

@@ -12,11 +12,20 @@
 #include <linux/miscdevice.h>
 #include <linux/types.h>
 
+#include <uapi/asm/diag.h>
+#include "diag_ioctl.h"
+
 static long diag_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
 	long rc;
 
 	switch (cmd) {
+	case DIAG324_GET_PIBLEN:
+		rc = diag324_piblen(arg);
+		break;
+	case DIAG324_GET_PIBBUF:
+		rc = diag324_pibbuf(arg);
+		break;
 	default:
 		rc = -ENOIOCTLCMD;
 		break;

drivers/s390/char/sclp.h

@@ -196,7 +196,9 @@ struct read_info_sccb {
 	u8	byte_134;			/* 134 */
 	u8	cpudirq;		/* 135 */
 	u16	cbl;			/* 136-137 */
-	u8	_pad_138[EXT_SCCB_READ_SCP - 138];
+	u8	byte_138;		/* 138 */
+	u8	byte_139;		/* 139 */
+	u8	_pad_140[EXT_SCCB_READ_SCP - 140];
 } __packed __aligned(PAGE_SIZE);
 
 struct read_storage_sccb {

drivers/s390/char/sclp_early.c

@@ -64,6 +64,8 @@ static void __init sclp_early_facilities_detect(void)
 		sclp.has_sipl = !!(sccb->cbl & 0x4000);
 		sclp.has_sipl_eckd = !!(sccb->cbl & 0x2000);
 	}
+	if (sccb->cpuoff > 139)
+		sclp.has_diag324 = !!(sccb->byte_139 & 0x80);
 	sclp.rnmax = sccb->rnmax ? sccb->rnmax : sccb->rnmax2;
 	sclp.rzm = sccb->rnsize ? sccb->rnsize : sccb->rnsize2;
 	sclp.rzm <<= 20;