2 * Hypervisor filesystem for Linux on s390. Diag 204 and 224
5 * Copyright IBM Corp. 2006, 2008
6 * Author(s): Michael Holzheu <holzheu@de.ibm.com>
9 #define KMSG_COMPONENT "hypfs"
10 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12 #include <linux/types.h>
13 #include <linux/errno.h>
14 #include <linux/slab.h>
15 #include <linux/string.h>
16 #include <linux/vmalloc.h>
19 #include <asm/ebcdic.h>
22 #define LPAR_NAME_LEN 8 /* lpar name len in diag 204 data */
23 #define CPU_NAME_LEN 16 /* type name len of cpus in diag224 name table */
24 #define TMP_SIZE 64 /* size of temporary buffers */
26 #define DBFS_D204_HDR_VERSION 0
28 /* diag 204 subcodes */
36 /* The two available diag 204 data formats */
42 /* bit is set in flags, when physical cpu info is included in diag 204 data */
43 #define LPAR_PHYS_FLG 0x80
45 static char *diag224_cpu_names
; /* diag 224 name table */
46 static enum diag204_sc diag204_store_sc
; /* used subcode for store */
47 static enum diag204_format diag204_info_type
; /* used diag 204 data format */
49 static void *diag204_buf
; /* 4K aligned buffer for diag204 data */
50 static void *diag204_buf_vmalloc
; /* vmalloc pointer for diag204 data */
51 static int diag204_buf_pages
; /* number of pages for diag204 data */
53 static struct dentry
*dbfs_d204_file
;
56 * DIAG 204 data structures and member access functions.
58 * Since we have two different diag 204 data formats for old and new s390
59 * machines, we do not access the structs directly, but use getter functions for
60 * each struct member instead. This should make the code more readable.
63 /* Time information block */
72 } __attribute__ ((packed
));
74 struct x_info_blk_hdr
{
83 } __attribute__ ((packed
));
85 static inline int info_blk_hdr__size(enum diag204_format type
)
87 if (type
== INFO_SIMPLE
)
88 return sizeof(struct info_blk_hdr
);
90 return sizeof(struct x_info_blk_hdr
);
93 static inline __u8
info_blk_hdr__npar(enum diag204_format type
, void *hdr
)
95 if (type
== INFO_SIMPLE
)
96 return ((struct info_blk_hdr
*)hdr
)->npar
;
98 return ((struct x_info_blk_hdr
*)hdr
)->npar
;
101 static inline __u8
info_blk_hdr__flags(enum diag204_format type
, void *hdr
)
103 if (type
== INFO_SIMPLE
)
104 return ((struct info_blk_hdr
*)hdr
)->flags
;
106 return ((struct x_info_blk_hdr
*)hdr
)->flags
;
109 static inline __u16
info_blk_hdr__pcpus(enum diag204_format type
, void *hdr
)
111 if (type
== INFO_SIMPLE
)
112 return ((struct info_blk_hdr
*)hdr
)->phys_cpus
;
114 return ((struct x_info_blk_hdr
*)hdr
)->phys_cpus
;
117 /* Partition header */
123 char part_name
[LPAR_NAME_LEN
];
124 } __attribute__ ((packed
));
132 char part_name
[LPAR_NAME_LEN
];
142 } __attribute__ ((packed
));
144 static inline int part_hdr__size(enum diag204_format type
)
146 if (type
== INFO_SIMPLE
)
147 return sizeof(struct part_hdr
);
149 return sizeof(struct x_part_hdr
);
152 static inline __u8
part_hdr__rcpus(enum diag204_format type
, void *hdr
)
154 if (type
== INFO_SIMPLE
)
155 return ((struct part_hdr
*)hdr
)->cpus
;
157 return ((struct x_part_hdr
*)hdr
)->rcpus
;
160 static inline void part_hdr__part_name(enum diag204_format type
, void *hdr
,
163 if (type
== INFO_SIMPLE
)
164 memcpy(name
, ((struct part_hdr
*)hdr
)->part_name
,
167 memcpy(name
, ((struct x_part_hdr
*)hdr
)->part_name
,
169 EBCASC(name
, LPAR_NAME_LEN
);
170 name
[LPAR_NAME_LEN
] = 0;
182 } __attribute__ ((packed
));
201 } __attribute__ ((packed
));
205 static inline int cpu_info__size(enum diag204_format type
)
207 if (type
== INFO_SIMPLE
)
208 return sizeof(struct cpu_info
);
210 return sizeof(struct x_cpu_info
);
213 static inline __u8
cpu_info__ctidx(enum diag204_format type
, void *hdr
)
215 if (type
== INFO_SIMPLE
)
216 return ((struct cpu_info
*)hdr
)->ctidx
;
218 return ((struct x_cpu_info
*)hdr
)->ctidx
;
221 static inline __u16
cpu_info__cpu_addr(enum diag204_format type
, void *hdr
)
223 if (type
== INFO_SIMPLE
)
224 return ((struct cpu_info
*)hdr
)->cpu_addr
;
226 return ((struct x_cpu_info
*)hdr
)->cpu_addr
;
229 static inline __u64
cpu_info__acc_time(enum diag204_format type
, void *hdr
)
231 if (type
== INFO_SIMPLE
)
232 return ((struct cpu_info
*)hdr
)->acc_time
;
234 return ((struct x_cpu_info
*)hdr
)->acc_time
;
237 static inline __u64
cpu_info__lp_time(enum diag204_format type
, void *hdr
)
239 if (type
== INFO_SIMPLE
)
240 return ((struct cpu_info
*)hdr
)->lp_time
;
242 return ((struct x_cpu_info
*)hdr
)->lp_time
;
245 static inline __u64
cpu_info__online_time(enum diag204_format type
, void *hdr
)
247 if (type
== INFO_SIMPLE
)
248 return 0; /* online_time not available in simple info */
250 return ((struct x_cpu_info
*)hdr
)->online_time
;
253 /* Physical header */
260 } __attribute__ ((packed
));
268 } __attribute__ ((packed
));
270 static inline int phys_hdr__size(enum diag204_format type
)
272 if (type
== INFO_SIMPLE
)
273 return sizeof(struct phys_hdr
);
275 return sizeof(struct x_phys_hdr
);
278 static inline __u8
phys_hdr__cpus(enum diag204_format type
, void *hdr
)
280 if (type
== INFO_SIMPLE
)
281 return ((struct phys_hdr
*)hdr
)->cpus
;
283 return ((struct x_phys_hdr
*)hdr
)->cpus
;
286 /* Physical CPU info block */
295 } __attribute__ ((packed
));
304 } __attribute__ ((packed
));
306 static inline int phys_cpu__size(enum diag204_format type
)
308 if (type
== INFO_SIMPLE
)
309 return sizeof(struct phys_cpu
);
311 return sizeof(struct x_phys_cpu
);
314 static inline __u16
phys_cpu__cpu_addr(enum diag204_format type
, void *hdr
)
316 if (type
== INFO_SIMPLE
)
317 return ((struct phys_cpu
*)hdr
)->cpu_addr
;
319 return ((struct x_phys_cpu
*)hdr
)->cpu_addr
;
322 static inline __u64
phys_cpu__mgm_time(enum diag204_format type
, void *hdr
)
324 if (type
== INFO_SIMPLE
)
325 return ((struct phys_cpu
*)hdr
)->mgm_time
;
327 return ((struct x_phys_cpu
*)hdr
)->mgm_time
;
330 static inline __u64
phys_cpu__ctidx(enum diag204_format type
, void *hdr
)
332 if (type
== INFO_SIMPLE
)
333 return ((struct phys_cpu
*)hdr
)->ctidx
;
335 return ((struct x_phys_cpu
*)hdr
)->ctidx
;
338 /* Diagnose 204 functions */
340 static inline int __diag204(unsigned long subcode
, unsigned long size
, void *addr
)
342 register unsigned long _subcode
asm("0") = subcode
;
343 register unsigned long _size
asm("1") = size
;
346 " diag %2,%0,0x204\n"
349 : "+d" (_subcode
), "+d" (_size
) : "d" (addr
) : "memory");
355 static int diag204(unsigned long subcode
, unsigned long size
, void *addr
)
357 diag_stat_inc(DIAG_STAT_X204
);
358 return __diag204(subcode
, size
, addr
);
362 * For the old diag subcode 4 with simple data format we have to use real
363 * memory. If we use subcode 6 or 7 with extended data format, we can (and
364 * should) use vmalloc, since we need a lot of memory in that case. Currently
368 static void diag204_free_buffer(void)
372 if (diag204_buf_vmalloc
) {
373 vfree(diag204_buf_vmalloc
);
374 diag204_buf_vmalloc
= NULL
;
376 free_pages((unsigned long) diag204_buf
, 0);
381 static void *page_align_ptr(void *ptr
)
383 return (void *) PAGE_ALIGN((unsigned long) ptr
);
386 static void *diag204_alloc_vbuf(int pages
)
388 /* The buffer has to be page aligned! */
389 diag204_buf_vmalloc
= vmalloc(PAGE_SIZE
* (pages
+ 1));
390 if (!diag204_buf_vmalloc
)
391 return ERR_PTR(-ENOMEM
);
392 diag204_buf
= page_align_ptr(diag204_buf_vmalloc
);
393 diag204_buf_pages
= pages
;
397 static void *diag204_alloc_rbuf(void)
399 diag204_buf
= (void*)__get_free_pages(GFP_KERNEL
,0);
401 return ERR_PTR(-ENOMEM
);
402 diag204_buf_pages
= 1;
406 static void *diag204_get_buffer(enum diag204_format fmt
, int *pages
)
409 *pages
= diag204_buf_pages
;
412 if (fmt
== INFO_SIMPLE
) {
414 return diag204_alloc_rbuf();
415 } else {/* INFO_EXT */
416 *pages
= diag204((unsigned long)SUBC_RSI
|
417 (unsigned long)INFO_EXT
, 0, NULL
);
419 return ERR_PTR(-ENOSYS
);
421 return diag204_alloc_vbuf(*pages
);
426 * diag204_probe() has to find out, which type of diagnose 204 implementation
427 * we have on our machine. Currently there are three possible scanarios:
428 * - subcode 4 + simple data format (only one page)
429 * - subcode 4-6 + extended data format
430 * - subcode 4-7 + extended data format
432 * Subcode 5 is used to retrieve the size of the data, provided by subcodes
433 * 6 and 7. Subcode 7 basically has the same function as subcode 6. In addition
434 * to subcode 6 it provides also information about secondary cpus.
435 * In order to get as much information as possible, we first try
436 * subcode 7, then 6 and if both fail, we use subcode 4.
439 static int diag204_probe(void)
444 buf
= diag204_get_buffer(INFO_EXT
, &pages
);
446 if (diag204((unsigned long)SUBC_STIB7
|
447 (unsigned long)INFO_EXT
, pages
, buf
) >= 0) {
448 diag204_store_sc
= SUBC_STIB7
;
449 diag204_info_type
= INFO_EXT
;
452 if (diag204((unsigned long)SUBC_STIB6
|
453 (unsigned long)INFO_EXT
, pages
, buf
) >= 0) {
454 diag204_store_sc
= SUBC_STIB6
;
455 diag204_info_type
= INFO_EXT
;
458 diag204_free_buffer();
461 /* subcodes 6 and 7 failed, now try subcode 4 */
463 buf
= diag204_get_buffer(INFO_SIMPLE
, &pages
);
468 if (diag204((unsigned long)SUBC_STIB4
|
469 (unsigned long)INFO_SIMPLE
, pages
, buf
) >= 0) {
470 diag204_store_sc
= SUBC_STIB4
;
471 diag204_info_type
= INFO_SIMPLE
;
480 diag204_free_buffer();
485 static int diag204_do_store(void *buf
, int pages
)
489 rc
= diag204((unsigned long) diag204_store_sc
|
490 (unsigned long) diag204_info_type
, pages
, buf
);
491 return rc
< 0 ? -ENOSYS
: 0;
494 static void *diag204_store(void)
499 buf
= diag204_get_buffer(diag204_info_type
, &pages
);
502 rc
= diag204_do_store(buf
, pages
);
509 /* Diagnose 224 functions */
511 static int diag224(void *ptr
)
513 int rc
= -EOPNOTSUPP
;
515 diag_stat_inc(DIAG_STAT_X224
);
517 " diag %1,%2,0x224\n"
521 : "+d" (rc
) :"d" (0), "d" (ptr
) : "memory");
525 static int diag224_get_name_table(void)
527 /* memory must be below 2GB */
528 diag224_cpu_names
= kmalloc(PAGE_SIZE
, GFP_KERNEL
| GFP_DMA
);
529 if (!diag224_cpu_names
)
531 if (diag224(diag224_cpu_names
)) {
532 kfree(diag224_cpu_names
);
535 EBCASC(diag224_cpu_names
+ 16, (*diag224_cpu_names
+ 1) * 16);
539 static void diag224_delete_name_table(void)
541 kfree(diag224_cpu_names
);
544 static int diag224_idx2name(int index
, char *name
)
546 memcpy(name
, diag224_cpu_names
+ ((index
+ 1) * CPU_NAME_LEN
),
548 name
[CPU_NAME_LEN
] = 0;
553 struct dbfs_d204_hdr
{
554 u64 len
; /* Length of d204 buffer without header */
555 u16 version
; /* Version of header */
556 u8 sc
; /* Used subcode */
558 } __attribute__ ((packed
));
561 struct dbfs_d204_hdr hdr
; /* 64 byte header */
562 char buf
[]; /* d204 buffer */
563 } __attribute__ ((packed
));
565 static int dbfs_d204_create(void **data
, void **data_free_ptr
, size_t *size
)
567 struct dbfs_d204
*d204
;
571 buf_size
= PAGE_SIZE
* (diag204_buf_pages
+ 1) + sizeof(d204
->hdr
);
572 base
= vzalloc(buf_size
);
575 d204
= page_align_ptr(base
+ sizeof(d204
->hdr
)) - sizeof(d204
->hdr
);
576 rc
= diag204_do_store(d204
->buf
, diag204_buf_pages
);
581 d204
->hdr
.version
= DBFS_D204_HDR_VERSION
;
582 d204
->hdr
.len
= PAGE_SIZE
* diag204_buf_pages
;
583 d204
->hdr
.sc
= diag204_store_sc
;
585 *data_free_ptr
= base
;
586 *size
= d204
->hdr
.len
+ sizeof(struct dbfs_d204_hdr
);
590 static struct hypfs_dbfs_file dbfs_file_d204
= {
592 .data_create
= dbfs_d204_create
,
596 __init
int hypfs_diag_init(void)
600 if (diag204_probe()) {
601 pr_err("The hardware system does not support hypfs\n");
604 if (diag204_info_type
== INFO_EXT
) {
605 rc
= hypfs_dbfs_create_file(&dbfs_file_d204
);
609 if (MACHINE_IS_LPAR
) {
610 rc
= diag224_get_name_table();
612 pr_err("The hardware system does not provide all "
613 "functions required by hypfs\n");
614 debugfs_remove(dbfs_d204_file
);
621 void hypfs_diag_exit(void)
623 debugfs_remove(dbfs_d204_file
);
624 diag224_delete_name_table();
625 diag204_free_buffer();
626 hypfs_dbfs_remove_file(&dbfs_file_d204
);
630 * Functions to create the directory structure
631 * *******************************************
634 static int hypfs_create_cpu_files(struct dentry
*cpus_dir
, void *cpu_info
)
636 struct dentry
*cpu_dir
;
637 char buffer
[TMP_SIZE
];
640 snprintf(buffer
, TMP_SIZE
, "%d", cpu_info__cpu_addr(diag204_info_type
,
642 cpu_dir
= hypfs_mkdir(cpus_dir
, buffer
);
643 rc
= hypfs_create_u64(cpu_dir
, "mgmtime",
644 cpu_info__acc_time(diag204_info_type
, cpu_info
) -
645 cpu_info__lp_time(diag204_info_type
, cpu_info
));
648 rc
= hypfs_create_u64(cpu_dir
, "cputime",
649 cpu_info__lp_time(diag204_info_type
, cpu_info
));
652 if (diag204_info_type
== INFO_EXT
) {
653 rc
= hypfs_create_u64(cpu_dir
, "onlinetime",
654 cpu_info__online_time(diag204_info_type
,
659 diag224_idx2name(cpu_info__ctidx(diag204_info_type
, cpu_info
), buffer
);
660 rc
= hypfs_create_str(cpu_dir
, "type", buffer
);
664 static void *hypfs_create_lpar_files(struct dentry
*systems_dir
, void *part_hdr
)
666 struct dentry
*cpus_dir
;
667 struct dentry
*lpar_dir
;
668 char lpar_name
[LPAR_NAME_LEN
+ 1];
672 part_hdr__part_name(diag204_info_type
, part_hdr
, lpar_name
);
673 lpar_name
[LPAR_NAME_LEN
] = 0;
674 lpar_dir
= hypfs_mkdir(systems_dir
, lpar_name
);
675 if (IS_ERR(lpar_dir
))
677 cpus_dir
= hypfs_mkdir(lpar_dir
, "cpus");
678 if (IS_ERR(cpus_dir
))
680 cpu_info
= part_hdr
+ part_hdr__size(diag204_info_type
);
681 for (i
= 0; i
< part_hdr__rcpus(diag204_info_type
, part_hdr
); i
++) {
683 rc
= hypfs_create_cpu_files(cpus_dir
, cpu_info
);
686 cpu_info
+= cpu_info__size(diag204_info_type
);
691 static int hypfs_create_phys_cpu_files(struct dentry
*cpus_dir
, void *cpu_info
)
693 struct dentry
*cpu_dir
;
694 char buffer
[TMP_SIZE
];
697 snprintf(buffer
, TMP_SIZE
, "%i", phys_cpu__cpu_addr(diag204_info_type
,
699 cpu_dir
= hypfs_mkdir(cpus_dir
, buffer
);
701 return PTR_ERR(cpu_dir
);
702 rc
= hypfs_create_u64(cpu_dir
, "mgmtime",
703 phys_cpu__mgm_time(diag204_info_type
, cpu_info
));
706 diag224_idx2name(phys_cpu__ctidx(diag204_info_type
, cpu_info
), buffer
);
707 rc
= hypfs_create_str(cpu_dir
, "type", buffer
);
711 static void *hypfs_create_phys_files(struct dentry
*parent_dir
, void *phys_hdr
)
715 struct dentry
*cpus_dir
;
717 cpus_dir
= hypfs_mkdir(parent_dir
, "cpus");
718 if (IS_ERR(cpus_dir
))
720 cpu_info
= phys_hdr
+ phys_hdr__size(diag204_info_type
);
721 for (i
= 0; i
< phys_hdr__cpus(diag204_info_type
, phys_hdr
); i
++) {
723 rc
= hypfs_create_phys_cpu_files(cpus_dir
, cpu_info
);
726 cpu_info
+= phys_cpu__size(diag204_info_type
);
731 int hypfs_diag_create_files(struct dentry
*root
)
733 struct dentry
*systems_dir
, *hyp_dir
;
734 void *time_hdr
, *part_hdr
;
738 buffer
= diag204_store();
740 return PTR_ERR(buffer
);
742 systems_dir
= hypfs_mkdir(root
, "systems");
743 if (IS_ERR(systems_dir
)) {
744 rc
= PTR_ERR(systems_dir
);
747 time_hdr
= (struct x_info_blk_hdr
*)buffer
;
748 part_hdr
= time_hdr
+ info_blk_hdr__size(diag204_info_type
);
749 for (i
= 0; i
< info_blk_hdr__npar(diag204_info_type
, time_hdr
); i
++) {
750 part_hdr
= hypfs_create_lpar_files(systems_dir
, part_hdr
);
751 if (IS_ERR(part_hdr
)) {
752 rc
= PTR_ERR(part_hdr
);
756 if (info_blk_hdr__flags(diag204_info_type
, time_hdr
) & LPAR_PHYS_FLG
) {
757 ptr
= hypfs_create_phys_files(root
, part_hdr
);
763 hyp_dir
= hypfs_mkdir(root
, "hyp");
764 if (IS_ERR(hyp_dir
)) {
765 rc
= PTR_ERR(hyp_dir
);
768 ptr
= hypfs_create_str(hyp_dir
, "type", "LPAR Hypervisor");