2 * PowerPC64 LPAR Configuration Information Driver
4 * Dave Engebretsen engebret@us.ibm.com
5 * Copyright (c) 2003 Dave Engebretsen
6 * Will Schmidt willschm@us.ibm.com
7 * SPLPAR updates, Copyright (c) 2003 Will Schmidt IBM Corporation.
8 * seq_file updates, Copyright (c) 2004 Will Schmidt IBM Corporation.
9 * Nathan Lynch nathanl@austin.ibm.com
10 * Added lparcfg_write, Copyright (C) 2004 Nathan Lynch IBM Corporation.
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * This driver creates a proc file at /proc/ppc64/lparcfg which contains
18 * keyword - value pairs that specify the configuration of the partition.
21 #include <linux/module.h>
22 #include <linux/types.h>
23 #include <linux/errno.h>
24 #include <linux/proc_fs.h>
25 #include <linux/init.h>
26 #include <linux/seq_file.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <asm/iseries/hv_lp_config.h>
30 #include <asm/lppaca.h>
31 #include <asm/hvcall.h>
32 #include <asm/firmware.h>
34 #include <asm/system.h>
37 #include <asm/vdso_datapage.h>
40 #include <asm/machdep.h>
44 * This isn't a module but we expose that to userspace
45 * via /proc so leave the definitions here
47 #define MODULE_VERS "1.9"
48 #define MODULE_NAME "lparcfg"
50 /* #define LPARCFG_DEBUG */
52 static struct proc_dir_entry *proc_ppc64_lparcfg;
55 * Track sum of all purrs across all processors. This is used to further
56 * calculate usage values by different applications
58 static unsigned long get_purr(void)
60 unsigned long sum_purr = 0;
63 for_each_possible_cpu(cpu) {
64 if (firmware_has_feature(FW_FEATURE_ISERIES))
65 sum_purr += lppaca_of(cpu).emulated_time_base;
69 cu = &per_cpu(cpu_usage_array, cpu);
70 sum_purr += cu->current_tb;
76 #ifdef CONFIG_PPC_ISERIES
79 * Methods used to fetch LPAR data when running on an iSeries platform.
81 static int iseries_lparcfg_data(struct seq_file *m, void *v)
83 unsigned long pool_id;
84 int shared, entitled_capacity, max_entitled_capacity;
85 int processors, max_processors;
86 unsigned long purr = get_purr();
88 shared = (int)(local_paca->lppaca_ptr->shared_proc);
90 seq_printf(m, "system_active_processors=%d\n",
91 (int)HvLpConfig_getSystemPhysicalProcessors());
93 seq_printf(m, "system_potential_processors=%d\n",
94 (int)HvLpConfig_getSystemPhysicalProcessors());
96 processors = (int)HvLpConfig_getPhysicalProcessors();
97 seq_printf(m, "partition_active_processors=%d\n", processors);
99 max_processors = (int)HvLpConfig_getMaxPhysicalProcessors();
100 seq_printf(m, "partition_potential_processors=%d\n", max_processors);
103 entitled_capacity = HvLpConfig_getSharedProcUnits();
104 max_entitled_capacity = HvLpConfig_getMaxSharedProcUnits();
106 entitled_capacity = processors * 100;
107 max_entitled_capacity = max_processors * 100;
109 seq_printf(m, "partition_entitled_capacity=%d\n", entitled_capacity);
111 seq_printf(m, "partition_max_entitled_capacity=%d\n",
112 max_entitled_capacity);
115 pool_id = HvLpConfig_getSharedPoolIndex();
116 seq_printf(m, "pool=%d\n", (int)pool_id);
117 seq_printf(m, "pool_capacity=%d\n",
118 (int)(HvLpConfig_getNumProcsInSharedPool(pool_id) *
120 seq_printf(m, "purr=%ld\n", purr);
123 seq_printf(m, "shared_processor_mode=%d\n", shared);
128 #else /* CONFIG_PPC_ISERIES */
130 static int iseries_lparcfg_data(struct seq_file *m, void *v)
135 #endif /* CONFIG_PPC_ISERIES */
137 #ifdef CONFIG_PPC_PSERIES
139 * Methods used to fetch LPAR data when running on a pSeries platform.
142 struct hvcall_ppp_data {
144 u64 unallocated_entitlement;
149 u8 unallocated_weight;
150 u16 active_procs_in_pool;
151 u16 active_system_procs;
152 u16 phys_platform_procs;
153 u32 max_proc_cap_avail;
154 u32 entitled_proc_cap_avail;
158 * H_GET_PPP hcall returns info in 4 parms.
159 * entitled_capacity,unallocated_capacity,
160 * aggregation, resource_capability).
162 * R4 = Entitled Processor Capacity Percentage.
163 * R5 = Unallocated Processor Capacity Percentage.
164 * R6 (AABBCCDDEEFFGGHH).
165 * XXXX - reserved (0)
166 * XXXX - reserved (0)
167 * XXXX - Group Number
168 * XXXX - Pool Number.
169 * R7 (IIJJKKLLMMNNOOPP).
171 * XX - bit 0-6 reserved (0). bit 7 is Capped indicator.
172 * XX - variable processor Capacity Weight
173 * XX - Unallocated Variable Processor Capacity Weight.
174 * XXXX - Active processors in Physical Processor Pool.
175 * XXXX - Processors active on platform.
176 * R8 (QQQQRRRRRRSSSSSS). if ibm,partition-performance-parameters-level >= 1
177 * XXXX - Physical platform procs allocated to virtualization.
178 * XXXXXX - Max procs capacity % available to the partitions pool.
179 * XXXXXX - Entitled procs capacity % available to the
182 static unsigned int h_get_ppp(struct hvcall_ppp_data *ppp_data)
185 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
187 rc = plpar_hcall9(H_GET_PPP, retbuf);
189 ppp_data->entitlement = retbuf[0];
190 ppp_data->unallocated_entitlement = retbuf[1];
192 ppp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
193 ppp_data->pool_num = retbuf[2] & 0xffff;
195 ppp_data->capped = (retbuf[3] >> 6 * 8) & 0x01;
196 ppp_data->weight = (retbuf[3] >> 5 * 8) & 0xff;
197 ppp_data->unallocated_weight = (retbuf[3] >> 4 * 8) & 0xff;
198 ppp_data->active_procs_in_pool = (retbuf[3] >> 2 * 8) & 0xffff;
199 ppp_data->active_system_procs = retbuf[3] & 0xffff;
201 ppp_data->phys_platform_procs = retbuf[4] >> 6 * 8;
202 ppp_data->max_proc_cap_avail = (retbuf[4] >> 3 * 8) & 0xffffff;
203 ppp_data->entitled_proc_cap_avail = retbuf[4] & 0xffffff;
208 static unsigned h_pic(unsigned long *pool_idle_time,
209 unsigned long *num_procs)
212 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
214 rc = plpar_hcall(H_PIC, retbuf);
216 *pool_idle_time = retbuf[0];
217 *num_procs = retbuf[1];
224 * Parse out the data returned from h_get_ppp and h_pic
226 static void parse_ppp_data(struct seq_file *m)
228 struct hvcall_ppp_data ppp_data;
229 struct device_node *root;
230 const int *perf_level;
233 rc = h_get_ppp(&ppp_data);
237 seq_printf(m, "partition_entitled_capacity=%lld\n",
238 ppp_data.entitlement);
239 seq_printf(m, "group=%d\n", ppp_data.group_num);
240 seq_printf(m, "system_active_processors=%d\n",
241 ppp_data.active_system_procs);
243 /* pool related entries are appropriate for shared configs */
244 if (lppaca_of(0).shared_proc) {
245 unsigned long pool_idle_time, pool_procs;
247 seq_printf(m, "pool=%d\n", ppp_data.pool_num);
249 /* report pool_capacity in percentage */
250 seq_printf(m, "pool_capacity=%d\n",
251 ppp_data.active_procs_in_pool * 100);
253 h_pic(&pool_idle_time, &pool_procs);
254 seq_printf(m, "pool_idle_time=%ld\n", pool_idle_time);
255 seq_printf(m, "pool_num_procs=%ld\n", pool_procs);
258 seq_printf(m, "unallocated_capacity_weight=%d\n",
259 ppp_data.unallocated_weight);
260 seq_printf(m, "capacity_weight=%d\n", ppp_data.weight);
261 seq_printf(m, "capped=%d\n", ppp_data.capped);
262 seq_printf(m, "unallocated_capacity=%lld\n",
263 ppp_data.unallocated_entitlement);
265 /* The last bits of information returned from h_get_ppp are only
266 * valid if the ibm,partition-performance-parameters-level
269 root = of_find_node_by_path("/");
271 perf_level = of_get_property(root,
272 "ibm,partition-performance-parameters-level",
274 if (perf_level && (*perf_level >= 1)) {
276 "physical_procs_allocated_to_virtualization=%d\n",
277 ppp_data.phys_platform_procs);
278 seq_printf(m, "max_proc_capacity_available=%d\n",
279 ppp_data.max_proc_cap_avail);
280 seq_printf(m, "entitled_proc_capacity_available=%d\n",
281 ppp_data.entitled_proc_cap_avail);
290 * Parse out data returned from h_get_mpp
292 static void parse_mpp_data(struct seq_file *m)
294 struct hvcall_mpp_data mpp_data;
297 rc = h_get_mpp(&mpp_data);
301 seq_printf(m, "entitled_memory=%ld\n", mpp_data.entitled_mem);
303 if (mpp_data.mapped_mem != -1)
304 seq_printf(m, "mapped_entitled_memory=%ld\n",
305 mpp_data.mapped_mem);
307 seq_printf(m, "entitled_memory_group_number=%d\n", mpp_data.group_num);
308 seq_printf(m, "entitled_memory_pool_number=%d\n", mpp_data.pool_num);
310 seq_printf(m, "entitled_memory_weight=%d\n", mpp_data.mem_weight);
311 seq_printf(m, "unallocated_entitled_memory_weight=%d\n",
312 mpp_data.unallocated_mem_weight);
313 seq_printf(m, "unallocated_io_mapping_entitlement=%ld\n",
314 mpp_data.unallocated_entitlement);
316 if (mpp_data.pool_size != -1)
317 seq_printf(m, "entitled_memory_pool_size=%ld bytes\n",
320 seq_printf(m, "entitled_memory_loan_request=%ld\n",
321 mpp_data.loan_request);
323 seq_printf(m, "backing_memory=%ld bytes\n", mpp_data.backing_mem);
328 * Parse out data returned from h_get_mpp_x
330 static void parse_mpp_x_data(struct seq_file *m)
332 struct hvcall_mpp_x_data mpp_x_data;
334 if (!firmware_has_feature(FW_FEATURE_XCMO))
336 if (h_get_mpp_x(&mpp_x_data))
339 seq_printf(m, "coalesced_bytes=%ld\n", mpp_x_data.coalesced_bytes);
341 if (mpp_x_data.pool_coalesced_bytes)
342 seq_printf(m, "pool_coalesced_bytes=%ld\n",
343 mpp_x_data.pool_coalesced_bytes);
344 if (mpp_x_data.pool_purr_cycles)
345 seq_printf(m, "coalesce_pool_purr=%ld\n", mpp_x_data.pool_purr_cycles);
346 if (mpp_x_data.pool_spurr_cycles)
347 seq_printf(m, "coalesce_pool_spurr=%ld\n", mpp_x_data.pool_spurr_cycles);
350 #define SPLPAR_CHARACTERISTICS_TOKEN 20
351 #define SPLPAR_MAXLENGTH 1026*(sizeof(char))
354 * parse_system_parameter_string()
355 * Retrieve the potential_processors, max_entitled_capacity and friends
356 * through the get-system-parameter rtas call. Replace keyword strings as
359 static void parse_system_parameter_string(struct seq_file *m)
363 unsigned char *local_buffer = kmalloc(SPLPAR_MAXLENGTH, GFP_KERNEL);
365 printk(KERN_ERR "%s %s kmalloc failure at line %d\n",
366 __FILE__, __func__, __LINE__);
370 spin_lock(&rtas_data_buf_lock);
371 memset(rtas_data_buf, 0, SPLPAR_MAXLENGTH);
372 call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
374 SPLPAR_CHARACTERISTICS_TOKEN,
377 memcpy(local_buffer, rtas_data_buf, SPLPAR_MAXLENGTH);
378 local_buffer[SPLPAR_MAXLENGTH - 1] = '\0';
379 spin_unlock(&rtas_data_buf_lock);
381 if (call_status != 0) {
383 "%s %s Error calling get-system-parameter (0x%x)\n",
384 __FILE__, __func__, call_status);
388 char *workbuffer = kzalloc(SPLPAR_MAXLENGTH, GFP_KERNEL);
390 printk(KERN_ERR "%s %s kmalloc failure at line %d\n",
391 __FILE__, __func__, __LINE__);
396 printk(KERN_INFO "success calling get-system-parameter\n");
398 splpar_strlen = local_buffer[0] * 256 + local_buffer[1];
399 local_buffer += 2; /* step over strlen value */
403 while ((*local_buffer) && (idx < splpar_strlen)) {
404 workbuffer[w_idx++] = local_buffer[idx++];
405 if ((local_buffer[idx] == ',')
406 || (local_buffer[idx] == '\0')) {
407 workbuffer[w_idx] = '\0';
409 /* avoid the empty string */
410 seq_printf(m, "%s\n", workbuffer);
412 memset(workbuffer, 0, SPLPAR_MAXLENGTH);
413 idx++; /* skip the comma */
415 } else if (local_buffer[idx] == '=') {
416 /* code here to replace workbuffer contents
417 with different keyword strings */
418 if (0 == strcmp(workbuffer, "MaxEntCap")) {
420 "partition_max_entitled_capacity");
421 w_idx = strlen(workbuffer);
423 if (0 == strcmp(workbuffer, "MaxPlatProcs")) {
425 "system_potential_processors");
426 w_idx = strlen(workbuffer);
431 local_buffer -= 2; /* back up over strlen value */
436 /* Return the number of processors in the system.
437 * This function reads through the device tree and counts
438 * the virtual processors, this does not include threads.
440 static int lparcfg_count_active_processors(void)
442 struct device_node *cpus_dn = NULL;
445 while ((cpus_dn = of_find_node_by_type(cpus_dn, "cpu"))) {
447 printk(KERN_ERR "cpus_dn %p\n", cpus_dn);
454 static void pseries_cmo_data(struct seq_file *m)
457 unsigned long cmo_faults = 0;
458 unsigned long cmo_fault_time = 0;
460 seq_printf(m, "cmo_enabled=%d\n", firmware_has_feature(FW_FEATURE_CMO));
462 if (!firmware_has_feature(FW_FEATURE_CMO))
465 for_each_possible_cpu(cpu) {
466 cmo_faults += lppaca_of(cpu).cmo_faults;
467 cmo_fault_time += lppaca_of(cpu).cmo_fault_time;
470 seq_printf(m, "cmo_faults=%lu\n", cmo_faults);
471 seq_printf(m, "cmo_fault_time_usec=%lu\n",
472 cmo_fault_time / tb_ticks_per_usec);
473 seq_printf(m, "cmo_primary_psp=%d\n", cmo_get_primary_psp());
474 seq_printf(m, "cmo_secondary_psp=%d\n", cmo_get_secondary_psp());
475 seq_printf(m, "cmo_page_size=%lu\n", cmo_get_page_size());
478 static void splpar_dispatch_data(struct seq_file *m)
481 unsigned long dispatches = 0;
482 unsigned long dispatch_dispersions = 0;
484 for_each_possible_cpu(cpu) {
485 dispatches += lppaca_of(cpu).yield_count;
486 dispatch_dispersions += lppaca_of(cpu).dispersion_count;
489 seq_printf(m, "dispatches=%lu\n", dispatches);
490 seq_printf(m, "dispatch_dispersions=%lu\n", dispatch_dispersions);
493 static void parse_em_data(struct seq_file *m)
495 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
497 if (firmware_has_feature(FW_FEATURE_LPAR) &&
498 plpar_hcall(H_GET_EM_PARMS, retbuf) == H_SUCCESS)
499 seq_printf(m, "power_mode_data=%016lx\n", retbuf[0]);
502 static int pseries_lparcfg_data(struct seq_file *m, void *v)
504 int partition_potential_processors;
505 int partition_active_processors;
506 struct device_node *rtas_node;
507 const int *lrdrp = NULL;
509 rtas_node = of_find_node_by_path("/rtas");
511 lrdrp = of_get_property(rtas_node, "ibm,lrdr-capacity", NULL);
514 partition_potential_processors = vdso_data->processorCount;
516 partition_potential_processors = *(lrdrp + 4);
518 of_node_put(rtas_node);
520 partition_active_processors = lparcfg_count_active_processors();
522 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
523 /* this call handles the ibm,get-system-parameter contents */
524 parse_system_parameter_string(m);
529 splpar_dispatch_data(m);
531 seq_printf(m, "purr=%ld\n", get_purr());
532 } else { /* non SPLPAR case */
534 seq_printf(m, "system_active_processors=%d\n",
535 partition_potential_processors);
537 seq_printf(m, "system_potential_processors=%d\n",
538 partition_potential_processors);
540 seq_printf(m, "partition_max_entitled_capacity=%d\n",
541 partition_potential_processors * 100);
543 seq_printf(m, "partition_entitled_capacity=%d\n",
544 partition_active_processors * 100);
547 seq_printf(m, "partition_active_processors=%d\n",
548 partition_active_processors);
550 seq_printf(m, "partition_potential_processors=%d\n",
551 partition_potential_processors);
553 seq_printf(m, "shared_processor_mode=%d\n", lppaca_of(0).shared_proc);
555 seq_printf(m, "slb_size=%d\n", mmu_slb_size);
562 static ssize_t update_ppp(u64 *entitlement, u8 *weight)
564 struct hvcall_ppp_data ppp_data;
569 /* Get our current parameters */
570 retval = h_get_ppp(&ppp_data);
575 new_weight = ppp_data.weight;
576 new_entitled = *entitlement;
578 new_weight = *weight;
579 new_entitled = ppp_data.entitlement;
583 pr_debug("%s: current_entitled = %llu, current_weight = %u\n",
584 __func__, ppp_data.entitlement, ppp_data.weight);
586 pr_debug("%s: new_entitled = %llu, new_weight = %u\n",
587 __func__, new_entitled, new_weight);
589 retval = plpar_hcall_norets(H_SET_PPP, new_entitled, new_weight);
596 * Update the memory entitlement and weight for the partition. Caller must
597 * specify either a new entitlement or weight, not both, to be updated
598 * since the h_set_mpp call takes both entitlement and weight as parameters.
600 static ssize_t update_mpp(u64 *entitlement, u8 *weight)
602 struct hvcall_mpp_data mpp_data;
608 /* Check with vio to ensure the new memory entitlement
611 rc = vio_cmo_entitlement_update(*entitlement);
616 rc = h_get_mpp(&mpp_data);
621 new_weight = mpp_data.mem_weight;
622 new_entitled = *entitlement;
624 new_weight = *weight;
625 new_entitled = mpp_data.entitled_mem;
629 pr_debug("%s: current_entitled = %lu, current_weight = %u\n",
630 __func__, mpp_data.entitled_mem, mpp_data.mem_weight);
632 pr_debug("%s: new_entitled = %llu, new_weight = %u\n",
633 __func__, new_entitled, new_weight);
635 rc = plpar_hcall_norets(H_SET_MPP, new_entitled, new_weight);
640 * Interface for changing system parameters (variable capacity weight
641 * and entitled capacity). Format of input is "param_name=value";
642 * anything after value is ignored. Valid parameters at this time are
643 * "partition_entitled_capacity" and "capacity_weight". We use
644 * H_SET_PPP to alter parameters.
646 * This function should be invoked only on systems with
649 static ssize_t lparcfg_write(struct file *file, const char __user * buf,
650 size_t count, loff_t * off)
655 u64 new_entitled, *new_entitled_ptr = &new_entitled;
656 u8 new_weight, *new_weight_ptr = &new_weight;
659 if (!firmware_has_feature(FW_FEATURE_SPLPAR) ||
660 firmware_has_feature(FW_FEATURE_ISERIES))
666 if (copy_from_user(kbuf, buf, count))
669 kbuf[count - 1] = '\0';
670 tmp = strchr(kbuf, '=');
676 if (!strcmp(kbuf, "partition_entitled_capacity")) {
678 *new_entitled_ptr = (u64) simple_strtoul(tmp, &endp, 10);
682 retval = update_ppp(new_entitled_ptr, NULL);
683 } else if (!strcmp(kbuf, "capacity_weight")) {
685 *new_weight_ptr = (u8) simple_strtoul(tmp, &endp, 10);
689 retval = update_ppp(NULL, new_weight_ptr);
690 } else if (!strcmp(kbuf, "entitled_memory")) {
692 *new_entitled_ptr = (u64) simple_strtoul(tmp, &endp, 10);
696 retval = update_mpp(new_entitled_ptr, NULL);
697 } else if (!strcmp(kbuf, "entitled_memory_weight")) {
699 *new_weight_ptr = (u8) simple_strtoul(tmp, &endp, 10);
703 retval = update_mpp(NULL, new_weight_ptr);
707 if (retval == H_SUCCESS || retval == H_CONSTRAINED) {
709 } else if (retval == H_BUSY) {
711 } else if (retval == H_HARDWARE) {
713 } else if (retval == H_PARAMETER) {
720 #else /* CONFIG_PPC_PSERIES */
722 static int pseries_lparcfg_data(struct seq_file *m, void *v)
727 static ssize_t lparcfg_write(struct file *file, const char __user * buf,
728 size_t count, loff_t * off)
733 #endif /* CONFIG_PPC_PSERIES */
735 static int lparcfg_data(struct seq_file *m, void *v)
737 struct device_node *rootdn;
738 const char *model = "";
739 const char *system_id = "";
741 const unsigned int *lp_index_ptr;
742 unsigned int lp_index = 0;
744 seq_printf(m, "%s %s\n", MODULE_NAME, MODULE_VERS);
746 rootdn = of_find_node_by_path("/");
748 tmp = of_get_property(rootdn, "model", NULL);
751 /* Skip "IBM," - see platforms/iseries/dt.c */
752 if (firmware_has_feature(FW_FEATURE_ISERIES))
755 tmp = of_get_property(rootdn, "system-id", NULL);
758 /* Skip "IBM," - see platforms/iseries/dt.c */
759 if (firmware_has_feature(FW_FEATURE_ISERIES))
762 lp_index_ptr = of_get_property(rootdn, "ibm,partition-no",
765 lp_index = *lp_index_ptr;
768 seq_printf(m, "serial_number=%s\n", system_id);
769 seq_printf(m, "system_type=%s\n", model);
770 seq_printf(m, "partition_id=%d\n", (int)lp_index);
772 if (firmware_has_feature(FW_FEATURE_ISERIES))
773 return iseries_lparcfg_data(m, v);
774 return pseries_lparcfg_data(m, v);
777 static int lparcfg_open(struct inode *inode, struct file *file)
779 return single_open(file, lparcfg_data, NULL);
782 static const struct file_operations lparcfg_fops = {
784 .write = lparcfg_write,
785 .open = lparcfg_open,
786 .release = single_release,
790 static int __init lparcfg_init(void)
792 struct proc_dir_entry *ent;
793 mode_t mode = S_IRUSR | S_IRGRP | S_IROTH;
795 /* Allow writing if we have FW_FEATURE_SPLPAR */
796 if (firmware_has_feature(FW_FEATURE_SPLPAR) &&
797 !firmware_has_feature(FW_FEATURE_ISERIES))
800 ent = proc_create("powerpc/lparcfg", mode, NULL, &lparcfg_fops);
802 printk(KERN_ERR "Failed to create powerpc/lparcfg\n");
806 proc_ppc64_lparcfg = ent;
809 machine_device_initcall(pseries, lparcfg_init);