2 * File: arch/blackfin/mm/sram-alloc.c
7 * Description: SRAM allocator for Blackfin L1 and L2 memory
10 * Copyright 2004-2008 Analog Devices Inc.
12 * Bugs: Enter bugs at http://blackfin.uclinux.org/
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License as published by
16 * the Free Software Foundation; either version 2 of the License, or
17 * (at your option) any later version.
19 * This program is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with this program; if not, see the file COPYING, or write
26 * to the Free Software Foundation, Inc.,
27 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
30 #include <linux/module.h>
31 #include <linux/kernel.h>
32 #include <linux/types.h>
33 #include <linux/miscdevice.h>
34 #include <linux/ioport.h>
35 #include <linux/fcntl.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/proc_fs.h>
39 #include <linux/spinlock.h>
40 #include <linux/rtc.h>
41 #include <asm/blackfin.h>
42 #include <asm/mem_map.h>
43 #include "blackfin_sram.h"
45 static DEFINE_PER_CPU(spinlock_t, l1sram_lock) ____cacheline_aligned_in_smp;
46 static DEFINE_PER_CPU(spinlock_t, l1_data_sram_lock) ____cacheline_aligned_in_smp;
47 static DEFINE_PER_CPU(spinlock_t, l1_inst_sram_lock) ____cacheline_aligned_in_smp;
48 static spinlock_t l2_sram_lock ____cacheline_aligned_in_smp;
50 /* the data structure for L1 scratchpad and DATA SRAM */
55 struct sram_piece *next;
58 static DEFINE_PER_CPU(struct sram_piece, free_l1_ssram_head);
59 static DEFINE_PER_CPU(struct sram_piece, used_l1_ssram_head);
61 #if L1_DATA_A_LENGTH != 0
62 static DEFINE_PER_CPU(struct sram_piece, free_l1_data_A_sram_head);
63 static DEFINE_PER_CPU(struct sram_piece, used_l1_data_A_sram_head);
66 #if L1_DATA_B_LENGTH != 0
67 static DEFINE_PER_CPU(struct sram_piece, free_l1_data_B_sram_head);
68 static DEFINE_PER_CPU(struct sram_piece, used_l1_data_B_sram_head);
71 #if L1_CODE_LENGTH != 0
72 static DEFINE_PER_CPU(struct sram_piece, free_l1_inst_sram_head);
73 static DEFINE_PER_CPU(struct sram_piece, used_l1_inst_sram_head);
77 static struct sram_piece free_l2_sram_head, used_l2_sram_head;
80 static struct kmem_cache *sram_piece_cache;
82 /* L1 Scratchpad SRAM initialization function */
83 static void __init l1sram_init(void)
86 for (cpu = 0; cpu < num_possible_cpus(); ++cpu) {
87 per_cpu(free_l1_ssram_head, cpu).next =
88 kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
89 if (!per_cpu(free_l1_ssram_head, cpu).next) {
90 printk(KERN_INFO "Fail to initialize Scratchpad data SRAM.\n");
94 per_cpu(free_l1_ssram_head, cpu).next->paddr = (void *)get_l1_scratch_start_cpu(cpu);
95 per_cpu(free_l1_ssram_head, cpu).next->size = L1_SCRATCH_LENGTH;
96 per_cpu(free_l1_ssram_head, cpu).next->pid = 0;
97 per_cpu(free_l1_ssram_head, cpu).next->next = NULL;
99 per_cpu(used_l1_ssram_head, cpu).next = NULL;
101 /* mutex initialize */
102 spin_lock_init(&per_cpu(l1sram_lock, cpu));
103 printk(KERN_INFO "Blackfin Scratchpad data SRAM: %d KB\n",
104 L1_SCRATCH_LENGTH >> 10);
108 static void __init l1_data_sram_init(void)
110 #if L1_DATA_A_LENGTH != 0 || L1_DATA_B_LENGTH != 0
113 #if L1_DATA_A_LENGTH != 0
114 for (cpu = 0; cpu < num_possible_cpus(); ++cpu) {
115 per_cpu(free_l1_data_A_sram_head, cpu).next =
116 kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
117 if (!per_cpu(free_l1_data_A_sram_head, cpu).next) {
118 printk(KERN_INFO "Fail to initialize L1 Data A SRAM.\n");
122 per_cpu(free_l1_data_A_sram_head, cpu).next->paddr =
123 (void *)get_l1_data_a_start_cpu(cpu) + (_ebss_l1 - _sdata_l1);
124 per_cpu(free_l1_data_A_sram_head, cpu).next->size =
125 L1_DATA_A_LENGTH - (_ebss_l1 - _sdata_l1);
126 per_cpu(free_l1_data_A_sram_head, cpu).next->pid = 0;
127 per_cpu(free_l1_data_A_sram_head, cpu).next->next = NULL;
129 per_cpu(used_l1_data_A_sram_head, cpu).next = NULL;
131 printk(KERN_INFO "Blackfin L1 Data A SRAM: %d KB (%d KB free)\n",
132 L1_DATA_A_LENGTH >> 10,
133 per_cpu(free_l1_data_A_sram_head, cpu).next->size >> 10);
136 #if L1_DATA_B_LENGTH != 0
137 for (cpu = 0; cpu < num_possible_cpus(); ++cpu) {
138 per_cpu(free_l1_data_B_sram_head, cpu).next =
139 kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
140 if (!per_cpu(free_l1_data_B_sram_head, cpu).next) {
141 printk(KERN_INFO "Fail to initialize L1 Data B SRAM.\n");
145 per_cpu(free_l1_data_B_sram_head, cpu).next->paddr =
146 (void *)get_l1_data_b_start_cpu(cpu) + (_ebss_b_l1 - _sdata_b_l1);
147 per_cpu(free_l1_data_B_sram_head, cpu).next->size =
148 L1_DATA_B_LENGTH - (_ebss_b_l1 - _sdata_b_l1);
149 per_cpu(free_l1_data_B_sram_head, cpu).next->pid = 0;
150 per_cpu(free_l1_data_B_sram_head, cpu).next->next = NULL;
152 per_cpu(used_l1_data_B_sram_head, cpu).next = NULL;
154 printk(KERN_INFO "Blackfin L1 Data B SRAM: %d KB (%d KB free)\n",
155 L1_DATA_B_LENGTH >> 10,
156 per_cpu(free_l1_data_B_sram_head, cpu).next->size >> 10);
157 /* mutex initialize */
161 #if L1_DATA_A_LENGTH != 0 || L1_DATA_B_LENGTH != 0
162 for (cpu = 0; cpu < num_possible_cpus(); ++cpu)
163 spin_lock_init(&per_cpu(l1_data_sram_lock, cpu));
167 static void __init l1_inst_sram_init(void)
169 #if L1_CODE_LENGTH != 0
171 for (cpu = 0; cpu < num_possible_cpus(); ++cpu) {
172 per_cpu(free_l1_inst_sram_head, cpu).next =
173 kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
174 if (!per_cpu(free_l1_inst_sram_head, cpu).next) {
175 printk(KERN_INFO "Failed to initialize L1 Instruction SRAM\n");
179 per_cpu(free_l1_inst_sram_head, cpu).next->paddr =
180 (void *)get_l1_code_start_cpu(cpu) + (_etext_l1 - _stext_l1);
181 per_cpu(free_l1_inst_sram_head, cpu).next->size =
182 L1_CODE_LENGTH - (_etext_l1 - _stext_l1);
183 per_cpu(free_l1_inst_sram_head, cpu).next->pid = 0;
184 per_cpu(free_l1_inst_sram_head, cpu).next->next = NULL;
186 per_cpu(used_l1_inst_sram_head, cpu).next = NULL;
188 printk(KERN_INFO "Blackfin L1 Instruction SRAM: %d KB (%d KB free)\n",
189 L1_CODE_LENGTH >> 10,
190 per_cpu(free_l1_inst_sram_head, cpu).next->size >> 10);
192 /* mutex initialize */
193 spin_lock_init(&per_cpu(l1_inst_sram_lock, cpu));
198 static void __init l2_sram_init(void)
201 free_l2_sram_head.next =
202 kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
203 if (!free_l2_sram_head.next) {
204 printk(KERN_INFO "Fail to initialize L2 SRAM.\n");
208 free_l2_sram_head.next->paddr =
209 (void *)L2_START + (_ebss_l2 - _stext_l2);
210 free_l2_sram_head.next->size =
211 L2_LENGTH - (_ebss_l2 - _stext_l2);
212 free_l2_sram_head.next->pid = 0;
213 free_l2_sram_head.next->next = NULL;
215 used_l2_sram_head.next = NULL;
217 printk(KERN_INFO "Blackfin L2 SRAM: %d KB (%d KB free)\n",
219 free_l2_sram_head.next->size >> 10);
222 /* mutex initialize */
223 spin_lock_init(&l2_sram_lock);
226 void __init bfin_sram_init(void)
228 sram_piece_cache = kmem_cache_create("sram_piece_cache",
229 sizeof(struct sram_piece),
230 0, SLAB_PANIC, NULL);
238 /* SRAM allocate function */
239 static void *_sram_alloc(size_t size, struct sram_piece *pfree_head,
240 struct sram_piece *pused_head)
242 struct sram_piece *pslot, *plast, *pavail;
244 if (size <= 0 || !pfree_head || !pused_head)
248 size = (size + 3) & ~3;
250 pslot = pfree_head->next;
253 /* search an available piece slot */
254 while (pslot != NULL && size > pslot->size) {
262 if (pslot->size == size) {
263 plast->next = pslot->next;
266 pavail = kmem_cache_alloc(sram_piece_cache, GFP_KERNEL);
271 pavail->paddr = pslot->paddr;
273 pslot->paddr += size;
277 pavail->pid = current->pid;
279 pslot = pused_head->next;
282 /* insert new piece into used piece list !!! */
283 while (pslot != NULL && pavail->paddr < pslot->paddr) {
288 pavail->next = pslot;
289 plast->next = pavail;
291 return pavail->paddr;
294 /* Allocate the largest available block. */
295 static void *_sram_alloc_max(struct sram_piece *pfree_head,
296 struct sram_piece *pused_head,
297 unsigned long *psize)
299 struct sram_piece *pslot, *pmax;
301 if (!pfree_head || !pused_head)
304 pmax = pslot = pfree_head->next;
306 /* search an available piece slot */
307 while (pslot != NULL) {
308 if (pslot->size > pmax->size)
318 return _sram_alloc(*psize, pfree_head, pused_head);
321 /* SRAM free function */
322 static int _sram_free(const void *addr,
323 struct sram_piece *pfree_head,
324 struct sram_piece *pused_head)
326 struct sram_piece *pslot, *plast, *pavail;
328 if (!pfree_head || !pused_head)
331 /* search the relevant memory slot */
332 pslot = pused_head->next;
335 /* search an available piece slot */
336 while (pslot != NULL && pslot->paddr != addr) {
344 plast->next = pslot->next;
348 /* insert free pieces back to the free list */
349 pslot = pfree_head->next;
352 while (pslot != NULL && addr > pslot->paddr) {
357 if (plast != pfree_head && plast->paddr + plast->size == pavail->paddr) {
358 plast->size += pavail->size;
359 kmem_cache_free(sram_piece_cache, pavail);
361 pavail->next = plast->next;
362 plast->next = pavail;
366 if (pslot && plast->paddr + plast->size == pslot->paddr) {
367 plast->size += pslot->size;
368 plast->next = pslot->next;
369 kmem_cache_free(sram_piece_cache, pslot);
375 int sram_free(const void *addr)
378 #if L1_CODE_LENGTH != 0
379 if (addr >= (void *)get_l1_code_start()
380 && addr < (void *)(get_l1_code_start() + L1_CODE_LENGTH))
381 return l1_inst_sram_free(addr);
384 #if L1_DATA_A_LENGTH != 0
385 if (addr >= (void *)get_l1_data_a_start()
386 && addr < (void *)(get_l1_data_a_start() + L1_DATA_A_LENGTH))
387 return l1_data_A_sram_free(addr);
390 #if L1_DATA_B_LENGTH != 0
391 if (addr >= (void *)get_l1_data_b_start()
392 && addr < (void *)(get_l1_data_b_start() + L1_DATA_B_LENGTH))
393 return l1_data_B_sram_free(addr);
397 if (addr >= (void *)L2_START
398 && addr < (void *)(L2_START + L2_LENGTH))
399 return l2_sram_free(addr);
404 EXPORT_SYMBOL(sram_free);
406 void *l1_data_A_sram_alloc(size_t size)
413 /* add mutex operation */
414 spin_lock_irqsave(&per_cpu(l1_data_sram_lock, cpu), flags);
416 #if L1_DATA_A_LENGTH != 0
417 addr = _sram_alloc(size, &per_cpu(free_l1_data_A_sram_head, cpu),
418 &per_cpu(used_l1_data_A_sram_head, cpu));
421 /* add mutex operation */
422 spin_unlock_irqrestore(&per_cpu(l1_data_sram_lock, cpu), flags);
425 pr_debug("Allocated address in l1_data_A_sram_alloc is 0x%lx+0x%lx\n",
426 (long unsigned int)addr, size);
430 EXPORT_SYMBOL(l1_data_A_sram_alloc);
432 int l1_data_A_sram_free(const void *addr)
439 /* add mutex operation */
440 spin_lock_irqsave(&per_cpu(l1_data_sram_lock, cpu), flags);
442 #if L1_DATA_A_LENGTH != 0
443 ret = _sram_free(addr, &per_cpu(free_l1_data_A_sram_head, cpu),
444 &per_cpu(used_l1_data_A_sram_head, cpu));
449 /* add mutex operation */
450 spin_unlock_irqrestore(&per_cpu(l1_data_sram_lock, cpu), flags);
455 EXPORT_SYMBOL(l1_data_A_sram_free);
457 void *l1_data_B_sram_alloc(size_t size)
459 #if L1_DATA_B_LENGTH != 0
465 /* add mutex operation */
466 spin_lock_irqsave(&per_cpu(l1_data_sram_lock, cpu), flags);
468 addr = _sram_alloc(size, &per_cpu(free_l1_data_B_sram_head, cpu),
469 &per_cpu(used_l1_data_B_sram_head, cpu));
471 /* add mutex operation */
472 spin_unlock_irqrestore(&per_cpu(l1_data_sram_lock, cpu), flags);
475 pr_debug("Allocated address in l1_data_B_sram_alloc is 0x%lx+0x%lx\n",
476 (long unsigned int)addr, size);
483 EXPORT_SYMBOL(l1_data_B_sram_alloc);
485 int l1_data_B_sram_free(const void *addr)
487 #if L1_DATA_B_LENGTH != 0
493 /* add mutex operation */
494 spin_lock_irqsave(&per_cpu(l1_data_sram_lock, cpu), flags);
496 ret = _sram_free(addr, &per_cpu(free_l1_data_B_sram_head, cpu),
497 &per_cpu(used_l1_data_B_sram_head, cpu));
499 /* add mutex operation */
500 spin_unlock_irqrestore(&per_cpu(l1_data_sram_lock, cpu), flags);
508 EXPORT_SYMBOL(l1_data_B_sram_free);
510 void *l1_data_sram_alloc(size_t size)
512 void *addr = l1_data_A_sram_alloc(size);
515 addr = l1_data_B_sram_alloc(size);
519 EXPORT_SYMBOL(l1_data_sram_alloc);
521 void *l1_data_sram_zalloc(size_t size)
523 void *addr = l1_data_sram_alloc(size);
526 memset(addr, 0x00, size);
530 EXPORT_SYMBOL(l1_data_sram_zalloc);
532 int l1_data_sram_free(const void *addr)
535 ret = l1_data_A_sram_free(addr);
537 ret = l1_data_B_sram_free(addr);
540 EXPORT_SYMBOL(l1_data_sram_free);
542 void *l1_inst_sram_alloc(size_t size)
544 #if L1_CODE_LENGTH != 0
550 /* add mutex operation */
551 spin_lock_irqsave(&per_cpu(l1_inst_sram_lock, cpu), flags);
553 addr = _sram_alloc(size, &per_cpu(free_l1_inst_sram_head, cpu),
554 &per_cpu(used_l1_inst_sram_head, cpu));
556 /* add mutex operation */
557 spin_unlock_irqrestore(&per_cpu(l1_inst_sram_lock, cpu), flags);
560 pr_debug("Allocated address in l1_inst_sram_alloc is 0x%lx+0x%lx\n",
561 (long unsigned int)addr, size);
568 EXPORT_SYMBOL(l1_inst_sram_alloc);
570 int l1_inst_sram_free(const void *addr)
572 #if L1_CODE_LENGTH != 0
578 /* add mutex operation */
579 spin_lock_irqsave(&per_cpu(l1_inst_sram_lock, cpu), flags);
581 ret = _sram_free(addr, &per_cpu(free_l1_inst_sram_head, cpu),
582 &per_cpu(used_l1_inst_sram_head, cpu));
584 /* add mutex operation */
585 spin_unlock_irqrestore(&per_cpu(l1_inst_sram_lock, cpu), flags);
593 EXPORT_SYMBOL(l1_inst_sram_free);
595 /* L1 Scratchpad memory allocate function */
596 void *l1sram_alloc(size_t size)
603 /* add mutex operation */
604 spin_lock_irqsave(&per_cpu(l1sram_lock, cpu), flags);
606 addr = _sram_alloc(size, &per_cpu(free_l1_ssram_head, cpu),
607 &per_cpu(used_l1_ssram_head, cpu));
609 /* add mutex operation */
610 spin_unlock_irqrestore(&per_cpu(l1sram_lock, cpu), flags);
616 /* L1 Scratchpad memory allocate function */
617 void *l1sram_alloc_max(size_t *psize)
624 /* add mutex operation */
625 spin_lock_irqsave(&per_cpu(l1sram_lock, cpu), flags);
627 addr = _sram_alloc_max(&per_cpu(free_l1_ssram_head, cpu),
628 &per_cpu(used_l1_ssram_head, cpu), psize);
630 /* add mutex operation */
631 spin_unlock_irqrestore(&per_cpu(l1sram_lock, cpu), flags);
637 /* L1 Scratchpad memory free function */
638 int l1sram_free(const void *addr)
645 /* add mutex operation */
646 spin_lock_irqsave(&per_cpu(l1sram_lock, cpu), flags);
648 ret = _sram_free(addr, &per_cpu(free_l1_ssram_head, cpu),
649 &per_cpu(used_l1_ssram_head, cpu));
651 /* add mutex operation */
652 spin_unlock_irqrestore(&per_cpu(l1sram_lock, cpu), flags);
658 void *l2_sram_alloc(size_t size)
664 /* add mutex operation */
665 spin_lock_irqsave(&l2_sram_lock, flags);
667 addr = _sram_alloc(size, &free_l2_sram_head,
670 /* add mutex operation */
671 spin_unlock_irqrestore(&l2_sram_lock, flags);
673 pr_debug("Allocated address in l2_sram_alloc is 0x%lx+0x%lx\n",
674 (long unsigned int)addr, size);
681 EXPORT_SYMBOL(l2_sram_alloc);
683 void *l2_sram_zalloc(size_t size)
685 void *addr = l2_sram_alloc(size);
688 memset(addr, 0x00, size);
692 EXPORT_SYMBOL(l2_sram_zalloc);
694 int l2_sram_free(const void *addr)
700 /* add mutex operation */
701 spin_lock_irqsave(&l2_sram_lock, flags);
703 ret = _sram_free(addr, &free_l2_sram_head,
706 /* add mutex operation */
707 spin_unlock_irqrestore(&l2_sram_lock, flags);
714 EXPORT_SYMBOL(l2_sram_free);
716 int sram_free_with_lsl(const void *addr)
718 struct sram_list_struct *lsl, **tmp;
719 struct mm_struct *mm = current->mm;
721 for (tmp = &mm->context.sram_list; *tmp; tmp = &(*tmp)->next)
722 if ((*tmp)->addr == addr)
733 EXPORT_SYMBOL(sram_free_with_lsl);
735 void *sram_alloc_with_lsl(size_t size, unsigned long flags)
738 struct sram_list_struct *lsl = NULL;
739 struct mm_struct *mm = current->mm;
741 lsl = kzalloc(sizeof(struct sram_list_struct), GFP_KERNEL);
745 if (flags & L1_INST_SRAM)
746 addr = l1_inst_sram_alloc(size);
748 if (addr == NULL && (flags & L1_DATA_A_SRAM))
749 addr = l1_data_A_sram_alloc(size);
751 if (addr == NULL && (flags & L1_DATA_B_SRAM))
752 addr = l1_data_B_sram_alloc(size);
754 if (addr == NULL && (flags & L2_SRAM))
755 addr = l2_sram_alloc(size);
763 lsl->next = mm->context.sram_list;
764 mm->context.sram_list = lsl;
767 EXPORT_SYMBOL(sram_alloc_with_lsl);
769 #ifdef CONFIG_PROC_FS
770 /* Once we get a real allocator, we'll throw all of this away.
771 * Until then, we need some sort of visibility into the L1 alloc.
773 /* Need to keep line of output the same. Currently, that is 44 bytes
774 * (including newline).
776 static int _sram_proc_read(char *buf, int *len, int count, const char *desc,
777 struct sram_piece *pfree_head,
778 struct sram_piece *pused_head)
780 struct sram_piece *pslot;
782 if (!pfree_head || !pused_head)
785 *len += sprintf(&buf[*len], "--- SRAM %-14s Size PID State \n", desc);
787 /* search the relevant memory slot */
788 pslot = pused_head->next;
790 while (pslot != NULL) {
791 *len += sprintf(&buf[*len], "%p-%p %10i %5i %-10s\n",
792 pslot->paddr, pslot->paddr + pslot->size,
793 pslot->size, pslot->pid, "ALLOCATED");
798 pslot = pfree_head->next;
800 while (pslot != NULL) {
801 *len += sprintf(&buf[*len], "%p-%p %10i %5i %-10s\n",
802 pslot->paddr, pslot->paddr + pslot->size,
803 pslot->size, pslot->pid, "FREE");
810 static int sram_proc_read(char *buf, char **start, off_t offset, int count,
811 int *eof, void *data)
816 for (cpu = 0; cpu < num_possible_cpus(); ++cpu) {
817 if (_sram_proc_read(buf, &len, count, "Scratchpad",
818 &per_cpu(free_l1_ssram_head, cpu), &per_cpu(used_l1_ssram_head, cpu)))
820 #if L1_DATA_A_LENGTH != 0
821 if (_sram_proc_read(buf, &len, count, "L1 Data A",
822 &per_cpu(free_l1_data_A_sram_head, cpu),
823 &per_cpu(used_l1_data_A_sram_head, cpu)))
826 #if L1_DATA_B_LENGTH != 0
827 if (_sram_proc_read(buf, &len, count, "L1 Data B",
828 &per_cpu(free_l1_data_B_sram_head, cpu),
829 &per_cpu(used_l1_data_B_sram_head, cpu)))
832 #if L1_CODE_LENGTH != 0
833 if (_sram_proc_read(buf, &len, count, "L1 Instruction",
834 &per_cpu(free_l1_inst_sram_head, cpu),
835 &per_cpu(used_l1_inst_sram_head, cpu)))
840 if (_sram_proc_read(buf, &len, count, "L2", &free_l2_sram_head,
849 static int __init sram_proc_init(void)
851 struct proc_dir_entry *ptr;
852 ptr = create_proc_entry("sram", S_IFREG | S_IRUGO, NULL);
854 printk(KERN_WARNING "unable to create /proc/sram\n");
857 ptr->owner = THIS_MODULE;
858 ptr->read_proc = sram_proc_read;
861 late_initcall(sram_proc_init);