diff options
author | Matt Fleming <matt@console-pimps.org> | 2009-08-16 10:44:08 -0400 |
---|---|---|
committer | Matt Fleming <matt@console-pimps.org> | 2009-08-21 08:02:43 -0400 |
commit | fb3f3e7fc6d4afb32f9eba32124beaf40313de3c (patch) | |
tree | 69ed6cfbc123a188aabbe4a98209c935680c27a2 /arch/sh/kernel/dwarf.c | |
parent | 97f361e2498ada54b48a235619eaf5af8e46427e (diff) |
sh: unwinder: Fix memory leak and create our own kmem cache
Plug a memory leak in dwarf_unwinder_dump() where we didn't free the
memory that we had previously allocated for the DWARF frames and DWARF
registers.
Now is also a opportune time to implement our own mempool and kmem
cache. It's a good idea to have a certain number of frame and register
objects in reserve at all times, so that we are guaranteed to have our
allocation satisfied even when memory is scarce. Since we have pools to
allocate from we can implement the registers for each frame as a linked
list as opposed to a sparsely populated array. Whilst it's true that the
lookup time for a linked list is larger than for arrays, there's only
usually a maximum of 8 registers per frame. So the overhead isn't that
much of a concern.
Signed-off-by: Matt Fleming <matt@console-pimps.org>
Diffstat (limited to 'arch/sh/kernel/dwarf.c')
-rw-r--r-- | arch/sh/kernel/dwarf.c | 201 |
1 files changed, 136 insertions, 65 deletions
diff --git a/arch/sh/kernel/dwarf.c b/arch/sh/kernel/dwarf.c index d0652153f576..e4810375207d 100644 --- a/arch/sh/kernel/dwarf.c +++ b/arch/sh/kernel/dwarf.c | |||
@@ -17,6 +17,7 @@ | |||
17 | #include <linux/kernel.h> | 17 | #include <linux/kernel.h> |
18 | #include <linux/io.h> | 18 | #include <linux/io.h> |
19 | #include <linux/list.h> | 19 | #include <linux/list.h> |
20 | #include <linux/mempool.h> | ||
20 | #include <linux/mm.h> | 21 | #include <linux/mm.h> |
21 | #include <asm/dwarf.h> | 22 | #include <asm/dwarf.h> |
22 | #include <asm/unwinder.h> | 23 | #include <asm/unwinder.h> |
@@ -25,6 +26,17 @@ | |||
25 | #include <asm/dwarf.h> | 26 | #include <asm/dwarf.h> |
26 | #include <asm/stacktrace.h> | 27 | #include <asm/stacktrace.h> |
27 | 28 | ||
29 | /* Reserve enough memory for two stack frames */ | ||
30 | #define DWARF_FRAME_MIN_REQ 2 | ||
31 | /* ... with 4 registers per frame. */ | ||
32 | #define DWARF_REG_MIN_REQ (DWARF_FRAME_MIN_REQ * 4) | ||
33 | |||
34 | static struct kmem_cache *dwarf_frame_cachep; | ||
35 | static mempool_t *dwarf_frame_pool; | ||
36 | |||
37 | static struct kmem_cache *dwarf_reg_cachep; | ||
38 | static mempool_t *dwarf_reg_pool; | ||
39 | |||
28 | static LIST_HEAD(dwarf_cie_list); | 40 | static LIST_HEAD(dwarf_cie_list); |
29 | static DEFINE_SPINLOCK(dwarf_cie_lock); | 41 | static DEFINE_SPINLOCK(dwarf_cie_lock); |
30 | 42 | ||
@@ -33,33 +45,25 @@ static DEFINE_SPINLOCK(dwarf_fde_lock); | |||
33 | 45 | ||
34 | static struct dwarf_cie *cached_cie; | 46 | static struct dwarf_cie *cached_cie; |
35 | 47 | ||
36 | /* | 48 | /** |
37 | * Figure out whether we need to allocate some dwarf registers. If dwarf | 49 | * dwarf_frame_alloc_reg - allocate memory for a DWARF register |
38 | * registers have already been allocated then we may need to realloc | 50 | * @frame: the DWARF frame whose list of registers we insert on |
39 | * them. "reg" is a register number that we need to be able to access | 51 | * @reg_num: the register number |
40 | * after this call. | 52 | * |
53 | * Allocate space for, and initialise, a dwarf reg from | ||
54 | * dwarf_reg_pool and insert it onto the (unsorted) linked-list of | ||
55 | * dwarf registers for @frame. | ||
41 | * | 56 | * |
42 | * Register numbers start at zero, therefore we need to allocate space | 57 | * Return the initialised DWARF reg. |
43 | * for "reg" + 1 registers. | ||
44 | */ | 58 | */ |
45 | static void dwarf_frame_alloc_regs(struct dwarf_frame *frame, | 59 | static struct dwarf_reg *dwarf_frame_alloc_reg(struct dwarf_frame *frame, |
46 | unsigned int reg) | 60 | unsigned int reg_num) |
47 | { | 61 | { |
48 | struct dwarf_reg *regs; | 62 | struct dwarf_reg *reg; |
49 | unsigned int num_regs = reg + 1; | ||
50 | size_t new_size; | ||
51 | size_t old_size; | ||
52 | 63 | ||
53 | new_size = num_regs * sizeof(*regs); | 64 | reg = mempool_alloc(dwarf_reg_pool, GFP_ATOMIC); |
54 | old_size = frame->num_regs * sizeof(*regs); | 65 | if (!reg) { |
55 | 66 | printk(KERN_WARNING "Unable to allocate a DWARF register\n"); | |
56 | /* Fast path: don't allocate any regs if we've already got enough. */ | ||
57 | if (frame->num_regs >= num_regs) | ||
58 | return; | ||
59 | |||
60 | regs = kzalloc(new_size, GFP_ATOMIC); | ||
61 | if (!regs) { | ||
62 | printk(KERN_WARNING "Unable to allocate DWARF registers\n"); | ||
63 | /* | 67 | /* |
64 | * Let's just bomb hard here, we have no way to | 68 | * Let's just bomb hard here, we have no way to |
65 | * gracefully recover. | 69 | * gracefully recover. |
@@ -67,13 +71,44 @@ static void dwarf_frame_alloc_regs(struct dwarf_frame *frame, | |||
67 | BUG(); | 71 | BUG(); |
68 | } | 72 | } |
69 | 73 | ||
70 | if (frame->regs) { | 74 | reg->number = reg_num; |
71 | memcpy(regs, frame->regs, old_size); | 75 | reg->addr = 0; |
72 | kfree(frame->regs); | 76 | reg->flags = 0; |
77 | |||
78 | list_add(®->link, &frame->reg_list); | ||
79 | |||
80 | return reg; | ||
81 | } | ||
82 | |||
83 | static void dwarf_frame_free_regs(struct dwarf_frame *frame) | ||
84 | { | ||
85 | struct dwarf_reg *reg, *n; | ||
86 | |||
87 | list_for_each_entry_safe(reg, n, &frame->reg_list, link) { | ||
88 | list_del(®->link); | ||
89 | mempool_free(reg, dwarf_reg_pool); | ||
90 | } | ||
91 | } | ||
92 | |||
93 | /** | ||
94 | * dwarf_frame_reg - return a DWARF register | ||
95 | * @frame: the DWARF frame to search in for @reg_num | ||
96 | * @reg_num: the register number to search for | ||
97 | * | ||
98 | * Lookup and return the dwarf reg @reg_num for this frame. Return | ||
99 | * NULL if @reg_num is an register invalid number. | ||
100 | */ | ||
101 | static struct dwarf_reg *dwarf_frame_reg(struct dwarf_frame *frame, | ||
102 | unsigned int reg_num) | ||
103 | { | ||
104 | struct dwarf_reg *reg; | ||
105 | |||
106 | list_for_each_entry(reg, &frame->reg_list, link) { | ||
107 | if (reg->number == reg_num) | ||
108 | return reg; | ||
73 | } | 109 | } |
74 | 110 | ||
75 | frame->regs = regs; | 111 | return NULL; |
76 | frame->num_regs = num_regs; | ||
77 | } | 112 | } |
78 | 113 | ||
79 | /** | 114 | /** |
@@ -347,6 +382,7 @@ static int dwarf_cfa_execute_insns(unsigned char *insn_start, | |||
347 | unsigned char insn; | 382 | unsigned char insn; |
348 | unsigned char *current_insn; | 383 | unsigned char *current_insn; |
349 | unsigned int count, delta, reg, expr_len, offset; | 384 | unsigned int count, delta, reg, expr_len, offset; |
385 | struct dwarf_reg *regp; | ||
350 | 386 | ||
351 | current_insn = insn_start; | 387 | current_insn = insn_start; |
352 | 388 | ||
@@ -369,9 +405,9 @@ static int dwarf_cfa_execute_insns(unsigned char *insn_start, | |||
369 | count = dwarf_read_uleb128(current_insn, &offset); | 405 | count = dwarf_read_uleb128(current_insn, &offset); |
370 | current_insn += count; | 406 | current_insn += count; |
371 | offset *= cie->data_alignment_factor; | 407 | offset *= cie->data_alignment_factor; |
372 | dwarf_frame_alloc_regs(frame, reg); | 408 | regp = dwarf_frame_alloc_reg(frame, reg); |
373 | frame->regs[reg].addr = offset; | 409 | regp->addr = offset; |
374 | frame->regs[reg].flags |= DWARF_REG_OFFSET; | 410 | regp->flags |= DWARF_REG_OFFSET; |
375 | continue; | 411 | continue; |
376 | /* NOTREACHED */ | 412 | /* NOTREACHED */ |
377 | case DW_CFA_restore: | 413 | case DW_CFA_restore: |
@@ -453,17 +489,18 @@ static int dwarf_cfa_execute_insns(unsigned char *insn_start, | |||
453 | count = dwarf_read_leb128(current_insn, &offset); | 489 | count = dwarf_read_leb128(current_insn, &offset); |
454 | current_insn += count; | 490 | current_insn += count; |
455 | offset *= cie->data_alignment_factor; | 491 | offset *= cie->data_alignment_factor; |
456 | dwarf_frame_alloc_regs(frame, reg); | 492 | regp = dwarf_frame_alloc_reg(frame, reg); |
457 | frame->regs[reg].flags |= DWARF_REG_OFFSET; | 493 | regp->flags |= DWARF_REG_OFFSET; |
458 | frame->regs[reg].addr = offset; | 494 | regp->addr = offset; |
459 | break; | 495 | break; |
460 | case DW_CFA_val_offset: | 496 | case DW_CFA_val_offset: |
461 | count = dwarf_read_uleb128(current_insn, ®); | 497 | count = dwarf_read_uleb128(current_insn, ®); |
462 | current_insn += count; | 498 | current_insn += count; |
463 | count = dwarf_read_leb128(current_insn, &offset); | 499 | count = dwarf_read_leb128(current_insn, &offset); |
464 | offset *= cie->data_alignment_factor; | 500 | offset *= cie->data_alignment_factor; |
465 | frame->regs[reg].flags |= DWARF_REG_OFFSET; | 501 | regp = dwarf_frame_alloc_reg(frame, reg); |
466 | frame->regs[reg].addr = offset; | 502 | regp->flags |= DWARF_REG_OFFSET; |
503 | regp->addr = offset; | ||
467 | break; | 504 | break; |
468 | case DW_CFA_GNU_args_size: | 505 | case DW_CFA_GNU_args_size: |
469 | count = dwarf_read_uleb128(current_insn, &offset); | 506 | count = dwarf_read_uleb128(current_insn, &offset); |
@@ -474,9 +511,10 @@ static int dwarf_cfa_execute_insns(unsigned char *insn_start, | |||
474 | current_insn += count; | 511 | current_insn += count; |
475 | count = dwarf_read_uleb128(current_insn, &offset); | 512 | count = dwarf_read_uleb128(current_insn, &offset); |
476 | offset *= cie->data_alignment_factor; | 513 | offset *= cie->data_alignment_factor; |
477 | dwarf_frame_alloc_regs(frame, reg); | 514 | |
478 | frame->regs[reg].flags |= DWARF_REG_OFFSET; | 515 | regp = dwarf_frame_alloc_reg(frame, reg); |
479 | frame->regs[reg].addr = -offset; | 516 | regp->flags |= DWARF_REG_OFFSET; |
517 | regp->addr = -offset; | ||
480 | break; | 518 | break; |
481 | default: | 519 | default: |
482 | pr_debug("unhandled DWARF instruction 0x%x\n", insn); | 520 | pr_debug("unhandled DWARF instruction 0x%x\n", insn); |
@@ -502,8 +540,8 @@ struct dwarf_frame *dwarf_unwind_stack(unsigned long pc, | |||
502 | struct dwarf_frame *frame; | 540 | struct dwarf_frame *frame; |
503 | struct dwarf_cie *cie; | 541 | struct dwarf_cie *cie; |
504 | struct dwarf_fde *fde; | 542 | struct dwarf_fde *fde; |
543 | struct dwarf_reg *reg; | ||
505 | unsigned long addr; | 544 | unsigned long addr; |
506 | int i, offset; | ||
507 | 545 | ||
508 | /* | 546 | /* |
509 | * If this is the first invocation of this recursive function we | 547 | * If this is the first invocation of this recursive function we |
@@ -516,11 +554,16 @@ struct dwarf_frame *dwarf_unwind_stack(unsigned long pc, | |||
516 | if (!pc && !prev) | 554 | if (!pc && !prev) |
517 | pc = (unsigned long)current_text_addr(); | 555 | pc = (unsigned long)current_text_addr(); |
518 | 556 | ||
519 | frame = kzalloc(sizeof(*frame), GFP_ATOMIC); | 557 | frame = mempool_alloc(dwarf_frame_pool, GFP_ATOMIC); |
520 | if (!frame) | 558 | if (!frame) { |
521 | return NULL; | 559 | printk(KERN_ERR "Unable to allocate a dwarf frame\n"); |
560 | BUG(); | ||
561 | } | ||
522 | 562 | ||
563 | INIT_LIST_HEAD(&frame->reg_list); | ||
564 | frame->flags = 0; | ||
523 | frame->prev = prev; | 565 | frame->prev = prev; |
566 | frame->return_addr = 0; | ||
524 | 567 | ||
525 | fde = dwarf_lookup_fde(pc); | 568 | fde = dwarf_lookup_fde(pc); |
526 | if (!fde) { | 569 | if (!fde) { |
@@ -540,7 +583,7 @@ struct dwarf_frame *dwarf_unwind_stack(unsigned long pc, | |||
540 | * case above, which sucks because we could print a | 583 | * case above, which sucks because we could print a |
541 | * warning here. | 584 | * warning here. |
542 | */ | 585 | */ |
543 | return NULL; | 586 | goto bail; |
544 | } | 587 | } |
545 | 588 | ||
546 | cie = dwarf_lookup_cie(fde->cie_pointer); | 589 | cie = dwarf_lookup_cie(fde->cie_pointer); |
@@ -560,10 +603,10 @@ struct dwarf_frame *dwarf_unwind_stack(unsigned long pc, | |||
560 | switch (frame->flags) { | 603 | switch (frame->flags) { |
561 | case DWARF_FRAME_CFA_REG_OFFSET: | 604 | case DWARF_FRAME_CFA_REG_OFFSET: |
562 | if (prev) { | 605 | if (prev) { |
563 | BUG_ON(!prev->regs[frame->cfa_register].flags); | 606 | reg = dwarf_frame_reg(prev, frame->cfa_register); |
607 | BUG_ON(!reg); | ||
564 | 608 | ||
565 | addr = prev->cfa; | 609 | addr = prev->cfa + reg->addr; |
566 | addr += prev->regs[frame->cfa_register].addr; | ||
567 | frame->cfa = __raw_readl(addr); | 610 | frame->cfa = __raw_readl(addr); |
568 | 611 | ||
569 | } else { | 612 | } else { |
@@ -584,23 +627,18 @@ struct dwarf_frame *dwarf_unwind_stack(unsigned long pc, | |||
584 | } | 627 | } |
585 | 628 | ||
586 | /* If we haven't seen the return address reg, we're screwed. */ | 629 | /* If we haven't seen the return address reg, we're screwed. */ |
587 | BUG_ON(!frame->regs[DWARF_ARCH_RA_REG].flags); | 630 | reg = dwarf_frame_reg(frame, DWARF_ARCH_RA_REG); |
588 | 631 | BUG_ON(!reg); | |
589 | for (i = 0; i <= frame->num_regs; i++) { | ||
590 | struct dwarf_reg *reg = &frame->regs[i]; | ||
591 | |||
592 | if (!reg->flags) | ||
593 | continue; | ||
594 | 632 | ||
595 | offset = reg->addr; | 633 | addr = frame->cfa + reg->addr; |
596 | offset += frame->cfa; | ||
597 | } | ||
598 | |||
599 | addr = frame->cfa + frame->regs[DWARF_ARCH_RA_REG].addr; | ||
600 | frame->return_addr = __raw_readl(addr); | 634 | frame->return_addr = __raw_readl(addr); |
601 | 635 | ||
602 | frame->next = dwarf_unwind_stack(frame->return_addr, frame); | ||
603 | return frame; | 636 | return frame; |
637 | |||
638 | bail: | ||
639 | dwarf_frame_free_regs(frame); | ||
640 | mempool_free(frame, dwarf_frame_pool); | ||
641 | return NULL; | ||
604 | } | 642 | } |
605 | 643 | ||
606 | static int dwarf_parse_cie(void *entry, void *p, unsigned long len, | 644 | static int dwarf_parse_cie(void *entry, void *p, unsigned long len, |
@@ -770,14 +808,29 @@ static void dwarf_unwinder_dump(struct task_struct *task, struct pt_regs *regs, | |||
770 | unsigned long *sp, | 808 | unsigned long *sp, |
771 | const struct stacktrace_ops *ops, void *data) | 809 | const struct stacktrace_ops *ops, void *data) |
772 | { | 810 | { |
773 | struct dwarf_frame *frame; | 811 | struct dwarf_frame *frame, *_frame; |
812 | unsigned long return_addr; | ||
813 | |||
814 | _frame = NULL; | ||
815 | return_addr = 0; | ||
774 | 816 | ||
775 | frame = dwarf_unwind_stack(0, NULL); | 817 | while (1) { |
818 | frame = dwarf_unwind_stack(return_addr, _frame); | ||
819 | |||
820 | if (_frame) { | ||
821 | dwarf_frame_free_regs(_frame); | ||
822 | mempool_free(_frame, dwarf_frame_pool); | ||
823 | } | ||
824 | |||
825 | _frame = frame; | ||
826 | |||
827 | if (!frame || !frame->return_addr) | ||
828 | break; | ||
776 | 829 | ||
777 | while (frame && frame->return_addr) { | 830 | return_addr = frame->return_addr; |
778 | ops->address(data, frame->return_addr, 1); | 831 | ops->address(data, return_addr, 1); |
779 | frame = frame->next; | ||
780 | } | 832 | } |
833 | |||
781 | } | 834 | } |
782 | 835 | ||
783 | static struct unwinder dwarf_unwinder = { | 836 | static struct unwinder dwarf_unwinder = { |
@@ -801,6 +854,9 @@ static void dwarf_unwinder_cleanup(void) | |||
801 | 854 | ||
802 | list_for_each_entry(fde, &dwarf_fde_list, link) | 855 | list_for_each_entry(fde, &dwarf_fde_list, link) |
803 | kfree(fde); | 856 | kfree(fde); |
857 | |||
858 | kmem_cache_destroy(dwarf_reg_cachep); | ||
859 | kmem_cache_destroy(dwarf_frame_cachep); | ||
804 | } | 860 | } |
805 | 861 | ||
806 | /** | 862 | /** |
@@ -827,6 +883,21 @@ static int __init dwarf_unwinder_init(void) | |||
827 | f_entries = 0; | 883 | f_entries = 0; |
828 | entry = &__start_eh_frame; | 884 | entry = &__start_eh_frame; |
829 | 885 | ||
886 | dwarf_frame_cachep = kmem_cache_create("dwarf_frames", | ||
887 | sizeof(struct dwarf_frame), 0, SLAB_PANIC, NULL); | ||
888 | dwarf_reg_cachep = kmem_cache_create("dwarf_regs", | ||
889 | sizeof(struct dwarf_reg), 0, SLAB_PANIC, NULL); | ||
890 | |||
891 | dwarf_frame_pool = mempool_create(DWARF_FRAME_MIN_REQ, | ||
892 | mempool_alloc_slab, | ||
893 | mempool_free_slab, | ||
894 | dwarf_frame_cachep); | ||
895 | |||
896 | dwarf_reg_pool = mempool_create(DWARF_REG_MIN_REQ, | ||
897 | mempool_alloc_slab, | ||
898 | mempool_free_slab, | ||
899 | dwarf_reg_cachep); | ||
900 | |||
830 | while ((char *)entry < __stop_eh_frame) { | 901 | while ((char *)entry < __stop_eh_frame) { |
831 | p = entry; | 902 | p = entry; |
832 | 903 | ||