diff options
author | Ananth N Mavinakayanahalli <ananth@in.ibm.com> | 2005-06-27 18:17:01 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-06-27 18:23:52 -0400 |
commit | 9ec4b1f356b3bad928ae8e2aa9caebfa737d52df (patch) | |
tree | 24d27ffed66595a9d864448ec53200ca1745f62c /arch | |
parent | d3b8a1a8496c83bc4a3cc76505c29255af15572c (diff) |
[PATCH] kprobes: fix single-step out of line - take2
Now that PPC64 has no-execute support, here is a second try to fix the
single step out of line during kprobe execution. Kprobes on x86_64 already
solved this problem by allocating an executable page and using it as the
scratch area for stepping out of line. Reuse that.
Signed-off-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'arch')
-rw-r--r-- | arch/ppc64/kernel/kprobes.c | 26 | ||||
-rw-r--r-- | arch/x86_64/kernel/kprobes.c | 113 |
2 files changed, 23 insertions, 116 deletions
diff --git a/arch/ppc64/kernel/kprobes.c b/arch/ppc64/kernel/kprobes.c index 782ce3efa2c1..86cc5496db9f 100644 --- a/arch/ppc64/kernel/kprobes.c +++ b/arch/ppc64/kernel/kprobes.c | |||
@@ -36,6 +36,8 @@ | |||
36 | #include <asm/kdebug.h> | 36 | #include <asm/kdebug.h> |
37 | #include <asm/sstep.h> | 37 | #include <asm/sstep.h> |
38 | 38 | ||
39 | static DECLARE_MUTEX(kprobe_mutex); | ||
40 | |||
39 | static struct kprobe *current_kprobe; | 41 | static struct kprobe *current_kprobe; |
40 | static unsigned long kprobe_status, kprobe_saved_msr; | 42 | static unsigned long kprobe_status, kprobe_saved_msr; |
41 | static struct kprobe *kprobe_prev; | 43 | static struct kprobe *kprobe_prev; |
@@ -54,6 +56,15 @@ int arch_prepare_kprobe(struct kprobe *p) | |||
54 | printk("Cannot register a kprobe on rfid or mtmsrd\n"); | 56 | printk("Cannot register a kprobe on rfid or mtmsrd\n"); |
55 | ret = -EINVAL; | 57 | ret = -EINVAL; |
56 | } | 58 | } |
59 | |||
60 | /* insn must be on a special executable page on ppc64 */ | ||
61 | if (!ret) { | ||
62 | up(&kprobe_mutex); | ||
63 | p->ainsn.insn = get_insn_slot(); | ||
64 | down(&kprobe_mutex); | ||
65 | if (!p->ainsn.insn) | ||
66 | ret = -ENOMEM; | ||
67 | } | ||
57 | return ret; | 68 | return ret; |
58 | } | 69 | } |
59 | 70 | ||
@@ -79,16 +90,22 @@ void arch_disarm_kprobe(struct kprobe *p) | |||
79 | 90 | ||
80 | void arch_remove_kprobe(struct kprobe *p) | 91 | void arch_remove_kprobe(struct kprobe *p) |
81 | { | 92 | { |
93 | up(&kprobe_mutex); | ||
94 | free_insn_slot(p->ainsn.insn); | ||
95 | down(&kprobe_mutex); | ||
82 | } | 96 | } |
83 | 97 | ||
84 | static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs) | 98 | static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs) |
85 | { | 99 | { |
100 | kprobe_opcode_t insn = *p->ainsn.insn; | ||
101 | |||
86 | regs->msr |= MSR_SE; | 102 | regs->msr |= MSR_SE; |
87 | /*single step inline if it a breakpoint instruction*/ | 103 | |
88 | if (p->opcode == BREAKPOINT_INSTRUCTION) | 104 | /* single step inline if it is a trap variant */ |
105 | if (IS_TW(insn) || IS_TD(insn) || IS_TWI(insn) || IS_TDI(insn)) | ||
89 | regs->nip = (unsigned long)p->addr; | 106 | regs->nip = (unsigned long)p->addr; |
90 | else | 107 | else |
91 | regs->nip = (unsigned long)&p->ainsn.insn; | 108 | regs->nip = (unsigned long)p->ainsn.insn; |
92 | } | 109 | } |
93 | 110 | ||
94 | static inline void save_previous_kprobe(void) | 111 | static inline void save_previous_kprobe(void) |
@@ -205,9 +222,10 @@ no_kprobe: | |||
205 | static void resume_execution(struct kprobe *p, struct pt_regs *regs) | 222 | static void resume_execution(struct kprobe *p, struct pt_regs *regs) |
206 | { | 223 | { |
207 | int ret; | 224 | int ret; |
225 | unsigned int insn = *p->ainsn.insn; | ||
208 | 226 | ||
209 | regs->nip = (unsigned long)p->addr; | 227 | regs->nip = (unsigned long)p->addr; |
210 | ret = emulate_step(regs, p->ainsn.insn[0]); | 228 | ret = emulate_step(regs, insn); |
211 | if (ret == 0) | 229 | if (ret == 0) |
212 | regs->nip = (unsigned long)p->addr + 4; | 230 | regs->nip = (unsigned long)p->addr + 4; |
213 | } | 231 | } |
diff --git a/arch/x86_64/kernel/kprobes.c b/arch/x86_64/kernel/kprobes.c index 4e680f87a75f..6a1c88376bef 100644 --- a/arch/x86_64/kernel/kprobes.c +++ b/arch/x86_64/kernel/kprobes.c | |||
@@ -38,7 +38,7 @@ | |||
38 | #include <linux/string.h> | 38 | #include <linux/string.h> |
39 | #include <linux/slab.h> | 39 | #include <linux/slab.h> |
40 | #include <linux/preempt.h> | 40 | #include <linux/preempt.h> |
41 | #include <linux/moduleloader.h> | 41 | |
42 | #include <asm/cacheflush.h> | 42 | #include <asm/cacheflush.h> |
43 | #include <asm/pgtable.h> | 43 | #include <asm/pgtable.h> |
44 | #include <asm/kdebug.h> | 44 | #include <asm/kdebug.h> |
@@ -51,8 +51,6 @@ static struct kprobe *kprobe_prev; | |||
51 | static unsigned long kprobe_status_prev, kprobe_old_rflags_prev, kprobe_saved_rflags_prev; | 51 | static unsigned long kprobe_status_prev, kprobe_old_rflags_prev, kprobe_saved_rflags_prev; |
52 | static struct pt_regs jprobe_saved_regs; | 52 | static struct pt_regs jprobe_saved_regs; |
53 | static long *jprobe_saved_rsp; | 53 | static long *jprobe_saved_rsp; |
54 | static kprobe_opcode_t *get_insn_slot(void); | ||
55 | static void free_insn_slot(kprobe_opcode_t *slot); | ||
56 | void jprobe_return_end(void); | 54 | void jprobe_return_end(void); |
57 | 55 | ||
58 | /* copy of the kernel stack at the probe fire time */ | 56 | /* copy of the kernel stack at the probe fire time */ |
@@ -681,112 +679,3 @@ int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs) | |||
681 | } | 679 | } |
682 | return 0; | 680 | return 0; |
683 | } | 681 | } |
684 | |||
685 | /* | ||
686 | * kprobe->ainsn.insn points to the copy of the instruction to be single-stepped. | ||
687 | * By default on x86_64, pages we get from kmalloc or vmalloc are not | ||
688 | * executable. Single-stepping an instruction on such a page yields an | ||
689 | * oops. So instead of storing the instruction copies in their respective | ||
690 | * kprobe objects, we allocate a page, map it executable, and store all the | ||
691 | * instruction copies there. (We can allocate additional pages if somebody | ||
692 | * inserts a huge number of probes.) Each page can hold up to INSNS_PER_PAGE | ||
693 | * instruction slots, each of which is MAX_INSN_SIZE*sizeof(kprobe_opcode_t) | ||
694 | * bytes. | ||
695 | */ | ||
696 | #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE*sizeof(kprobe_opcode_t))) | ||
697 | struct kprobe_insn_page { | ||
698 | struct hlist_node hlist; | ||
699 | kprobe_opcode_t *insns; /* page of instruction slots */ | ||
700 | char slot_used[INSNS_PER_PAGE]; | ||
701 | int nused; | ||
702 | }; | ||
703 | |||
704 | static struct hlist_head kprobe_insn_pages; | ||
705 | |||
706 | /** | ||
707 | * get_insn_slot() - Find a slot on an executable page for an instruction. | ||
708 | * We allocate an executable page if there's no room on existing ones. | ||
709 | */ | ||
710 | static kprobe_opcode_t *get_insn_slot(void) | ||
711 | { | ||
712 | struct kprobe_insn_page *kip; | ||
713 | struct hlist_node *pos; | ||
714 | |||
715 | hlist_for_each(pos, &kprobe_insn_pages) { | ||
716 | kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | ||
717 | if (kip->nused < INSNS_PER_PAGE) { | ||
718 | int i; | ||
719 | for (i = 0; i < INSNS_PER_PAGE; i++) { | ||
720 | if (!kip->slot_used[i]) { | ||
721 | kip->slot_used[i] = 1; | ||
722 | kip->nused++; | ||
723 | return kip->insns + (i*MAX_INSN_SIZE); | ||
724 | } | ||
725 | } | ||
726 | /* Surprise! No unused slots. Fix kip->nused. */ | ||
727 | kip->nused = INSNS_PER_PAGE; | ||
728 | } | ||
729 | } | ||
730 | |||
731 | /* All out of space. Need to allocate a new page. Use slot 0.*/ | ||
732 | kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL); | ||
733 | if (!kip) { | ||
734 | return NULL; | ||
735 | } | ||
736 | |||
737 | /* | ||
738 | * For the %rip-relative displacement fixups to be doable, we | ||
739 | * need our instruction copy to be within +/- 2GB of any data it | ||
740 | * might access via %rip. That is, within 2GB of where the | ||
741 | * kernel image and loaded module images reside. So we allocate | ||
742 | * a page in the module loading area. | ||
743 | */ | ||
744 | kip->insns = module_alloc(PAGE_SIZE); | ||
745 | if (!kip->insns) { | ||
746 | kfree(kip); | ||
747 | return NULL; | ||
748 | } | ||
749 | INIT_HLIST_NODE(&kip->hlist); | ||
750 | hlist_add_head(&kip->hlist, &kprobe_insn_pages); | ||
751 | memset(kip->slot_used, 0, INSNS_PER_PAGE); | ||
752 | kip->slot_used[0] = 1; | ||
753 | kip->nused = 1; | ||
754 | return kip->insns; | ||
755 | } | ||
756 | |||
757 | /** | ||
758 | * free_insn_slot() - Free instruction slot obtained from get_insn_slot(). | ||
759 | */ | ||
760 | static void free_insn_slot(kprobe_opcode_t *slot) | ||
761 | { | ||
762 | struct kprobe_insn_page *kip; | ||
763 | struct hlist_node *pos; | ||
764 | |||
765 | hlist_for_each(pos, &kprobe_insn_pages) { | ||
766 | kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | ||
767 | if (kip->insns <= slot | ||
768 | && slot < kip->insns+(INSNS_PER_PAGE*MAX_INSN_SIZE)) { | ||
769 | int i = (slot - kip->insns) / MAX_INSN_SIZE; | ||
770 | kip->slot_used[i] = 0; | ||
771 | kip->nused--; | ||
772 | if (kip->nused == 0) { | ||
773 | /* | ||
774 | * Page is no longer in use. Free it unless | ||
775 | * it's the last one. We keep the last one | ||
776 | * so as not to have to set it up again the | ||
777 | * next time somebody inserts a probe. | ||
778 | */ | ||
779 | hlist_del(&kip->hlist); | ||
780 | if (hlist_empty(&kprobe_insn_pages)) { | ||
781 | INIT_HLIST_NODE(&kip->hlist); | ||
782 | hlist_add_head(&kip->hlist, | ||
783 | &kprobe_insn_pages); | ||
784 | } else { | ||
785 | module_free(NULL, kip->insns); | ||
786 | kfree(kip); | ||
787 | } | ||
788 | } | ||
789 | return; | ||
790 | } | ||
791 | } | ||
792 | } | ||