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.
Showing the changes to kernel (1 of 6 files). Show all
Unified · Split
| @@ -36,6 +36,7 @@ · 35 unmodified lines | ||
| 36 | 36 | #include <linux/hash.h> |
| 37 | 37 | #include <linux/init.h> |
| 38 | 38 | #include <linux/module.h> |
| 39 | +#include <linux/moduleloader.h> | |
| 39 | 40 | #include <asm/cacheflush.h> |
| 40 | 41 | #include <asm/errno.h> |
| 41 | 42 | #include <asm/kdebug.h> |
| @@ -50,6 +51,106 @@ unsigned int kprobe_cpu = NR_CPUS; · 8 unmodified lines | ||
| 50 | 51 | static DEFINE_SPINLOCK(kprobe_lock); |
| 51 | 52 | static struct kprobe *curr_kprobe; |
| 52 | 53 | |
| 54 | +/* | |
| 55 | + * kprobe->ainsn.insn points to the copy of the instruction to be | |
| 56 | + * single-stepped. x86_64, POWER4 and above have no-exec support and | |
| 57 | + * stepping on the instruction on a vmalloced/kmalloced/data page | |
| 58 | + * is a recipe for disaster | |
| 59 | + */ | |
| 60 | +#define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t))) | |
| 61 | + | |
| 62 | +struct kprobe_insn_page { | |
| 63 | + struct hlist_node hlist; | |
| 64 | + kprobe_opcode_t *insns; /* Page of instruction slots */ | |
| 65 | + char slot_used[INSNS_PER_PAGE]; | |
| 66 | + int nused; | |
| 67 | +}; | |
| 68 | + | |
| 69 | +static struct hlist_head kprobe_insn_pages; | |
| 70 | + | |
| 71 | +/** | |
| 72 | + * get_insn_slot() - Find a slot on an executable page for an instruction. | |
| 73 | + * We allocate an executable page if there's no room on existing ones. | |
| 74 | + */ | |
| 75 | +kprobe_opcode_t *get_insn_slot(void) | |
| 76 | +{ | |
| 77 | + struct kprobe_insn_page *kip; | |
| 78 | + struct hlist_node *pos; | |
| 79 | + | |
| 80 | + hlist_for_each(pos, &kprobe_insn_pages) { | |
| 81 | + kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | |
| 82 | + if (kip->nused < INSNS_PER_PAGE) { | |
| 83 | + int i; | |
| 84 | + for (i = 0; i < INSNS_PER_PAGE; i++) { | |
| 85 | + if (!kip->slot_used[i]) { | |
| 86 | + kip->slot_used[i] = 1; | |
| 87 | + kip->nused++; | |
| 88 | + return kip->insns + (i * MAX_INSN_SIZE); | |
| 89 | + } | |
| 90 | + } | |
| 91 | + /* Surprise! No unused slots. Fix kip->nused. */ | |
| 92 | + kip->nused = INSNS_PER_PAGE; | |
| 93 | + } | |
| 94 | + } | |
| 95 | + | |
| 96 | + /* All out of space. Need to allocate a new page. Use slot 0.*/ | |
| 97 | + kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL); | |
| 98 | + if (!kip) { | |
| 99 | + return NULL; | |
| 100 | + } | |
| 101 | + | |
| 102 | + /* | |
| 103 | + * Use module_alloc so this page is within +/- 2GB of where the | |
| 104 | + * kernel image and loaded module images reside. This is required | |
| 105 | + * so x86_64 can correctly handle the %rip-relative fixups. | |
| 106 | + */ | |
| 107 | + kip->insns = module_alloc(PAGE_SIZE); | |
| 108 | + if (!kip->insns) { | |
| 109 | + kfree(kip); | |
| 110 | + return NULL; | |
| 111 | + } | |
| 112 | + INIT_HLIST_NODE(&kip->hlist); | |
| 113 | + hlist_add_head(&kip->hlist, &kprobe_insn_pages); | |
| 114 | + memset(kip->slot_used, 0, INSNS_PER_PAGE); | |
| 115 | + kip->slot_used[0] = 1; | |
| 116 | + kip->nused = 1; | |
| 117 | + return kip->insns; | |
| 118 | +} | |
| 119 | + | |
| 120 | +void free_insn_slot(kprobe_opcode_t *slot) | |
| 121 | +{ | |
| 122 | + struct kprobe_insn_page *kip; | |
| 123 | + struct hlist_node *pos; | |
| 124 | + | |
| 125 | + hlist_for_each(pos, &kprobe_insn_pages) { | |
| 126 | + kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | |
| 127 | + if (kip->insns <= slot && | |
| 128 | + slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) { | |
| 129 | + int i = (slot - kip->insns) / MAX_INSN_SIZE; | |
| 130 | + kip->slot_used[i] = 0; | |
| 131 | + kip->nused--; | |
| 132 | + if (kip->nused == 0) { | |
| 133 | + /* | |
| 134 | + * Page is no longer in use. Free it unless | |
| 135 | + * it's the last one. We keep the last one | |
| 136 | + * so as not to have to set it up again the | |
| 137 | + * next time somebody inserts a probe. | |
| 138 | + */ | |
| 139 | + hlist_del(&kip->hlist); | |
| 140 | + if (hlist_empty(&kprobe_insn_pages)) { | |
| 141 | + INIT_HLIST_NODE(&kip->hlist); | |
| 142 | + hlist_add_head(&kip->hlist, | |
| 143 | + &kprobe_insn_pages); | |
| 144 | + } else { | |
| 145 | + module_free(NULL, kip->insns); | |
| 146 | + kfree(kip); | |
| 147 | + } | |
| 148 | + } | |
| 149 | + return; | |
| 150 | + } | |
| 151 | + } | |
| 152 | +} | |
| 153 | + | |
| 53 | 154 | /* Locks kprobe: irqs must be disabled */ |
| 54 | 155 | void lock_kprobes(void) |
| 55 | 156 | { |