diff options
author | Ingo Molnar <mingo@elte.hu> | 2008-07-10 05:43:06 -0400 |
---|---|---|
committer | Ingo Molnar <mingo@elte.hu> | 2008-07-10 05:43:06 -0400 |
commit | 5373fdbdc1dba69aa956098650f71b731d471885 (patch) | |
tree | 8d9f07539896a696352818820c9c5f6612370882 /arch | |
parent | bac0c9103b31c3dd83ad9d731dd9834e2ba75e4f (diff) | |
parent | 4d51c7587bb13dbb2fafcad6c0b5231bd864b55f (diff) |
Merge branch 'tracing/mmiotrace' into auto-ftrace-next
Diffstat (limited to 'arch')
-rw-r--r-- | arch/x86/Kconfig.debug | 32 | ||||
-rw-r--r-- | arch/x86/mm/Makefile | 5 | ||||
-rw-r--r-- | arch/x86/mm/fault.c | 57 | ||||
-rw-r--r-- | arch/x86/mm/ioremap.c | 11 | ||||
-rw-r--r-- | arch/x86/mm/kmmio.c | 510 | ||||
-rw-r--r-- | arch/x86/mm/mmio-mod.c | 515 | ||||
-rw-r--r-- | arch/x86/mm/pageattr.c | 1 | ||||
-rw-r--r-- | arch/x86/mm/pf_in.c | 489 | ||||
-rw-r--r-- | arch/x86/mm/pf_in.h | 39 | ||||
-rw-r--r-- | arch/x86/mm/testmmiotrace.c | 71 |
10 files changed, 1673 insertions, 57 deletions
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug index f395fd537c5c..f7169edfbeab 100644 --- a/arch/x86/Kconfig.debug +++ b/arch/x86/Kconfig.debug | |||
@@ -172,13 +172,33 @@ config IOMMU_LEAK | |||
172 | Add a simple leak tracer to the IOMMU code. This is useful when you | 172 | Add a simple leak tracer to the IOMMU code. This is useful when you |
173 | are debugging a buggy device driver that leaks IOMMU mappings. | 173 | are debugging a buggy device driver that leaks IOMMU mappings. |
174 | 174 | ||
175 | config PAGE_FAULT_HANDLERS | 175 | config MMIOTRACE_HOOKS |
176 | bool "Custom page fault handlers" | 176 | bool |
177 | depends on DEBUG_KERNEL | 177 | |
178 | config MMIOTRACE | ||
179 | bool "Memory mapped IO tracing" | ||
180 | depends on DEBUG_KERNEL && PCI | ||
181 | select TRACING | ||
182 | select MMIOTRACE_HOOKS | ||
183 | default y | ||
184 | help | ||
185 | Mmiotrace traces Memory Mapped I/O access and is meant for | ||
186 | debugging and reverse engineering. It is called from the ioremap | ||
187 | implementation and works via page faults. Tracing is disabled by | ||
188 | default and can be enabled at run-time. | ||
189 | |||
190 | See Documentation/tracers/mmiotrace.txt. | ||
191 | If you are not helping to develop drivers, say N. | ||
192 | |||
193 | config MMIOTRACE_TEST | ||
194 | tristate "Test module for mmiotrace" | ||
195 | depends on MMIOTRACE && m | ||
178 | help | 196 | help |
179 | Allow the use of custom page fault handlers. A kernel module may | 197 | This is a dumb module for testing mmiotrace. It is very dangerous |
180 | register a function that is called on every page fault. Custom | 198 | as it will write garbage to IO memory starting at a given address. |
181 | handlers are used by some debugging and reverse engineering tools. | 199 | However, it should be safe to use on e.g. unused portion of VRAM. |
200 | |||
201 | Say N, unless you absolutely know what you are doing. | ||
182 | 202 | ||
183 | # | 203 | # |
184 | # IO delay types: | 204 | # IO delay types: |
diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile index b7b3e4c7cfc9..07dab503c9e3 100644 --- a/arch/x86/mm/Makefile +++ b/arch/x86/mm/Makefile | |||
@@ -8,6 +8,11 @@ obj-$(CONFIG_X86_PTDUMP) += dump_pagetables.o | |||
8 | 8 | ||
9 | obj-$(CONFIG_HIGHMEM) += highmem_32.o | 9 | obj-$(CONFIG_HIGHMEM) += highmem_32.o |
10 | 10 | ||
11 | obj-$(CONFIG_MMIOTRACE_HOOKS) += kmmio.o | ||
12 | obj-$(CONFIG_MMIOTRACE) += mmiotrace.o | ||
13 | mmiotrace-y := pf_in.o mmio-mod.o | ||
14 | obj-$(CONFIG_MMIOTRACE_TEST) += testmmiotrace.o | ||
15 | |||
11 | ifeq ($(CONFIG_X86_32),y) | 16 | ifeq ($(CONFIG_X86_32),y) |
12 | obj-$(CONFIG_NUMA) += discontig_32.o | 17 | obj-$(CONFIG_NUMA) += discontig_32.o |
13 | else | 18 | else |
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c index 42394b353c6a..0a778e3c43ee 100644 --- a/arch/x86/mm/fault.c +++ b/arch/x86/mm/fault.c | |||
@@ -10,6 +10,7 @@ | |||
10 | #include <linux/string.h> | 10 | #include <linux/string.h> |
11 | #include <linux/types.h> | 11 | #include <linux/types.h> |
12 | #include <linux/ptrace.h> | 12 | #include <linux/ptrace.h> |
13 | #include <linux/mmiotrace.h> | ||
13 | #include <linux/mman.h> | 14 | #include <linux/mman.h> |
14 | #include <linux/mm.h> | 15 | #include <linux/mm.h> |
15 | #include <linux/smp.h> | 16 | #include <linux/smp.h> |
@@ -49,58 +50,14 @@ | |||
49 | #define PF_RSVD (1<<3) | 50 | #define PF_RSVD (1<<3) |
50 | #define PF_INSTR (1<<4) | 51 | #define PF_INSTR (1<<4) |
51 | 52 | ||
52 | #ifdef CONFIG_PAGE_FAULT_HANDLERS | 53 | static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr) |
53 | static HLIST_HEAD(pf_handlers); /* protected by RCU */ | ||
54 | static DEFINE_SPINLOCK(pf_handlers_writer); | ||
55 | |||
56 | void register_page_fault_handler(struct pf_handler *new_pfh) | ||
57 | { | ||
58 | unsigned long flags; | ||
59 | spin_lock_irqsave(&pf_handlers_writer, flags); | ||
60 | hlist_add_head_rcu(&new_pfh->hlist, &pf_handlers); | ||
61 | spin_unlock_irqrestore(&pf_handlers_writer, flags); | ||
62 | } | ||
63 | EXPORT_SYMBOL_GPL(register_page_fault_handler); | ||
64 | |||
65 | /** | ||
66 | * unregister_page_fault_handler: | ||
67 | * The caller must ensure @old_pfh is not in use anymore before freeing it. | ||
68 | * This function does not guarantee it. The list of handlers is protected by | ||
69 | * RCU, so you can do this by e.g. calling synchronize_rcu(). | ||
70 | */ | ||
71 | void unregister_page_fault_handler(struct pf_handler *old_pfh) | ||
72 | { | 54 | { |
73 | unsigned long flags; | 55 | #ifdef CONFIG_MMIOTRACE_HOOKS |
74 | spin_lock_irqsave(&pf_handlers_writer, flags); | 56 | if (unlikely(is_kmmio_active())) |
75 | hlist_del_rcu(&old_pfh->hlist); | 57 | if (kmmio_handler(regs, addr) == 1) |
76 | spin_unlock_irqrestore(&pf_handlers_writer, flags); | 58 | return -1; |
77 | } | ||
78 | EXPORT_SYMBOL_GPL(unregister_page_fault_handler); | ||
79 | #endif | 59 | #endif |
80 | |||
81 | /* returns non-zero if do_page_fault() should return */ | ||
82 | static int handle_custom_pf(struct pt_regs *regs, unsigned long error_code, | ||
83 | unsigned long address) | ||
84 | { | ||
85 | #ifdef CONFIG_PAGE_FAULT_HANDLERS | ||
86 | int ret = 0; | ||
87 | struct pf_handler *cur; | ||
88 | struct hlist_node *ncur; | ||
89 | |||
90 | if (hlist_empty(&pf_handlers)) | ||
91 | return 0; | ||
92 | |||
93 | rcu_read_lock(); | ||
94 | hlist_for_each_entry_rcu(cur, ncur, &pf_handlers, hlist) { | ||
95 | ret = cur->handler(regs, error_code, address); | ||
96 | if (ret) | ||
97 | break; | ||
98 | } | ||
99 | rcu_read_unlock(); | ||
100 | return ret; | ||
101 | #else | ||
102 | return 0; | 60 | return 0; |
103 | #endif | ||
104 | } | 61 | } |
105 | 62 | ||
106 | static inline int notify_page_fault(struct pt_regs *regs) | 63 | static inline int notify_page_fault(struct pt_regs *regs) |
@@ -660,7 +617,7 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) | |||
660 | 617 | ||
661 | if (notify_page_fault(regs)) | 618 | if (notify_page_fault(regs)) |
662 | return; | 619 | return; |
663 | if (handle_custom_pf(regs, error_code, address)) | 620 | if (unlikely(kmmio_fault(regs, address))) |
664 | return; | 621 | return; |
665 | 622 | ||
666 | /* | 623 | /* |
diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c index 2b2bb3f9b683..e92aa461f4d6 100644 --- a/arch/x86/mm/ioremap.c +++ b/arch/x86/mm/ioremap.c | |||
@@ -12,6 +12,7 @@ | |||
12 | #include <linux/module.h> | 12 | #include <linux/module.h> |
13 | #include <linux/slab.h> | 13 | #include <linux/slab.h> |
14 | #include <linux/vmalloc.h> | 14 | #include <linux/vmalloc.h> |
15 | #include <linux/mmiotrace.h> | ||
15 | 16 | ||
16 | #include <asm/cacheflush.h> | 17 | #include <asm/cacheflush.h> |
17 | #include <asm/e820.h> | 18 | #include <asm/e820.h> |
@@ -122,10 +123,13 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr, | |||
122 | { | 123 | { |
123 | unsigned long pfn, offset, vaddr; | 124 | unsigned long pfn, offset, vaddr; |
124 | resource_size_t last_addr; | 125 | resource_size_t last_addr; |
126 | const resource_size_t unaligned_phys_addr = phys_addr; | ||
127 | const unsigned long unaligned_size = size; | ||
125 | struct vm_struct *area; | 128 | struct vm_struct *area; |
126 | unsigned long new_prot_val; | 129 | unsigned long new_prot_val; |
127 | pgprot_t prot; | 130 | pgprot_t prot; |
128 | int retval; | 131 | int retval; |
132 | void __iomem *ret_addr; | ||
129 | 133 | ||
130 | /* Don't allow wraparound or zero size */ | 134 | /* Don't allow wraparound or zero size */ |
131 | last_addr = phys_addr + size - 1; | 135 | last_addr = phys_addr + size - 1; |
@@ -233,7 +237,10 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr, | |||
233 | return NULL; | 237 | return NULL; |
234 | } | 238 | } |
235 | 239 | ||
236 | return (void __iomem *) (vaddr + offset); | 240 | ret_addr = (void __iomem *) (vaddr + offset); |
241 | mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr); | ||
242 | |||
243 | return ret_addr; | ||
237 | } | 244 | } |
238 | 245 | ||
239 | /** | 246 | /** |
@@ -325,6 +332,8 @@ void iounmap(volatile void __iomem *addr) | |||
325 | addr = (volatile void __iomem *) | 332 | addr = (volatile void __iomem *) |
326 | (PAGE_MASK & (unsigned long __force)addr); | 333 | (PAGE_MASK & (unsigned long __force)addr); |
327 | 334 | ||
335 | mmiotrace_iounmap(addr); | ||
336 | |||
328 | /* Use the vm area unlocked, assuming the caller | 337 | /* Use the vm area unlocked, assuming the caller |
329 | ensures there isn't another iounmap for the same address | 338 | ensures there isn't another iounmap for the same address |
330 | in parallel. Reuse of the virtual address is prevented by | 339 | in parallel. Reuse of the virtual address is prevented by |
diff --git a/arch/x86/mm/kmmio.c b/arch/x86/mm/kmmio.c new file mode 100644 index 000000000000..93d82038af4b --- /dev/null +++ b/arch/x86/mm/kmmio.c | |||
@@ -0,0 +1,510 @@ | |||
1 | /* Support for MMIO probes. | ||
2 | * Benfit many code from kprobes | ||
3 | * (C) 2002 Louis Zhuang <louis.zhuang@intel.com>. | ||
4 | * 2007 Alexander Eichner | ||
5 | * 2008 Pekka Paalanen <pq@iki.fi> | ||
6 | */ | ||
7 | |||
8 | #include <linux/list.h> | ||
9 | #include <linux/rculist.h> | ||
10 | #include <linux/spinlock.h> | ||
11 | #include <linux/hash.h> | ||
12 | #include <linux/init.h> | ||
13 | #include <linux/module.h> | ||
14 | #include <linux/kernel.h> | ||
15 | #include <linux/uaccess.h> | ||
16 | #include <linux/ptrace.h> | ||
17 | #include <linux/preempt.h> | ||
18 | #include <linux/percpu.h> | ||
19 | #include <linux/kdebug.h> | ||
20 | #include <linux/mutex.h> | ||
21 | #include <linux/io.h> | ||
22 | #include <asm/cacheflush.h> | ||
23 | #include <asm/tlbflush.h> | ||
24 | #include <linux/errno.h> | ||
25 | #include <asm/debugreg.h> | ||
26 | #include <linux/mmiotrace.h> | ||
27 | |||
28 | #define KMMIO_PAGE_HASH_BITS 4 | ||
29 | #define KMMIO_PAGE_TABLE_SIZE (1 << KMMIO_PAGE_HASH_BITS) | ||
30 | |||
31 | struct kmmio_fault_page { | ||
32 | struct list_head list; | ||
33 | struct kmmio_fault_page *release_next; | ||
34 | unsigned long page; /* location of the fault page */ | ||
35 | |||
36 | /* | ||
37 | * Number of times this page has been registered as a part | ||
38 | * of a probe. If zero, page is disarmed and this may be freed. | ||
39 | * Used only by writers (RCU). | ||
40 | */ | ||
41 | int count; | ||
42 | }; | ||
43 | |||
44 | struct kmmio_delayed_release { | ||
45 | struct rcu_head rcu; | ||
46 | struct kmmio_fault_page *release_list; | ||
47 | }; | ||
48 | |||
49 | struct kmmio_context { | ||
50 | struct kmmio_fault_page *fpage; | ||
51 | struct kmmio_probe *probe; | ||
52 | unsigned long saved_flags; | ||
53 | unsigned long addr; | ||
54 | int active; | ||
55 | }; | ||
56 | |||
57 | static DEFINE_SPINLOCK(kmmio_lock); | ||
58 | |||
59 | /* Protected by kmmio_lock */ | ||
60 | unsigned int kmmio_count; | ||
61 | |||
62 | /* Read-protected by RCU, write-protected by kmmio_lock. */ | ||
63 | static struct list_head kmmio_page_table[KMMIO_PAGE_TABLE_SIZE]; | ||
64 | static LIST_HEAD(kmmio_probes); | ||
65 | |||
66 | static struct list_head *kmmio_page_list(unsigned long page) | ||
67 | { | ||
68 | return &kmmio_page_table[hash_long(page, KMMIO_PAGE_HASH_BITS)]; | ||
69 | } | ||
70 | |||
71 | /* Accessed per-cpu */ | ||
72 | static DEFINE_PER_CPU(struct kmmio_context, kmmio_ctx); | ||
73 | |||
74 | /* | ||
75 | * this is basically a dynamic stabbing problem: | ||
76 | * Could use the existing prio tree code or | ||
77 | * Possible better implementations: | ||
78 | * The Interval Skip List: A Data Structure for Finding All Intervals That | ||
79 | * Overlap a Point (might be simple) | ||
80 | * Space Efficient Dynamic Stabbing with Fast Queries - Mikkel Thorup | ||
81 | */ | ||
82 | /* Get the kmmio at this addr (if any). You must be holding RCU read lock. */ | ||
83 | static struct kmmio_probe *get_kmmio_probe(unsigned long addr) | ||
84 | { | ||
85 | struct kmmio_probe *p; | ||
86 | list_for_each_entry_rcu(p, &kmmio_probes, list) { | ||
87 | if (addr >= p->addr && addr <= (p->addr + p->len)) | ||
88 | return p; | ||
89 | } | ||
90 | return NULL; | ||
91 | } | ||
92 | |||
93 | /* You must be holding RCU read lock. */ | ||
94 | static struct kmmio_fault_page *get_kmmio_fault_page(unsigned long page) | ||
95 | { | ||
96 | struct list_head *head; | ||
97 | struct kmmio_fault_page *p; | ||
98 | |||
99 | page &= PAGE_MASK; | ||
100 | head = kmmio_page_list(page); | ||
101 | list_for_each_entry_rcu(p, head, list) { | ||
102 | if (p->page == page) | ||
103 | return p; | ||
104 | } | ||
105 | return NULL; | ||
106 | } | ||
107 | |||
108 | static void set_page_present(unsigned long addr, bool present, | ||
109 | unsigned int *pglevel) | ||
110 | { | ||
111 | pteval_t pteval; | ||
112 | pmdval_t pmdval; | ||
113 | unsigned int level; | ||
114 | pmd_t *pmd; | ||
115 | pte_t *pte = lookup_address(addr, &level); | ||
116 | |||
117 | if (!pte) { | ||
118 | pr_err("kmmio: no pte for page 0x%08lx\n", addr); | ||
119 | return; | ||
120 | } | ||
121 | |||
122 | if (pglevel) | ||
123 | *pglevel = level; | ||
124 | |||
125 | switch (level) { | ||
126 | case PG_LEVEL_2M: | ||
127 | pmd = (pmd_t *)pte; | ||
128 | pmdval = pmd_val(*pmd) & ~_PAGE_PRESENT; | ||
129 | if (present) | ||
130 | pmdval |= _PAGE_PRESENT; | ||
131 | set_pmd(pmd, __pmd(pmdval)); | ||
132 | break; | ||
133 | |||
134 | case PG_LEVEL_4K: | ||
135 | pteval = pte_val(*pte) & ~_PAGE_PRESENT; | ||
136 | if (present) | ||
137 | pteval |= _PAGE_PRESENT; | ||
138 | set_pte_atomic(pte, __pte(pteval)); | ||
139 | break; | ||
140 | |||
141 | default: | ||
142 | pr_err("kmmio: unexpected page level 0x%x.\n", level); | ||
143 | return; | ||
144 | } | ||
145 | |||
146 | __flush_tlb_one(addr); | ||
147 | } | ||
148 | |||
149 | /** Mark the given page as not present. Access to it will trigger a fault. */ | ||
150 | static void arm_kmmio_fault_page(unsigned long page, unsigned int *pglevel) | ||
151 | { | ||
152 | set_page_present(page & PAGE_MASK, false, pglevel); | ||
153 | } | ||
154 | |||
155 | /** Mark the given page as present. */ | ||
156 | static void disarm_kmmio_fault_page(unsigned long page, unsigned int *pglevel) | ||
157 | { | ||
158 | set_page_present(page & PAGE_MASK, true, pglevel); | ||
159 | } | ||
160 | |||
161 | /* | ||
162 | * This is being called from do_page_fault(). | ||
163 | * | ||
164 | * We may be in an interrupt or a critical section. Also prefecthing may | ||
165 | * trigger a page fault. We may be in the middle of process switch. | ||
166 | * We cannot take any locks, because we could be executing especially | ||
167 | * within a kmmio critical section. | ||
168 | * | ||
169 | * Local interrupts are disabled, so preemption cannot happen. | ||
170 | * Do not enable interrupts, do not sleep, and watch out for other CPUs. | ||
171 | */ | ||
172 | /* | ||
173 | * Interrupts are disabled on entry as trap3 is an interrupt gate | ||
174 | * and they remain disabled thorough out this function. | ||
175 | */ | ||
176 | int kmmio_handler(struct pt_regs *regs, unsigned long addr) | ||
177 | { | ||
178 | struct kmmio_context *ctx; | ||
179 | struct kmmio_fault_page *faultpage; | ||
180 | int ret = 0; /* default to fault not handled */ | ||
181 | |||
182 | /* | ||
183 | * Preemption is now disabled to prevent process switch during | ||
184 | * single stepping. We can only handle one active kmmio trace | ||
185 | * per cpu, so ensure that we finish it before something else | ||
186 | * gets to run. We also hold the RCU read lock over single | ||
187 | * stepping to avoid looking up the probe and kmmio_fault_page | ||
188 | * again. | ||
189 | */ | ||
190 | preempt_disable(); | ||
191 | rcu_read_lock(); | ||
192 | |||
193 | faultpage = get_kmmio_fault_page(addr); | ||
194 | if (!faultpage) { | ||
195 | /* | ||
196 | * Either this page fault is not caused by kmmio, or | ||
197 | * another CPU just pulled the kmmio probe from under | ||
198 | * our feet. The latter case should not be possible. | ||
199 | */ | ||
200 | goto no_kmmio; | ||
201 | } | ||
202 | |||
203 | ctx = &get_cpu_var(kmmio_ctx); | ||
204 | if (ctx->active) { | ||
205 | disarm_kmmio_fault_page(faultpage->page, NULL); | ||
206 | if (addr == ctx->addr) { | ||
207 | /* | ||
208 | * On SMP we sometimes get recursive probe hits on the | ||
209 | * same address. Context is already saved, fall out. | ||
210 | */ | ||
211 | pr_debug("kmmio: duplicate probe hit on CPU %d, for " | ||
212 | "address 0x%08lx.\n", | ||
213 | smp_processor_id(), addr); | ||
214 | ret = 1; | ||
215 | goto no_kmmio_ctx; | ||
216 | } | ||
217 | /* | ||
218 | * Prevent overwriting already in-flight context. | ||
219 | * This should not happen, let's hope disarming at least | ||
220 | * prevents a panic. | ||
221 | */ | ||
222 | pr_emerg("kmmio: recursive probe hit on CPU %d, " | ||
223 | "for address 0x%08lx. Ignoring.\n", | ||
224 | smp_processor_id(), addr); | ||
225 | pr_emerg("kmmio: previous hit was at 0x%08lx.\n", | ||
226 | ctx->addr); | ||
227 | goto no_kmmio_ctx; | ||
228 | } | ||
229 | ctx->active++; | ||
230 | |||
231 | ctx->fpage = faultpage; | ||
232 | ctx->probe = get_kmmio_probe(addr); | ||
233 | ctx->saved_flags = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF)); | ||
234 | ctx->addr = addr; | ||
235 | |||
236 | if (ctx->probe && ctx->probe->pre_handler) | ||
237 | ctx->probe->pre_handler(ctx->probe, regs, addr); | ||
238 | |||
239 | /* | ||
240 | * Enable single-stepping and disable interrupts for the faulting | ||
241 | * context. Local interrupts must not get enabled during stepping. | ||
242 | */ | ||
243 | regs->flags |= X86_EFLAGS_TF; | ||
244 | regs->flags &= ~X86_EFLAGS_IF; | ||
245 | |||
246 | /* Now we set present bit in PTE and single step. */ | ||
247 | disarm_kmmio_fault_page(ctx->fpage->page, NULL); | ||
248 | |||
249 | /* | ||
250 | * If another cpu accesses the same page while we are stepping, | ||
251 | * the access will not be caught. It will simply succeed and the | ||
252 | * only downside is we lose the event. If this becomes a problem, | ||
253 | * the user should drop to single cpu before tracing. | ||
254 | */ | ||
255 | |||
256 | put_cpu_var(kmmio_ctx); | ||
257 | return 1; /* fault handled */ | ||
258 | |||
259 | no_kmmio_ctx: | ||
260 | put_cpu_var(kmmio_ctx); | ||
261 | no_kmmio: | ||
262 | rcu_read_unlock(); | ||
263 | preempt_enable_no_resched(); | ||
264 | return ret; | ||
265 | } | ||
266 | |||
267 | /* | ||
268 | * Interrupts are disabled on entry as trap1 is an interrupt gate | ||
269 | * and they remain disabled thorough out this function. | ||
270 | * This must always get called as the pair to kmmio_handler(). | ||
271 | */ | ||
272 | static int post_kmmio_handler(unsigned long condition, struct pt_regs *regs) | ||
273 | { | ||
274 | int ret = 0; | ||
275 | struct kmmio_context *ctx = &get_cpu_var(kmmio_ctx); | ||
276 | |||
277 | if (!ctx->active) { | ||
278 | pr_debug("kmmio: spurious debug trap on CPU %d.\n", | ||
279 | smp_processor_id()); | ||
280 | goto out; | ||
281 | } | ||
282 | |||
283 | if (ctx->probe && ctx->probe->post_handler) | ||
284 | ctx->probe->post_handler(ctx->probe, condition, regs); | ||
285 | |||
286 | arm_kmmio_fault_page(ctx->fpage->page, NULL); | ||
287 | |||
288 | regs->flags &= ~X86_EFLAGS_TF; | ||
289 | regs->flags |= ctx->saved_flags; | ||
290 | |||
291 | /* These were acquired in kmmio_handler(). */ | ||
292 | ctx->active--; | ||
293 | BUG_ON(ctx->active); | ||
294 | rcu_read_unlock(); | ||
295 | preempt_enable_no_resched(); | ||
296 | |||
297 | /* | ||
298 | * if somebody else is singlestepping across a probe point, flags | ||
299 | * will have TF set, in which case, continue the remaining processing | ||
300 | * of do_debug, as if this is not a probe hit. | ||
301 | */ | ||
302 | if (!(regs->flags & X86_EFLAGS_TF)) | ||
303 | ret = 1; | ||
304 | out: | ||
305 | put_cpu_var(kmmio_ctx); | ||
306 | return ret; | ||
307 | } | ||
308 | |||
309 | /* You must be holding kmmio_lock. */ | ||
310 | static int add_kmmio_fault_page(unsigned long page) | ||
311 | { | ||
312 | struct kmmio_fault_page *f; | ||
313 | |||
314 | page &= PAGE_MASK; | ||
315 | f = get_kmmio_fault_page(page); | ||
316 | if (f) { | ||
317 | if (!f->count) | ||
318 | arm_kmmio_fault_page(f->page, NULL); | ||
319 | f->count++; | ||
320 | return 0; | ||
321 | } | ||
322 | |||
323 | f = kmalloc(sizeof(*f), GFP_ATOMIC); | ||
324 | if (!f) | ||
325 | return -1; | ||
326 | |||
327 | f->count = 1; | ||
328 | f->page = page; | ||
329 | list_add_rcu(&f->list, kmmio_page_list(f->page)); | ||
330 | |||
331 | arm_kmmio_fault_page(f->page, NULL); | ||
332 | |||
333 | return 0; | ||
334 | } | ||
335 | |||
336 | /* You must be holding kmmio_lock. */ | ||
337 | static void release_kmmio_fault_page(unsigned long page, | ||
338 | struct kmmio_fault_page **release_list) | ||
339 | { | ||
340 | struct kmmio_fault_page *f; | ||
341 | |||
342 | page &= PAGE_MASK; | ||
343 | f = get_kmmio_fault_page(page); | ||
344 | if (!f) | ||
345 | return; | ||
346 | |||
347 | f->count--; | ||
348 | BUG_ON(f->count < 0); | ||
349 | if (!f->count) { | ||
350 | disarm_kmmio_fault_page(f->page, NULL); | ||
351 | f->release_next = *release_list; | ||
352 | *release_list = f; | ||
353 | } | ||
354 | } | ||
355 | |||
356 | /* | ||
357 | * With page-unaligned ioremaps, one or two armed pages may contain | ||
358 | * addresses from outside the intended mapping. Events for these addresses | ||
359 | * are currently silently dropped. The events may result only from programming | ||
360 | * mistakes by accessing addresses before the beginning or past the end of a | ||
361 | * mapping. | ||
362 | */ | ||
363 | int register_kmmio_probe(struct kmmio_probe *p) | ||
364 | { | ||
365 | unsigned long flags; | ||
366 | int ret = 0; | ||
367 | unsigned long size = 0; | ||
368 | const unsigned long size_lim = p->len + (p->addr & ~PAGE_MASK); | ||
369 | |||
370 | spin_lock_irqsave(&kmmio_lock, flags); | ||
371 | if (get_kmmio_probe(p->addr)) { | ||
372 | ret = -EEXIST; | ||
373 | goto out; | ||
374 | } | ||
375 | kmmio_count++; | ||
376 | list_add_rcu(&p->list, &kmmio_probes); | ||
377 | while (size < size_lim) { | ||
378 | if (add_kmmio_fault_page(p->addr + size)) | ||
379 | pr_err("kmmio: Unable to set page fault.\n"); | ||
380 | size += PAGE_SIZE; | ||
381 | } | ||
382 | out: | ||
383 | spin_unlock_irqrestore(&kmmio_lock, flags); | ||
384 | /* | ||
385 | * XXX: What should I do here? | ||
386 | * Here was a call to global_flush_tlb(), but it does not exist | ||
387 | * anymore. It seems it's not needed after all. | ||
388 | */ | ||
389 | return ret; | ||
390 | } | ||
391 | EXPORT_SYMBOL(register_kmmio_probe); | ||
392 | |||
393 | static void rcu_free_kmmio_fault_pages(struct rcu_head *head) | ||
394 | { | ||
395 | struct kmmio_delayed_release *dr = container_of( | ||
396 | head, | ||
397 | struct kmmio_delayed_release, | ||
398 | rcu); | ||
399 | struct kmmio_fault_page *p = dr->release_list; | ||
400 | while (p) { | ||
401 | struct kmmio_fault_page *next = p->release_next; | ||
402 | BUG_ON(p->count); | ||
403 | kfree(p); | ||
404 | p = next; | ||
405 | } | ||
406 | kfree(dr); | ||
407 | } | ||
408 | |||
409 | static void remove_kmmio_fault_pages(struct rcu_head *head) | ||
410 | { | ||
411 | struct kmmio_delayed_release *dr = container_of( | ||
412 | head, | ||
413 | struct kmmio_delayed_release, | ||
414 | rcu); | ||
415 | struct kmmio_fault_page *p = dr->release_list; | ||
416 | struct kmmio_fault_page **prevp = &dr->release_list; | ||
417 | unsigned long flags; | ||
418 | spin_lock_irqsave(&kmmio_lock, flags); | ||
419 | while (p) { | ||
420 | if (!p->count) | ||
421 | list_del_rcu(&p->list); | ||
422 | else | ||
423 | *prevp = p->release_next; | ||
424 | prevp = &p->release_next; | ||
425 | p = p->release_next; | ||
426 | } | ||
427 | spin_unlock_irqrestore(&kmmio_lock, flags); | ||
428 | /* This is the real RCU destroy call. */ | ||
429 | call_rcu(&dr->rcu, rcu_free_kmmio_fault_pages); | ||
430 | } | ||
431 | |||
432 | /* | ||
433 | * Remove a kmmio probe. You have to synchronize_rcu() before you can be | ||
434 | * sure that the callbacks will not be called anymore. Only after that | ||
435 | * you may actually release your struct kmmio_probe. | ||
436 | * | ||
437 | * Unregistering a kmmio fault page has three steps: | ||
438 | * 1. release_kmmio_fault_page() | ||
439 | * Disarm the page, wait a grace period to let all faults finish. | ||
440 | * 2. remove_kmmio_fault_pages() | ||
441 | * Remove the pages from kmmio_page_table. | ||
442 | * 3. rcu_free_kmmio_fault_pages() | ||
443 | * Actally free the kmmio_fault_page structs as with RCU. | ||
444 | */ | ||
445 | void unregister_kmmio_probe(struct kmmio_probe *p) | ||
446 | { | ||
447 | unsigned long flags; | ||
448 | unsigned long size = 0; | ||
449 | const unsigned long size_lim = p->len + (p->addr & ~PAGE_MASK); | ||
450 | struct kmmio_fault_page *release_list = NULL; | ||
451 | struct kmmio_delayed_release *drelease; | ||
452 | |||
453 | spin_lock_irqsave(&kmmio_lock, flags); | ||
454 | while (size < size_lim) { | ||
455 | release_kmmio_fault_page(p->addr + size, &release_list); | ||
456 | size += PAGE_SIZE; | ||
457 | } | ||
458 | list_del_rcu(&p->list); | ||
459 | kmmio_count--; | ||
460 | spin_unlock_irqrestore(&kmmio_lock, flags); | ||
461 | |||
462 | drelease = kmalloc(sizeof(*drelease), GFP_ATOMIC); | ||
463 | if (!drelease) { | ||
464 | pr_crit("kmmio: leaking kmmio_fault_page objects.\n"); | ||
465 | return; | ||
466 | } | ||
467 | drelease->release_list = release_list; | ||
468 | |||
469 | /* | ||
470 | * This is not really RCU here. We have just disarmed a set of | ||
471 | * pages so that they cannot trigger page faults anymore. However, | ||
472 | * we cannot remove the pages from kmmio_page_table, | ||
473 | * because a probe hit might be in flight on another CPU. The | ||
474 | * pages are collected into a list, and they will be removed from | ||
475 | * kmmio_page_table when it is certain that no probe hit related to | ||
476 | * these pages can be in flight. RCU grace period sounds like a | ||
477 | * good choice. | ||
478 | * | ||
479 | * If we removed the pages too early, kmmio page fault handler might | ||
480 | * not find the respective kmmio_fault_page and determine it's not | ||
481 | * a kmmio fault, when it actually is. This would lead to madness. | ||
482 | */ | ||
483 | call_rcu(&drelease->rcu, remove_kmmio_fault_pages); | ||
484 | } | ||
485 | EXPORT_SYMBOL(unregister_kmmio_probe); | ||
486 | |||
487 | static int kmmio_die_notifier(struct notifier_block *nb, unsigned long val, | ||
488 | void *args) | ||
489 | { | ||
490 | struct die_args *arg = args; | ||
491 | |||
492 | if (val == DIE_DEBUG && (arg->err & DR_STEP)) | ||
493 | if (post_kmmio_handler(arg->err, arg->regs) == 1) | ||
494 | return NOTIFY_STOP; | ||
495 | |||
496 | return NOTIFY_DONE; | ||
497 | } | ||
498 | |||
499 | static struct notifier_block nb_die = { | ||
500 | .notifier_call = kmmio_die_notifier | ||
501 | }; | ||
502 | |||
503 | static int __init init_kmmio(void) | ||
504 | { | ||
505 | int i; | ||
506 | for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++) | ||
507 | INIT_LIST_HEAD(&kmmio_page_table[i]); | ||
508 | return register_die_notifier(&nb_die); | ||
509 | } | ||
510 | fs_initcall(init_kmmio); /* should be before device_initcall() */ | ||
diff --git a/arch/x86/mm/mmio-mod.c b/arch/x86/mm/mmio-mod.c new file mode 100644 index 000000000000..e7397e108beb --- /dev/null +++ b/arch/x86/mm/mmio-mod.c | |||
@@ -0,0 +1,515 @@ | |||
1 | /* | ||
2 | * This program is free software; you can redistribute it and/or modify | ||
3 | * it under the terms of the GNU General Public License as published by | ||
4 | * the Free Software Foundation; either version 2 of the License, or | ||
5 | * (at your option) any later version. | ||
6 | * | ||
7 | * This program is distributed in the hope that it will be useful, | ||
8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
10 | * GNU General Public License for more details. | ||
11 | * | ||
12 | * You should have received a copy of the GNU General Public License | ||
13 | * along with this program; if not, write to the Free Software | ||
14 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | ||
15 | * | ||
16 | * Copyright (C) IBM Corporation, 2005 | ||
17 | * Jeff Muizelaar, 2006, 2007 | ||
18 | * Pekka Paalanen, 2008 <pq@iki.fi> | ||
19 | * | ||
20 | * Derived from the read-mod example from relay-examples by Tom Zanussi. | ||
21 | */ | ||
22 | #define DEBUG 1 | ||
23 | |||
24 | #include <linux/module.h> | ||
25 | #include <linux/debugfs.h> | ||
26 | #include <linux/uaccess.h> | ||
27 | #include <linux/io.h> | ||
28 | #include <linux/version.h> | ||
29 | #include <linux/kallsyms.h> | ||
30 | #include <asm/pgtable.h> | ||
31 | #include <linux/mmiotrace.h> | ||
32 | #include <asm/e820.h> /* for ISA_START_ADDRESS */ | ||
33 | #include <asm/atomic.h> | ||
34 | #include <linux/percpu.h> | ||
35 | #include <linux/cpu.h> | ||
36 | |||
37 | #include "pf_in.h" | ||
38 | |||
39 | #define NAME "mmiotrace: " | ||
40 | |||
41 | struct trap_reason { | ||
42 | unsigned long addr; | ||
43 | unsigned long ip; | ||
44 | enum reason_type type; | ||
45 | int active_traces; | ||
46 | }; | ||
47 | |||
48 | struct remap_trace { | ||
49 | struct list_head list; | ||
50 | struct kmmio_probe probe; | ||
51 | resource_size_t phys; | ||
52 | unsigned long id; | ||
53 | }; | ||
54 | |||
55 | /* Accessed per-cpu. */ | ||
56 | static DEFINE_PER_CPU(struct trap_reason, pf_reason); | ||
57 | static DEFINE_PER_CPU(struct mmiotrace_rw, cpu_trace); | ||
58 | |||
59 | #if 0 /* XXX: no way gather this info anymore */ | ||
60 | /* Access to this is not per-cpu. */ | ||
61 | static DEFINE_PER_CPU(atomic_t, dropped); | ||
62 | #endif | ||
63 | |||
64 | static struct dentry *marker_file; | ||
65 | |||
66 | static DEFINE_MUTEX(mmiotrace_mutex); | ||
67 | static DEFINE_SPINLOCK(trace_lock); | ||
68 | static atomic_t mmiotrace_enabled; | ||
69 | static LIST_HEAD(trace_list); /* struct remap_trace */ | ||
70 | |||
71 | /* | ||
72 | * Locking in this file: | ||
73 | * - mmiotrace_mutex enforces enable/disable_mmiotrace() critical sections. | ||
74 | * - mmiotrace_enabled may be modified only when holding mmiotrace_mutex | ||
75 | * and trace_lock. | ||
76 | * - Routines depending on is_enabled() must take trace_lock. | ||
77 | * - trace_list users must hold trace_lock. | ||
78 | * - is_enabled() guarantees that mmio_trace_record is allowed. | ||
79 | * - pre/post callbacks assume the effect of is_enabled() being true. | ||
80 | */ | ||
81 | |||
82 | /* module parameters */ | ||
83 | static unsigned long filter_offset; | ||
84 | static int nommiotrace; | ||
85 | static int trace_pc; | ||
86 | |||
87 | module_param(filter_offset, ulong, 0); | ||
88 | module_param(nommiotrace, bool, 0); | ||
89 | module_param(trace_pc, bool, 0); | ||
90 | |||
91 | MODULE_PARM_DESC(filter_offset, "Start address of traced mappings."); | ||
92 | MODULE_PARM_DESC(nommiotrace, "Disable actual MMIO tracing."); | ||
93 | MODULE_PARM_DESC(trace_pc, "Record address of faulting instructions."); | ||
94 | |||
95 | static bool is_enabled(void) | ||
96 | { | ||
97 | return atomic_read(&mmiotrace_enabled); | ||
98 | } | ||
99 | |||
100 | #if 0 /* XXX: needs rewrite */ | ||
101 | /* | ||
102 | * Write callback for the debugfs entry: | ||
103 | * Read a marker and write it to the mmio trace log | ||
104 | */ | ||
105 | static ssize_t write_marker(struct file *file, const char __user *buffer, | ||
106 | size_t count, loff_t *ppos) | ||
107 | { | ||
108 | char *event = NULL; | ||
109 | struct mm_io_header *headp; | ||
110 | ssize_t len = (count > 65535) ? 65535 : count; | ||
111 | |||
112 | event = kzalloc(sizeof(*headp) + len, GFP_KERNEL); | ||
113 | if (!event) | ||
114 | return -ENOMEM; | ||
115 | |||
116 | headp = (struct mm_io_header *)event; | ||
117 | headp->type = MMIO_MAGIC | (MMIO_MARKER << MMIO_OPCODE_SHIFT); | ||
118 | headp->data_len = len; | ||
119 | |||
120 | if (copy_from_user(event + sizeof(*headp), buffer, len)) { | ||
121 | kfree(event); | ||
122 | return -EFAULT; | ||
123 | } | ||
124 | |||
125 | spin_lock_irq(&trace_lock); | ||
126 | #if 0 /* XXX: convert this to use tracing */ | ||
127 | if (is_enabled()) | ||
128 | relay_write(chan, event, sizeof(*headp) + len); | ||
129 | else | ||
130 | #endif | ||
131 | len = -EINVAL; | ||
132 | spin_unlock_irq(&trace_lock); | ||
133 | kfree(event); | ||
134 | return len; | ||
135 | } | ||
136 | #endif | ||
137 | |||
138 | static void print_pte(unsigned long address) | ||
139 | { | ||
140 | unsigned int level; | ||
141 | pte_t *pte = lookup_address(address, &level); | ||
142 | |||
143 | if (!pte) { | ||
144 | pr_err(NAME "Error in %s: no pte for page 0x%08lx\n", | ||
145 | __func__, address); | ||
146 | return; | ||
147 | } | ||
148 | |||
149 | if (level == PG_LEVEL_2M) { | ||
150 | pr_emerg(NAME "4MB pages are not currently supported: " | ||
151 | "0x%08lx\n", address); | ||
152 | BUG(); | ||
153 | } | ||
154 | pr_info(NAME "pte for 0x%lx: 0x%llx 0x%llx\n", address, | ||
155 | (unsigned long long)pte_val(*pte), | ||
156 | (unsigned long long)pte_val(*pte) & _PAGE_PRESENT); | ||
157 | } | ||
158 | |||
159 | /* | ||
160 | * For some reason the pre/post pairs have been called in an | ||
161 | * unmatched order. Report and die. | ||
162 | */ | ||
163 | static void die_kmmio_nesting_error(struct pt_regs *regs, unsigned long addr) | ||
164 | { | ||
165 | const struct trap_reason *my_reason = &get_cpu_var(pf_reason); | ||
166 | pr_emerg(NAME "unexpected fault for address: 0x%08lx, " | ||
167 | "last fault for address: 0x%08lx\n", | ||
168 | addr, my_reason->addr); | ||
169 | print_pte(addr); | ||
170 | print_symbol(KERN_EMERG "faulting IP is at %s\n", regs->ip); | ||
171 | print_symbol(KERN_EMERG "last faulting IP was at %s\n", my_reason->ip); | ||
172 | #ifdef __i386__ | ||
173 | pr_emerg("eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n", | ||
174 | regs->ax, regs->bx, regs->cx, regs->dx); | ||
175 | pr_emerg("esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n", | ||
176 | regs->si, regs->di, regs->bp, regs->sp); | ||
177 | #else | ||
178 | pr_emerg("rax: %016lx rcx: %016lx rdx: %016lx\n", | ||
179 | regs->ax, regs->cx, regs->dx); | ||
180 | pr_emerg("rsi: %016lx rdi: %016lx rbp: %016lx rsp: %016lx\n", | ||
181 | regs->si, regs->di, regs->bp, regs->sp); | ||
182 | #endif | ||
183 | put_cpu_var(pf_reason); | ||
184 | BUG(); | ||
185 | } | ||
186 | |||
187 | static void pre(struct kmmio_probe *p, struct pt_regs *regs, | ||
188 | unsigned long addr) | ||
189 | { | ||
190 | struct trap_reason *my_reason = &get_cpu_var(pf_reason); | ||
191 | struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace); | ||
192 | const unsigned long instptr = instruction_pointer(regs); | ||
193 | const enum reason_type type = get_ins_type(instptr); | ||
194 | struct remap_trace *trace = p->private; | ||
195 | |||
196 | /* it doesn't make sense to have more than one active trace per cpu */ | ||
197 | if (my_reason->active_traces) | ||
198 | die_kmmio_nesting_error(regs, addr); | ||
199 | else | ||
200 | my_reason->active_traces++; | ||
201 | |||
202 | my_reason->type = type; | ||
203 | my_reason->addr = addr; | ||
204 | my_reason->ip = instptr; | ||
205 | |||
206 | my_trace->phys = addr - trace->probe.addr + trace->phys; | ||
207 | my_trace->map_id = trace->id; | ||
208 | |||
209 | /* | ||
210 | * Only record the program counter when requested. | ||
211 | * It may taint clean-room reverse engineering. | ||
212 | */ | ||
213 | if (trace_pc) | ||
214 | my_trace->pc = instptr; | ||
215 | else | ||
216 | my_trace->pc = 0; | ||
217 | |||
218 | /* | ||
219 | * XXX: the timestamp recorded will be *after* the tracing has been | ||
220 | * done, not at the time we hit the instruction. SMP implications | ||
221 | * on event ordering? | ||
222 | */ | ||
223 | |||
224 | switch (type) { | ||
225 | case REG_READ: | ||
226 | my_trace->opcode = MMIO_READ; | ||
227 | my_trace->width = get_ins_mem_width(instptr); | ||
228 | break; | ||
229 | case REG_WRITE: | ||
230 | my_trace->opcode = MMIO_WRITE; | ||
231 | my_trace->width = get_ins_mem_width(instptr); | ||
232 | my_trace->value = get_ins_reg_val(instptr, regs); | ||
233 | break; | ||
234 | case IMM_WRITE: | ||
235 | my_trace->opcode = MMIO_WRITE; | ||
236 | my_trace->width = get_ins_mem_width(instptr); | ||
237 | my_trace->value = get_ins_imm_val(instptr); | ||
238 | break; | ||
239 | default: | ||
240 | { | ||
241 | unsigned char *ip = (unsigned char *)instptr; | ||
242 | my_trace->opcode = MMIO_UNKNOWN_OP; | ||
243 | my_trace->width = 0; | ||
244 | my_trace->value = (*ip) << 16 | *(ip + 1) << 8 | | ||
245 | *(ip + 2); | ||
246 | } | ||
247 | } | ||
248 | put_cpu_var(cpu_trace); | ||
249 | put_cpu_var(pf_reason); | ||
250 | } | ||
251 | |||
252 | static void post(struct kmmio_probe *p, unsigned long condition, | ||
253 | struct pt_regs *regs) | ||
254 | { | ||
255 | struct trap_reason *my_reason = &get_cpu_var(pf_reason); | ||
256 | struct mmiotrace_rw *my_trace = &get_cpu_var(cpu_trace); | ||
257 | |||
258 | /* this should always return the active_trace count to 0 */ | ||
259 | my_reason->active_traces--; | ||
260 | if (my_reason->active_traces) { | ||
261 | pr_emerg(NAME "unexpected post handler"); | ||
262 | BUG(); | ||
263 | } | ||
264 | |||
265 | switch (my_reason->type) { | ||
266 | case REG_READ: | ||
267 | my_trace->value = get_ins_reg_val(my_reason->ip, regs); | ||
268 | break; | ||
269 | default: | ||
270 | break; | ||
271 | } | ||
272 | |||
273 | mmio_trace_rw(my_trace); | ||
274 | put_cpu_var(cpu_trace); | ||
275 | put_cpu_var(pf_reason); | ||
276 | } | ||
277 | |||
278 | static void ioremap_trace_core(resource_size_t offset, unsigned long size, | ||
279 | void __iomem *addr) | ||
280 | { | ||
281 | static atomic_t next_id; | ||
282 | struct remap_trace *trace = kmalloc(sizeof(*trace), GFP_KERNEL); | ||
283 | /* These are page-unaligned. */ | ||
284 | struct mmiotrace_map map = { | ||
285 | .phys = offset, | ||
286 | .virt = (unsigned long)addr, | ||
287 | .len = size, | ||
288 | .opcode = MMIO_PROBE | ||
289 | }; | ||
290 | |||
291 | if (!trace) { | ||
292 | pr_err(NAME "kmalloc failed in ioremap\n"); | ||
293 | return; | ||
294 | } | ||
295 | |||
296 | *trace = (struct remap_trace) { | ||
297 | .probe = { | ||
298 | .addr = (unsigned long)addr, | ||
299 | .len = size, | ||
300 | .pre_handler = pre, | ||
301 | .post_handler = post, | ||
302 | .private = trace | ||
303 | }, | ||
304 | .phys = offset, | ||
305 | .id = atomic_inc_return(&next_id) | ||
306 | }; | ||
307 | map.map_id = trace->id; | ||
308 | |||
309 | spin_lock_irq(&trace_lock); | ||
310 | if (!is_enabled()) | ||
311 | goto not_enabled; | ||
312 | |||
313 | mmio_trace_mapping(&map); | ||
314 | list_add_tail(&trace->list, &trace_list); | ||
315 | if (!nommiotrace) | ||
316 | register_kmmio_probe(&trace->probe); | ||
317 | |||
318 | not_enabled: | ||
319 | spin_unlock_irq(&trace_lock); | ||
320 | } | ||
321 | |||
322 | void mmiotrace_ioremap(resource_size_t offset, unsigned long size, | ||
323 | void __iomem *addr) | ||
324 | { | ||
325 | if (!is_enabled()) /* recheck and proper locking in *_core() */ | ||
326 | return; | ||
327 | |||
328 | pr_debug(NAME "ioremap_*(0x%llx, 0x%lx) = %p\n", | ||
329 | (unsigned long long)offset, size, addr); | ||
330 | if ((filter_offset) && (offset != filter_offset)) | ||
331 | return; | ||
332 | ioremap_trace_core(offset, size, addr); | ||
333 | } | ||
334 | |||
335 | static void iounmap_trace_core(volatile void __iomem *addr) | ||
336 | { | ||
337 | struct mmiotrace_map map = { | ||
338 | .phys = 0, | ||
339 | .virt = (unsigned long)addr, | ||
340 | .len = 0, | ||
341 | .opcode = MMIO_UNPROBE | ||
342 | }; | ||
343 | struct remap_trace *trace; | ||
344 | struct remap_trace *tmp; | ||
345 | struct remap_trace *found_trace = NULL; | ||
346 | |||
347 | pr_debug(NAME "Unmapping %p.\n", addr); | ||
348 | |||
349 | spin_lock_irq(&trace_lock); | ||
350 | if (!is_enabled()) | ||
351 | goto not_enabled; | ||
352 | |||
353 | list_for_each_entry_safe(trace, tmp, &trace_list, list) { | ||
354 | if ((unsigned long)addr == trace->probe.addr) { | ||
355 | if (!nommiotrace) | ||
356 | unregister_kmmio_probe(&trace->probe); | ||
357 | list_del(&trace->list); | ||
358 | found_trace = trace; | ||
359 | break; | ||
360 | } | ||
361 | } | ||
362 | map.map_id = (found_trace) ? found_trace->id : -1; | ||
363 | mmio_trace_mapping(&map); | ||
364 | |||
365 | not_enabled: | ||
366 | spin_unlock_irq(&trace_lock); | ||
367 | if (found_trace) { | ||
368 | synchronize_rcu(); /* unregister_kmmio_probe() requirement */ | ||
369 | kfree(found_trace); | ||
370 | } | ||
371 | } | ||
372 | |||
373 | void mmiotrace_iounmap(volatile void __iomem *addr) | ||
374 | { | ||
375 | might_sleep(); | ||
376 | if (is_enabled()) /* recheck and proper locking in *_core() */ | ||
377 | iounmap_trace_core(addr); | ||
378 | } | ||
379 | |||
380 | static void clear_trace_list(void) | ||
381 | { | ||
382 | struct remap_trace *trace; | ||
383 | struct remap_trace *tmp; | ||
384 | |||
385 | /* | ||
386 | * No locking required, because the caller ensures we are in a | ||
387 | * critical section via mutex, and is_enabled() is false, | ||
388 | * i.e. nothing can traverse or modify this list. | ||
389 | * Caller also ensures is_enabled() cannot change. | ||
390 | */ | ||
391 | list_for_each_entry(trace, &trace_list, list) { | ||
392 | pr_notice(NAME "purging non-iounmapped " | ||
393 | "trace @0x%08lx, size 0x%lx.\n", | ||
394 | trace->probe.addr, trace->probe.len); | ||
395 | if (!nommiotrace) | ||
396 | unregister_kmmio_probe(&trace->probe); | ||
397 | } | ||
398 | synchronize_rcu(); /* unregister_kmmio_probe() requirement */ | ||
399 | |||
400 | list_for_each_entry_safe(trace, tmp, &trace_list, list) { | ||
401 | list_del(&trace->list); | ||
402 | kfree(trace); | ||
403 | } | ||
404 | } | ||
405 | |||
406 | #ifdef CONFIG_HOTPLUG_CPU | ||
407 | static cpumask_t downed_cpus; | ||
408 | |||
409 | static void enter_uniprocessor(void) | ||
410 | { | ||
411 | int cpu; | ||
412 | int err; | ||
413 | |||
414 | get_online_cpus(); | ||
415 | downed_cpus = cpu_online_map; | ||
416 | cpu_clear(first_cpu(cpu_online_map), downed_cpus); | ||
417 | if (num_online_cpus() > 1) | ||
418 | pr_notice(NAME "Disabling non-boot CPUs...\n"); | ||
419 | put_online_cpus(); | ||
420 | |||
421 | for_each_cpu_mask(cpu, downed_cpus) { | ||
422 | err = cpu_down(cpu); | ||
423 | if (!err) | ||
424 | pr_info(NAME "CPU%d is down.\n", cpu); | ||
425 | else | ||
426 | pr_err(NAME "Error taking CPU%d down: %d\n", cpu, err); | ||
427 | } | ||
428 | if (num_online_cpus() > 1) | ||
429 | pr_warning(NAME "multiple CPUs still online, " | ||
430 | "may miss events.\n"); | ||
431 | } | ||
432 | |||
433 | static void leave_uniprocessor(void) | ||
434 | { | ||
435 | int cpu; | ||
436 | int err; | ||
437 | |||
438 | if (cpus_weight(downed_cpus) == 0) | ||
439 | return; | ||
440 | pr_notice(NAME "Re-enabling CPUs...\n"); | ||
441 | for_each_cpu_mask(cpu, downed_cpus) { | ||
442 | err = cpu_up(cpu); | ||
443 | if (!err) | ||
444 | pr_info(NAME "enabled CPU%d.\n", cpu); | ||
445 | else | ||
446 | pr_err(NAME "cannot re-enable CPU%d: %d\n", cpu, err); | ||
447 | } | ||
448 | } | ||
449 | |||
450 | #else /* !CONFIG_HOTPLUG_CPU */ | ||
451 | static void enter_uniprocessor(void) | ||
452 | { | ||
453 | if (num_online_cpus() > 1) | ||
454 | pr_warning(NAME "multiple CPUs are online, may miss events. " | ||
455 | "Suggest booting with maxcpus=1 kernel argument.\n"); | ||
456 | } | ||
457 | |||
458 | static void leave_uniprocessor(void) | ||
459 | { | ||
460 | } | ||
461 | #endif | ||
462 | |||
463 | #if 0 /* XXX: out of order */ | ||
464 | static struct file_operations fops_marker = { | ||
465 | .owner = THIS_MODULE, | ||
466 | .write = write_marker | ||
467 | }; | ||
468 | #endif | ||
469 | |||
470 | void enable_mmiotrace(void) | ||
471 | { | ||
472 | mutex_lock(&mmiotrace_mutex); | ||
473 | if (is_enabled()) | ||
474 | goto out; | ||
475 | |||
476 | #if 0 /* XXX: tracing does not support text entries */ | ||
477 | marker_file = debugfs_create_file("marker", 0660, dir, NULL, | ||
478 | &fops_marker); | ||
479 | if (!marker_file) | ||
480 | pr_err(NAME "marker file creation failed.\n"); | ||
481 | #endif | ||
482 | |||
483 | if (nommiotrace) | ||
484 | pr_info(NAME "MMIO tracing disabled.\n"); | ||
485 | enter_uniprocessor(); | ||
486 | spin_lock_irq(&trace_lock); | ||
487 | atomic_inc(&mmiotrace_enabled); | ||
488 | spin_unlock_irq(&trace_lock); | ||
489 | pr_info(NAME "enabled.\n"); | ||
490 | out: | ||
491 | mutex_unlock(&mmiotrace_mutex); | ||
492 | } | ||
493 | |||
494 | void disable_mmiotrace(void) | ||
495 | { | ||
496 | mutex_lock(&mmiotrace_mutex); | ||
497 | if (!is_enabled()) | ||
498 | goto out; | ||
499 | |||
500 | spin_lock_irq(&trace_lock); | ||
501 | atomic_dec(&mmiotrace_enabled); | ||
502 | BUG_ON(is_enabled()); | ||
503 | spin_unlock_irq(&trace_lock); | ||
504 | |||
505 | clear_trace_list(); /* guarantees: no more kmmio callbacks */ | ||
506 | leave_uniprocessor(); | ||
507 | if (marker_file) { | ||
508 | debugfs_remove(marker_file); | ||
509 | marker_file = NULL; | ||
510 | } | ||
511 | |||
512 | pr_info(NAME "disabled.\n"); | ||
513 | out: | ||
514 | mutex_unlock(&mmiotrace_mutex); | ||
515 | } | ||
diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c index 60bcb5b6a37e..57970f2935c0 100644 --- a/arch/x86/mm/pageattr.c +++ b/arch/x86/mm/pageattr.c | |||
@@ -227,6 +227,7 @@ pte_t *lookup_address(unsigned long address, unsigned int *level) | |||
227 | 227 | ||
228 | return pte_offset_kernel(pmd, address); | 228 | return pte_offset_kernel(pmd, address); |
229 | } | 229 | } |
230 | EXPORT_SYMBOL_GPL(lookup_address); | ||
230 | 231 | ||
231 | /* | 232 | /* |
232 | * Set the new pmd in all the pgds we know about: | 233 | * Set the new pmd in all the pgds we know about: |
diff --git a/arch/x86/mm/pf_in.c b/arch/x86/mm/pf_in.c new file mode 100644 index 000000000000..efa1911e20ca --- /dev/null +++ b/arch/x86/mm/pf_in.c | |||
@@ -0,0 +1,489 @@ | |||
1 | /* | ||
2 | * Fault Injection Test harness (FI) | ||
3 | * Copyright (C) Intel Crop. | ||
4 | * | ||
5 | * This program is free software; you can redistribute it and/or | ||
6 | * modify it under the terms of the GNU General Public License | ||
7 | * as published by the Free Software Foundation; either version 2 | ||
8 | * of the License, or (at your option) any later version. | ||
9 | * | ||
10 | * This program is distributed in the hope that it will be useful, | ||
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
13 | * GNU General Public License for more details. | ||
14 | * | ||
15 | * You should have received a copy of the GNU General Public License | ||
16 | * along with this program; if not, write to the Free Software | ||
17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, | ||
18 | * USA. | ||
19 | * | ||
20 | */ | ||
21 | |||
22 | /* Id: pf_in.c,v 1.1.1.1 2002/11/12 05:56:32 brlock Exp | ||
23 | * Copyright by Intel Crop., 2002 | ||
24 | * Louis Zhuang (louis.zhuang@intel.com) | ||
25 | * | ||
26 | * Bjorn Steinbrink (B.Steinbrink@gmx.de), 2007 | ||
27 | */ | ||
28 | |||
29 | #include <linux/module.h> | ||
30 | #include <linux/ptrace.h> /* struct pt_regs */ | ||
31 | #include "pf_in.h" | ||
32 | |||
33 | #ifdef __i386__ | ||
34 | /* IA32 Manual 3, 2-1 */ | ||
35 | static unsigned char prefix_codes[] = { | ||
36 | 0xF0, 0xF2, 0xF3, 0x2E, 0x36, 0x3E, 0x26, 0x64, | ||
37 | 0x65, 0x2E, 0x3E, 0x66, 0x67 | ||
38 | }; | ||
39 | /* IA32 Manual 3, 3-432*/ | ||
40 | static unsigned int reg_rop[] = { | ||
41 | 0x8A, 0x8B, 0xB60F, 0xB70F, 0xBE0F, 0xBF0F | ||
42 | }; | ||
43 | static unsigned int reg_wop[] = { 0x88, 0x89 }; | ||
44 | static unsigned int imm_wop[] = { 0xC6, 0xC7 }; | ||
45 | /* IA32 Manual 3, 3-432*/ | ||
46 | static unsigned int rw8[] = { 0x88, 0x8A, 0xC6 }; | ||
47 | static unsigned int rw32[] = { | ||
48 | 0x89, 0x8B, 0xC7, 0xB60F, 0xB70F, 0xBE0F, 0xBF0F | ||
49 | }; | ||
50 | static unsigned int mw8[] = { 0x88, 0x8A, 0xC6, 0xB60F, 0xBE0F }; | ||
51 | static unsigned int mw16[] = { 0xB70F, 0xBF0F }; | ||
52 | static unsigned int mw32[] = { 0x89, 0x8B, 0xC7 }; | ||
53 | static unsigned int mw64[] = {}; | ||
54 | #else /* not __i386__ */ | ||
55 | static unsigned char prefix_codes[] = { | ||
56 | 0x66, 0x67, 0x2E, 0x3E, 0x26, 0x64, 0x65, 0x36, | ||
57 | 0xF0, 0xF3, 0xF2, | ||
58 | /* REX Prefixes */ | ||
59 | 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, | ||
60 | 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f | ||
61 | }; | ||
62 | /* AMD64 Manual 3, Appendix A*/ | ||
63 | static unsigned int reg_rop[] = { | ||
64 | 0x8A, 0x8B, 0xB60F, 0xB70F, 0xBE0F, 0xBF0F | ||
65 | }; | ||
66 | static unsigned int reg_wop[] = { 0x88, 0x89 }; | ||
67 | static unsigned int imm_wop[] = { 0xC6, 0xC7 }; | ||
68 | static unsigned int rw8[] = { 0xC6, 0x88, 0x8A }; | ||
69 | static unsigned int rw32[] = { | ||
70 | 0xC7, 0x89, 0x8B, 0xB60F, 0xB70F, 0xBE0F, 0xBF0F | ||
71 | }; | ||
72 | /* 8 bit only */ | ||
73 | static unsigned int mw8[] = { 0xC6, 0x88, 0x8A, 0xB60F, 0xBE0F }; | ||
74 | /* 16 bit only */ | ||
75 | static unsigned int mw16[] = { 0xB70F, 0xBF0F }; | ||
76 | /* 16 or 32 bit */ | ||
77 | static unsigned int mw32[] = { 0xC7 }; | ||
78 | /* 16, 32 or 64 bit */ | ||
79 | static unsigned int mw64[] = { 0x89, 0x8B }; | ||
80 | #endif /* not __i386__ */ | ||
81 | |||
82 | static int skip_prefix(unsigned char *addr, int *shorted, int *enlarged, | ||
83 | int *rexr) | ||
84 | { | ||
85 | int i; | ||
86 | unsigned char *p = addr; | ||
87 | *shorted = 0; | ||
88 | *enlarged = 0; | ||
89 | *rexr = 0; | ||
90 | |||
91 | restart: | ||
92 | for (i = 0; i < ARRAY_SIZE(prefix_codes); i++) { | ||
93 | if (*p == prefix_codes[i]) { | ||
94 | if (*p == 0x66) | ||
95 | *shorted = 1; | ||
96 | #ifdef __amd64__ | ||
97 | if ((*p & 0xf8) == 0x48) | ||
98 | *enlarged = 1; | ||
99 | if ((*p & 0xf4) == 0x44) | ||
100 | *rexr = 1; | ||
101 | #endif | ||
102 | p++; | ||
103 | goto restart; | ||
104 | } | ||
105 | } | ||
106 | |||
107 | return (p - addr); | ||
108 | } | ||
109 | |||
110 | static int get_opcode(unsigned char *addr, unsigned int *opcode) | ||
111 | { | ||
112 | int len; | ||
113 | |||
114 | if (*addr == 0x0F) { | ||
115 | /* 0x0F is extension instruction */ | ||
116 | *opcode = *(unsigned short *)addr; | ||
117 | len = 2; | ||
118 | } else { | ||
119 | *opcode = *addr; | ||
120 | len = 1; | ||
121 | } | ||
122 | |||
123 | return len; | ||
124 | } | ||
125 | |||
126 | #define CHECK_OP_TYPE(opcode, array, type) \ | ||
127 | for (i = 0; i < ARRAY_SIZE(array); i++) { \ | ||
128 | if (array[i] == opcode) { \ | ||
129 | rv = type; \ | ||
130 | goto exit; \ | ||
131 | } \ | ||
132 | } | ||
133 | |||
134 | enum reason_type get_ins_type(unsigned long ins_addr) | ||
135 | { | ||
136 | unsigned int opcode; | ||
137 | unsigned char *p; | ||
138 | int shorted, enlarged, rexr; | ||
139 | int i; | ||
140 | enum reason_type rv = OTHERS; | ||
141 | |||
142 | p = (unsigned char *)ins_addr; | ||
143 | p += skip_prefix(p, &shorted, &enlarged, &rexr); | ||
144 | p += get_opcode(p, &opcode); | ||
145 | |||
146 | CHECK_OP_TYPE(opcode, reg_rop, REG_READ); | ||
147 | CHECK_OP_TYPE(opcode, reg_wop, REG_WRITE); | ||
148 | CHECK_OP_TYPE(opcode, imm_wop, IMM_WRITE); | ||
149 | |||
150 | exit: | ||
151 | return rv; | ||
152 | } | ||
153 | #undef CHECK_OP_TYPE | ||
154 | |||
155 | static unsigned int get_ins_reg_width(unsigned long ins_addr) | ||
156 | { | ||
157 | unsigned int opcode; | ||
158 | unsigned char *p; | ||
159 | int i, shorted, enlarged, rexr; | ||
160 | |||
161 | p = (unsigned char *)ins_addr; | ||
162 | p += skip_prefix(p, &shorted, &enlarged, &rexr); | ||
163 | p += get_opcode(p, &opcode); | ||
164 | |||
165 | for (i = 0; i < ARRAY_SIZE(rw8); i++) | ||
166 | if (rw8[i] == opcode) | ||
167 | return 1; | ||
168 | |||
169 | for (i = 0; i < ARRAY_SIZE(rw32); i++) | ||
170 | if (rw32[i] == opcode) | ||
171 | return (shorted ? 2 : (enlarged ? 8 : 4)); | ||
172 | |||
173 | printk(KERN_ERR "mmiotrace: Unknown opcode 0x%02x\n", opcode); | ||
174 | return 0; | ||
175 | } | ||
176 | |||
177 | unsigned int get_ins_mem_width(unsigned long ins_addr) | ||
178 | { | ||
179 | unsigned int opcode; | ||
180 | unsigned char *p; | ||
181 | int i, shorted, enlarged, rexr; | ||
182 | |||
183 | p = (unsigned char *)ins_addr; | ||
184 | p += skip_prefix(p, &shorted, &enlarged, &rexr); | ||
185 | p += get_opcode(p, &opcode); | ||
186 | |||
187 | for (i = 0; i < ARRAY_SIZE(mw8); i++) | ||
188 | if (mw8[i] == opcode) | ||
189 | return 1; | ||
190 | |||
191 | for (i = 0; i < ARRAY_SIZE(mw16); i++) | ||
192 | if (mw16[i] == opcode) | ||
193 | return 2; | ||
194 | |||
195 | for (i = 0; i < ARRAY_SIZE(mw32); i++) | ||
196 | if (mw32[i] == opcode) | ||
197 | return shorted ? 2 : 4; | ||
198 | |||
199 | for (i = 0; i < ARRAY_SIZE(mw64); i++) | ||
200 | if (mw64[i] == opcode) | ||
201 | return shorted ? 2 : (enlarged ? 8 : 4); | ||
202 | |||
203 | printk(KERN_ERR "mmiotrace: Unknown opcode 0x%02x\n", opcode); | ||
204 | return 0; | ||
205 | } | ||
206 | |||
207 | /* | ||
208 | * Define register ident in mod/rm byte. | ||
209 | * Note: these are NOT the same as in ptrace-abi.h. | ||
210 | */ | ||
211 | enum { | ||
212 | arg_AL = 0, | ||
213 | arg_CL = 1, | ||
214 | arg_DL = 2, | ||
215 | arg_BL = 3, | ||
216 | arg_AH = 4, | ||
217 | arg_CH = 5, | ||
218 | arg_DH = 6, | ||
219 | arg_BH = 7, | ||
220 | |||
221 | arg_AX = 0, | ||
222 | arg_CX = 1, | ||
223 | arg_DX = 2, | ||
224 | arg_BX = 3, | ||
225 | arg_SP = 4, | ||
226 | arg_BP = 5, | ||
227 | arg_SI = 6, | ||
228 | arg_DI = 7, | ||
229 | #ifdef __amd64__ | ||
230 | arg_R8 = 8, | ||
231 | arg_R9 = 9, | ||
232 | arg_R10 = 10, | ||
233 | arg_R11 = 11, | ||
234 | arg_R12 = 12, | ||
235 | arg_R13 = 13, | ||
236 | arg_R14 = 14, | ||
237 | arg_R15 = 15 | ||
238 | #endif | ||
239 | }; | ||
240 | |||
241 | static unsigned char *get_reg_w8(int no, struct pt_regs *regs) | ||
242 | { | ||
243 | unsigned char *rv = NULL; | ||
244 | |||
245 | switch (no) { | ||
246 | case arg_AL: | ||
247 | rv = (unsigned char *)®s->ax; | ||
248 | break; | ||
249 | case arg_BL: | ||
250 | rv = (unsigned char *)®s->bx; | ||
251 | break; | ||
252 | case arg_CL: | ||
253 | rv = (unsigned char *)®s->cx; | ||
254 | break; | ||
255 | case arg_DL: | ||
256 | rv = (unsigned char *)®s->dx; | ||
257 | break; | ||
258 | case arg_AH: | ||
259 | rv = 1 + (unsigned char *)®s->ax; | ||
260 | break; | ||
261 | case arg_BH: | ||
262 | rv = 1 + (unsigned char *)®s->bx; | ||
263 | break; | ||
264 | case arg_CH: | ||
265 | rv = 1 + (unsigned char *)®s->cx; | ||
266 | break; | ||
267 | case arg_DH: | ||
268 | rv = 1 + (unsigned char *)®s->dx; | ||
269 | break; | ||
270 | #ifdef __amd64__ | ||
271 | case arg_R8: | ||
272 | rv = (unsigned char *)®s->r8; | ||
273 | break; | ||
274 | case arg_R9: | ||
275 | rv = (unsigned char *)®s->r9; | ||
276 | break; | ||
277 | case arg_R10: | ||
278 | rv = (unsigned char *)®s->r10; | ||
279 | break; | ||
280 | case arg_R11: | ||
281 | rv = (unsigned char *)®s->r11; | ||
282 | break; | ||
283 | case arg_R12: | ||
284 | rv = (unsigned char *)®s->r12; | ||
285 | break; | ||
286 | case arg_R13: | ||
287 | rv = (unsigned char *)®s->r13; | ||
288 | break; | ||
289 | case arg_R14: | ||
290 | rv = (unsigned char *)®s->r14; | ||
291 | break; | ||
292 | case arg_R15: | ||
293 | rv = (unsigned char *)®s->r15; | ||
294 | break; | ||
295 | #endif | ||
296 | default: | ||
297 | printk(KERN_ERR "mmiotrace: Error reg no# %d\n", no); | ||
298 | break; | ||
299 | } | ||
300 | return rv; | ||
301 | } | ||
302 | |||
303 | static unsigned long *get_reg_w32(int no, struct pt_regs *regs) | ||
304 | { | ||
305 | unsigned long *rv = NULL; | ||
306 | |||
307 | switch (no) { | ||
308 | case arg_AX: | ||
309 | rv = ®s->ax; | ||
310 | break; | ||
311 | case arg_BX: | ||
312 | rv = ®s->bx; | ||
313 | break; | ||
314 | case arg_CX: | ||
315 | rv = ®s->cx; | ||
316 | break; | ||
317 | case arg_DX: | ||
318 | rv = ®s->dx; | ||
319 | break; | ||
320 | case arg_SP: | ||
321 | rv = ®s->sp; | ||
322 | break; | ||
323 | case arg_BP: | ||
324 | rv = ®s->bp; | ||
325 | break; | ||
326 | case arg_SI: | ||
327 | rv = ®s->si; | ||
328 | break; | ||
329 | case arg_DI: | ||
330 | rv = ®s->di; | ||
331 | break; | ||
332 | #ifdef __amd64__ | ||
333 | case arg_R8: | ||
334 | rv = ®s->r8; | ||
335 | break; | ||
336 | case arg_R9: | ||
337 | rv = ®s->r9; | ||
338 | break; | ||
339 | case arg_R10: | ||
340 | rv = ®s->r10; | ||
341 | break; | ||
342 | case arg_R11: | ||
343 | rv = ®s->r11; | ||
344 | break; | ||
345 | case arg_R12: | ||
346 | rv = ®s->r12; | ||
347 | break; | ||
348 | case arg_R13: | ||
349 | rv = ®s->r13; | ||
350 | break; | ||
351 | case arg_R14: | ||
352 | rv = ®s->r14; | ||
353 | break; | ||
354 | case arg_R15: | ||
355 | rv = ®s->r15; | ||
356 | break; | ||
357 | #endif | ||
358 | default: | ||
359 | printk(KERN_ERR "mmiotrace: Error reg no# %d\n", no); | ||
360 | } | ||
361 | |||
362 | return rv; | ||
363 | } | ||
364 | |||
365 | unsigned long get_ins_reg_val(unsigned long ins_addr, struct pt_regs *regs) | ||
366 | { | ||
367 | unsigned int opcode; | ||
368 | unsigned char mod_rm; | ||
369 | int reg; | ||
370 | unsigned char *p; | ||
371 | int i, shorted, enlarged, rexr; | ||
372 | unsigned long rv; | ||
373 | |||
374 | p = (unsigned char *)ins_addr; | ||
375 | p += skip_prefix(p, &shorted, &enlarged, &rexr); | ||
376 | p += get_opcode(p, &opcode); | ||
377 | for (i = 0; i < ARRAY_SIZE(reg_rop); i++) | ||
378 | if (reg_rop[i] == opcode) { | ||
379 | rv = REG_READ; | ||
380 | goto do_work; | ||
381 | } | ||
382 | |||
383 | for (i = 0; i < ARRAY_SIZE(reg_wop); i++) | ||
384 | if (reg_wop[i] == opcode) { | ||
385 | rv = REG_WRITE; | ||
386 | goto do_work; | ||
387 | } | ||
388 | |||
389 | printk(KERN_ERR "mmiotrace: Not a register instruction, opcode " | ||
390 | "0x%02x\n", opcode); | ||
391 | goto err; | ||
392 | |||
393 | do_work: | ||
394 | mod_rm = *p; | ||
395 | reg = ((mod_rm >> 3) & 0x7) | (rexr << 3); | ||
396 | switch (get_ins_reg_width(ins_addr)) { | ||
397 | case 1: | ||
398 | return *get_reg_w8(reg, regs); | ||
399 | |||
400 | case 2: | ||
401 | return *(unsigned short *)get_reg_w32(reg, regs); | ||
402 | |||
403 | case 4: | ||
404 | return *(unsigned int *)get_reg_w32(reg, regs); | ||
405 | |||
406 | #ifdef __amd64__ | ||
407 | case 8: | ||
408 | return *(unsigned long *)get_reg_w32(reg, regs); | ||
409 | #endif | ||
410 | |||
411 | default: | ||
412 | printk(KERN_ERR "mmiotrace: Error width# %d\n", reg); | ||
413 | } | ||
414 | |||
415 | err: | ||
416 | return 0; | ||
417 | } | ||
418 | |||
419 | unsigned long get_ins_imm_val(unsigned long ins_addr) | ||
420 | { | ||
421 | unsigned int opcode; | ||
422 | unsigned char mod_rm; | ||
423 | unsigned char mod; | ||
424 | unsigned char *p; | ||
425 | int i, shorted, enlarged, rexr; | ||
426 | unsigned long rv; | ||
427 | |||
428 | p = (unsigned char *)ins_addr; | ||
429 | p += skip_prefix(p, &shorted, &enlarged, &rexr); | ||
430 | p += get_opcode(p, &opcode); | ||
431 | for (i = 0; i < ARRAY_SIZE(imm_wop); i++) | ||
432 | if (imm_wop[i] == opcode) { | ||
433 | rv = IMM_WRITE; | ||
434 | goto do_work; | ||
435 | } | ||
436 | |||
437 | printk(KERN_ERR "mmiotrace: Not an immediate instruction, opcode " | ||
438 | "0x%02x\n", opcode); | ||
439 | goto err; | ||
440 | |||
441 | do_work: | ||
442 | mod_rm = *p; | ||
443 | mod = mod_rm >> 6; | ||
444 | p++; | ||
445 | switch (mod) { | ||
446 | case 0: | ||
447 | /* if r/m is 5 we have a 32 disp (IA32 Manual 3, Table 2-2) */ | ||
448 | /* AMD64: XXX Check for address size prefix? */ | ||
449 | if ((mod_rm & 0x7) == 0x5) | ||
450 | p += 4; | ||
451 | break; | ||
452 | |||
453 | case 1: | ||
454 | p += 1; | ||
455 | break; | ||
456 | |||
457 | case 2: | ||
458 | p += 4; | ||
459 | break; | ||
460 | |||
461 | case 3: | ||
462 | default: | ||
463 | printk(KERN_ERR "mmiotrace: not a memory access instruction " | ||
464 | "at 0x%lx, rm_mod=0x%02x\n", | ||
465 | ins_addr, mod_rm); | ||
466 | } | ||
467 | |||
468 | switch (get_ins_reg_width(ins_addr)) { | ||
469 | case 1: | ||
470 | return *(unsigned char *)p; | ||
471 | |||
472 | case 2: | ||
473 | return *(unsigned short *)p; | ||
474 | |||
475 | case 4: | ||
476 | return *(unsigned int *)p; | ||
477 | |||
478 | #ifdef __amd64__ | ||
479 | case 8: | ||
480 | return *(unsigned long *)p; | ||
481 | #endif | ||
482 | |||
483 | default: | ||
484 | printk(KERN_ERR "mmiotrace: Error: width.\n"); | ||
485 | } | ||
486 | |||
487 | err: | ||
488 | return 0; | ||
489 | } | ||
diff --git a/arch/x86/mm/pf_in.h b/arch/x86/mm/pf_in.h new file mode 100644 index 000000000000..e05341a51a27 --- /dev/null +++ b/arch/x86/mm/pf_in.h | |||
@@ -0,0 +1,39 @@ | |||
1 | /* | ||
2 | * Fault Injection Test harness (FI) | ||
3 | * Copyright (C) Intel Crop. | ||
4 | * | ||
5 | * This program is free software; you can redistribute it and/or | ||
6 | * modify it under the terms of the GNU General Public License | ||
7 | * as published by the Free Software Foundation; either version 2 | ||
8 | * of the License, or (at your option) any later version. | ||
9 | * | ||
10 | * This program is distributed in the hope that it will be useful, | ||
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
13 | * GNU General Public License for more details. | ||
14 | * | ||
15 | * You should have received a copy of the GNU General Public License | ||
16 | * along with this program; if not, write to the Free Software | ||
17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, | ||
18 | * USA. | ||
19 | * | ||
20 | */ | ||
21 | |||
22 | #ifndef __PF_H_ | ||
23 | #define __PF_H_ | ||
24 | |||
25 | enum reason_type { | ||
26 | NOT_ME, /* page fault is not in regions */ | ||
27 | NOTHING, /* access others point in regions */ | ||
28 | REG_READ, /* read from addr to reg */ | ||
29 | REG_WRITE, /* write from reg to addr */ | ||
30 | IMM_WRITE, /* write from imm to addr */ | ||
31 | OTHERS /* Other instructions can not intercept */ | ||
32 | }; | ||
33 | |||
34 | enum reason_type get_ins_type(unsigned long ins_addr); | ||
35 | unsigned int get_ins_mem_width(unsigned long ins_addr); | ||
36 | unsigned long get_ins_reg_val(unsigned long ins_addr, struct pt_regs *regs); | ||
37 | unsigned long get_ins_imm_val(unsigned long ins_addr); | ||
38 | |||
39 | #endif /* __PF_H_ */ | ||
diff --git a/arch/x86/mm/testmmiotrace.c b/arch/x86/mm/testmmiotrace.c new file mode 100644 index 000000000000..d877c5b423ef --- /dev/null +++ b/arch/x86/mm/testmmiotrace.c | |||
@@ -0,0 +1,71 @@ | |||
1 | /* | ||
2 | * Written by Pekka Paalanen, 2008 <pq@iki.fi> | ||
3 | */ | ||
4 | #include <linux/module.h> | ||
5 | #include <linux/io.h> | ||
6 | |||
7 | #define MODULE_NAME "testmmiotrace" | ||
8 | |||
9 | static unsigned long mmio_address; | ||
10 | module_param(mmio_address, ulong, 0); | ||
11 | MODULE_PARM_DESC(mmio_address, "Start address of the mapping of 16 kB."); | ||
12 | |||
13 | static void do_write_test(void __iomem *p) | ||
14 | { | ||
15 | unsigned int i; | ||
16 | for (i = 0; i < 256; i++) | ||
17 | iowrite8(i, p + i); | ||
18 | for (i = 1024; i < (5 * 1024); i += 2) | ||
19 | iowrite16(i * 12 + 7, p + i); | ||
20 | for (i = (5 * 1024); i < (16 * 1024); i += 4) | ||
21 | iowrite32(i * 212371 + 13, p + i); | ||
22 | } | ||
23 | |||
24 | static void do_read_test(void __iomem *p) | ||
25 | { | ||
26 | unsigned int i; | ||
27 | for (i = 0; i < 256; i++) | ||
28 | ioread8(p + i); | ||
29 | for (i = 1024; i < (5 * 1024); i += 2) | ||
30 | ioread16(p + i); | ||
31 | for (i = (5 * 1024); i < (16 * 1024); i += 4) | ||
32 | ioread32(p + i); | ||
33 | } | ||
34 | |||
35 | static void do_test(void) | ||
36 | { | ||
37 | void __iomem *p = ioremap_nocache(mmio_address, 0x4000); | ||
38 | if (!p) { | ||
39 | pr_err(MODULE_NAME ": could not ioremap, aborting.\n"); | ||
40 | return; | ||
41 | } | ||
42 | do_write_test(p); | ||
43 | do_read_test(p); | ||
44 | iounmap(p); | ||
45 | } | ||
46 | |||
47 | static int __init init(void) | ||
48 | { | ||
49 | if (mmio_address == 0) { | ||
50 | pr_err(MODULE_NAME ": you have to use the module argument " | ||
51 | "mmio_address.\n"); | ||
52 | pr_err(MODULE_NAME ": DO NOT LOAD THIS MODULE UNLESS" | ||
53 | " YOU REALLY KNOW WHAT YOU ARE DOING!\n"); | ||
54 | return -ENXIO; | ||
55 | } | ||
56 | |||
57 | pr_warning(MODULE_NAME ": WARNING: mapping 16 kB @ 0x%08lx " | ||
58 | "in PCI address space, and writing " | ||
59 | "rubbish in there.\n", mmio_address); | ||
60 | do_test(); | ||
61 | return 0; | ||
62 | } | ||
63 | |||
64 | static void __exit cleanup(void) | ||
65 | { | ||
66 | pr_debug(MODULE_NAME ": unloaded.\n"); | ||
67 | } | ||
68 | |||
69 | module_init(init); | ||
70 | module_exit(cleanup); | ||
71 | MODULE_LICENSE("GPL"); | ||