Re: [patch 4/8] Immediate Value - i386 Optimization; kprobes

From: Andrew Morton
Date: Mon Jun 18 2007 - 14:57:47 EST


On Mon, 18 Jun 2007 14:44:57 -0400
Chuck Ebbert <cebbert@xxxxxxxxxx> wrote:

> > I fancy it's done by the kprobe_page_fault handler, but I do not see
> > clearly how writing the breakpoint from arch_arm_kprobe() in
> > non-writeable memory is done.
>
> Looks like it's not merged yet:
>
> http://lkml.org/lkml/2007/6/7/2
>
> This needs to go in before 2.6.22-final

Andi, I'll include the below two patches in the next batch, OK?



From: "S. P. Prasanna" <prasanna@xxxxxxxxxx>

Fix the problem of page protection introduced by CONFIG_DEBUG_RODATA for
x86_64 architecture. As per Andi Kleen's suggestion, the kernel text pages
are marked writeable only for a short duration to insert or remove the
breakpoints.

Signed-off-by: Prasanna S P<prasanna@xxxxxxxxxx>
Acked-by: Jim Keniston <jkenisto@xxxxxxxxxx>
Cc: Andi Kleen <ak@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---

arch/x86_64/kernel/kprobes.c | 26 ++++++++++++++++++++++++++
arch/x86_64/mm/init.c | 6 +++++-
include/asm-x86_64/kprobes.h | 10 ++++++++++
3 files changed, 41 insertions(+), 1 deletion(-)

diff -puN arch/x86_64/kernel/kprobes.c~kprobes-x86_64-fix-for-mark-ro-data arch/x86_64/kernel/kprobes.c
--- a/arch/x86_64/kernel/kprobes.c~kprobes-x86_64-fix-for-mark-ro-data
+++ a/arch/x86_64/kernel/kprobes.c
@@ -209,16 +209,42 @@ static void __kprobes arch_copy_kprobe(s

void __kprobes arch_arm_kprobe(struct kprobe *p)
{
+ unsigned long addr = (unsigned long)p->addr;
+ int page_readonly = 0;
+
+ if (kernel_readonly_text(addr)) {
+ change_page_attr_addr(addr, 1, PAGE_KERNEL_EXEC);
+ global_flush_tlb();
+ page_readonly = 1;
+ }
*p->addr = BREAKPOINT_INSTRUCTION;
flush_icache_range((unsigned long) p->addr,
(unsigned long) p->addr + sizeof(kprobe_opcode_t));
+ if (page_readonly) {
+ change_page_attr_addr(addr, 1, PAGE_KERNEL_RO);
+ global_flush_tlb();
+ }
}

void __kprobes arch_disarm_kprobe(struct kprobe *p)
{
+ unsigned long addr = (unsigned long)p->addr;
+ int page_readonly = 0;
+
+ if (kernel_readonly_text(addr)) {
+ change_page_attr_addr(addr, 1, PAGE_KERNEL_EXEC);
+ global_flush_tlb();
+ page_readonly = 1;
+ }
+
*p->addr = p->opcode;
flush_icache_range((unsigned long) p->addr,
(unsigned long) p->addr + sizeof(kprobe_opcode_t));
+
+ if (page_readonly) {
+ change_page_attr_addr(addr, 1, PAGE_KERNEL_RO);
+ global_flush_tlb();
+ }
}

void __kprobes arch_remove_kprobe(struct kprobe *p)
diff -puN arch/x86_64/mm/init.c~kprobes-x86_64-fix-for-mark-ro-data arch/x86_64/mm/init.c
--- a/arch/x86_64/mm/init.c~kprobes-x86_64-fix-for-mark-ro-data
+++ a/arch/x86_64/mm/init.c
@@ -48,6 +48,7 @@
#define Dprintk(x...)
#endif

+int kernel_text_is_ro;
const struct dma_mapping_ops* dma_ops;
EXPORT_SYMBOL(dma_ops);

@@ -600,10 +601,13 @@ void mark_rodata_ro(void)
{
unsigned long start = (unsigned long)_stext, end;

+ kernel_text_is_ro = 1;
#ifdef CONFIG_HOTPLUG_CPU
/* It must still be possible to apply SMP alternatives. */
- if (num_possible_cpus() > 1)
+ if (num_possible_cpus() > 1) {
start = (unsigned long)_etext;
+ kernel_text_is_ro = 0;
+ }
#endif
end = (unsigned long)__end_rodata;
start = (start + PAGE_SIZE - 1) & PAGE_MASK;
diff -puN include/asm-x86_64/kprobes.h~kprobes-x86_64-fix-for-mark-ro-data include/asm-x86_64/kprobes.h
--- a/include/asm-x86_64/kprobes.h~kprobes-x86_64-fix-for-mark-ro-data
+++ a/include/asm-x86_64/kprobes.h
@@ -26,6 +26,7 @@
#include <linux/types.h>
#include <linux/ptrace.h>
#include <linux/percpu.h>
+#include <asm-generic/sections.h>

#define __ARCH_WANT_KPROBES_INSN_SLOT

@@ -88,4 +89,13 @@ extern int kprobe_handler(struct pt_regs

extern int kprobe_exceptions_notify(struct notifier_block *self,
unsigned long val, void *data);
+extern int kernel_text_is_ro;
+static inline int kernel_readonly_text(unsigned long address)
+{
+ if (kernel_text_is_ro && ((address >= (unsigned long)_stext)
+ && (address < (unsigned long) _etext)))
+ return 1;
+
+ return 0;
+}
#endif /* _ASM_KPROBES_H */
_



From: "S. P. Prasanna" <prasanna@xxxxxxxxxx>

Fix the problem of page protection introduced by CONFIG_DEBUG_RODATA.
CONFIG_DEBUG_RODATA marks the text pages as read-only, hence kprobes is
unable to insert breakpoints in the kernel text. This patch overrides the
page protection when adding or removing a probe for the i386 architecture.

Signed-off-by: Prasanna S P<prasanna@xxxxxxxxxx>
Acked-by: Jim Keniston <jkenisto@xxxxxxxxxx>
Cc: Andi Kleen <ak@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---

arch/i386/kernel/kprobes.c | 26 ++++++++++++++++++++++++++
arch/i386/mm/init.c | 2 ++
include/asm-i386/kprobes.h | 12 ++++++++++++
include/asm-i386/pgtable.h | 2 ++
4 files changed, 42 insertions(+)

diff -puN arch/i386/kernel/kprobes.c~kprobes-i386-fix-for-mark-ro-data arch/i386/kernel/kprobes.c
--- a/arch/i386/kernel/kprobes.c~kprobes-i386-fix-for-mark-ro-data
+++ a/arch/i386/kernel/kprobes.c
@@ -169,16 +169,42 @@ int __kprobes arch_prepare_kprobe(struct

void __kprobes arch_arm_kprobe(struct kprobe *p)
{
+ unsigned long addr = (unsigned long) p->addr;
+ int page_readonly = 0;
+
+ if (kernel_readonly_text(addr)) {
+ page_readonly = 1;
+ change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RWX);
+ global_flush_tlb();
+ }
+
*p->addr = BREAKPOINT_INSTRUCTION;
flush_icache_range((unsigned long) p->addr,
(unsigned long) p->addr + sizeof(kprobe_opcode_t));
+
+ if (page_readonly) {
+ change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RX);
+ global_flush_tlb();
+ }
}

void __kprobes arch_disarm_kprobe(struct kprobe *p)
{
+ unsigned long addr = (unsigned long) p->addr;
+ int page_readonly = 0;
+
+ if (kernel_readonly_text(addr)) {
+ page_readonly = 1;
+ change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RWX);
+ global_flush_tlb();
+ }
*p->addr = p->opcode;
flush_icache_range((unsigned long) p->addr,
(unsigned long) p->addr + sizeof(kprobe_opcode_t));
+ if (page_readonly) {
+ change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RX);
+ global_flush_tlb();
+ }
}

void __kprobes arch_remove_kprobe(struct kprobe *p)
diff -puN arch/i386/mm/init.c~kprobes-i386-fix-for-mark-ro-data arch/i386/mm/init.c
--- a/arch/i386/mm/init.c~kprobes-i386-fix-for-mark-ro-data
+++ a/arch/i386/mm/init.c
@@ -45,6 +45,7 @@
#include <asm/sections.h>
#include <asm/paravirt.h>

+int kernel_text_is_ro;
unsigned int __VMALLOC_RESERVE = 128 << 20;

DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
@@ -808,6 +809,7 @@ void mark_rodata_ro(void)
change_page_attr(virt_to_page(start),
size >> PAGE_SHIFT, PAGE_KERNEL_RX);
printk("Write protecting the kernel text: %luk\n", size >> 10);
+ kernel_text_is_ro = 1;
}

start += size;
diff -puN include/asm-i386/kprobes.h~kprobes-i386-fix-for-mark-ro-data include/asm-i386/kprobes.h
--- a/include/asm-i386/kprobes.h~kprobes-i386-fix-for-mark-ro-data
+++ a/include/asm-i386/kprobes.h
@@ -26,6 +26,8 @@
*/
#include <linux/types.h>
#include <linux/ptrace.h>
+#include <linux/pfn.h>
+#include <asm-generic/sections.h>

#define __ARCH_WANT_KPROBES_INSN_SLOT

@@ -90,4 +92,14 @@ static inline void restore_interrupts(st

extern int kprobe_exceptions_notify(struct notifier_block *self,
unsigned long val, void *data);
+extern int kernel_text_is_ro;
+static inline int kernel_readonly_text(unsigned long address)
+{
+
+ if (kernel_text_is_ro && ((address >= PFN_ALIGN(_text))
+ && (address < PFN_ALIGN(_etext))))
+ return 1;
+
+ return 0;
+}
#endif /* _ASM_KPROBES_H */
diff -puN include/asm-i386/pgtable.h~kprobes-i386-fix-for-mark-ro-data include/asm-i386/pgtable.h
--- a/include/asm-i386/pgtable.h~kprobes-i386-fix-for-mark-ro-data
+++ a/include/asm-i386/pgtable.h
@@ -159,6 +159,7 @@ void paging_init(void);
extern unsigned long long __PAGE_KERNEL, __PAGE_KERNEL_EXEC;
#define __PAGE_KERNEL_RO (__PAGE_KERNEL & ~_PAGE_RW)
#define __PAGE_KERNEL_RX (__PAGE_KERNEL_EXEC & ~_PAGE_RW)
+#define __PAGE_KERNEL_RWX (__PAGE_KERNEL_EXEC | _PAGE_RW)
#define __PAGE_KERNEL_NOCACHE (__PAGE_KERNEL | _PAGE_PCD)
#define __PAGE_KERNEL_LARGE (__PAGE_KERNEL | _PAGE_PSE)
#define __PAGE_KERNEL_LARGE_EXEC (__PAGE_KERNEL_EXEC | _PAGE_PSE)
@@ -167,6 +168,7 @@ extern unsigned long long __PAGE_KERNEL,
#define PAGE_KERNEL_RO __pgprot(__PAGE_KERNEL_RO)
#define PAGE_KERNEL_EXEC __pgprot(__PAGE_KERNEL_EXEC)
#define PAGE_KERNEL_RX __pgprot(__PAGE_KERNEL_RX)
+#define PAGE_KERNEL_RWX __pgprot(__PAGE_KERNEL_RWX)
#define PAGE_KERNEL_NOCACHE __pgprot(__PAGE_KERNEL_NOCACHE)
#define PAGE_KERNEL_LARGE __pgprot(__PAGE_KERNEL_LARGE)
#define PAGE_KERNEL_LARGE_EXEC __pgprot(__PAGE_KERNEL_LARGE_EXEC)
_

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/