Skip to contentSkip to conversation

fix search_extable() to find correct entry

RFC PATCH In review Masami Hiramatsu 3 participants 5 messages Jun 15, 2006 – Jun 16, 2006

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 fix search_extable() to find correct entry

Hi, Tony

I found a suspicious buggy code in the linux kernel on IA64 arch. As far as I can see, search_extable() doesn't work correctly, because the lookup routine expects that the address format of the exception_table_entry is "IP + slot", but the compiler (gcc-3.4.5) generates it as "IP + (slot << 2)". Thus the lookup routine always fails to find the corresponding entry.
You can check it by dumping __ex_table section of vmlinux.

I made a patch to fix this bug attached in this mail. This patch is against 2.6.17-rc6-mm2.
Please review it.

Description:
Fix search_extable() and ia64_handle_exception() to handle the
address format of exception_table_entry correctly.

Thanks,

Signature · 57 lines
Masami HIRAMATSU
2nd Research Dept.
Hitachi, Ltd., Systems Development Laboratory
E-mail: [email protected]

Signed-off-by: Masami Hiramatsu <[email protected]>

 arch/ia64/mm/extable.c     |    7 ++++---
 include/asm-ia64/uaccess.h |    4 ++--
 2 files changed, 6 insertions(+), 5 deletions(-)
diff --exclude=CVS -Narup a/arch/ia64/mm/extable.c b/arch/ia64/mm/extable.c
--- a/arch/ia64/mm/extable.c	2005-10-28 09:02:08.000000000 +0900
+++ b/arch/ia64/mm/extable.c	2006-06-14 14:59:11.000000000 +0900
@@ -63,9 +63,10 @@ search_extable (const struct exception_t
 	unsigned long mid_ip;
 	long diff;

+	ip = (ip & ~0xf) + ((ip & 0x3) << 2);
         while (first <= last) {
 		mid = &first[(last - first)/2];
-		mid_ip = (u64) &mid->addr + mid->addr;
+		mid_ip = ((u64) &mid->addr + mid->addr) & ~0x3;
 		diff = mid_ip - ip;
                 if (diff = 0)
                         return mid;
@@ -83,8 +84,8 @@ ia64_handle_exception (struct pt_regs *r
 	long fix = (u64) &e->cont + e->cont;

 	regs->r8 = -EFAULT;
-	if (fix & 4)
+	if (fix & 2)
 		regs->r9 = 0;
 	regs->cr_iip = fix & ~0xf;
-	ia64_psr(regs)->ri = fix & 0x3;		/* set continuation slot number */
+	ia64_psr(regs)->ri = (fix & 0xc) >> 2;		/* set continuation slot number */
 }
diff --exclude=CVS -Narup a/include/asm-ia64/uaccess.h b/include/asm-ia64/uaccess.h
--- a/include/asm-ia64/uaccess.h	2005-10-28 09:02:08.000000000 +0900
+++ b/include/asm-ia64/uaccess.h	2006-06-14 23:54:55.000000000 +0900
@@ -139,7 +139,7 @@ do {												\
 	register long __gu_r8 asm ("r8") = 0;							\
 	register long __gu_r9 asm ("r9");							\
 	asm ("\n[1:]\tld"#n" %0=%2%P2\t// %0 and %1 get overwritten by exception handler\n"	\
-	     "\t.xdata4 \"__ex_table\", 1b-., 1f-.+4\n"						\
+	     "\t.xdata4 \"__ex_table\", 1b-., 1f-.+2\n"						\
 	     "[1:]"										\
 	     : "=r"(__gu_r9), "=r"(__gu_r8) : "m"(__m(addr)), "1"(__gu_r8));			\
 	(err) = __gu_r8;									\
@@ -346,7 +346,7 @@ extern unsigned long __strnlen_user (con

 struct exception_table_entry {
 	int addr;	/* location-relative address of insn this fixup is for */
-	int cont;	/* location-relative continuation addr.; if bit 2 is set, r9 is set to 0 */
+	int cont;	/* location-relative continuation addr.; if bit 1 is set, r9 is set to 0 */
 };

 extern void ia64_handle_exception (struct pt_regs *regs, const struct exception_table_entry *e);
#1 lore raw

What about EXCLR() in asmmacro.h? Doesn't that need to be updated as well? Did you verify that the equivalent isn't open-coded in any assembly-file?

--david

Quoted 87 lines · Masami Hiramatsu

Hi, Tony

I found a suspicious buggy code in the linux kernel on IA64 arch. As far as I can see, search_extable() doesn't work correctly, because the lookup routine expects that the address format of the exception_table_entry is "IP + slot", but the compiler (gcc-3.4.5) generates it as "IP + (slot << 2)". Thus the lookup routine always fails to find the corresponding entry.
You can check it by dumping __ex_table section of vmlinux.

I made a patch to fix this bug attached in this mail. This patch is against 2.6.17-rc6-mm2.
Please review it.

Description:
Fix search_extable() and ia64_handle_exception() to handle the
address format of exception_table_entry correctly.

Thanks,

--
Masami HIRAMATSU
2nd Research Dept.
Hitachi, Ltd., Systems Development Laboratory
E-mail: [email protected]

 arch/ia64/mm/extable.c     |    7 ++++---
 include/asm-ia64/uaccess.h |    4 ++--
 2 files changed, 6 insertions(+), 5 deletions(-)

diff --exclude=CVS -Narup a/arch/ia64/mm/extable.c b/arch/ia64/mm/extable.c

arch/ia64/mm/extable.c 2006-06-14 14:59:11.000000000 +0900 renamed from arch/ia64/mm/extable.c 2005-10-28 09:02:08.000000000 +0900 +4 -3
@@ -63,9 +63,10 @@ search_extable (const struct exception_t
        unsigned long mid_ip;
        long diff;
 
+       ip = (ip & ~0xf) + ((ip & 0x3) << 2);
         while (first <= last) {
                mid = &first[(last - first)/2];
-               mid_ip = (u64) &mid->addr + mid->addr;
+               mid_ip = ((u64) &mid->addr + mid->addr) & ~0x3;
                diff = mid_ip - ip;
                 if (diff = 0)
                         return mid;
@@ -83,8 +84,8 @@ ia64_handle_exception (struct pt_regs *r
        long fix = (u64) &e->cont + e->cont;
 
        regs->r8 = -EFAULT;
-       if (fix & 4)
+       if (fix & 2)
                regs->r9 = 0;
        regs->cr_iip = fix & ~0xf;
-       ia64_psr(regs)->ri = fix & 0x3;         /* set continuation slot number */
+       ia64_psr(regs)->ri = (fix & 0xc) >> 2;          /* set continuation slot number */
 }

diff --exclude=CVS -Narup a/include/asm-ia64/uaccess.h b/include/asm-ia64/uaccess.h

include/asm-ia64/uaccess.h 2006-06-14 23:54:55.000000000 +0900 renamed from include/asm-ia64/uaccess.h 2005-10-28 09:02:08.000000000 +0900 +2 -3
@@ -139,7 +139,7 @@ do {                                                                                                \
        register long __gu_r8 asm ("r8") = 0;                                                   \
        register long __gu_r9 asm ("r9");                                                       \
        asm ("\n[1:]\tld"#n" %0=%2%P2\t// %0 and %1 get overwritten by exception handler\n"     \
-            "\t.xdata4 \"__ex_table\", 1b-., 1f-.+4\n"                                         \
+            "\t.xdata4 \"__ex_table\", 1b-., 1f-.+2\n"                                         \
             "[1:]"                                                                             \
             : "=r"(__gu_r9), "=r"(__gu_r8) : "m"(__m(addr)), "1"(__gu_r8));                    \
        (err) = __gu_r8;                                                                        \
@@ -346,7 +346,7 @@ extern unsigned long __strnlen_user (con
 
 struct exception_table_entry {
        int addr;       /* location-relative address of insn this fixup is for */
-       int cont;       /* location-relative continuation addr.; if bit 2 is set, r9 is set to 0 */
+       int cont;       /* location-relative continuation addr.; if bit 1 is set, r9 is set to 0 */
 };
 
 extern void ia64_handle_exception (struct pt_regs *regs, const struct exception_table_entry *e);
 
 
 
 
-

To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 1 line
Mosberger Consulting LLC, http://www.mosberger-consulting.com/
#2 lore raw
In reply to Masami Hiramatsu

Masami Hiramatsu wrote on Thursday, June 15, 2006 6:27 AM

Quoted 7 lines

I found a suspicious buggy code in the linux kernel on IA64 arch. As far as I can see, search_extable() doesn't work correctly, because the lookup routine expects that the address format of the exception_table_entry is "IP + slot", but the compiler (gcc-3.4.5) generates it as "IP + (slot << 2)". Thus the lookup routine always fails to find the corresponding entry.
You can check it by dumping __ex_table section of vmlinux.

Are you sure about the exception table being wrong? A quick look on my system indicates that the compiler only generate bundle address, there is no inter-bundle address in the exception table.

I can explain the data in continuation address. Since there is no exception code that starts at inter-bundle address, it should always be bundle aligned, i.e., slot will always be zero.

But the tag address in the table is also bundle aligned, which I will look a bit more.

# gcc -v
Reading specs from /usr/lib/gcc/ia64-redhat-linux/3.4.5/specs
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix
--disable-checking --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-java-awt=gtk
--host=ia64-redhat-linux
Thread model: posix
gcc version 3.4.5 20051201 (Red Hat 3.4.5-2)
#3 lore raw
In reply to Masami Hiramatsu

Masami Hiramatsu wrote on Thursday, June 15, 2006 6:27 AM

Quoted 7 lines

I found a suspicious buggy code in the linux kernel on IA64 arch. As far as I can see, search_extable() doesn't work correctly, because the lookup routine expects that the address format of the exception_table_entry is "IP + slot", but the compiler (gcc-3.4.5) generates it as "IP + (slot << 2)". Thus the lookup routine always fails to find the corresponding entry.
You can check it by dumping __ex_table section of vmlinux.

Chen, Kenneth W wrote on Thursday, June 15, 2006 2:22 PM

Quoted 2 lines

But the tag address in the table is also bundle aligned, which I will look a bit more.

I've double checked (triple checked with sample test code below). It's a false alarm. Everyone should rest assure that exception table and compiler do match and generates correct code. There is no bug AFAICT.

Test code:

#include <linux/config.h>
#include <asm/asmmacro.h>

GLOBAL_ENTRY(ken)

        EX(eh0, ld1 r32 = [r0])
        EX(eh1, ld1 r33 = [r0])
        nop 0
        ;;

[eh0:] br.ret.sptk b0
[eh1:] br.ret.sptk b0

END(ken)

Exception table entry for eh0 and eh1:

0xa0000001004bc828 ffff7518 ffff7524 ffff7511 ffff751d

objdump of test code:

a0000001004b3d40:    ld1 r32=[r0]
a0000001004b3d46:    ld1 r33=[r0]
a0000001004b3d4c:    nop.i 0x0;;
a0000001004b3d50:    br.ret.sptk.few b0
a0000001004b3d56:    br.ret.sptk.few b0
a0000001004b3d5c:    nop.b 0x0;;
#4 lore raw
In reply to Masami Hiramatsu

Hi,

Quoted 9 lines · Kenneth W

Masami Hiramatsu wrote on Thursday, June 15, 2006 6:27 AM

Quoted 7 lines

I found a suspicious buggy code in the linux kernel on IA64 arch. As far as I can see, search_extable() doesn't work correctly, because the lookup routine expects that the address format of the exception_table_entry is "IP + slot", but the compiler (gcc-3.4.5) generates it as "IP + (slot << 2)". Thus the lookup routine always fails to find the corresponding entry.
You can check it by dumping __ex_table section of vmlinux.

Quoted 7 lines

Chen, Kenneth W wrote on Thursday, June 15, 2006 2:22 PM

Quoted 2 lines

But the tag address in the table is also bundle aligned, which I will look a bit more.

I've double checked (triple checked with sample test code below). It's a false alarm. Everyone should rest assure that exception table and compiler do match and generates correct code. There is no bug AFAICT.

Thank you for your comments.
I had just checked the dump of __ex_table section of vmlinux. This time, I checked the __ex_table of running kernel. It is true that those format is "IP + slot". So you are right. I admit my mistake.
Please dispose the previous patch.

Thanks again.

Signature · 4 lines
Masami HIRAMATSU
2nd Research Dept.
Hitachi, Ltd., Systems Development Laboratory
E-mail: [email protected]