diff mbox

powerpc: optprobes: fix TOC handling in optprobes trampoline

Message ID 20170221150047.7460-1-naveen.n.rao@linux.vnet.ibm.com (mailing list archive)
State Accepted
Headers show

Commit Message

Naveen N. Rao Feb. 21, 2017, 3 p.m. UTC 
Optprobes on powerpc is limited to kernel text area. We decided to also
optimize kretprobe_trampoline since that is also in kernel text area.
However,we failed to take into consideration the fact that the same
trampoline is also used to catch function returns from kernel modules.
As an example:

    $ sudo modprobe kobject-example
    $ sudo bash -c "echo 'r foo_show+8' > /sys/kernel/debug/tracing/kprobe_events"
    $ sudo bash -c "echo 1 > /sys/kernel/debug/tracing/events/kprobes/enable"
    $ sudo cat /sys/kernel/debug/kprobes/list
    c000000000041350  k  kretprobe_trampoline+0x0    [OPTIMIZED]
    d000000000e00200  r  foo_show+0x8  kobject_example
    $ cat /sys/kernel/kobject_example/foo
    Segmentation fault

With the below (trimmed) splat in dmesg:

    [70646.248029] Unable to handle kernel paging request for data at address 0xfec40000
    [70646.248730] Faulting instruction address: 0xc000000000041540
    [70646.249210] Oops: Kernel access of bad area, sig: 11 [#1]
    [snip]
    [70646.259635] NIP [c000000000041540] optimized_callback+0x70/0xe0
    [70646.259962] LR [c000000000041e60] optinsn_slot+0xf8/0x10000
    [70646.260268] Call Trace:
    [70646.260583] [c0000000c7327850] [c000000000289af4] alloc_set_pte+0x1c4/0x860 (unreliable)
    [70646.260910] [c0000000c7327890] [c000000000041e60] optinsn_slot+0xf8/0x10000
    [70646.261223] --- interrupt: 700 at 0xc0000000c7327a80
		       LR = kretprobe_trampoline+0x0/0x10
    [70646.261849] [c0000000c7327ba0] [c0000000003a30d4] sysfs_kf_seq_show+0x104/0x1d0
    [70646.262135] [c0000000c7327bf0] [c0000000003a0bb4] kernfs_seq_show+0x44/0x60
    [70646.264211] [c0000000c7327c10] [c000000000330578] seq_read+0xf8/0x560
    [70646.265142] [c0000000c7327cb0] [c0000000003a1e64] kernfs_fop_read+0x194/0x260
    [70646.266070] [c0000000c7327d00] [c0000000002f9954] __vfs_read+0x44/0x1a0
    [70646.266977] [c0000000c7327d90] [c0000000002fb4cc] vfs_read+0xbc/0x1b0
    [70646.267860] [c0000000c7327de0] [c0000000002fd138] SyS_read+0x68/0x110
    [70646.268701] [c0000000c7327e30] [c00000000000b8e0] system_call+0x38/0xfc
    [snip]

Fix this by loading up the kernel TOC before calling into the kernel.
The original TOC gets restored as part of the usual pt_regs restore.

Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
---
 arch/powerpc/kernel/optprobes_head.S | 7 +++++++
 1 file changed, 7 insertions(+)

Comments

Naveen N. Rao Feb. 22, 2017, 2 p.m. UTC | #1 
On 2017/02/21 08:30PM, Naveen N Rao wrote:
> Optprobes on powerpc is limited to kernel text area. We decided to also
> optimize kretprobe_trampoline since that is also in kernel text area.
> However,we failed to take into consideration the fact that the same
> trampoline is also used to catch function returns from kernel modules.
> As an example:
> 
>     $ sudo modprobe kobject-example
>     $ sudo bash -c "echo 'r foo_show+8' > /sys/kernel/debug/tracing/kprobe_events"
>     $ sudo bash -c "echo 1 > /sys/kernel/debug/tracing/events/kprobes/enable"
>     $ sudo cat /sys/kernel/debug/kprobes/list
>     c000000000041350  k  kretprobe_trampoline+0x0    [OPTIMIZED]
>     d000000000e00200  r  foo_show+0x8  kobject_example
>     $ cat /sys/kernel/kobject_example/foo
>     Segmentation fault
> 
> With the below (trimmed) splat in dmesg:
> 
>     [70646.248029] Unable to handle kernel paging request for data at address 0xfec40000
>     [70646.248730] Faulting instruction address: 0xc000000000041540
>     [70646.249210] Oops: Kernel access of bad area, sig: 11 [#1]
>     [snip]
>     [70646.259635] NIP [c000000000041540] optimized_callback+0x70/0xe0
>     [70646.259962] LR [c000000000041e60] optinsn_slot+0xf8/0x10000
>     [70646.260268] Call Trace:
>     [70646.260583] [c0000000c7327850] [c000000000289af4] alloc_set_pte+0x1c4/0x860 (unreliable)
>     [70646.260910] [c0000000c7327890] [c000000000041e60] optinsn_slot+0xf8/0x10000
>     [70646.261223] --- interrupt: 700 at 0xc0000000c7327a80
> 		       LR = kretprobe_trampoline+0x0/0x10
>     [70646.261849] [c0000000c7327ba0] [c0000000003a30d4] sysfs_kf_seq_show+0x104/0x1d0
>     [70646.262135] [c0000000c7327bf0] [c0000000003a0bb4] kernfs_seq_show+0x44/0x60
>     [70646.264211] [c0000000c7327c10] [c000000000330578] seq_read+0xf8/0x560
>     [70646.265142] [c0000000c7327cb0] [c0000000003a1e64] kernfs_fop_read+0x194/0x260
>     [70646.266070] [c0000000c7327d00] [c0000000002f9954] __vfs_read+0x44/0x1a0
>     [70646.266977] [c0000000c7327d90] [c0000000002fb4cc] vfs_read+0xbc/0x1b0
>     [70646.267860] [c0000000c7327de0] [c0000000002fd138] SyS_read+0x68/0x110
>     [70646.268701] [c0000000c7327e30] [c00000000000b8e0] system_call+0x38/0xfc
>     [snip]
> 
> Fix this by loading up the kernel TOC before calling into the kernel.
> The original TOC gets restored as part of the usual pt_regs restore.
> 
> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>

Forgot to add:
Fixes: 762df10bad69 ("powerpc/kprobes: Optimize kprobe in 
kretprobe_trampoline()")

- Naveen

> ---
>  arch/powerpc/kernel/optprobes_head.S | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/arch/powerpc/kernel/optprobes_head.S b/arch/powerpc/kernel/optprobes_head.S
> index 53e429b5a29d..bf28188f308c 100644
> --- a/arch/powerpc/kernel/optprobes_head.S
> +++ b/arch/powerpc/kernel/optprobes_head.S
> @@ -65,6 +65,13 @@ optprobe_template_entry:
>  	mfdsisr	r5
>  	std	r5,_DSISR(r1)
> 
> +	/*
> +	 * We may get here from a module, so load the kernel TOC in r2.
> +	 * The original TOC gets restored when pt_regs is restored
> +	 *  further below.
> +	 */
> +	ld	r2,PACATOC(r13)
> +
>  	.global optprobe_template_op_address
>  optprobe_template_op_address:
>  	/*
> -- 
> 2.11.0
>
Michael Ellerman Feb. 27, 2017, 10:11 a.m. UTC | #2 
On Tue, 2017-02-21 at 15:00:47 UTC, "Naveen N. Rao" wrote:
> Optprobes on powerpc is limited to kernel text area. We decided to also
> optimize kretprobe_trampoline since that is also in kernel text area.
> However,we failed to take into consideration the fact that the same
> trampoline is also used to catch function returns from kernel modules.
> As an example:
> 
>     $ sudo modprobe kobject-example
>     $ sudo bash -c "echo 'r foo_show+8' > /sys/kernel/debug/tracing/kprobe_events"
>     $ sudo bash -c "echo 1 > /sys/kernel/debug/tracing/events/kprobes/enable"
>     $ sudo cat /sys/kernel/debug/kprobes/list
>     c000000000041350  k  kretprobe_trampoline+0x0    [OPTIMIZED]
>     d000000000e00200  r  foo_show+0x8  kobject_example
>     $ cat /sys/kernel/kobject_example/foo
>     Segmentation fault
> 
> With the below (trimmed) splat in dmesg:
> 
>     [70646.248029] Unable to handle kernel paging request for data at address 0xfec40000
>     [70646.248730] Faulting instruction address: 0xc000000000041540
>     [70646.249210] Oops: Kernel access of bad area, sig: 11 [#1]
>     [snip]
>     [70646.259635] NIP [c000000000041540] optimized_callback+0x70/0xe0
>     [70646.259962] LR [c000000000041e60] optinsn_slot+0xf8/0x10000
>     [70646.260268] Call Trace:
>     [70646.260583] [c0000000c7327850] [c000000000289af4] alloc_set_pte+0x1c4/0x860 (unreliable)
>     [70646.260910] [c0000000c7327890] [c000000000041e60] optinsn_slot+0xf8/0x10000
>     [70646.261223] --- interrupt: 700 at 0xc0000000c7327a80
> 		       LR = kretprobe_trampoline+0x0/0x10
>     [70646.261849] [c0000000c7327ba0] [c0000000003a30d4] sysfs_kf_seq_show+0x104/0x1d0
>     [70646.262135] [c0000000c7327bf0] [c0000000003a0bb4] kernfs_seq_show+0x44/0x60
>     [70646.264211] [c0000000c7327c10] [c000000000330578] seq_read+0xf8/0x560
>     [70646.265142] [c0000000c7327cb0] [c0000000003a1e64] kernfs_fop_read+0x194/0x260
>     [70646.266070] [c0000000c7327d00] [c0000000002f9954] __vfs_read+0x44/0x1a0
>     [70646.266977] [c0000000c7327d90] [c0000000002fb4cc] vfs_read+0xbc/0x1b0
>     [70646.267860] [c0000000c7327de0] [c0000000002fd138] SyS_read+0x68/0x110
>     [70646.268701] [c0000000c7327e30] [c00000000000b8e0] system_call+0x38/0xfc
>     [snip]
> 
> Fix this by loading up the kernel TOC before calling into the kernel.
> The original TOC gets restored as part of the usual pt_regs restore.
> 
> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/f558b37bf4c35a54e1949f6533f39c

cheers
diff mbox

Patch

diff --git a/arch/powerpc/kernel/optprobes_head.S b/arch/powerpc/kernel/optprobes_head.S
index 53e429b5a29d..bf28188f308c 100644
--- a/arch/powerpc/kernel/optprobes_head.S
+++ b/arch/powerpc/kernel/optprobes_head.S
@@ -65,6 +65,13 @@  optprobe_template_entry:
 	mfdsisr	r5
 	std	r5,_DSISR(r1)
 
+	/*
+	 * We may get here from a module, so load the kernel TOC in r2.
+	 * The original TOC gets restored when pt_regs is restored
+	 *  further below.
+	 */
+	ld	r2,PACATOC(r13)
+
 	.global optprobe_template_op_address
 optprobe_template_op_address:
 	/*