On Fri, Feb 12, 2016 at 06:00:10PM +0100, Borislav Petkov wrote:
> Something for me to try when I get a chance.
Ok, so I wanted to know what happens in detail. Here's some ftracing
(debug patch at the end).
Now pay attention to this udevadm thing
[ 3.816977] rcu_pree-7 0d..2 4058241us : __switch_to: prev:
rcu_preempt <-> next: udevadm
[ 3.816977] rcu_pree-7 0d..2 4058241us : __switch_to: set
->fpregs_active
[ 3.816977] udevadm-982 0.... 4058258us : __fpu__restore_sig:
fpregs_active 0, f443d7c0
We're in __fpu__restore_sig() about to call schedule()
[ 3.816977] udevadm-982 0d..2 4058260us : __switch_to: prev: udevadm
<-> next: usb_id
[ 3.816977] udevadm-982 0d..2 4058260us : __switch_to: set
->fpregs_active
[ 3.816977] usb_id-987 0d..2 4059684us : __switch_to: prev: usb_id <->
next: udevd
[ 3.816977] usb_id-987 0d..2 4059685us : __switch_to: set
->fpregs_active
[ 3.816977] udevd-843 0d..2 4059697us : __switch_to: prev: udevd <->
next: udevd
[ 3.816977] udevd-843 0d..2 4059697us : __switch_to: set
->fpregs_active
[ 3.816977] alsa-uti-989 0d..2 4060452us : __switch_to: prev: alsa-utils
<-> next: udevd
[ 3.816977] alsa-uti-989 0d..2 4060452us : __switch_to: set
->fpregs_active
[ 3.816977] udevd-840 0d..2 4060521us : __switch_to: prev: udevd <->
next: udevd
[ 3.816977] udevd-840 0d..2 4060522us : __switch_to: set
->fpregs_active
[ 3.816977] udevd-829 0d..2 4060557us : __switch_to: prev: udevd <->
next: udevd
[ 3.816977] udevd-829 0d..2 4060558us : __switch_to: set
->fpregs_active
[ 3.816977] udevd-840 0d..2 4060862us : __switch_to: prev: udevd <->
next: blkid
[ 3.816977] udevd-840 0d..2 4060862us : __switch_to: set
->fpregs_active
[ 3.816977] blkid-985 0d..2 4061148us : __switch_to: prev: blkid <->
next: udevadm
[ 3.816977] blkid-985 0d..2 4061148us : __switch_to: set
->fpregs_active
Now we're switching back to udevadm which is @next_p of __switch_to().
There we do:
fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
which does:
/* Don't change CR0.TS if we just switch! */
if (fpu.preload) {
new_fpu->counter++;
__fpregs_activate(new_fpu);
prefetch(&new_fpu->state);
__fpregs_activate() sets ->fpregs_active of @new_fpu, i.e. udevadm's one.
[ 3.816977] udevadm-982 0.... 4061149us : __fpu__restore_sig: after
schedule: fpregs_active: 1 f443d7c0
__fpu__restore_sig() -> fpu__restore() sets ->fpregs_active again.
[ 3.816977] udevadm-982 0.N.1 4185386us : fpu__restore: WARN: fpu:
f443d7c0
Boom!
[ 3.816977] udevadm-982 0.N.1 4185392us : <stack trace>
[ 3.816977] => fpu__restore
[ 3.816977] => __fpu__restore_sig
[ 3.816977] => fpu__restore_sig
[ 3.816977] => restore_sigcontext
[ 3.816977] => sys_sigreturn
[ 3.816977] => do_syscall_32_irqs_on
[ 3.816977] => restore_all
[ 3.816977] ---------------------------------
[ 3.816977] Kernel Offset: disabled
[ 3.816977] ---[ end Kernel panic - not syncing: Outta here...
So yeah, we probably should enlarge the preemption-off region to contain
->fpstate_active. Here's what you basically suggested but with a
*looot* of explanatory text. Which might be really wrong or completely
unparseable or both. So holler what should be changed.
Thanks!
---
From: Borislav Petkov <b...@suse.de>
Date: Mon, 15 Feb 2016 19:50:33 +0100
Subject: [RFC PATCH] x86/FPU: Fix double FPU regs activation
On the entry_INT80_32->do_syscall_32_irqs_on path on 32-bit we run with
interrupts enabled. And it can happen that we get preempted right after
setting ->fpstate_active in a task's FPU.
After we get preempted, we switch between tasks merrily and eventually
are about to switch to that task above whose ->fpstate_active we
set. We enter __switch_to() and do switch_fpu_prepare(). Our task gets
->fpregs_active set, we find ourselves back on the call stack below and
especially in __fpu__restore_sig() which sets ->fpregs_active again.
Leading to that whoops below.
So let's enlarge the preemption-off region so that we set ->fpstate_active with
preemption disabled and thus not trigger fpu.preload:
switch_fpu_prepare
...
fpu.preload = static_cpu_has(X86_FEATURE_FPU) &&
new_fpu->fpstate_active &&
^^^^^^^^^^^^^^^^^^^^^^
prematurely.
WARNING: CPU: 0 PID: 3031 at ./arch/x86/include/asm/fpu/internal.h:530
fpu__restore+0x90/0x130()
Modules linked in: hid_generic usbhid hid snd_hda_codec_hdmi
snd_hda_codec_realtek snd_hda_codec_generic iTCO_wdt iTCO_vendor_support
emp_thermal coretemp kvm_intel kvm irqbypass crc32_pclmul crc32c_intel
iwldvm mac80211 aesni_intel xts snd_hda_intel input_leds aes_i586
sdhci_pci lrw iwlwifi snd_hwdep gf128mul snd_hda_core ablk_helper cryptd
ehci_pci pcspkr serio_raw xhci_pci sdhci snd_pcm sg mmc_core 211 lpc_ich
mfd_core e1000e snd_timer ehci_hcd xhci_hcd thinkpad_acpi nvram wmi snd
battery soundcore led_class ac thermal
CPU: 0 PID: 3031 Comm: bash Not tainted 4.5.0-rc3+ #1
Hardware name: LENOVO 2320CTO/2320CTO, BIOS G2ET86WW (2.06 ) 11/13/2012
00000000 00000286 f158be4c c12cce56 00000000 00000000 f158be80 c10567fb
c1866c2c 00000000 00000bd7 c1859e8c 00000212 c1025ab0 00000212 c1025ab0
f2012b00 f2011f00 f2012d80 f158be90 c10568d2 00000009 00000000 f158bea4
Call Trace:
dump_stack
warn_slowpath_common
? fpu__restore
? fpu__restore
warn_slowpath_null
fpu__restore
__fpu__restore_sig
fpu__restore_sig
restore_sigcontext
sys_sigreturn
do_syscall_32_irqs_on
entry_INT80_32
Suggested-by: Andy Lutomirski <l...@amacapital.net>
Signed-off-by: Borislav Petkov <b...@suse.de>
---
arch/x86/kernel/fpu/signal.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/arch/x86/kernel/fpu/signal.c b/arch/x86/kernel/fpu/signal.c
index 31c6a60505e6..408e5a1c6fdd 100644
--- a/arch/x86/kernel/fpu/signal.c
+++ b/arch/x86/kernel/fpu/signal.c
@@ -316,12 +316,11 @@ static int __fpu__restore_sig(void __user *buf, void
__user *buf_fx, int size)
sanitize_restored_xstate(tsk, &env, xfeatures, fx_only);
}
+ preempt_disable();
fpu->fpstate_active = 1;
- if (use_eager_fpu()) {
- preempt_disable();
+ if (use_eager_fpu())
fpu__restore(fpu);
- preempt_enable();
- }
+ preempt_enable();
return err;
} else {
--
2.3.5
Tracing patch:
---
diff --git a/arch/x86/include/asm/fpu/internal.h
b/arch/x86/include/asm/fpu/internal.h
index a2124343edf5..2cbc3bf34928 100644
--- a/arch/x86/include/asm/fpu/internal.h
+++ b/arch/x86/include/asm/fpu/internal.h
@@ -527,9 +527,16 @@ static inline void __fpregs_deactivate(struct fpu *fpu)
/* Must be paired with a 'clts' (fpregs_activate_hw()) before! */
static inline void __fpregs_activate(struct fpu *fpu)
{
- WARN_ON_FPU(fpu->fpregs_active);
+ if (WARN_ON_FPU(fpu->fpregs_active)) {
+ trace_printk("WARN: fpu: %p\n", fpu);
+ trace_dump_stack(0);
+ tracing_off();
+ panic("Outta here...\n");
+ }
fpu->fpregs_active = 1;
+ trace_printk("set ->fpregs_active\n");
+
this_cpu_write(fpu_fpregs_owner_ctx, fpu);
}
diff --git a/arch/x86/kernel/fpu/signal.c b/arch/x86/kernel/fpu/signal.c
index 31c6a60505e6..bb40f02cdfdd 100644
--- a/arch/x86/kernel/fpu/signal.c
+++ b/arch/x86/kernel/fpu/signal.c
@@ -317,6 +317,14 @@ static int __fpu__restore_sig(void __user *buf, void
__user *buf_fx, int size)
}
fpu->fpstate_active = 1;
+
+ trace_printk("fpregs_active %d, %p\n", fpu->fpregs_active, fpu);
+
+ schedule();
+
+ trace_printk("after schedule: fpregs_active: %d %p\n",
+ fpu->fpregs_active, fpu);
+
if (use_eager_fpu()) {
preempt_disable();
fpu__restore(fpu);
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 9f950917528b..ce768c728f38 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -249,6 +249,8 @@ __switch_to(struct task_struct *prev_p, struct task_struct
*next_p)
struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
fpu_switch_t fpu_switch;
+ trace_printk("prev: %s <-> next: %s\n", prev_p->comm, next_p->comm);
+
/* never put a printk in __switch_to... printk() calls wake_up*()
indirectly */
fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
--
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
