Re: [PATCH 3/9] x86/fpu: Make task_struct::thread constant size

From: Oleg Nesterov
Date: Wed Jun 12 2024 - 05:42:55 EST


On 06/12, Ingo Molnar wrote:
>
> * Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
>
> > > + per_cpu(fpu_fpregs_owner_ctx, this_cpu) = &x86_init_fpu;
> > > + x86_init_fpu.last_cpu = this_cpu;
> >
> > Why? I think it should do
> >
> > x86_init_fpu.last_cpu = -1;
> > set_thread_flag(TIF_NEED_FPU_LOAD);
> >
> > And the next patch should kill x86_init_fpu altogether, but keep
> > TIF_NEED_FPU_LOAD. It should be never cleared if PF_KTHREAD.
>
> So I applied the patch further below on top of:
>
> 4f4a9b399357 x86/fpu: Make task_struct::thread constant size
>
> And Nathan's 32-bit kernel testcase [but running with 1 CPU to simplify it]
> still crashes in a similar fashion

Yes, I didn't expect it can fix the problem. Still makes sense, I think.

> in the (first?) modprobe instance with a
> bad FPU state exception:

OK, I reproduced it too. I see nothing wrong in the usermodehelper or
kernel_execve paths... and fpu_clone() looks fine, "minimal" is still
true if init_task or another PF_KTHREAD calls user_mode_thread().

So I appiled the patch below and save_fpregs_to_fpstate() in
fpu__init_system() triggers the WARN_ON_FPU(err) in os_xsave()

[ 0.014609] RESTORED !!!!!!!!!!!!!!!!!!!!!!!!!!!!!
[ 0.014958] ------------[ cut here ]------------
[ 0.014958] WARNING: CPU: 0 PID: 0 at arch/x86/kernel/fpu/xstate.h:189 save_fpregs_to_fpstate+0x74/0x80
...

so I _think_ we can probably forget about modprobe/etc.

Oleg.


diff --git a/arch/x86/kernel/fpu/context.h b/arch/x86/kernel/fpu/context.h
index 10d0a720659c..9fa78f75b2e5 100644
--- a/arch/x86/kernel/fpu/context.h
+++ b/arch/x86/kernel/fpu/context.h
@@ -56,8 +56,8 @@ static inline void fpregs_restore_userregs(void)
struct fpu *fpu = x86_task_fpu(current);
int cpu = smp_processor_id();

- if (WARN_ON_ONCE(current->flags & (PF_KTHREAD | PF_USER_WORKER)))
- return;
+// if (WARN_ON_ONCE(current->flags & (PF_KTHREAD | PF_USER_WORKER)))
+// return;

if (!fpregs_state_valid(fpu, cpu)) {
/*
diff --git a/arch/x86/kernel/fpu/init.c b/arch/x86/kernel/fpu/init.c
index 4e8d37b5a90b..0e63d54595aa 100644
--- a/arch/x86/kernel/fpu/init.c
+++ b/arch/x86/kernel/fpu/init.c
@@ -5,11 +5,11 @@
#include <asm/fpu/api.h>
#include <asm/tlbflush.h>
#include <asm/setup.h>
-
+#include <asm/fpu/signal.h>
#include <linux/sched.h>
#include <linux/sched/task.h>
#include <linux/init.h>
-
+#include "context.h"
#include "internal.h"
#include "legacy.h"
#include "xstate.h"
@@ -75,12 +75,12 @@ static struct fpu x86_init_fpu __read_mostly;

static void __init fpu__init_system_early_generic(void)
{
- int this_cpu = smp_processor_id();
+// int this_cpu = smp_processor_id();

fpstate_reset(&x86_init_fpu);
current->thread.fpu = &x86_init_fpu;
- per_cpu(fpu_fpregs_owner_ctx, this_cpu) = &x86_init_fpu;
- x86_init_fpu.last_cpu = this_cpu;
+ set_thread_flag(TIF_NEED_FPU_LOAD);
+ x86_init_fpu.last_cpu = -1;

if (!boot_cpu_has(X86_FEATURE_CPUID) &&
!test_bit(X86_FEATURE_FPU, (unsigned long *)cpu_caps_cleared)) {
@@ -217,6 +217,7 @@ static void __init fpu__init_system_xstate_size_legacy(void)
* Called on the boot CPU once per system bootup, to set up the initial
* FPU state that is later cloned into all processes:
*/
+void save_fpregs_to_fpstate(struct fpu *fpu);
void __init fpu__init_system(void)
{
fpu__init_system_early_generic();
@@ -231,4 +232,10 @@ void __init fpu__init_system(void)
fpu__init_system_xstate_size_legacy();
fpu__init_system_xstate(fpu_kernel_cfg.max_size);
fpu__init_task_struct_size();
+
+ BUG_ON(x86_task_fpu(current) != &x86_init_fpu);
+ fpregs_restore_userregs();
+ pr_crit("RESTORED !!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
+ save_fpregs_to_fpstate(&x86_init_fpu);
+ pr_crit("SAVED !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n");
}