[PATCH v2] mips: Call lose_fpu(0) before initializing fcr31 in mips_set_personality_nan

[PATCH v2] mips: Call lose_fpu(0) before initializing fcr31 in mips_set_personality_nan

[Xi Ruoyao]

If we still own the FPU after initializing fcr31, when we are preempted
the dirty value in the FPU will be read out and stored into fcr31,
clobbering our setting.  This can cause an improper floating-point
environment after execve().  For example:

    zsh% cat measure.c
    #include <fenv.h>
    int main() { return fetestexcept(FE_INEXACT); }
    zsh% cc measure.c -o measure -lm
    zsh% echo $((1.0/3)) # raising FE_INEXACT
    0.33333333333333331
    zsh% while ./measure; do ; done
    (stopped in seconds)

Call lose_fpu(0) before setting fcr31 to prevent this.

Closes: https://lore.kernel.org/linux-mips/7a6aa1bbdbbe2e63ae96ff163fab0349f58f1b9e.camel@xry111.site/
Fixes: 9b26616c8d9d ("MIPS: Respect the ISA level in FCSR handling")
Cc: stable@vger.kernel.org
Signed-off-by: Xi Ruoyao <xry111@xry111.site>
---

v1 -> v2: Fix stable list address in Cc line.

 arch/mips/kernel/elf.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/mips/kernel/elf.c b/arch/mips/kernel/elf.c
index 5582a4ca1e9e..7aa2c2360ff6 100644
--- a/arch/mips/kernel/elf.c
+++ b/arch/mips/kernel/elf.c
@@ -11,6 +11,7 @@
 
 #include <asm/cpu-features.h>
 #include <asm/cpu-info.h>
+#include <asm/fpu.h>
 
 #ifdef CONFIG_MIPS_FP_SUPPORT
 
@@ -309,6 +310,11 @@ void mips_set_personality_nan(struct arch_elf_state *state)
 	struct cpuinfo_mips *c = &boot_cpu_data;
 	struct task_struct *t = current;
 
+	/* Do this early so t->thread.fpu.fcr31 won't be clobbered in case
+	 * we are preempted before the lose_fpu(0) in start_thread.
+	 */
+	lose_fpu(0);
+
 	t->thread.fpu.fcr31 = c->fpu_csr31;
 	switch (state->nan_2008) {
 	case 0:
-- 
2.43.0

Re: [PATCH v2] mips: Call lose_fpu(0) before initializing fcr31 in mips_set_personality_nan

[Thomas Bogendoerfer]

On Sat, Jan 27, 2024 at 05:05:57AM +0800, Xi Ruoyao wrote:
> If we still own the FPU after initializing fcr31, when we are preempted
> the dirty value in the FPU will be read out and stored into fcr31,
> clobbering our setting.  This can cause an improper floating-point
> environment after execve().  For example:
> 
>     zsh% cat measure.c
>     #include <fenv.h>
>     int main() { return fetestexcept(FE_INEXACT); }
>     zsh% cc measure.c -o measure -lm
>     zsh% echo $((1.0/3)) # raising FE_INEXACT
>     0.33333333333333331
>     zsh% while ./measure; do ; done
>     (stopped in seconds)
> 
> Call lose_fpu(0) before setting fcr31 to prevent this.
> 
> Closes: https://lore.kernel.org/linux-mips/7a6aa1bbdbbe2e63ae96ff163fab0349f58f1b9e.camel@xry111.site/
> Fixes: 9b26616c8d9d ("MIPS: Respect the ISA level in FCSR handling")
> Cc: stable@vger.kernel.org
> Signed-off-by: Xi Ruoyao <xry111@xry111.site>
> ---
> 
> v1 -> v2: Fix stable list address in Cc line.
> 
>  arch/mips/kernel/elf.c | 6 ++++++
>  1 file changed, 6 insertions(+)

applied to mips-fixes.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]