]> git.hungrycats.org Git - linux/commitdiff
arm64: flush FP/SIMD state correctly after execve()
authorArd Biesheuvel <ard.biesheuvel@linaro.org>
Thu, 27 Aug 2015 06:12:33 +0000 (07:12 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 2 Sep 2017 05:05:46 +0000 (07:05 +0200)
commit 674c242c9323d3c293fc4f9a3a3a619fe3063290 upstream.

When a task calls execve(), its FP/SIMD state is flushed so that
none of the original program state is observeable by the incoming
program.

However, since this flushing consists of setting the in-memory copy
of the FP/SIMD state to all zeroes, the CPU field is set to CPU 0 as
well, which indicates to the lazy FP/SIMD preserve/restore code that
the FP/SIMD state does not need to be reread from memory if the task
is scheduled again on CPU 0 without any other tasks having entered
userland (or used the FP/SIMD in kernel mode) on the same CPU in the
mean time. If this happens, the FP/SIMD state of the old program will
still be present in the registers when the new program starts.

So set the CPU field to the invalid value of NR_CPUS when performing
the flush, by calling fpsimd_flush_task_state().

Reported-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>
Reported-by: Janet Liu <janet.liu@spreadtrum.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/arm64/kernel/fpsimd.c

index 3dca15634e69c6b1b34db6e00b3564089129326b..c31e59fe2cb881dd5623e802aeeda3228e4ff528 100644 (file)
@@ -157,6 +157,7 @@ void fpsimd_thread_switch(struct task_struct *next)
 void fpsimd_flush_thread(void)
 {
        memset(&current->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
+       fpsimd_flush_task_state(current);
        set_thread_flag(TIF_FOREIGN_FPSTATE);
 }