CODEWITHSUNDEEP
linux-kernellinux-device-driverx86-64
feeling_lonely
feeling_lonely

Reputation: 6863

What fpu_xrstor_checking does? extra commentary?

I am doing my own version of do_fork() (for many purposes). So, basically, I am copying the process from one place to another, possibly to another machine. Copying the code and pasting it here would be tough. But this explanation should good enough, I believe.

My code works most of the times, but in other times the function fpu_xrstor_checking() returns an error (value = -1). Can anyone please explain what this function is supposed to do and extra commentary?

Here is the function pasted here for convenience:

 45 static inline int fpu_xrstor_checking(struct fpu *fpu)
 46 {
 47         struct xsave_struct *fx = &fpu->state->xsave;
 48         int err;
 49 
 50         asm volatile("1: .byte " REX_PREFIX "0x0f,0xae,0x2f\n\t"
 51                      "2:\n"
 52                      ".section .fixup,\"ax\"\n"
 53                      "3:  movl $-1,%[err]\n"
 54                      "    jmp  2b\n"
 55                      ".previous\n"
 56                      _ASM_EXTABLE(1b, 3b)
 57                      : [err] "=r" (err)
 58                      : "D" (fx), "m" (*fx), "a" (-1), "d" (-1), "" (0)
 59                      : "memory");
 60 
 61         return err;
 62 }

Thank you!

Upvotes: 0

Views: 65

Answers (1)

user781847
user781847

Reputation:

The inline assembly make use of a feature of Linux kernel that allow developer to "catch" CPU exceptions. The instruction at label 1 is a XRSTOR (more on this later).
The code at label 3 is emitted in the .fixup section that contain code used to handle exceptions.
The _ASM_EXTABLE tell the assembler to generate a table structure to inform the kernel that the instruction at 1 may generate an exception and that its handler is at 3.

The handler just set the err to -1.
The XRSTOR instruction (coded with opcodes maybe because the assembler does not support it yet?) restore the following part of the CPU architectural state: x87 (FPU), SSE, AVX.
The instruction take EDX:EAX as a mask (called the instruction mask) and it is quite elaborate, it can generate a #GP for a lot of reasons, listing them here would be pointless (one cause being its operand not aligned on 64 byte boundary).
When this instruction faults the function return -1.

I suggest the reading of the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 1 Section 13 for a full description of the topic (general understanding of Section 8-12 are required). You can also read the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2b for the instruction XRSTOR reference with a full list of reasons for the exceptions that can be generated.

Upvotes: 2

Related Questions