Re: Please review a important patch abort fix setting of CPUX86State::gdt::base

[Richard Henderson]

On 1/30/23 03:12, fanwj@mail.ustc.edu.cn wrote:
> 1. "The memcpy is definitely wrong, because you're casting a guest address into a 
> host address, which is incorrect. You have to use g2h()."
> There is no need to use g2h(), Because there are both guest address whether source or 
> dest memory. refer to "linux-user/i386/cpu_loop.c" target_cpu_copy_regs 
> function, Only use g2h_untagged when convert gdt::base to gdt_table. I don't use and 
> modify gdt_table, Only copy gdt::base from source CPU to dest CPU. They are same type so 
> no needed to convert by g2h.

This is *not* about the type, this is about the location in host memory, as input to 
memcpy.  The g2h function is 1-to-1, but it is not the identity function.


> 2. "I'm actually surprised that you need this for TARGET_X86_64 at all ..."
> GDT on QEMU User Mode is Pseudorandom GDT，It is NOT kernel private data 
> structures. It is NOT Global Descriptor Table. It IS index table of fs and gs. 
> And It is Thread local data. The Memory which gdt::base point can be modified 
> by syscall SYS_set_thread_data.

Well, then you'll need to fix other assumptions in target/i386/tcg/translate.c, beginning 
with

#if defined(CONFIG_USER_ONLY) && defined(TARGET_X86_64)
#define VM86(S)   false
#define CODE32(S) true
#define SS32(S)   true
#define ADDSEG(S) false

which currently means that whatever you do with set_thread_data won't be recognized at 
translation time.


r~