Re: [PATCH] linux-user: fix bug about incorrect base addresss of gdt on i386 and x86_64

[Richard Henderson]

On 1/3/23 01:38, fanwenjie wrote:
> On linux user mode, CPUX86State::gdt::base from Different CPUX86State Objects 
> have same value, It is incorrect! Every CPUX86State::gdt::base Must points to 
> independent memory space.
>
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1405
> Signed-off-by: fanwenjie <fanwj@mail.ustc.edu.cn>
> ---
>   linux-user/i386/cpu_loop.c | 9 +++++++++
>   linux-user/main.c          | 7 +++++++
>   2 files changed, 16 insertions(+)
>
> diff --git a/linux-user/i386/cpu_loop.c b/linux-user/i386/cpu_loop.c
> index 865413c..48511cd 100644
> --- a/linux-user/i386/cpu_loop.c
> +++ b/linux-user/i386/cpu_loop.c
> @@ -314,8 +314,17 @@ void cpu_loop(CPUX86State *env)
>       }
>   }
>   
> +static void target_cpu_free(void *obj)
> +{
> +    CPUArchState* env = ((CPUState*)obj)->env_ptr;
> +    target_munmap(env->gdt.base, sizeof(uint64_t) * TARGET_GDT_ENTRIES);
> +    g_free(obj);
> +}
> +
>   void target_cpu_copy_regs(CPUArchState *env, struct target_pt_regs *regs)
>   {
> +    CPUState* cpu = env_cpu(env);
> +    OBJECT(cpu)->free = target_cpu_free;
>       env->cr[0] = CR0_PG_MASK | CR0_WP_MASK | CR0_PE_MASK;
>       env->hflags |= HF_PE_MASK | HF_CPL_MASK;
>       if (env->features[FEAT_1_EDX] & CPUID_SSE) {
> diff --git a/linux-user/main.c b/linux-user/main.c
> index a17fed0..3acd9b4 100644
> --- a/linux-user/main.c
> +++ b/linux-user/main.c
> @@ -234,6 +234,13 @@ CPUArchState *cpu_copy(CPUArchState *env)
>   
>       new_cpu->tcg_cflags = cpu->tcg_cflags;
>       memcpy(new_env, env, sizeof(CPUArchState));
> +#if defined(TARGET_I386) || defined(TARGET_X86_64)
> +    new_env->gdt.base = target_mmap(0, sizeof(uint64_t) * TARGET_GDT_ENTRIES,
> +                                    PROT_READ|PROT_WRITE,
> +                                    MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
> +    memcpy((void*)new_env->gdt.base, (void*)env->gdt.base, sizeof(uint64_t) * 
> TARGET_GDT_ENTRIES);
> +    OBJECT(new_cpu)->free = OBJECT(cpu)->free;
> +#endif

This isn't a fantastic solution, because neither the ldt nor the gdt should be mapped into 
the user address space -- these are kernel private data structures. But cpu.h uses a 
target_ulong, and seg_helper.c is set up to load data from the guest, and it would be a 
medium sized job to address that.

The memcpy is definitely wrong, because you're casting a guest address into a host 
address, which is incorrect.  You have to use g2h().

I'm actually surprised that you need this for TARGET_X86_64 at all -- the two TLS segments 
don't really use the GDT at all, since fs_base and gs_base may be set directly.


r~