Re: [PATCH v2] x86/sgx: Fix free_cnt counting logic in epc section

[Sean Christopherson]

On Wed, Jan 20, 2021, Tianjia Zhang wrote:
> Increase `section->free_cnt` in sgx_sanitize_section() is more
> reasonable, which is called in ksgxd kernel thread, instead of
> assigning it to epc section pages number at initialization.
> Although this is unlikely to fail, these pages cannot be
> allocated after initialization, and which need to be reset
> by ksgxd.
> 
> At the same time, taking section->lock could be moved inside
> the !ret flow so that EREMOVE is done without holding the lock.
> it's theoretically possible that ksgxd hasn't finished
> sanitizing the EPC when userspace starts creating enclaves.

Moving the lock should be in a separate patch, they are clearly two different
functional changes.

> Reported-by: Jia Zhang <zhang....@linux.alibaba.com>
> Suggested-by: Sean Christopherson <sea...@google.com>

Moving lock was suggested by me, the original patch was not.

> Reviewed-by: Sean Christopherson <sea...@google.com>
> Signed-off-by: Tianjia Zhang <tianjia.zh...@linux.alibaba.com>
> ---
>  arch/x86/kernel/cpu/sgx/main.c | 13 +++++++------
>  1 file changed, 7 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/x86/kernel/cpu/sgx/main.c b/arch/x86/kernel/cpu/sgx/main.c
> index c519fc5f6948..34a72a147983 100644
> --- a/arch/x86/kernel/cpu/sgx/main.c
> +++ b/arch/x86/kernel/cpu/sgx/main.c
> @@ -41,16 +41,18 @@ static void sgx_sanitize_section(struct sgx_epc_section 
> *section)
>               if (kthread_should_stop())
>                       return;
>  
> -             /* needed for access to ->page_list: */
> -             spin_lock(&section->lock);
> -
>               page = list_first_entry(&section->init_laundry_list,
>                                       struct sgx_epc_page, list);
>  
>               ret = __eremove(sgx_get_epc_virt_addr(page));
> -             if (!ret)
> +
> +             /* needed for access to ->page_list: */
> +             spin_lock(&section->lock);

This can actually be even more precise, as the lock doesn't need to be taken
if __eremove() fails.  The lock protects section->page_list, not page->list.
At that point, the comment about why the lock is needed can probably be dropped?

> +
> +             if (!ret) {
>                       list_move(&page->list, &section->page_list);
> -             else
> +                     section->free_cnt += 1;

Belated feedback, this can use "++".

> +             } else

Need curly braces here.

E.g. when all is said and done, this code can be:

                if (!ret) {
                        spin_lock(&section->lock);
                        list_move(&page->list, &section->page_list);
                        section->free_cnt++;
                        spin_unlock(&section->lock);
                } else {
                        list_move_tail(&page->list, &dirty);
                }

>                       list_move_tail(&page->list, &dirty);
>  
>               spin_unlock(&section->lock);
> @@ -646,7 +648,6 @@ static bool __init sgx_setup_epc_section(u64 phys_addr, 
> u64 size,
>               list_add_tail(&section->pages[i].list, 
> &section->init_laundry_list);
>       }
>  
> -     section->free_cnt = nr_pages;
>       return true;
>  }
>  
> -- 
> 2.19.1.3.ge56e4f7
>