Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 20 Jan 2021 08:27:02 -0800
From:   Sean Christopherson <seanjc@google.com>
To:     Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Cc:     Jarkko Sakkinen <jarkko@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H. Peter Anvin" <hpa@zytor.com>, x86@kernel.org,
        linux-sgx@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jia Zhang <zhang.jia@linux.alibaba.com>
Subject: Re: [PATCH v2] x86/sgx: Fix free_cnt counting logic in epc section
Message-ID: <YAhZ1vydPiLtDOcj@google.com>
References: <20210120035320.19709-1-tianjia.zhang@linux.alibaba.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210120035320.19709-1-tianjia.zhang@linux.alibaba.com>
Precedence: bulk

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.jia@linux.alibaba.com>
> Suggested-by: Sean Christopherson <seanjc@google.com>

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

> Reviewed-by: Sean Christopherson <seanjc@google.com>
> Signed-off-by: Tianjia Zhang <tianjia.zhang@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
>