Re: [PATCH 1/5] x86/sgx: Document permission handling better

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Mon, Aug 19, 2019 at 06:25:40PM +0300, Jarkko Sakkinen wrote:
> The way permissions are managed inside the driver is not trivial and
> intuitive. The non-obvious parts were not properly remarked in the
> source code. This patch refines them a bit.
> 
> Cc: Sean Christopherson <sean.j.christpherson@intel.com>
> Cc: Shay Katz-zamir <shay.katz-zamir@intel.com>
> Cc: Serge Ayoun <serge.ayoun@intel.com>
> Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> ---
>  arch/x86/kernel/cpu/sgx/driver/ioctl.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kernel/cpu/sgx/driver/ioctl.c b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> index 9b784a061a47..64d3286f3324 100644
> --- a/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> +++ b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> @@ -259,7 +259,10 @@ static struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
>  		return ERR_PTR(-ENOMEM);
>  	encl_page->desc = addr;
>  	encl_page->encl = encl;
> +
> +	/* Calculate maximum of the VM flags for the page. */
>  	encl_page->vm_prot_bits = calc_vm_prot_bits(prot, 0);
> +
>  	ret = radix_tree_insert(&encl->page_tree, PFN_DOWN(encl_page->desc),
>  				encl_page);
>  	if (ret) {
> @@ -640,11 +643,15 @@ static long sgx_ioc_enclave_add_page(struct file *filep, void __user *arg)
>  
>  	data = kmap(data_page);
>  
> +	/* Set prot bits matching to the SECINFO permissions. */
>  	prot = _calc_vm_trans(secinfo.flags, SGX_SECINFO_R, PROT_READ)  |
>  	       _calc_vm_trans(secinfo.flags, SGX_SECINFO_W, PROT_WRITE) |
>  	       _calc_vm_trans(secinfo.flags, SGX_SECINFO_X, PROT_EXEC);
>  
> -	/* TCS pages need to be RW in the PTEs, but can be 0 in the EPCM. */
> +	/* TCS pages must always RW set for CPU access while the SECINFO

Should be

	/*
	 * TCS pages ...

> +	 * permissions are *always* zero - the CPU ignores the user provided
> +	 * values and silently overwrites zero permissions.

Maybe 'overwrites with zero permissions'?

> +	 */
>  	if ((secinfo.flags & SGX_SECINFO_PAGE_TYPE_MASK) == SGX_SECINFO_TCS)
>  		prot |= PROT_READ | PROT_WRITE;
>  
> -- 
> 2.20.1
>