Re: [RFC PATCH v4 2/4] x86/sgx: Implement support for MADV_WILLNEED

From: "Huang, Kai" <kai.huang@intel.com>
To: "jarkko@kernel.org" <jarkko@kernel.org>
Cc: "linux-sgx@vger.kernel.org" <linux-sgx@vger.kernel.org>,
	"Chatre, Reinette" <reinette.chatre@intel.com>,
	"Dhanraj, Vijay" <vijay.dhanraj@intel.com>,
	"haitao.huang@linux.intel.com" <haitao.huang@linux.intel.com>,
	"dave.hansen@linux.intel.com" <dave.hansen@linux.intel.com>
Subject: Re: [RFC PATCH v4 2/4] x86/sgx: Implement support for MADV_WILLNEED
Date: Sun, 12 Mar 2023 22:25:45 +0000	[thread overview]
Message-ID: <d37990018617357537456c4554edf8c7f3fc71b0.camel@intel.com> (raw)
In-Reply-To: <ZA0p4Kf8sLcGBAJb@kernel.org>

On Sun, 2023-03-12 at 03:25 +0200, jarkko@kernel.org wrote:
> On Tue, Mar 07, 2023 at 11:32:15PM +0000, Huang, Kai wrote:
> > And in following mmap()s in which we want to map a small range of enclave:
> > 
> > 	encl_addr = mmap(encl_addr, MAP_SHARED|MAP_FIXED, encl_fd,
> > 			(encl_addr - encl_base) >> PAGE_SHIFT);
> > 
> > ?
> > 
> > Anything wrong above?
> 
> I'm not sure I fully comprehended your response because it was
> honestly a bit scattered so please correct me if I'm missing
> something but: why a process would want to map a small range
> of an enclave?

For SGX2, if mmap(MAP_ANONYMOUS) was used to get the enclave base address before
ECREATE, you will need to mmap(encl_fd) for any regions that are not pre-
populated (via EADD), no matter whether the first mmap(MAP_ANONYMOUS) has
covered all enclave range or not.  Otherwise, the fault to SGX2 regions won't be
handled by SGX driver.

You can find such mmap() in patch 4 in this series too.

Also, architecturally, for SGX2 the first mmap() (before ECREATE) doesn't have
to map the entire enclave range.  For SGX2 it's fine to pass a larger enclave
range in ioctl(ECREATE) than the range got from the first mmap().  Userspace can
later on choose to only mmap() the dynamic ranges that it wants to use.