On 2020-07-07 05:37, Jarkko Sakkinen wrote:
> From: Sean Christopherson <sean.j.christopherson@intel.com>
> 
> An SGX runtime must be aware of the exceptions, which happen inside an
> enclave. Introduce a vDSO call that wraps EENTER/ERESUME cycle and returns
> the CPU exception back to the caller exactly when it happens.
> 
> Kernel fixups the exception information to RDI, RSI and RDX. The SGX call
> vDSO handler fills this information to the user provided buffer or
> alternatively trigger user provided callback at the time of the exception.
> 
> The calling convention is custom and does not follow System V x86-64 ABI.
> 
> Suggested-by: Andy Lutomirski <luto@amacapital.net>
> Acked-by: Jethro Beekman <jethro@fortanix.com>
> Tested-by: Jethro Beekman <jethro@fortanix.com>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> Co-developed-by: Cedric Xing <cedric.xing@intel.com>
> Signed-off-by: Cedric Xing <cedric.xing@intel.com>
> Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> ---
>  arch/x86/entry/vdso/Makefile             |   2 +
>  arch/x86/entry/vdso/vdso.lds.S           |   1 +
>  arch/x86/entry/vdso/vsgx_enter_enclave.S | 131 +++++++++++++++++++++++
>  arch/x86/include/asm/enclu.h             |   8 ++
>  arch/x86/include/uapi/asm/sgx.h          |  98 +++++++++++++++++
>  5 files changed, 240 insertions(+)
>  create mode 100644 arch/x86/entry/vdso/vsgx_enter_enclave.S
>  create mode 100644 arch/x86/include/asm/enclu.h
> 
> diff --git a/arch/x86/entry/vdso/Makefile b/arch/x86/entry/vdso/Makefile
> index ebe82b7aecda..f71ad5ebd0c4 100644
> --- a/arch/x86/entry/vdso/Makefile
> +++ b/arch/x86/entry/vdso/Makefile
> @@ -29,6 +29,7 @@ VDSO32-$(CONFIG_IA32_EMULATION)	:= y
>  vobjs-y := vdso-note.o vclock_gettime.o vgetcpu.o
>  vobjs32-y := vdso32/note.o vdso32/system_call.o vdso32/sigreturn.o
>  vobjs32-y += vdso32/vclock_gettime.o
> +vobjs-$(VDSO64-y)		+= vsgx_enter_enclave.o
>  
>  # files to link into kernel
>  obj-y				+= vma.o extable.o
> @@ -100,6 +101,7 @@ $(vobjs): KBUILD_CFLAGS := $(filter-out $(GCC_PLUGINS_CFLAGS) $(RETPOLINE_CFLAGS
>  CFLAGS_REMOVE_vclock_gettime.o = -pg
>  CFLAGS_REMOVE_vdso32/vclock_gettime.o = -pg
>  CFLAGS_REMOVE_vgetcpu.o = -pg
> +CFLAGS_REMOVE_vsgx_enter_enclave.o = -pg
>  
>  #
>  # X32 processes use x32 vDSO to access 64bit kernel data.
> diff --git a/arch/x86/entry/vdso/vdso.lds.S b/arch/x86/entry/vdso/vdso.lds.S
> index 36b644e16272..4bf48462fca7 100644
> --- a/arch/x86/entry/vdso/vdso.lds.S
> +++ b/arch/x86/entry/vdso/vdso.lds.S
> @@ -27,6 +27,7 @@ VERSION {
>  		__vdso_time;
>  		clock_getres;
>  		__vdso_clock_getres;
> +		__vdso_sgx_enter_enclave;
>  	local: *;
>  	};
>  }
> diff --git a/arch/x86/entry/vdso/vsgx_enter_enclave.S b/arch/x86/entry/vdso/vsgx_enter_enclave.S
> new file mode 100644
> index 000000000000..be7e467e1efb
> --- /dev/null
> +++ b/arch/x86/entry/vdso/vsgx_enter_enclave.S
> @@ -0,0 +1,131 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +
> +#include <linux/linkage.h>
> +#include <asm/export.h>
> +#include <asm/errno.h>
> +#include <asm/enclu.h>
> +
> +#include "extable.h"
> +
> +#define EX_LEAF		0*8
> +#define EX_TRAPNR	0*8+4
> +#define EX_ERROR_CODE	0*8+6
> +#define EX_ADDRESS	1*8
> +
> +.code64
> +.section .text, "ax"
> +
> +SYM_FUNC_START(__vdso_sgx_enter_enclave)
> +	/* Prolog */
> +	.cfi_startproc
> +	push	%rbp
> +	.cfi_adjust_cfa_offset	8
> +	.cfi_rel_offset		%rbp, 0
> +	mov	%rsp, %rbp
> +	.cfi_def_cfa_register	%rbp
> +	push	%rbx
> +	.cfi_rel_offset		%rbx, -8
> +
> +	mov	%ecx, %eax
> +.Lenter_enclave:
> +	/* EENTER <= leaf <= ERESUME */
> +	cmp	$EENTER, %eax
> +	jb	.Linvalid_leaf
> +	cmp	$ERESUME, %eax
> +	ja	.Linvalid_leaf
> +
> +	/* Load TCS and AEP */
> +	mov	0x10(%rbp), %rbx
> +	lea	.Lasync_exit_pointer(%rip), %rcx
> +
> +	/* Single ENCLU serving as both EENTER and AEP (ERESUME) */
> +.Lasync_exit_pointer:
> +.Lenclu_eenter_eresume:
> +	enclu

After thinking about this some more, I'd like to come back to this setup. Prior discussion at https://lkml.org/lkml/2018/11/2/597 . I hope I'm not derailing the discussion so much as to delay the patch set :(

I previously mentioned “Userspace may want fine-grained control over enclave scheduling” as a reason userspace may want to specify a different AEP, but gave a bad example. Here's a better example: If I'm running my enclave in an M:N threading model (where M user threads run N TCSs, with N > M), an AEX is a good oppurtunity to switch contexts. Yes, I could implement this with alarm() or so, but that adds overhead while missing out on a lot of opportunities for context switching.

--
Jethro Beekman | Fortanix