Skip to content

RFC x86/vdso: x86/sgx: Rework SGX vDSO API

Sean Christopherson linux-sgx 1 version NAK

Status from this message: NAK.

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest4 patches 39 repliesNAK

Patches in v1

#Subject and filesSizeRepliesApplied
1/4x86/vdso: x86/sgx: Explicitly force 8-byte CMP for detecting user handlerarch/x86/entry/vdso/vsgx_enter_enclave.S+1 -12 replies
2/4x86/vdso: x86/sgx: Rework __vdso_sgx_enter_enclave() APIarch/x86/entry/vdso/vsgx_enter_enclave.S arch/x86/include/uapi/asm/sgx.h+107 -5520 replies
3/4x86/vdso: x86/sgx: Introduce dedicated SGX exit reasons for vDSOarch/x86/entry/vdso/vsgx_enter_enclave.S arch/x86/include/uapi/asm/sgx.h+8 -22 replies
4/4x86/vdso: x86/sgx: Allow the user to exit the vDSO loop on interruptsarch/x86/entry/vdso/vsgx_enter_enclave.S arch/x86/include/uapi/asm/sgx.h+26 -415 replies

Reviews trailers per patch

2 of 4 patches have a Reviewed-by or Acked-by.

PatchJarkko SakkinenJethro Beekman
1/4A-b A-b
2/4T-b
3/4A-b
4/4T-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v1 thread

Cover letter [RFC PATCH 0/4] x86/vdso: x86/sgx: Rework SGX vDSO API

Rework __vdso_sgx_enter_enclave() to move all input/output params, except for pass-through GPRs, into a single struct. With the new struct, add two new features (requested by Nathaniel and Jethro), and fix a long-standing nit (from Andy).

1. Add an opaque param to pass data from the runtime to its handler.
https://lkml.kernel.org/r/CAOASepOFh-vOrNZEVDFrDSuHs+9GEzzpXUTG-fZMuyjWAkpRWw@mail.gmail.com

2. Allow the runtime to exit the vDSO on interrupts, e.g. for context switching when doing M:N scheduling of enclave threads.
https://lkml.kernel.org/r/[email protected]

3. Use a dedicated exit reason instead of using -EFAULT for "exception" (and effectively -EINTR for interrupts, too).
https://lkml.kernel.org/r/[email protected]

Patch 1 is a bug fix I found by inspection when reworking the code.

Reworking so much of the code this late in the game is a bit scary, but the alternative is massive param lists for both the vDSO and the handler, especially if we add both a flags param and an opaque pointer. And IMO, the result is also a tiny bit cleaner than what we have today, even without adding @flags and @opaque.

typedef int (*vdso_sgx_enter_enclave_t)(unsigned long rdi, unsigned long rsi,
unsigned long rdx, unsigned int leaf, unsigned long r8, unsigned long r9, struct sgx_enclave_run *r);

typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx,
long ursp, long r8, long r9, struct sgx_enclave_run *r);

vs.

typedef int (*vdso_sgx_enter_enclave_t)(unsigned long rdi, unsigned long rsi,
unsigned long rdx, unsigned int leaf, unsigned long r8, unsigned long r9, void *tcs,
struct sgx_enclave_exception *e, sgx_enclave_exit_handler_t handler, unsigned long flags, unsigned long opaque);

typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx,
long ursp, long r8, long r9, void *tcs, int ret, struct sgx_enclave_exception *e, unsigned long opaque);

Sean Christopherson (4):
x86/vdso: x86/sgx: Explicitly force 8-byte CMP for detecting user
handler
x86/vdso: x86/sgx: Rework __vdso_sgx_enter_enclave() API x86/vdso: x86/sgx: Introduce dedicated SGX exit reasons for vDSO x86/vdso: x86/sgx: Allow the user to exit the vDSO loop on interrupts

 arch/x86/entry/vdso/vsgx_enter_enclave.S | 94 +++++++++++++++++------
 arch/x86/include/uapi/asm/sgx.h          | 96 ++++++++++++++++--------
 2 files changed, 135 insertions(+), 55 deletions(-)
Signature · 1 line
2.28.0