Skip to content

Commit 1a09a27

Browse files
Brian GerstIngo Molnar
authored andcommitted
x86/entry/32: Clean up syscall fast exit tests
Merge compat and native code and clarify comments. No change in functionality expected. Signed-off-by: Brian Gerst <[email protected]> Signed-off-by: Ingo Molnar <[email protected]> Cc: Andy Lutomirski <[email protected]> Cc: Borislav Petkov <[email protected]> Cc: Brian Gerst <[email protected]> Cc: Denys Vlasenko <[email protected]> Cc: H. Peter Anvin <[email protected]> Cc: Linus Torvalds <[email protected]> Cc: Peter Zijlstra <[email protected]> Cc: Thomas Gleixner <[email protected]> Cc: Josh Poimboeuf <[email protected]> Cc: Uros Bizjak <[email protected]> Link: https://lore.kernel.org/r/[email protected]
1 parent 58978b4 commit 1a09a27

File tree

1 file changed

+22
-26
lines changed

1 file changed

+22
-26
lines changed

arch/x86/entry/common.c

Lines changed: 22 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -255,34 +255,30 @@ __visible noinstr bool do_fast_syscall_32(struct pt_regs *regs)
255255
if (!__do_fast_syscall_32(regs))
256256
return false;
257257

258-
#ifdef CONFIG_X86_64
259258
/*
260-
* Opportunistic SYSRETL: if possible, try to return using SYSRETL.
261-
* SYSRETL is available on all 64-bit CPUs, so we don't need to
262-
* bother with SYSEXIT.
263-
*
264-
* Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
265-
* because the ECX fixup above will ensure that this is essentially
266-
* never the case.
259+
* Check that the register state is valid for using SYSRETL/SYSEXIT
260+
* to exit to userspace. Otherwise use the slower but fully capable
261+
* IRET exit path.
267262
*/
268-
return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
269-
regs->ip == landing_pad &&
270-
(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
271-
#else
272-
/*
273-
* Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
274-
*
275-
* Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
276-
* because the ECX fixup above will ensure that this is essentially
277-
* never the case.
278-
*
279-
* We don't allow syscalls at all from VM86 mode, but we still
280-
* need to check VM, because we might be returning from sys_vm86.
281-
*/
282-
return regs->cs == __USER_CS && regs->ss == __USER_DS &&
283-
regs->ip == landing_pad &&
284-
(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
285-
#endif
263+
264+
/* XEN PV guests always use the IRET path */
265+
if (cpu_feature_enabled(X86_FEATURE_XENPV))
266+
return false;
267+
268+
/* EIP must point to the VDSO landing pad */
269+
if (unlikely(regs->ip != landing_pad))
270+
return false;
271+
272+
/* CS and SS must match the values set in MSR_STAR */
273+
if (unlikely(regs->cs != __USER32_CS || regs->ss != __USER_DS))
274+
return false;
275+
276+
/* If the TF, RF, or VM flags are set, use IRET */
277+
if (unlikely(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)))
278+
return false;
279+
280+
/* Use SYSRETL/SYSEXIT to exit to userspace */
281+
return true;
286282
}
287283

288284
/* Returns true to return using SYSEXIT/SYSRETL, or false to use IRET */

0 commit comments

Comments
 (0)