lkml.org 
[lkml]   [2021]   [Sep]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH] x86/entry: move ptregs->rsp and IF fixups to asm, remove do_SYSENTER_32()
Date
This removes two loads, one store, and one level of function calls
from both 32- and 64-bit SYSENTER32 code paths:

do_SYSENTER_32() does two fixups:

/* SYSENTER loses RSP, but the vDSO saved it in RBP. */
regs->sp = regs->bp;
but we can just do this when we form regs->sp on the stack in the first place.

/* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
regs->flags |= X86_EFLAGS_IF;
but we already load regs->flags in asm code and peek at its bits in order
to clear TF et al. We can set IF and update regs->flags right there,
avoiding loading it again.

With this changed, do_SYSENTER_32() does not do anything except calling
do_fast_syscall_32(), so get rid of it.

While at it, tweak some outdated comments.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
---
arch/x86/entry/common.c | 12 ------------
arch/x86/entry/entry_32.S | 13 +++++++++----
arch/x86/entry/entry_64_compat.S | 19 ++++++++++++-------
3 files changed, 21 insertions(+), 23 deletions(-)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 6c2826417b33..7639b0fc7e8a 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -233,18 +233,6 @@ __visible noinstr long do_fast_syscall_32(struct pt_regs *regs)
(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
#endif
}
-
-/* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
-__visible noinstr long do_SYSENTER_32(struct pt_regs *regs)
-{
- /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
- regs->sp = regs->bp;
-
- /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
- regs->flags |= X86_EFLAGS_IF;
-
- return do_fast_syscall_32(regs);
-}
#endif

SYSCALL_DEFINE0(ni_syscall)
diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index ccb9d32768f3..e14b13e9c3e1 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -837,7 +837,8 @@ SYM_FUNC_START(entry_SYSENTER_32)

.Lsysenter_past_esp:
pushl $__USER_DS /* pt_regs->ss */
- pushl $0 /* pt_regs->sp (placeholder) */
+ /* SYSENTER loses RSP, but the vDSO saved it in EBP. */
+ pushl %ebp /* pt_regs->sp */
pushfl /* pt_regs->flags (except IF = 0) */
pushl $__USER_CS /* pt_regs->cs */
pushl $0 /* pt_regs->ip = 0 (placeholder) */
@@ -845,7 +846,7 @@ SYM_FUNC_START(entry_SYSENTER_32)
SAVE_ALL pt_regs_ax=$-ENOSYS /* save rest, stack already switched */

/*
- * SYSENTER doesn't filter flags, so we need to clear NT, AC
+ * SYSENTER filters only IF and VM, so we need to clear NT, AC
* and TF ourselves. To save a few cycles, we can check whether
* either was set instead of doing an unconditional popfq.
* This needs to happen before enabling interrupts so that
@@ -863,12 +864,16 @@ SYM_FUNC_START(entry_SYSENTER_32)
* we're keeping that code behind a branch which will predict as
* not-taken and therefore its instructions won't be fetched.
*/
- testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
+ movl PT_EFLAGS(%esp), %eax
+ testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, %eax
jnz .Lsysenter_fix_flags
.Lsysenter_flags_fixed:
+ /* SYSENTER cleared IF. Assume it was set in usermode. */
+ orl $X86_EFLAGS_IF, %eax
+ movl %eax, PT_EFLAGS(%esp)

movl %esp, %eax
- call do_SYSENTER_32
+ call do_fast_syscall_32
testl %eax, %eax
jz .Lsyscall_32_done

diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index 0051cf5c792d..71a9bc09b14d 100644
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -59,13 +59,14 @@ SYM_CODE_START(entry_SYSENTER_compat)

/* Construct struct pt_regs on stack */
pushq $__USER32_DS /* pt_regs->ss */
- pushq $0 /* pt_regs->sp = 0 (placeholder) */
+ /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
+ pushq %rbp /* pt_regs->sp */

/*
* Push flags. This is nasty. First, interrupts are currently
- * off, but we need pt_regs->flags to have IF set. Second, if TS
+ * off, but we need pt_regs->flags to have IF set. Second, if TF
* was set in usermode, it's still set, and we're singlestepping
- * through this code. do_SYSENTER_32() will fix up IF.
+ * through this code. We fix up IF before do_fast_syscall_32().
*/
pushfq /* pt_regs->flags (except IF = 0) */
pushq $__USER32_CS /* pt_regs->cs */
@@ -113,7 +114,7 @@ SYM_INNER_LABEL(entry_SYSENTER_compat_after_hwframe, SYM_L_GLOBAL)
cld

/*
- * SYSENTER doesn't filter flags, so we need to clear NT and AC
+ * SYSENTER filters only IF and VM, so we need to clear NT and AC
* ourselves. To save a few cycles, we can check whether
* either was set instead of doing an unconditional popfq.
* This needs to happen before enabling interrupts so that
@@ -131,12 +132,16 @@ SYM_INNER_LABEL(entry_SYSENTER_compat_after_hwframe, SYM_L_GLOBAL)
* we're keeping that code behind a branch which will predict as
* not-taken and therefore its instructions won't be fetched.
*/
- testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, EFLAGS(%rsp)
+ movl EFLAGS(%rsp), %edi
+ testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, %edi
jnz .Lsysenter_fix_flags
.Lsysenter_flags_fixed:
+ /* SYSENTER cleared IF. Assume it was set in usermode. */
+ orl $X86_EFLAGS_IF, %edi
+ movl %edi, EFLAGS(%rsp)

movq %rsp, %rdi
- call do_SYSENTER_32
+ call do_fast_syscall_32
/* XEN PV guests always use IRET path */
ALTERNATIVE "testl %eax, %eax; jz swapgs_restore_regs_and_return_to_usermode", \
"jmp swapgs_restore_regs_and_return_to_usermode", X86_FEATURE_XENPV
@@ -189,7 +194,7 @@ SYM_CODE_END(entry_SYSENTER_compat)
* eax system call number
* ecx return address
* ebx arg1
- * ebp arg2 (note: not saved in the stack frame, should not be touched)
+ * ebp arg2
* edx arg3
* esi arg4
* edi arg5
--
2.30.0
\
 
 \ /
  Last update: 2021-09-13 22:55    [W:0.069 / U:0.656 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site