[PATCHv2 0/4] x86, entry: some cleanup and simplification...

[PATCHv2 0/4] x86, entry: some cleanup and simplification...

[Alexander van Heukelum]

Hi Andy,

The last patchset did not compile on i386. Please ignore it. This one
should be better. Instead of removing KERNEL_STACK_OFFSET, it is now
used consistently on both i386 and x86_64.

Boot tested using qemu (using klibc for userspace)
	- x86_64, 32-bit userspace, core2duo (sysenter32)
	- x86_64, 32-bit userspace, phenom (syscall32)
	- x86_64, 32-bit userspace, vdso=0 (int 0x80)
	- x86_64, 64-bit userspace
	- i386, pentium3 (sysenter)
	- i386, athlon (syscall)
	- i386, vdso=0 (int 0x80)

They were tested on top of 22f2aa4a0361707a5cfb1de9d45260b39965dead
(x86/entry-devel in your tree) and this kernel is now running on my
laptop.

Greetings,
   Alexander

Alexander van Heukelum (4):
  x86_64: cleanup THREAD_INFO(reg,offset) macro
  x86_64: embrace KERNEL_STACK_OFFSET
  i386: clean up KERNEL_STACK_OFFSET
  x86_64, entry: Create IRET-compatible stack frame at syscall entry

 arch/x86/ia32/ia32entry.S          | 33 ++++++++--------
 arch/x86/include/asm/calling.h     |  1 +
 arch/x86/include/asm/processor.h   | 43 ++++++++-------------
 arch/x86/include/asm/thread_info.h | 17 +++++++-
 arch/x86/kernel/entry_32.S         |  5 ++-
 arch/x86/kernel/entry_64.S         | 79 +++++++++++---------------------------
 arch/x86/kernel/process_64.c       |  5 +--
 7 files changed, 75 insertions(+), 108 deletions(-)

-- 
2.1.0

[PATCHv2 1/4] x86_64: cleanup THREAD_INFO(reg,offset) macro

[Alexander van Heukelum]

The macro THREAD_INFO(reg,offset) is used in assembly to compute the
offset between the user ptregs and the thread_info struct. Change
the macro and all its uses so that offset is given as the current
top of stack in the pt_regs frame. The generated code is identical.

Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
---
 arch/x86/ia32/ia32entry.S          | 30 +++++++++++++++---------------
 arch/x86/include/asm/calling.h     |  1 +
 arch/x86/include/asm/thread_info.h |  2 +-
 arch/x86/kernel/entry_64.S         |  4 ++--
 4 files changed, 19 insertions(+), 18 deletions(-)

diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index 156ebca..1c74f39 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -135,7 +135,7 @@ ENTRY(ia32_sysenter_target)
 	CFI_REL_OFFSET rsp,0
 	pushfq_cfi
 	/*CFI_REL_OFFSET rflags,0*/
-	movl	TI_sysenter_return+THREAD_INFO(%rsp,3*8-KERNEL_STACK_OFFSET),%r10d
+	movl	TI_sysenter_return+THREAD_INFO(%rsp,EFLAGS),%r10d
 	CFI_REGISTER rip,r10
 	pushq_cfi $__USER32_CS
 	/*CFI_REL_OFFSET cs,0*/
@@ -161,8 +161,8 @@ ENTRY(ia32_sysenter_target)
 	jnz sysenter_fix_flags
 sysenter_flags_fixed:
 
-	orl     $TS_COMPAT,TI_status+THREAD_INFO(%rsp,RIP-ARGOFFSET)
-	testl   $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	orl     $TS_COMPAT,TI_status+THREAD_INFO(%rsp,ARGOFFSET)
+	testl   $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	CFI_REMEMBER_STATE
 	jnz  sysenter_tracesys
 	cmpq	$(IA32_NR_syscalls-1),%rax
@@ -174,10 +174,10 @@ sysenter_dispatch:
 	movq	%rax,RAX-ARGOFFSET(%rsp)
 	DISABLE_INTERRUPTS(CLBR_NONE)
 	TRACE_IRQS_OFF
-	testl	$_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl	$_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz	sysexit_audit
 sysexit_from_sys_call:
-	andl    $~TS_COMPAT,TI_status+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	andl    $~TS_COMPAT,TI_status+THREAD_INFO(%rsp,ARGOFFSET)
 	/* clear IF, that popfq doesn't enable interrupts early */
 	andl	$~0x200,EFLAGS-ARGOFFSET(%rsp)
 	movl	RIP-ARGOFFSET(%rsp),%edx		/* User %eip */
@@ -216,7 +216,7 @@ sysexit_from_sys_call:
 	.endm
 
 	.macro auditsys_exit exit
-	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz ia32_ret_from_sys_call
 	TRACE_IRQS_ON
 	ENABLE_INTERRUPTS(CLBR_NONE)
@@ -231,7 +231,7 @@ sysexit_from_sys_call:
 	movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
 	DISABLE_INTERRUPTS(CLBR_NONE)
 	TRACE_IRQS_OFF
-	testl %edi,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl %edi,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jz \exit
 	CLEAR_RREGS -ARGOFFSET
 	jmp int_with_check
@@ -253,7 +253,7 @@ sysenter_fix_flags:
 
 sysenter_tracesys:
 #ifdef CONFIG_AUDITSYSCALL
-	testl	$(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl	$(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jz	sysenter_auditsys
 #endif
 	SAVE_REST
@@ -324,8 +324,8 @@ ENTRY(ia32_cstar_target)
 1:	movl	(%r8),%r9d
 	_ASM_EXTABLE(1b,ia32_badarg)
 	ASM_CLAC
-	orl     $TS_COMPAT,TI_status+THREAD_INFO(%rsp,RIP-ARGOFFSET)
-	testl   $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	orl     $TS_COMPAT,TI_status+THREAD_INFO(%rsp,ARGOFFSET)
+	testl   $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	CFI_REMEMBER_STATE
 	jnz   cstar_tracesys
 	cmpq $IA32_NR_syscalls-1,%rax
@@ -337,10 +337,10 @@ cstar_dispatch:
 	movq %rax,RAX-ARGOFFSET(%rsp)
 	DISABLE_INTERRUPTS(CLBR_NONE)
 	TRACE_IRQS_OFF
-	testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz sysretl_audit
 sysretl_from_sys_call:
-	andl $~TS_COMPAT,TI_status+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	andl $~TS_COMPAT,TI_status+THREAD_INFO(%rsp,ARGOFFSET)
 	RESTORE_ARGS 0,-ARG_SKIP,0,0,0
 	movl RIP-ARGOFFSET(%rsp),%ecx
 	CFI_REGISTER rip,rcx
@@ -368,7 +368,7 @@ sysretl_audit:
 
 cstar_tracesys:
 #ifdef CONFIG_AUDITSYSCALL
-	testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jz cstar_auditsys
 #endif
 	xchgl %r9d,%ebp
@@ -434,8 +434,8 @@ ENTRY(ia32_syscall)
 	/* note the registers are not zero extended to the sf.
 	   this could be a problem. */
 	SAVE_ARGS 0,1,0
-	orl $TS_COMPAT,TI_status+THREAD_INFO(%rsp,RIP-ARGOFFSET)
-	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	orl $TS_COMPAT,TI_status+THREAD_INFO(%rsp,ARGOFFSET)
+	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz ia32_tracesys
 	cmpq $(IA32_NR_syscalls-1),%rax
 	ja ia32_badsys
diff --git a/arch/x86/include/asm/calling.h b/arch/x86/include/asm/calling.h
index 1f1297b..16ab13d 100644
--- a/arch/x86/include/asm/calling.h
+++ b/arch/x86/include/asm/calling.h
@@ -81,6 +81,7 @@ For 32-bit we have the following conventions - kernel is built with
 #define EFLAGS		144
 #define RSP		152
 #define SS		160
+#define PTREGS_SIZE	168
 
 #define ARGOFFSET	R11
 
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index e82e95a..471037d 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -190,7 +190,7 @@ static inline unsigned long current_stack_pointer(void)
  * Same if PER_CPU_VAR(kernel_stack) is, perhaps with some offset, already in
  * a certain register (to be used in assembler memory operands).
  */
-#define THREAD_INFO(reg, off) KERNEL_STACK_OFFSET+(off)-THREAD_SIZE(reg)
+#define THREAD_INFO(reg, off) PTREGS_SIZE-(off)-THREAD_SIZE(reg)
 
 #endif
 
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index db13655..9f9ca20 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -343,7 +343,7 @@ GLOBAL(system_call_after_swapgs)
 	movq_cfi rax,(ORIG_RAX-ARGOFFSET)
 	movq  %rcx,RIP-ARGOFFSET(%rsp)
 	CFI_REL_OFFSET rip,RIP-ARGOFFSET
-	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz tracesys
 system_call_fastpath:
 #if __SYSCALL_MASK == ~0
@@ -361,7 +361,7 @@ system_call_fastpath:
  * Has incomplete stack frame and undefined top of stack.
  */
 ret_from_sys_call:
-	testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
+	testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz int_ret_from_sys_call_fixup	/* Go the the slow path */
 
 	LOCKDEP_SYS_EXIT
-- 
2.1.0

Re: [PATCHv2 1/4] x86_64: cleanup THREAD_INFO(reg,offset) macro

[Denys Vlasenko]

On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
<heukelum@fastmail.fm> wrote:
> The macro THREAD_INFO(reg,offset) is used in assembly to compute the
> offset between the user ptregs and the thread_info struct. Change
> the macro and all its uses so that offset is given as the current
> top of stack in the pt_regs frame. The generated code is identical.

What is the purpose of doing this?

I don't mean to say that it's pointless, but it is also not obvious
why it's better than the code before patch.

Re: [PATCHv2 1/4] x86_64: cleanup THREAD_INFO(reg,offset) macro

[Alexander van Heukelum]

On Wed, Jan 21, 2015, at 14:40, Denys Vlasenko wrote:
> On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
> <heukelum@fastmail.fm> wrote:
> > The macro THREAD_INFO(reg,offset) is used in assembly to compute the
> > offset between the user ptregs and the thread_info struct. Change
> > the macro and all its uses so that offset is given as the current
> > top of stack in the pt_regs frame. The generated code is identical.
> 
> What is the purpose of doing this?
> 
> I don't mean to say that it's pointless, but it is also not obvious
> why it's better than the code before patch.

Just because I had a hard time understanding what the offset
parameter means. Turns out Borislav did another variant in his
patchset. Quoting him from the commit message which made
that the change:
> Semi-mysterious expressions THREAD_INFO(%rsp,RIP) - "why RIP??"
> are now replaced by more logical THREAD_INFO(%rsp,SIZEOF_PTREGS) -
> "rsp is SIZEOF_PTREGS bytes below the top, calculate
> thread_info's address using that information"

Borislav expresses "offset" as an offset from the stack's page
boundary,while I chose the offset from the start of struct
pt_regs on the stack.

Re: [PATCHv2 1/4] x86_64: cleanup THREAD_INFO(reg,offset) macro

[Borislav Petkov]

On Wed, Jan 21, 2015 at 05:20:09PM +0100, Alexander van Heukelum wrote:
> On Wed, Jan 21, 2015, at 14:40, Denys Vlasenko wrote:
> > On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
> > <heukelum@fastmail.fm> wrote:
> > > The macro THREAD_INFO(reg,offset) is used in assembly to compute the
> > > offset between the user ptregs and the thread_info struct. Change
> > > the macro and all its uses so that offset is given as the current
> > > top of stack in the pt_regs frame. The generated code is identical.
> > 
> > What is the purpose of doing this?
> > 
> > I don't mean to say that it's pointless, but it is also not obvious
> > why it's better than the code before patch.
> 
> Just because I had a hard time understanding what the offset
> parameter means. Turns out Borislav did another variant in his
> patchset. Quoting him from the commit message which made
> that the change:
> > Semi-mysterious expressions THREAD_INFO(%rsp,RIP) - "why RIP??"
> > are now replaced by more logical THREAD_INFO(%rsp,SIZEOF_PTREGS) -
> > "rsp is SIZEOF_PTREGS bytes below the top, calculate
> > thread_info's address using that information"
> 
> Borislav expresses "offset" as an offset from the stack's page
> boundary,while I chose the offset from the start of struct
> pt_regs on the stack.

I've never done any such thing :-)

s/Borislav/Denys/g.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCHv2 1/4] x86_64: cleanup THREAD_INFO(reg,offset) macro

[Alexander van Heukelum]

On Wed, Jan 21, 2015, at 19:04, Borislav Petkov wrote:
> On Wed, Jan 21, 2015 at 05:20:09PM +0100, Alexander van Heukelum wrote:
> > On Wed, Jan 21, 2015, at 14:40, Denys Vlasenko wrote:
> > > On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
> > > <heukelum@fastmail.fm> wrote:
> > > > The macro THREAD_INFO(reg,offset) is used in assembly to compute the
> > > > offset between the user ptregs and the thread_info struct. Change
> > > > the macro and all its uses so that offset is given as the current
> > > > top of stack in the pt_regs frame. The generated code is identical.
> > > 
> > > What is the purpose of doing this?
> > > 
> > > I don't mean to say that it's pointless, but it is also not obvious
> > > why it's better than the code before patch.
> > 
> > Just because I had a hard time understanding what the offset
> > parameter means. Turns out Borislav did another variant in his
> > patchset. Quoting him from the commit message which made
> > that the change:
> > > Semi-mysterious expressions THREAD_INFO(%rsp,RIP) - "why RIP??"
> > > are now replaced by more logical THREAD_INFO(%rsp,SIZEOF_PTREGS) -
> > > "rsp is SIZEOF_PTREGS bytes below the top, calculate
> > > thread_info's address using that information"
> > 
> > Borislav expresses "offset" as an offset from the stack's page
> > boundary,while I chose the offset from the start of struct
> > pt_regs on the stack.
> 
> I've never done any such thing :-)
> 
> s/Borislav/Denys/g.

twice ... and stop using third person too... Oops.

Err. sorry :-/.
 
> -- 
> Regards/Gruss,
>     Boris.
> 
> ECO tip #101: Trim your mails when you reply.
> --

[PATCHv2 2/4] x86_64: embrace KERNEL_STACK_OFFSET

[Alexander van Heukelum]

KERNEL_STACK_OFFSET is the offset from the top of the kernel stack
page to the value of the kernel_stack percpu variable. This patch
changes KERNEL_STACK_OFFSET to configure a reserved space of 16
bytes above the user ptregs frame. KERNEL_STACK_OFFSET must be
set to a multiple of 16 bytes due to the automatic stack alignment
of interrupts, traps, and exceptions on x86_64.

Also change task_pt_regs to be independant of the thread's current
sp0 setting, like i386, and use it to initialize thread.sp0 in
copy_thread.

Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
---
 arch/x86/ia32/ia32entry.S          |  3 +--
 arch/x86/include/asm/processor.h   | 11 ++++++++---
 arch/x86/include/asm/thread_info.h | 13 ++++++++++++-
 arch/x86/kernel/entry_64.S         |  2 +-
 arch/x86/kernel/process_64.c       |  5 ++---
 5 files changed, 24 insertions(+), 10 deletions(-)

diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index 1c74f39..4c6c5d9 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -122,7 +122,6 @@ ENTRY(ia32_sysenter_target)
 	CFI_REGISTER	rsp,rbp
 	SWAPGS_UNSAFE_STACK
 	movq	PER_CPU_VAR(kernel_stack), %rsp
-	addq	$(KERNEL_STACK_OFFSET),%rsp
 	/*
 	 * No need to follow this irqs on/off section: the syscall
 	 * disabled irqs, here we enable it straight after entry:
@@ -304,7 +303,7 @@ ENTRY(ia32_cstar_target)
 	 * disabled irqs and here we enable it straight after entry:
 	 */
 	ENABLE_INTERRUPTS(CLBR_NONE)
-	SAVE_ARGS 8,0,0
+	SAVE_ARGS 6*8,0,0	/* skip: hardware stackframe and orig_rax */
 	movl 	%eax,%eax	/* zero extension */
 	movq	%rax,ORIG_RAX-ARGOFFSET(%rsp)
 	movq	%rcx,RIP-ARGOFFSET(%rsp)
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index a092a0c..97117d1 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -919,11 +919,13 @@ extern unsigned long thread_saved_pc(struct task_struct *tsk);
 #define STACK_TOP_MAX		TASK_SIZE_MAX
 
 #define INIT_THREAD  { \
-	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
+	.sp0 = (unsigned long)&init_stack + \
+		sizeof(init_stack) - KERNEL_STACK_OFFSET \
 }
 
 #define INIT_TSS  { \
-	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
+	.x86_tss.sp0 = (unsigned long)&init_stack + \
+		sizeof(init_stack) - KERNEL_STACK_OFFSET \
 }
 
 /*
@@ -932,7 +934,10 @@ extern unsigned long thread_saved_pc(struct task_struct *tsk);
  */
 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
 
-#define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
+#define task_pt_regs(task) \
+	((struct pt_regs *)((unsigned long)task_stack_page(task) + \
+			THREAD_SIZE - KERNEL_STACK_OFFSET) - 1)
+
 extern unsigned long KSTK_ESP(struct task_struct *task);
 
 /*
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index 471037d..9f0c47f 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -149,7 +149,18 @@ struct thread_info {
 #define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW)
 
 #define STACK_WARN		(THREAD_SIZE/8)
+
+/*
+ * Amount of reserved space between the top of the kernel stack page and the
+ * user ptregs frame.
+ * On x86_64, KERNEL_STACK_OFFSET must be set to a multiple of 16 bytes due
+ * to its automatic stack alignment for interrupts, traps, and exceptions.
+ */
+#ifdef CONFIG_X86_32
 #define KERNEL_STACK_OFFSET	(5*(BITS_PER_LONG/8))
+#else
+#define KERNEL_STACK_OFFSET	(2*(BITS_PER_LONG/8))
+#endif
 
 /*
  * macros/functions for gaining access to the thread information structure
@@ -190,7 +201,7 @@ static inline unsigned long current_stack_pointer(void)
  * Same if PER_CPU_VAR(kernel_stack) is, perhaps with some offset, already in
  * a certain register (to be used in assembler memory operands).
  */
-#define THREAD_INFO(reg, off) PTREGS_SIZE-(off)-THREAD_SIZE(reg)
+#define THREAD_INFO(reg, off) PTREGS_SIZE+KERNEL_STACK_OFFSET-(off)-THREAD_SIZE(reg)
 
 #endif
 
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 9f9ca20..6b95c2f 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -339,7 +339,7 @@ GLOBAL(system_call_after_swapgs)
 	 * and short:
 	 */
 	ENABLE_INTERRUPTS(CLBR_NONE)
-	SAVE_ARGS 8, 0, rax_enosys=1
+	SAVE_ARGS 6*8, 0, rax_enosys=1	/* skip: hardware stackframe and orig_rax */
 	movq_cfi rax,(ORIG_RAX-ARGOFFSET)
 	movq  %rcx,RIP-ARGOFFSET(%rsp)
 	CFI_REL_OFFSET rip,RIP-ARGOFFSET
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 5a2c029..d579ebf 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -155,12 +155,11 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
 		unsigned long arg, struct task_struct *p)
 {
 	int err;
-	struct pt_regs *childregs;
+	struct pt_regs *childregs = task_pt_regs(p);
 	struct task_struct *me = current;
 
-	p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
-	childregs = task_pt_regs(p);
 	p->thread.sp = (unsigned long) childregs;
+	p->thread.sp0 = (unsigned long) (childregs + 1);
 	p->thread.usersp = me->thread.usersp;
 	set_tsk_thread_flag(p, TIF_FORK);
 	p->thread.io_bitmap_ptr = NULL;
-- 
2.1.0

Re: [PATCHv2 2/4] x86_64: embrace KERNEL_STACK_OFFSET

[Denys Vlasenko]

On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
<heukelum@fastmail.fm> wrote:
> KERNEL_STACK_OFFSET is the offset from the top of the kernel stack
> page to the value of the kernel_stack percpu variable. This patch
> changes KERNEL_STACK_OFFSET to configure a reserved space of 16
> bytes above the user ptregs frame. KERNEL_STACK_OFFSET must be
> set to a multiple of 16 bytes due to the automatic stack alignment
> of interrupts, traps, and exceptions on x86_64.

I propose to set kernel_stack percpu variable to point
to the top of kernel stack (obvious, isn't it?)
and eliminate KERNEL_STACK_OFFSET altogether.

Re: [PATCHv2 2/4] x86_64: embrace KERNEL_STACK_OFFSET

[Alexander van Heukelum]

On Wed, Jan 21, 2015, at 14:44, Denys Vlasenko wrote:
> On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
> <heukelum@fastmail.fm> wrote:
> > KERNEL_STACK_OFFSET is the offset from the top of the kernel stack
> > page to the value of the kernel_stack percpu variable. This patch
> > changes KERNEL_STACK_OFFSET to configure a reserved space of 16
> > bytes above the user ptregs frame. KERNEL_STACK_OFFSET must be
> > set to a multiple of 16 bytes due to the automatic stack alignment
> > of interrupts, traps, and exceptions on x86_64.
> 
> I propose to set kernel_stack percpu variable to point
> to the top of kernel stack (obvious, isn't it?)
> and eliminate KERNEL_STACK_OFFSET altogether.

By "top of kernel stack", do you mean the page boundary or the
top of struct pt_regs on the kernel stack? (is it really that obvious?)
I think Borislav did the latter for x86_64 in his patchset.

Eliminating KERNEL_STACK_OFFSET was what I did in my first
attempt, and I broke i386 by not thinking through what x86 common
code was really doing :).

Re: [PATCHv2 2/4] x86_64: embrace KERNEL_STACK_OFFSET

[Denys Vlasenko]

On Wed, Jan 21, 2015 at 5:29 PM, Alexander van Heukelum
<heukelum@fastmail.fm> wrote:
> On Wed, Jan 21, 2015, at 14:44, Denys Vlasenko wrote:
>> On Sun, Jan 18, 2015 at 12:45 PM, Alexander van Heukelum
>> <heukelum@fastmail.fm> wrote:
>> > KERNEL_STACK_OFFSET is the offset from the top of the kernel stack
>> > page to the value of the kernel_stack percpu variable. This patch
>> > changes KERNEL_STACK_OFFSET to configure a reserved space of 16
>> > bytes above the user ptregs frame. KERNEL_STACK_OFFSET must be
>> > set to a multiple of 16 bytes due to the automatic stack alignment
>> > of interrupts, traps, and exceptions on x86_64.
>>
>> I propose to set kernel_stack percpu variable to point
>> to the top of kernel stack (obvious, isn't it?)
>> and eliminate KERNEL_STACK_OFFSET altogether.
>
> By "top of kernel stack", do you mean the page boundary or the
> top of struct pt_regs on the kernel stack? (is it really that obvious?)
> I think Borislav did the latter for x86_64 in his patchset.

Page boundary.

kernel_stack is currently initialized as follows:

        this_cpu_write(kernel_stack,
                  (unsigned long)task_stack_page(next_p) +
                  THREAD_SIZE - KERNEL_STACK_OFFSET);

i.e. it points KERNEL_STACK_OFFSET bytes below top-of-stack,
which is two pages above task_struct.

Why do we have KERNEL_STACK_OFFSET?

The original idea was that on SYSCALL instruction entry, which
does not create iret stack, we can eliminate one "sub $5*8,%rsp"
instruction. This idea currently does not work, because we
have such instruction anyway (it allocates pr_regs). Nothing is saved there.

And here, in 32-bit compat code:

ENTRY(ia32_sysenter_target)
        CFI_STARTPROC32 simple
        CFI_SIGNAL_FRAME
        CFI_DEF_CFA     rsp,0
        CFI_REGISTER    rsp,rbp
        SWAPGS_UNSAFE_STACK
        movq    PER_CPU_VAR(kernel_stack), %rsp
        addq    $(KERNEL_STACK_OFFSET),%rsp

we even need to _undo_ the "KERNEL_STACK_OFFSET optimization"
(last insn).

My patch "[PATCH 09/11] x86: get rid of KERNEL_STACK_OFFSET"
simply drops the KERNEL_STACK_OFFSET thing.

[PATCHv2 3/4] i386: clean up KERNEL_STACK_OFFSET

[Alexander van Heukelum]

On i386, 8 bytes are reserved above the user ptregs frame. The area is
unused, but "necessary to guarantee that the entire "struct pt_regs" is
accessible even if the CPU haven't stored the SS/ESP registers on the
stack (interrupt gate does not save these registers when switching to
the same priv ring)."

Use KERNEL_STACK_OFFSET to make the size of this area configurable and
remove the difference between the sp0 setting in the tss and the percpu
variable kernel_stack.

For i386, KERNEL_STACK_OFFSET must be at least 8 bytes for the reason
mentioned above and must be a multiple of 4 bytes, the minimal stack
alignment.

Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
---
 arch/x86/include/asm/processor.h   | 32 +++++++-------------------------
 arch/x86/include/asm/thread_info.h | 10 ++++++----
 arch/x86/kernel/entry_32.S         |  5 +++--
 3 files changed, 16 insertions(+), 31 deletions(-)

diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 97117d1..f424e5f 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -842,7 +842,8 @@ static inline void spin_lock_prefetch(const void *x)
 #define STACK_TOP_MAX		STACK_TOP
 
 #define INIT_THREAD  {							  \
-	.sp0			= sizeof(init_stack) + (long)&init_stack, \
+	.sp0			= sizeof(init_stack) + (long)&init_stack  \
+						- KERNEL_STACK_OFFSET,    \
 	.vm86_info		= NULL,					  \
 	.sysenter_cs		= __KERNEL_CS,				  \
 	.io_bitmap_ptr		= NULL,					  \
@@ -856,7 +857,8 @@ static inline void spin_lock_prefetch(const void *x)
  */
 #define INIT_TSS  {							  \
 	.x86_tss = {							  \
-		.sp0		= sizeof(init_stack) + (long)&init_stack, \
+		.sp0		= sizeof(init_stack) + (long)&init_stack  \
+						- KERNEL_STACK_OFFSET,    \
 		.ss0		= __KERNEL_DS,				  \
 		.ss1		= __KERNEL_CS,				  \
 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
@@ -866,29 +868,9 @@ static inline void spin_lock_prefetch(const void *x)
 
 extern unsigned long thread_saved_pc(struct task_struct *tsk);
 
-#define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
-#define KSTK_TOP(info)                                                 \
-({                                                                     \
-       unsigned long *__ptr = (unsigned long *)(info);                 \
-       (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
-})
-
-/*
- * The below -8 is to reserve 8 bytes on top of the ring0 stack.
- * This is necessary to guarantee that the entire "struct pt_regs"
- * is accessible even if the CPU haven't stored the SS/ESP registers
- * on the stack (interrupt gate does not save these registers
- * when switching to the same priv ring).
- * Therefore beware: accessing the ss/esp fields of the
- * "struct pt_regs" is possible, but they may contain the
- * completely wrong values.
- */
-#define task_pt_regs(task)                                             \
-({                                                                     \
-       struct pt_regs *__regs__;                                       \
-       __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
-       __regs__ - 1;                                                   \
-})
+#define task_pt_regs(task) \
+	((struct pt_regs *)((unsigned long)task_stack_page(task) + \
+			THREAD_SIZE - KERNEL_STACK_OFFSET) - 1)
 
 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
 
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index 9f0c47f..36b8a10 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -153,14 +153,16 @@ struct thread_info {
 /*
  * Amount of reserved space between the top of the kernel stack page and the
  * user ptregs frame.
+ * On i386, this is necessary to guarantee that the entire "struct pt_regs"
+ * is accessible even if the CPU hasn't stored the SS/ESP registers on the
+ * stack (an interrupt gate does not save these registers when switching to
+ * the same priv ring). Therefore beware: accessing the ss/esp fields of the
+ * "struct pt_regs" is possible, but they may contain the completely wrong
+ * values.
  * On x86_64, KERNEL_STACK_OFFSET must be set to a multiple of 16 bytes due
  * to its automatic stack alignment for interrupts, traps, and exceptions.
  */
-#ifdef CONFIG_X86_32
-#define KERNEL_STACK_OFFSET	(5*(BITS_PER_LONG/8))
-#else
 #define KERNEL_STACK_OFFSET	(2*(BITS_PER_LONG/8))
-#endif
 
 /*
  * macros/functions for gaining access to the thread information structure
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index 000d419..e94b994 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -396,9 +396,10 @@ sysenter_past_esp:
 	/*
 	 * Push current_thread_info()->sysenter_return to the stack.
 	 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
-	 * pushed above; +8 corresponds to copy_thread's esp0 setting.
+	 * pushed above; KERNEL_STACK_OFFSET corresponds to copy_thread's
+	 * esp0 setting.
 	 */
-	pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
+	pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+KERNEL_STACK_OFFSET+4*4)(%esp)
 	CFI_REL_OFFSET eip, 0
 
 	pushl_cfi %eax
-- 
2.1.0

[PATCHv2 4/4] x86_64, entry: Create IRET-compatible stack frame at syscall entry

[Alexander van Heukelum]

Create an IRET-compatible top of stack at syscall entry and use this
information to return to user mode in the sysret path. This removes
the need for the FIXUP_TOP_OF_STACK and RESTORE_TOP_OF_STACK macros.

Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
---
 arch/x86/kernel/entry_64.S | 75 +++++++++++++---------------------------------
 1 file changed, 21 insertions(+), 54 deletions(-)

diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 6b95c2f..c4cb8f1 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -33,8 +33,6 @@
  * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
  * Gives a full stack frame.
  * - ENTRY/END Define functions in the symbol table.
- * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
- * frame that is otherwise undefined after a SYSCALL
  * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
  * - idtentry - Define exception entry points.
  */
@@ -130,33 +128,6 @@ ENDPROC(native_usergs_sysret64)
 #endif
 
 /*
- * C code is not supposed to know about undefined top of stack. Every time
- * a C function with an pt_regs argument is called from the SYSCALL based
- * fast path FIXUP_TOP_OF_STACK is needed.
- * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
- * manipulation.
- */
-
-	/* %rsp:at FRAMEEND */
-	.macro FIXUP_TOP_OF_STACK tmp offset=0
-	movq PER_CPU_VAR(old_rsp),\tmp
-	movq \tmp,RSP+\offset(%rsp)
-	movq $__USER_DS,SS+\offset(%rsp)
-	movq $__USER_CS,CS+\offset(%rsp)
-	movq RIP+\offset(%rsp),\tmp  /* get rip */
-	movq \tmp,RCX+\offset(%rsp)  /* copy it to rcx as sysret would do */
-	movq R11+\offset(%rsp),\tmp  /* get eflags */
-	movq \tmp,EFLAGS+\offset(%rsp)
-	.endm
-
-	.macro RESTORE_TOP_OF_STACK tmp offset=0
-	movq RSP+\offset(%rsp),\tmp
-	movq \tmp,PER_CPU_VAR(old_rsp)
-	movq EFLAGS+\offset(%rsp),\tmp
-	movq \tmp,R11+\offset(%rsp)
-	.endm
-
-/*
  * initial frame state for interrupts (and exceptions without error code)
  */
 	.macro EMPTY_FRAME start=1 offset=0
@@ -272,7 +243,6 @@ ENTRY(ret_from_fork)
 	testl $_TIF_IA32, TI_flags(%rcx)	# 32-bit compat task needs IRET
 	jnz  int_ret_from_sys_call
 
-	RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
 	jmp ret_from_sys_call			# go to the SYSRET fastpath
 
 1:
@@ -339,10 +309,24 @@ GLOBAL(system_call_after_swapgs)
 	 * and short:
 	 */
 	ENABLE_INTERRUPTS(CLBR_NONE)
-	SAVE_ARGS 6*8, 0, rax_enosys=1	/* skip: hardware stackframe and orig_rax */
+	/*
+	 * Save user mode rsp (temporarily saved above in old_rsp),
+	 * rflags (%r11), rip (%rcx) and segments (fixed values) on
+	 * the stack as a regular interrupt frame.
+	 */
+	pushq_cfi $__USER_DS
+	/* CFI_REL_OFFSET ss, 0 */
+	pushq_cfi PER_CPU_VAR(old_rsp)
+	CFI_REL_OFFSET rsp, 0
+	pushq_cfi %r11 /* %r11 clobbered (userspace %rflags) */
+	/* CFI_REL_OFFSET rflags, 0 */
+	pushq_cfi $__USER_CS
+	/* CFI_REL_OFFSET cs, 0 */
+	pushq_cfi %rcx /* %rcx clobbered (userspace %rip) */
+	CFI_REL_OFFSET rip, 0
+
+	SAVE_ARGS 8, rax_enosys=1
 	movq_cfi rax,(ORIG_RAX-ARGOFFSET)
-	movq  %rcx,RIP-ARGOFFSET(%rsp)
-	CFI_REL_OFFSET rip,RIP-ARGOFFSET
 	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
 	jnz tracesys
 system_call_fastpath:
@@ -362,7 +346,7 @@ system_call_fastpath:
  */
 ret_from_sys_call:
 	testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,ARGOFFSET)
-	jnz int_ret_from_sys_call_fixup	/* Go the the slow path */
+	jnz int_ret_from_sys_call	/* Go the the slow path */
 
 	LOCKDEP_SYS_EXIT
 	DISABLE_INTERRUPTS(CLBR_NONE)
@@ -372,19 +356,16 @@ ret_from_sys_call:
 	 * sysretq will re-enable interrupts:
 	 */
 	TRACE_IRQS_ON
+	RESTORE_ARGS addskip=-ARG_SKIP, rstor_rcx=0, rstor_r11=0
 	movq RIP-ARGOFFSET(%rsp),%rcx
 	CFI_REGISTER	rip,rcx
-	RESTORE_ARGS 1,-ARG_SKIP,0
+	mov EFLAGS-ARGOFFSET(%rsp), %r11
 	/*CFI_REGISTER	rflags,r11*/
-	movq	PER_CPU_VAR(old_rsp), %rsp
+	mov RSP-ARGOFFSET(%rsp), %rsp
 	USERGS_SYSRET64
 
 	CFI_RESTORE_STATE
 
-int_ret_from_sys_call_fixup:
-	FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
-	jmp int_ret_from_sys_call
-
 	/* Do syscall tracing */
 tracesys:
 	leaq -REST_SKIP(%rsp), %rdi
@@ -397,7 +378,6 @@ tracesys:
 
 tracesys_phase2:
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %rdi
 	movq %rsp, %rdi
 	movq $AUDIT_ARCH_X86_64, %rsi
 	movq %rax,%rdx
@@ -493,10 +473,8 @@ ENTRY(stub_\func)
 	PARTIAL_FRAME 0
 	SAVE_REST
 	pushq	%r11			/* put it back on stack */
-	FIXUP_TOP_OF_STACK %r11, 8
 	DEFAULT_FRAME 0 8		/* offset 8: return address */
 	call sys_\func
-	RESTORE_TOP_OF_STACK %r11, 8
 	ret $REST_SKIP		/* pop extended registers */
 	CFI_ENDPROC
 END(stub_\func)
@@ -506,9 +484,7 @@ END(stub_\func)
 ENTRY(\label)
 	CFI_STARTPROC
 	PARTIAL_FRAME 0 8		/* offset 8: return address */
-	FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET
 	call \func
-	RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET
 	ret
 	CFI_ENDPROC
 END(\label)
@@ -524,7 +500,6 @@ ENTRY(stub_execve)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call sys_execve
 	movq %rax,RAX(%rsp)
 	RESTORE_REST
@@ -537,9 +512,7 @@ ENTRY(stub_execveat)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call sys_execveat
-	RESTORE_TOP_OF_STACK %r11
 	movq %rax,RAX(%rsp)
 	RESTORE_REST
 	jmp int_ret_from_sys_call
@@ -555,7 +528,6 @@ ENTRY(stub_rt_sigreturn)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call sys_rt_sigreturn
 	movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 	RESTORE_REST
@@ -569,7 +541,6 @@ ENTRY(stub_x32_rt_sigreturn)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call sys32_x32_rt_sigreturn
 	movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 	RESTORE_REST
@@ -582,9 +553,7 @@ ENTRY(stub_x32_execve)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call compat_sys_execve
-	RESTORE_TOP_OF_STACK %r11
 	movq %rax,RAX(%rsp)
 	RESTORE_REST
 	jmp int_ret_from_sys_call
@@ -596,9 +565,7 @@ ENTRY(stub_x32_execveat)
 	addq $8, %rsp
 	PARTIAL_FRAME 0
 	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
 	call compat_sys_execveat
-	RESTORE_TOP_OF_STACK %r11
 	movq %rax,RAX(%rsp)
 	RESTORE_REST
 	jmp int_ret_from_sys_call
-- 
2.1.0

Re: [PATCHv2 4/4] x86_64, entry: Create IRET-compatible stack frame at syscall entry

[Andy Lutomirski]

On Sun, Jan 18, 2015 at 3:45 AM, Alexander van Heukelum
<heukelum@fastmail.fm> wrote:
> Create an IRET-compatible top of stack at syscall entry and use this
> information to return to user mode in the sysret path. This removes
> the need for the FIXUP_TOP_OF_STACK and RESTORE_TOP_OF_STACK macros.

Since I have limited bandwidth, I'd like to tackle these one at a time.

I like the idea of this patch, but it has some issues.

First, it needs to be benchmarked.  The syscall fast path entry code
is *very* hot in some workloads, and it needs to be fast.

Second, I think you're really making three changes here.

a) You're putting rsp where it belongs -- it's in pt_regs instead of
being magically shoved into a combination of per-cpu variables and
extra arch state (thread->usersp).  This ideally consists of (AFAICS)
two tiny asm changes: one extra mov (most likely cache-hot) on entry
and a change of where you're reading from when you reload rsp on exit.
The former change could easily add a cycle (or zero cycles, or lots of
cycles -- hardware can be complicated, and I have no idea how well
store forwarding works on gs-relative accesses).  The latter change is
probably a speedup -- we'd be reading from pt_regs (almost certainly
hot or at least easily detected by the hardware prefetcher) instead of
a random percpu variable on exit.

*However*, this change enables the removal of all the usersp crap when
context switching, and all of the old_rsp references need to be
audited, and (having added yet another of them a week or two ago) I
know that you missed at least one and probably three or four :)  Also,
removing the usersp crap could easily speed up context switches by a
cache line or so.

Can you do that and split out just the old_rsp, usersp, and rsp part
as its own patch?

b) You're putting the saved flags into the EFLAGS pt_regs slot, which
seems to me to be an unambiguous win -- it removes two instructions
from RESTORE_TOP_OF_STACK, and it adds nothing whatsoever (except to
the extent that you continue to initialize R11 on entry instead of in
FIXUP_TOP_OF_STACK).

(a) and (b) alone should be enough to eliminate RESTORE_TOP_OF_STACK.

c) You're initializing the rest of the "top of stack" (cs, ss, and
rcx) unconditionally.  This is simpler, but I'm not sure it's
worthwhile -- we still lazily save the caller-saved regs, and
FIXUP_TOP_OF_STACK fits right in.  It also may have a performance
impact.

I think that (a) and (b) are clear wins (a is a really nice cleanup
and I bet it's a speedup, too, and b seems to be better in all
respects).  (c) is much less clearly a win to me.

Would you be willing to send split-out patches along with benchmarks?

--Andy

Re: [PATCHv2 4/4] x86_64, entry: Create IRET-compatible stack frame at syscall entry

[Alexander van Heukelum]

On Sun, Jan 18, 2015, at 17:38, Andy Lutomirski wrote:
> On Sun, Jan 18, 2015 at 3:45 AM, Alexander van Heukelum
> <heukelum@fastmail.fm> wrote:
> > Create an IRET-compatible top of stack at syscall entry and use this
> > information to return to user mode in the sysret path. This removes
> > the need for the FIXUP_TOP_OF_STACK and RESTORE_TOP_OF_STACK macros.
> 
> Since I have limited bandwidth, I'd like to tackle these one at a time.
> 
> I like the idea of this patch, but it has some issues.
> 
> First, it needs to be benchmarked.  The syscall fast path entry code
> is *very* hot in some workloads, and it needs to be fast.

Yeah. I didn't even try to benchmark it. What do you propose to use in this case?

> Second, I think you're really making three changes here.
> 
> a) You're putting rsp where it belongs -- it's in pt_regs instead of
> being magically shoved into a combination of per-cpu variables and
> extra arch state (thread->usersp).  This ideally consists of (AFAICS)
> two tiny asm changes: one extra mov (most likely cache-hot) on entry
> and a change of where you're reading from when you reload rsp on exit.
> The former change could easily add a cycle (or zero cycles, or lots of
> cycles -- hardware can be complicated, and I have no idea how well
> store forwarding works on gs-relative accesses).  The latter change is
> probably a speedup -- we'd be reading from pt_regs (almost certainly
> hot or at least easily detected by the hardware prefetcher) instead of
> a random percpu variable on exit.
> 
> *However*, this change enables the removal of all the usersp crap when
> context switching, and all of the old_rsp references need to be
> audited, and (having added yet another of them a week or two ago) I
> know that you missed at least one and probably three or four :)  Also,
> removing the usersp crap could easily speed up context switches by a
> cache line or so.

Yes. I missed that part. I'll  look into it. But nothing seems to blow up :)

> Can you do that and split out just the old_rsp, usersp, and rsp part
> as its own patch?
> 
> b) You're putting the saved flags into the EFLAGS pt_regs slot, which
> seems to me to be an unambiguous win -- it removes two instructions
> from RESTORE_TOP_OF_STACK, and it adds nothing whatsoever (except to
> the extent that you continue to initialize R11 on entry instead of in
> FIXUP_TOP_OF_STACK).
> 
> (a) and (b) alone should be enough to eliminate RESTORE_TOP_OF_STACK.
> 
> c) You're initializing the rest of the "top of stack" (cs, ss, and
> rcx) unconditionally.  This is simpler, but I'm not sure it's
> worthwhile -- we still lazily save the caller-saved regs, and
> FIXUP_TOP_OF_STACK fits right in.  It also may have a performance
> impact.
> 
> I think that (a) and (b) are clear wins (a is a really nice cleanup
> and I bet it's a speedup, too, and b seems to be better in all
> respects).  (c) is much less clearly a win to me.
> 
> Would you be willing to send split-out patches along with benchmarks?

Timing will depend highly on amounts of spare time, but I will give it
a shot.

Thanks for your valuable input!

Greetings,
    Alexander 

> --Andy

Re: [PATCHv2 0/4] x86, entry: some cleanup and simplification...

[Borislav Petkov]

Hi,

On Sun, Jan 18, 2015 at 12:45:16PM +0100, Alexander van Heukelum wrote:
> Hi Andy,
> 
> The last patchset did not compile on i386. Please ignore it. This one
> should be better. Instead of removing KERNEL_STACK_OFFSET, it is now
> used consistently on both i386 and x86_64.
> 
> Boot tested using qemu (using klibc for userspace)
> 	- x86_64, 32-bit userspace, core2duo (sysenter32)
> 	- x86_64, 32-bit userspace, phenom (syscall32)
> 	- x86_64, 32-bit userspace, vdso=0 (int 0x80)
> 	- x86_64, 64-bit userspace
> 	- i386, pentium3 (sysenter)
> 	- i386, athlon (syscall)
> 	- i386, vdso=0 (int 0x80)
> 
> They were tested on top of 22f2aa4a0361707a5cfb1de9d45260b39965dead
> (x86/entry-devel in your tree) and this kernel is now running on my
> laptop.

btw, you might wanna sync with Denys who's doing cleanups in that area too:

https://lkml.kernel.org/r/1421272101-16847-1-git-send-email-dvlasenk@redhat.com

and touching some of the stuff you're changing too.

Thanks.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCHv2 0/4] x86, entry: some cleanup and simplification...

[Alexander van Heukelum]

On Sun, Jan 18, 2015, at 13:05, Borislav Petkov wrote:
> Hi,
> 
> btw, you might wanna sync with Denys who's doing cleanups in that area too:
> 
> https://lkml.kernel.org/r/1421272101-16847-1-git-send-email-dvlasenk@redhat.com
> 
> and touching some of the stuff you're changing too.

Thanks. FWIW, the change touches the same area in code, but is orthogonal in concept.

Greetings,
    Alexander 

> Thanks.
> 
> -- 
> Regards/Gruss,
>     Boris.
> 
> ECO tip #101: Trim your mails when you reply.
> --

Re: [PATCHv2 0/4] x86, entry: some cleanup and simplification...

[Denys Vlasenko]

On Sun, Jan 18, 2015 at 4:47 PM, Alexander van Heukelum
<heukelum@fastmail.fm> wrote:
> On Sun, Jan 18, 2015, at 13:05, Borislav Petkov wrote:
>> Hi,
>>
>> btw, you might wanna sync with Denys who's doing cleanups in that area too:
>>
>> https://lkml.kernel.org/r/1421272101-16847-1-git-send-email-dvlasenk@redhat.com
>>
>> and touching some of the stuff you're changing too.
>
> Thanks. FWIW, the change touches the same area in code, but is orthogonal in concept.

My patches eventually eliminate both ARGOFFSET and KERNEL_STACK_OFFSET.
For example, eventually TI_flags tests look like this after all my patches:

testl $const, TI_flags+THREAD_INFO(%rsp,SIZEOF_PTREGS)

Your patches massage the same defines in a different way than mine do.
We clearly have a conflict here.

Re: [PATCHv2 0/4] x86, entry: some cleanup and simplification...

[Alexander van Heukelum]

On Wed, Jan 21, 2015, at 14:26, Denys Vlasenko wrote:
> On Sun, Jan 18, 2015 at 4:47 PM, Alexander van Heukelum
> <heukelum@fastmail.fm> wrote:
> > On Sun, Jan 18, 2015, at 13:05, Borislav Petkov wrote:
> >> Hi,
> >>
> >> btw, you might wanna sync with Denys who's doing cleanups in that area too:
> >>
> >> https://lkml.kernel.org/r/1421272101-16847-1-git-send-email-dvlasenk@redhat.com
> >>
> >> and touching some of the stuff you're changing too.
> >
> > Thanks. FWIW, the change touches the same area in code, but is orthogonal in concept.
> 
> My patches eventually eliminate both ARGOFFSET and KERNEL_STACK_OFFSET.
> For example, eventually TI_flags tests look like this after all my patches:
> 
> testl $const, TI_flags+THREAD_INFO(%rsp,SIZEOF_PTREGS)
> 
> Your patches massage the same defines in a different way than mine do.
> We clearly have a conflict here.

Indeed. I missed that, because Borislav's link only pointed to a single
change and I failed to notice the "[01/11]" in the title.