Skip to contentSkip to conversation

Enable LKGS instruction

PATCH Superseded Xin Li 4 participants 34 messages Oct 6, 2022 – Oct 14, 2022

Subsystems (3)

v1 of 6 Superseded Latest is v6 →

lore
Share & tools
lore

Xin Li

author
#0 lore raw

cover Enable LKGS instruction

LKGS instruction is introduced with Intel FRED (flexible return and event delivery) specification https://cdrdv2.intel.com/v1/dl/getContent/678938.

LKGS is independent of FRED, so we enable it as a standalone CPU feature.

LKGS behaves like the MOV to GS instruction except that it loads the base address into the IA32_KERNEL_GS_BASE MSR instead of the GS segment’s descriptor cache, which is exactly what Linux kernel does to load user level GS base. Thus, with LKGS, there is no need to SWAPGS away from the kernel GS base.

H. Peter Anvin (Intel) (6):
x86/cpufeature: add cpu feature bit for LKGS
x86/opcode: add LKGS instruction to x86-opcode-map x86/gsseg: make asm_load_gs_index() take an u16 x86/gsseg: move local_irq_save/restore() into asm_load_gs_index() x86/gsseg: move load_gs_index() to its own header file x86/gsseg: use the LKGS instruction if available for load_gs_index()

 arch/x86/entry/entry_64.S                | 28 +++++++++---
 arch/x86/ia32/ia32_signal.c              |  1 +
 arch/x86/include/asm/cpufeatures.h       |  1 +
 arch/x86/include/asm/gsseg.h             | 58 ++++++++++++++++++++++++
 arch/x86/include/asm/mmu_context.h       |  1 +
 arch/x86/include/asm/special_insns.h     | 21 ---------
 arch/x86/kernel/paravirt.c               |  1 +
 arch/x86/kernel/tls.c                    |  1 +
 arch/x86/lib/x86-opcode-map.txt          |  1 +
 tools/arch/x86/include/asm/cpufeatures.h |  1 +
 tools/arch/x86/lib/x86-opcode-map.txt    |  1 +
 11 files changed, 88 insertions(+), 27 deletions(-)
 create mode 100644 arch/x86/include/asm/gsseg.h
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 1/6 x86/cpufeature: add cpu feature bit for LKGS

From: "H. Peter Anvin (Intel)" <[email protected]>

Add the CPU feature bit for LKGS (Load "Kernel" GS).

LKGS instruction is introduced with Intel FRED (flexible return and event delivery) specificaton
https://cdrdv2.intel.com/v1/dl/getContent/678938.

LKGS behaves like the MOV to GS instruction except that it loads the base address into the IA32_KERNEL_GS_BASE MSR instead of the GS segment’s descriptor cache, which is exactly what Linux kernel does to load a user level GS base. Thus, with LKGS, there is no need to SWAPGS away from the kernel GS base.

---
 arch/x86/include/asm/cpufeatures.h       | 1 +
 tools/arch/x86/include/asm/cpufeatures.h | 1 +
 2 files changed, 2 insertions(+)
arch/x86/include/asm/cpufeatures.h +1 -0
index ef4775c6db01..459fb0c21dd4 100644
@@ -308,6 +308,7 @@
 /* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */
 #define X86_FEATURE_AVX_VNNI		(12*32+ 4) /* AVX VNNI instructions */
 #define X86_FEATURE_AVX512_BF16		(12*32+ 5) /* AVX512 BFLOAT16 instructions */
+#define X86_FEATURE_LKGS		(12*32+ 18) /* Load "kernel" (userspace) gs */
 
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
tools/arch/x86/include/asm/cpufeatures.h +1 -0
index ef4775c6db01..459fb0c21dd4 100644
@@ -308,6 +308,7 @@
 /* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */
 #define X86_FEATURE_AVX_VNNI		(12*32+ 4) /* AVX VNNI instructions */
 #define X86_FEATURE_AVX512_BF16		(12*32+ 5) /* AVX512 BFLOAT16 instructions */
+#define X86_FEATURE_LKGS		(12*32+ 18) /* Load "kernel" (userspace) gs */
 
 /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */
 #define X86_FEATURE_CLZERO		(13*32+ 0) /* CLZERO instruction */
Signature · 1 line
2.34.1
author
#2 lore raw

PATCH 2/6 x86/opcode: add LKGS instruction to x86-opcode-map

In reply to Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

Add the instruction opcode used by LKGS.
Opcode number is per public FRED draft spec v3.0
https://cdrdv2.intel.com/v1/dl/getContent/678938.

---
 arch/x86/lib/x86-opcode-map.txt       | 1 +
 tools/arch/x86/lib/x86-opcode-map.txt | 1 +
 2 files changed, 2 insertions(+)
arch/x86/lib/x86-opcode-map.txt +1 -0
index d12d1358f96d..5168ee0360b2 100644
@@ -1047,6 +1047,7 @@ GrpTable: Grp6
 3: LTR Ew
 4: VERR Ew
 5: VERW Ew
+6: LKGS Ew (F2)
 EndTable
 
 GrpTable: Grp7
tools/arch/x86/lib/x86-opcode-map.txt +1 -0
index d12d1358f96d..5168ee0360b2 100644
@@ -1047,6 +1047,7 @@ GrpTable: Grp6
 3: LTR Ew
 4: VERR Ew
 5: VERW Ew
+6: LKGS Ew (F2)
 EndTable
 
 GrpTable: Grp7
Signature · 1 line
2.34.1
author
#3 lore raw

PATCH 3/6 x86/gsseg: make asm_load_gs_index() take an u16

In reply to Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

Let gcc know that only the low 16 bits of load_gs_index() argument actually matter. It might allow it to create slightly better code. However, do not propagate this into the prototypes of functions that end up being paravirtualized, to avoid unnecessary changes.

---
 arch/x86/entry/entry_64.S            | 2 +-
 arch/x86/include/asm/special_insns.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/x86/entry/entry_64.S +1 -1
index 9953d966d124..e0c48998d2fb 100644
@@ -779,7 +779,7 @@ _ASM_NOKPROBE(common_interrupt_return)
 
 /*
  * Reload gs selector with exception handling
- * edi:  new selector
+ *  di:  new selector
  *
  * Is in entry.text as it shouldn't be instrumented.
  */
arch/x86/include/asm/special_insns.h +1 -1
index 35f709f619fb..a71d0e8d4684 100644
@@ -120,7 +120,7 @@ static inline void native_wbinvd(void)
 	asm volatile("wbinvd": : :"memory");
 }
 
-extern asmlinkage void asm_load_gs_index(unsigned int selector);
+extern asmlinkage void asm_load_gs_index(u16 selector);
 
 static inline void native_load_gs_index(unsigned int selector)
 {
Signature · 1 line
2.34.1
author
#4 lore raw

PATCH 4/6 x86/gsseg: move local_irq_save/restore() into asm_load_gs_index()

In reply to Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

The need to disable/enable interrupts around asm_load_gs_index() is a consequence of the implementation. Prepare for using the LKGS instruction, which is locally atomic and therefore doesn't need interrupts disabled.

---
 arch/x86/entry/entry_64.S            | 26 +++++++++++++++++++++-----
 arch/x86/include/asm/special_insns.h |  4 ----
 2 files changed, 21 insertions(+), 9 deletions(-)
arch/x86/entry/entry_64.S +21 -5
index e0c48998d2fb..cc6ba6672418 100644
@@ -778,19 +778,35 @@ SYM_CODE_END(common_interrupt_return)
 _ASM_NOKPROBE(common_interrupt_return)
 
 /*
- * Reload gs selector with exception handling
+ * Reload gs selector with exception handling. This is used only on
+ * native, so using swapgs, pushf, popf, cli, sti, ... directly is fine.
+ *
  *  di:  new selector
+ * rax:  scratch register
  *
  * Is in entry.text as it shouldn't be instrumented.
+ *
+ * Note: popf is slow, so use pushf to read IF and then execute cli/sti
+ * if necessary.
  */
 SYM_FUNC_START(asm_load_gs_index)
 	FRAME_BEGIN
+	pushf
+	pop	%rax
+	andl	$X86_EFLAGS_IF, %eax	/* Interrupts enabled? */
+	jz	1f
+	cli
+1:
 	swapgs
 .Lgs_change:
 	ANNOTATE_NOENDBR // error_entry
 	movl	%edi, %gs
 2:	ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
 	swapgs
+	testl	%eax, %eax
+	jz	3f
+	sti
+3:
 	FRAME_END
 	RET
 
@@ -799,12 +815,12 @@ SYM_FUNC_START(asm_load_gs_index)
 	swapgs					/* switch back to user gs */
 .macro ZAP_GS
 	/* This can't be a string because the preprocessor needs to see it. */
-	movl $__USER_DS, %eax
-	movl %eax, %gs
+	movl $__USER_DS, %edi
+	movl %edi, %gs
 .endm
 	ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
-	xorl	%eax, %eax
-	movl	%eax, %gs
+	xorl	%edi, %edi
+	movl	%edi, %gs
 	jmp	2b
 
 	_ASM_EXTABLE(.Lgs_change, .Lbad_gs)
arch/x86/include/asm/special_insns.h +0 -4
index a71d0e8d4684..6de00dec6564 100644
@@ -124,11 +124,7 @@ extern asmlinkage void asm_load_gs_index(u16 selector);
 
 static inline void native_load_gs_index(unsigned int selector)
 {
-	unsigned long flags;
-
-	local_irq_save(flags);
 	asm_load_gs_index(selector);
-	local_irq_restore(flags);
 }
 
 static inline unsigned long __read_cr4(void)
Signature · 1 line
2.34.1
#5 lore raw

Re: [PATCH 4/6] x86/gsseg: move local_irq_save/restore() into asm_load_gs_index()

arch/x86/entry/entry_64.S SYM_CODE_END(common_interrupt_return) · line 799

Quoted 8 lines · Xin Li

SYM_FUNC_START(asm_load_gs_index)

arch/x86/entry/entry_64.S +6 -0
 	FRAME_BEGIN
+	pushf
+	pop	%rax
+	andl	$X86_EFLAGS_IF, %eax	/* Interrupts enabled? */
+	jz	1f
+	cli
+1:

Why the pop,andl,jz ? AFAICT our arch_local_irq_save() doesn't even bother with that, why does this function.

#6 lore raw

Re: [PATCH 4/6] x86/gsseg: move local_irq_save/restore() into asm_load_gs_index()

Quoted 12 lines · Peter Zijlstra

arch/x86/entry/entry_64.S SYM_CODE_END(common_interrupt_return) · line 799

Quoted 8 lines · Xin Li

SYM_FUNC_START(asm_load_gs_index)

arch/x86/entry/entry_64.S +6 -0
 	FRAME_BEGIN
+	pushf
+	pop	%rax
+	andl	$X86_EFLAGS_IF, %eax	/* Interrupts enabled? */
+	jz	1f
+	cli
+1:

Why the pop,andl,jz ? AFAICT our arch_local_irq_save() doesn't even bother with that, why does this function.

They pop and and are needed for the sti anyway, and so might as well be leveraged here, too.

author
#7 lore raw

PATCH 5/6 x86/gsseg: move load_gs_index() to its own header file

In reply to Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

<asm/cpufeature.h> depends on <asm/special_insns.h>, so in order to be able to use alternatives in native_load_gs_index(), factor it out into a separate header file.

---
 arch/x86/ia32/ia32_signal.c          |  1 +
 arch/x86/include/asm/gsseg.h         | 32 ++++++++++++++++++++++++++++
 arch/x86/include/asm/mmu_context.h   |  1 +
 arch/x86/include/asm/special_insns.h | 17 ---------------
 arch/x86/kernel/paravirt.c           |  1 +
 arch/x86/kernel/tls.c                |  1 +
 6 files changed, 36 insertions(+), 17 deletions(-)
 create mode 100644 arch/x86/include/asm/gsseg.h
arch/x86/ia32/ia32_signal.c +1 -0
index c9c3859322fa..14c739303099 100644
@@ -34,6 +34,7 @@
 #include <asm/sigframe.h>
 #include <asm/sighandling.h>
 #include <asm/smap.h>
+#include <asm/gsseg.h>
 
 static inline void reload_segments(struct sigcontext_32 *sc)
 {
arch/x86/include/asm/gsseg.h new file +32 -0
new file mode 100644
index 000000000000..5e3b56a17098
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+#ifndef _ASM_X86_GSSEG_H
+#define _ASM_X86_GSSEG_H
+
+#include <linux/types.h>
+#include <asm/processor.h>
+
+#ifdef CONFIG_X86_64
+
+extern asmlinkage void asm_load_gs_index(u16 selector);
+
+static inline void native_load_gs_index(unsigned int selector)
+{
+	asm_load_gs_index(selector);
+}
+
+#endif /* CONFIG_X86_64 */
+
+#ifndef CONFIG_PARAVIRT_XXL
+
+static inline void load_gs_index(unsigned int selector)
+{
+#ifdef CONFIG_X86_64
+	native_load_gs_index(selector);
+#else
+	loadsegment(gs, selector);
+#endif
+}
+
+#endif /* CONFIG_PARAVIRT_XXL */
+
+#endif /* _ASM_X86_GSSEG_H */
arch/x86/include/asm/mmu_context.h +1 -0
index b8d40ddeab00..e01aa74a6de7 100644
@@ -12,6 +12,7 @@
 #include <asm/tlbflush.h>
 #include <asm/paravirt.h>
 #include <asm/debugreg.h>
+#include <asm/gsseg.h>
 
 extern atomic64_t last_mm_ctx_id;
 
arch/x86/include/asm/special_insns.h +0 -17
index 6de00dec6564..cfd9499b617c 100644
@@ -120,13 +120,6 @@ static inline void native_wbinvd(void)
 	asm volatile("wbinvd": : :"memory");
 }
 
-extern asmlinkage void asm_load_gs_index(u16 selector);
-
-static inline void native_load_gs_index(unsigned int selector)
-{
-	asm_load_gs_index(selector);
-}
-
 static inline unsigned long __read_cr4(void)
 {
 	return native_read_cr4();
@@ -180,16 +173,6 @@ static inline void wbinvd(void)
 	native_wbinvd();
 }
 
-
-static inline void load_gs_index(unsigned int selector)
-{
-#ifdef CONFIG_X86_64
-	native_load_gs_index(selector);
-#else
-	loadsegment(gs, selector);
-#endif
-}
-
 #endif /* CONFIG_PARAVIRT_XXL */
 
 static inline void clflush(volatile void *__p)
arch/x86/kernel/paravirt.c +1 -0
index 7ca2d46c08cc..00f6a92551d2 100644
@@ -32,6 +32,7 @@
 #include <asm/special_insns.h>
 #include <asm/tlb.h>
 #include <asm/io_bitmap.h>
+#include <asm/gsseg.h>
 
 /*
  * nop stub, which must not clobber anything *including the stack* to
arch/x86/kernel/tls.c +1 -0
index 3c883e064242..3ffbab0081f4 100644
@@ -12,6 +12,7 @@
 #include <asm/ldt.h>
 #include <asm/processor.h>
 #include <asm/proto.h>
+#include <asm/gsseg.h>
 
 #include "tls.h"
 
Signature · 1 line
2.34.1
#8 lore raw

Re: [PATCH 5/6] x86/gsseg: move load_gs_index() to its own header file

Quoted 9 lines · Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

<asm/cpufeature.h> depends on <asm/special_insns.h>, so in order to be able to use alternatives in native_load_gs_index(), factor it out into a separate header file.

This could be moved into <asm/segment.h> instead of creating a new header file.

--
Brian Gerst

#9 lore raw

Re: [PATCH 5/6] x86/gsseg: move load_gs_index() to its own header file

Quoted 8 lines · Brian Gerst
Quoted 7 lines · Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

<asm/cpufeature.h> depends on <asm/special_insns.h>, so in order to be able to use alternatives in native_load_gs_index(), factor it out into a separate header file.

Quoted 7 lines
Quoted 2 lines

This could be moved into <asm/segment.h> instead of creating a new header file.

--
Brian Gerst

At least at the time I wrote the code, it could not, without creating yet another circular header file dependency.

author
#10 lore raw

RE: [PATCH 5/6] x86/gsseg: move load_gs_index() to its own header file

Quoted 9 lines · Brian Gerst
Quoted 8 lines · Xin Li
Quoted 7 lines

From: "H. Peter Anvin (Intel)" <[email protected]>

<asm/cpufeature.h> depends on <asm/special_insns.h>, so in order to be able to use alternatives in native_load_gs_index(), factor it out into a separate header file.

Quoted 5 lines
Quoted 4 lines
Quoted 2 lines

This could be moved into <asm/segment.h> instead of creating a new header

file.

Good suggestion. However I still prefer to keep GS segment in its own header file,
1) it's a special segment for x86_64 kernel, and it's more readable to keep it separated.
2) segment.h is included in too many files, but gsseg.h only 4 files. We avoid header pollution.

Maybe we should even factor GS segment out of segment.h for x86_64.

Xin

Quoted 6 lines
Quoted 3 lines

--
Brian Gerst

At least at the time I wrote the code, it could not, without creating yet another circular header file dependency.

author
#11 lore raw

PATCH 6/6 x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

The LKGS instruction atomically loads a segment descriptor into the %gs descriptor registers, *except* that %gs.base is unchanged, and the base is instead loaded into MSR_IA32_KERNEL_GS_BASE, which is exactly what we want this function to do.

---
 arch/x86/include/asm/gsseg.h | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)
arch/x86/include/asm/gsseg.h +27 -1
index 5e3b56a17098..b8a6a98d88b8 100644
@@ -3,15 +3,41 @@
 #define _ASM_X86_GSSEG_H
 
 #include <linux/types.h>
+
+#include <asm/asm.h>
+#include <asm/cpufeature.h>
+#include <asm/alternative.h>
 #include <asm/processor.h>
+#include <asm/nops.h>
 
 #ifdef CONFIG_X86_64
 
 extern asmlinkage void asm_load_gs_index(u16 selector);
 
+#define LKGS_DI	_ASM_BYTES(0xf2,0x0f,0x00,0xf7)
+
 static inline void native_load_gs_index(unsigned int selector)
 {
-	asm_load_gs_index(selector);
+	u16 sel = selector;
+
+	/*
+	 * Note: the fixup is used for the LKGS instruction, but
+	 * it needs to be attached to the primary instruction sequence
+	 * as it isn't something that gets patched.
+	 *
+	 * %rax is provided to the assembly routine as a scratch
+	 * register.
+	 */
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }
 
 #endif /* CONFIG_X86_64 */
Signature · 1 line
2.34.1
#12 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 13 lines · Xin Li

static inline void native_load_gs_index(unsigned int selector) {

arch/x86/include/asm/gsseg.h +11 -0
+	u16 sel = selector;
+
+	/*
+	 * Note: the fixup is used for the LKGS instruction, but
+	 * it needs to be attached to the primary instruction sequence
+	 * as it isn't something that gets patched.
+	 *
+	 * %rax is provided to the assembly routine as a scratch
+	 * register.
+	 */
+	alternative_io("1: call asm_load_gs_index\n"

arch/x86/include/asm/gsseg.h line 41

Quoted 10 lines
arch/x86/include/asm/gsseg.h +9 -0
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

e5eefda5aa51 ("x86: Remove .fixup section")

#13 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 29 lines · Peter Zijlstra
Quoted 13 lines · Xin Li

static inline void native_load_gs_index(unsigned int selector) {

arch/x86/include/asm/gsseg.h +11 -0
+	u16 sel = selector;
+
+	/*
+	 * Note: the fixup is used for the LKGS instruction, but
+	 * it needs to be attached to the primary instruction sequence
+	 * as it isn't something that gets patched.
+	 *
+	 * %rax is provided to the assembly routine as a scratch
+	 * register.
+	 */
+	alternative_io("1: call asm_load_gs_index\n"

arch/x86/include/asm/gsseg.h line 41

Quoted 10 lines
arch/x86/include/asm/gsseg.h +9 -0
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 1 line

e5eefda5aa51 ("x86: Remove .fixup section")

Xin, what did you use as the forward-porting baseline?

author
#14 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 15 lines
Quoted 15 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 11 lines
arch/x86/include/asm/gsseg.h +10 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 3 lines
Quoted 1 line

e5eefda5aa51 ("x86: Remove .fixup section")

Xin, what did you use as the forward-porting baseline?

6.0 release, and my kernel dump shows me a fixup section is there, and a fixup section is created anyway if we do "pushsection "\.fixup\"".

#15 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 15 lines · On October 7, 2022 11:07:34 AM PDT, "Li, Xin3" <[email protected]> wrote:
Quoted 15 lines
Quoted 15 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 11 lines
arch/x86/include/asm/gsseg.h +10 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

e5eefda5aa51 ("x86: Remove .fixup section")

Xin, what did you use as the forward-porting baseline?

6.0 release, and my kernel dump shows me a fixup section is there, and a fixup section is created anyway if we do "pushsection "\.fixup\"".

Yeah. .fixup is really Just Another Text Section ™, so it is probably not surprising if it has crept back in.

author
#16 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Peter Zijlstra
Quoted 6 lines
Quoted 6 lines
arch/x86/include/asm/gsseg.h +6 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
Quoted 8 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 5 lines
arch/x86/include/asm/gsseg.h +4 -0
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Weird, did you ever check a kernel dump?

Quoted 2 lines

e5eefda5aa51 ("x86: Remove .fixup section")

#17 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 15 lines · On Fri, Oct 07, 2022 at 06:01:06PM +0000, Li, Xin3 wrote:
Quoted 6 lines
Quoted 6 lines
arch/x86/include/asm/gsseg.h +6 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
Quoted 8 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 5 lines
arch/x86/include/asm/gsseg.h +4 -0
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 1 line

Weird, did you ever check a kernel dump?

$ readelf -WS defconfig-build/vmlinux | grep fixup
[ 5] .pci_fixup        PROGBITS        ffffffff826a5350 18a5350 003570 00   A  0   0 16
[ 6] .rela.pci_fixup   RELA            0000000000000000 360c388 005028 18   I 60   5  8

In fact, when I add one I get:

ld: warning: orphan section .fixup' from arch/x86/kernel/traps.o' being placed in section `.fixup'

#18 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 18 lines · Peter Zijlstra
Quoted 15 lines · On Fri, Oct 07, 2022 at 06:01:06PM +0000, Li, Xin3 wrote:
Quoted 6 lines
Quoted 6 lines
arch/x86/include/asm/gsseg.h +6 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
Quoted 8 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 5 lines
arch/x86/include/asm/gsseg.h +4 -0
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 1 line

Weird, did you ever check a kernel dump?

Quoted 7 lines
$ readelf -WS defconfig-build/vmlinux | grep fixup
[ 5] .pci_fixup        PROGBITS        ffffffff826a5350 18a5350 003570 00   A  0   0 16
[ 6] .rela.pci_fixup   RELA            0000000000000000 360c388 005028 18   I 60   5  8

In fact, when I add one I get:

ld: warning: orphan section .fixup' from arch/x86/kernel/traps.o' being placed in section `.fixup'

Perhaps the two of you need to compare confugurations?

#19 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 23 lines · H. Peter Anvin
Quoted 22 lines · Peter Zijlstra
Quoted 15 lines · On Fri, Oct 07, 2022 at 06:01:06PM +0000, Li, Xin3 wrote:
Quoted 6 lines
Quoted 6 lines
arch/x86/include/asm/gsseg.h +6 -0
+	alternative_io("1: call asm_load_gs_index\n"
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
Quoted 8 lines

arch/x86/include/asm/gsseg.h line 41

Quoted 5 lines
arch/x86/include/asm/gsseg.h +4 -0
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

I'm very sure none of this was tested... the .fixup section hasn't existed for almost a year now.

Quoted 1 line

Weird, did you ever check a kernel dump?

$ readelf -WS defconfig-build/vmlinux | grep fixup
[ 5] .pci_fixup        PROGBITS        ffffffff826a5350 18a5350 003570 00   A  0   0 16
[ 6] .rela.pci_fixup   RELA            0000000000000000 360c388 005028 18   I 60   5  8
Quoted 5 lines
Quoted 3 lines

In fact, when I add one I get:

ld: warning: orphan section .fixup' from arch/x86/kernel/traps.o' being placed in section `.fixup'

Perhaps the two of you need to compare confugurations?

Whatever for? I know the robots report this warning because there was one from the KVM cross-merge when the .fixup removal went in. It got reported and fixed and that was the last of it.

Anyway; try:

  $ git grep "\.fixup" arch/x86/

there isn't a single usage.

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

#20 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 7 lines · Peter Zijlstra
Quoted 2 lines

Perhaps the two of you need to compare confugurations?

Whatever for? I know the robots report this warning because there was one from the KVM cross-merge when the .fixup removal went in. It got reported and fixed and that was the last of it.

To help track down what appears to be a problem?

Quoted 9 lines

Anyway; try:

   $ git grep "\.fixup" arch/x86/

there isn't a single usage.

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

Agreed 1000%. This is a no-brainer. From IRC:

<andyhhp> -LDFLAGS_vmlinux += --orphan-handling=warn <andyhhp> +LDFLAGS_vmlinux += --orphan-handling=error

-hpa

author
#21 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 9 lines
Quoted 3 lines

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

Agreed 1000%. This is a no-brainer. From IRC:

<andyhhp> -LDFLAGS_vmlinux += --orphan-handling=warn <andyhhp> +LDFLAGS_vmlinux += --orphan-handling=error

Who is going to make the change?

#22 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 4 lines · On October 7, 2022 10:32:31 PM PDT, "Li, Xin3" <[email protected]> wrote:
Quoted 4 lines
Quoted 3 lines

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

Quoted 8 lines
Quoted 5 lines

Agreed 1000%. This is a no-brainer. From IRC:

<andyhhp> -LDFLAGS_vmlinux += --orphan-handling=warn <andyhhp> +LDFLAGS_vmlinux += --orphan-handling=error

Who is going to make the change?

Why don't you?

author
#23 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 8 lines · On October 7, 2022 10:32:31 PM PDT, "Li, Xin3" <[email protected]> wrote:
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

Agreed 1000%. This is a no-brainer. From IRC:

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

<andyhhp> -LDFLAGS_vmlinux += --orphan-handling=warn <andyhhp> +LDFLAGS_vmlinux += --orphan-handling=error

Who is going to make the change?

Why don't you?

Will do.
Xin

author
#24 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to H. Peter Anvin
Quoted 9 lines
Quoted 3 lines

Andrew Cooper suggested upgrading the orphan section warning to a hard link error, orphan sections are bad regardless.

Agreed 1000%. This is a no-brainer. From IRC:

<andyhhp> -LDFLAGS_vmlinux += --orphan-handling=warn <andyhhp> +LDFLAGS_vmlinux += --orphan-handling=error

There is an arch independent config CONFIG_LD_ORPHAN_WARN, which forces linker to warn on implicit named sections, or there is even no warning.

CONFIG_LD_ORPHAN_WARN depends on ARCH_WANT_LD_ORPHAN_WARN, and some archs (arm/arm64/mips/x86/...) have it defined, and then ld generates warnings on orphan sections.

Should we promote warning to error only on x86?

Quoted 2 lines

-hpa

#25 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Xin Li
Quoted 1 line · Xin Li

+#define LKGS_DI _ASM_BYTES(0xf2,0x0f,0x00,0xf7)

This should come with a comment that states the binutils version that
understands: "lkgs %di" (or however that ends being).

#26 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Xin Li
Quoted 13 lines · Xin Li

static inline void native_load_gs_index(unsigned int selector) {

arch/x86/include/asm/gsseg.h +11 -0
+	u16 sel = selector;
+
+	/*
+	 * Note: the fixup is used for the LKGS instruction, but
+	 * it needs to be attached to the primary instruction sequence
+	 * as it isn't something that gets patched.
+	 *
+	 * %rax is provided to the assembly routine as a scratch
+	 * register.
+	 */
+	alternative_io("1: call asm_load_gs_index\n"

arch/x86/include/asm/gsseg.h line 41

Quoted 10 lines
arch/x86/include/asm/gsseg.h +9 -0
+		       ".pushsection \".fixup\",\"ax\"\n"
+		       "2:	xorl %k[sel], %k[sel]\n"
+		       "	jmp 1b\n"
+		       ".popsection\n"
+		       _ASM_EXTABLE(1b, 2b),
+		       _ASM_BYTES(0x3e) LKGS_DI,
+		       X86_FEATURE_LKGS,
+		       ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+		       ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }

Something like so perhaps?

	asm_inline volatile("1:\n"
			    ALTERNATIVE("call asm_load_gs_index\n",
					LKGS_DI,
					X86_FEATURE_LKGS)
			    _ASM_EXTABLE_TYPE_REG(1b, 1b, EX_TYPE_ZERO_REG, %k[sel])
			    : ASM_CALL_CONSTRAINT
			    : [sel] "D" (sel)
			    : "memory", _ASM_AX);

(completely untested, not even seen a compiler upclose)

author
#27 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 13 lines

Something like so perhaps?

	asm_inline volatile("1:\n"
			    ALTERNATIVE("call asm_load_gs_index\n",
					LKGS_DI,
					X86_FEATURE_LKGS)
			    _ASM_EXTABLE_TYPE_REG(1b, 1b,
EX_TYPE_ZERO_REG, %k[sel])
			    : ASM_CALL_CONSTRAINT
			    : [sel] "D" (sel)
			    : "memory", _ASM_AX);
Quoted 1 line

(completely untested, not even seen a compiler upclose)

It compiles (after adding "0x3e" to make it 5 bytes to match the size of "call asm_load_gs_index") and passes my fixup test, thanks!

#28 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Xin Li
Quoted 62 lines · Xin Li

From: "H. Peter Anvin (Intel)" <[email protected]>

The LKGS instruction atomically loads a segment descriptor into the %gs descriptor registers, *except* that %gs.base is unchanged, and the base is instead loaded into MSR_IA32_KERNEL_GS_BASE, which is exactly what we want this function to do.

---
 arch/x86/include/asm/gsseg.h | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)
arch/x86/include/asm/gsseg.h +27 -2
index 5e3b56a17098..b8a6a98d88b8 100644
@@ -3,15 +3,41 @@
 #define _ASM_X86_GSSEG_H
 
 #include <linux/types.h>
+
+#include <asm/asm.h>
+#include <asm/cpufeature.h>
+#include <asm/alternative.h>
 #include <asm/processor.h>
+#include <asm/nops.h>
 
 #ifdef CONFIG_X86_64
 
 extern asmlinkage void asm_load_gs_index(u16 selector);
 
+#define LKGS_DI        _ASM_BYTES(0xf2,0x0f,0x00,0xf7)
+
 static inline void native_load_gs_index(unsigned int selector)
 {
-       asm_load_gs_index(selector);
+       u16 sel = selector;
+
+       /*
+        * Note: the fixup is used for the LKGS instruction, but
+        * it needs to be attached to the primary instruction sequence
+        * as it isn't something that gets patched.
+        *
+        * %rax is provided to the assembly routine as a scratch
+        * register.
+        */
+       alternative_io("1: call asm_load_gs_index\n"
+                      ".pushsection \".fixup\",\"ax\"\n"
+                      "2:      xorl %k[sel], %k[sel]\n"
+                      "        jmp 1b\n"
+                      ".popsection\n"
+                      _ASM_EXTABLE(1b, 2b),
+                      _ASM_BYTES(0x3e) LKGS_DI,
+                      X86_FEATURE_LKGS,
+                      ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+                      ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }
 
 #endif /* CONFIG_X86_64 */
--
Quoted 1 line

2.34.1

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

        if (static_cpu_has(X86_FEATURE_LKGS))
                asm volatile("1: " LKGS_DI
                             _ASM_EXTABLE_TYPE_REG(1b, 1b,
EX_TYPE_ZERO_REG, %k[sel])
                             : [sel] "+D" (sel) : : "memory");
        else
                asm_load_gs_index(sel);

--
Brian Gerst

author
#29 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 12 lines
Quoted 12 lines
arch/x86/include/asm/gsseg.h +10 -0
+       alternative_io("1: call asm_load_gs_index\n"
+                      ".pushsection \".fixup\",\"ax\"\n"
+                      "2:      xorl %k[sel], %k[sel]\n"
+                      "        jmp 1b\n"
+                      ".popsection\n"
+                      _ASM_EXTABLE(1b, 2b),
+                      _ASM_BYTES(0x3e) LKGS_DI,
+                      X86_FEATURE_LKGS,
+                      ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+                      ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }
Quoted 7 lines
Quoted 3 lines

#endif /* CONFIG_X86_64 */
--
2.34.1

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

The existing approach patches the binary code thus we don't need to check it at runtime.

Quoted 8 lines
        if (static_cpu_has(X86_FEATURE_LKGS))
                asm volatile("1: " LKGS_DI
                             _ASM_EXTABLE_TYPE_REG(1b, 1b, EX_TYPE_ZERO_REG,
%k[sel])
                             : [sel] "+D" (sel) : : "memory");
        else
                asm_load_gs_index(sel);
#30 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 17 lines · On Sat, Oct 8, 2022 at 1:40 AM Li, Xin3 <[email protected]> wrote:
Quoted 17 lines
Quoted 16 lines
arch/x86/include/asm/gsseg.h +10 -1
+       alternative_io("1: call asm_load_gs_index\n"
+                      ".pushsection \".fixup\",\"ax\"\n"
+                      "2:      xorl %k[sel], %k[sel]\n"
+                      "        jmp 1b\n"
+                      ".popsection\n"
+                      _ASM_EXTABLE(1b, 2b),
+                      _ASM_BYTES(0x3e) LKGS_DI,
+                      X86_FEATURE_LKGS,
+                      ASM_OUTPUT2([sel] "+D" (sel), ASM_CALL_CONSTRAINT),
+                      ASM_NO_INPUT_CLOBBER(_ASM_AX));
 }
 
 #endif /* CONFIG_X86_64 */
--

2.34.1

Quoted 5 lines
Quoted 3 lines

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

The existing approach patches the binary code thus we don't need to check it at runtime.

static_cpu_has() uses alternatives to patch the branch, so there is no runtime check after early boot.

--
Brian Gerst

author
#31 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 10 lines
Quoted 5 lines
Quoted 3 lines

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

The existing approach patches the binary code thus we don't need to check it

at runtime.

static_cpu_has() uses alternatives to patch the branch, so there is no runtime check after early boot.

Sorry, didn't know it, thanks for point it out.

If we prefer static_cpu_has, are you asking to replace all alternative macros with it?

Xin

#32 lore raw

RE: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

Quoted 11 lines · On October 9, 2022 9:32:34 PM PDT, "Li, Xin3" <[email protected]> wrote:
Quoted 10 lines
Quoted 5 lines
Quoted 3 lines

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

The existing approach patches the binary code thus we don't need to check it

at runtime.

static_cpu_has() uses alternatives to patch the branch, so there is no runtime check after early boot.

Quoted 7 lines

Sorry, didn't know it, thanks for point it out.

If we prefer static_cpu_has, are you asking to replace all alternative macros with it?

Xin

Honestly, it seems to me to be more than a bit excessive. The code might be nontrivial, but *with proper commenting* it should be perfectly understandable...

#33 lore raw

Re: [PATCH 6/6] x86/gsseg: use the LKGS instruction if available for load_gs_index()

In reply to Li, Xin3
Quoted 7 lines · On Mon, Oct 10, 2022 at 04:32:34AM +0000, Li, Xin3 wrote:
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

There are not that many call sites, so using something like this (incorporating Peter Z's suggestion for the exception handler) would be better from a code readability perspective vs. a tiny increase in code size.

The existing approach patches the binary code thus we don't need to check it

at runtime.

Quoted 7 lines
Quoted 3 lines

static_cpu_has() uses alternatives to patch the branch, so there is no runtime check after early boot.

Sorry, didn't know it, thanks for point it out.

If we prefer static_cpu_has, are you asking to replace all alternative macros with it?

No; the only reason to do it here would be to unconfuse that exception thing. But even there I'm not convinced.

Yes, Brian's code is much easier to read, but code-gen is quite terrible (also, my binutils can't seem to decode this -- GNU objdump (GNU Binutils for Debian) 2.38.90.20220713)