Skip to contentSkip to conversation

x86/xsave: Move xsave initialization to after parsing early parameters

PATCH In review Andi Kleen 3 participants 5 messages Mar 11, 2017 – Mar 12, 2017

Subsystems (2)
lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

PATCH 1/2 x86/xsave: Move xsave initialization to after parsing early parameters

From: Andi Kleen <[email protected]>

Move the XSAVE initialization code to be after parsing early parameters. I don't see any reason why the FPU code needs to be initialized that early, nothing else in the initialization phase uses XSAVE. This is useful to be able to handle command line parameters in the XSAVE initialization code.

---
 arch/x86/kernel/cpu/common.c | 1 -
 arch/x86/kernel/setup.c      | 3 +++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/common.c +0 -1
index 58094a1f9e9d..db16a28d49a1 100644
@@ -844,7 +844,6 @@ static void __init early_identify_cpu(struct cpuinfo_x86 *c)
 	}
 
 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
-	fpu__init_system(c);
 }
 
 void __init early_cpu_init(void)
arch/x86/kernel/setup.c +3 -0
index 4bf0c8926a1c..2a85efc37e1f 100644
@@ -90,6 +90,7 @@
 #include <asm/processor.h>
 #include <asm/bugs.h>
 #include <asm/kasan.h>
+#include <asm/fpu/internal.h>
 
 #include <asm/vsyscall.h>
 #include <asm/cpu.h>
@@ -987,6 +988,8 @@ void __init setup_arch(char **cmdline_p)
 
 	parse_early_param();
 
+	fpu__init_system(&boot_cpu_data);
+
 #ifdef CONFIG_MEMORY_HOTPLUG
 	/*
 	 * Memory used by the kernel cannot be hot-removed because Linux
Signature · 1 line
2.9.3
author
#1 lore raw

PATCH 2/2 x86/fpu: Support disabling AVX and AVX512

From: Andi Kleen <[email protected]>

For performance testing it is useful to be able to disable AVX and AVX512. User programs check in XGETBV if AVX is supported by the OS. If we don't initialize the XSAVE state for AVX it will appear as if the OS is not supporting AVX. For kernel users we can also clear the internal cpu feature bits.

This patch implements disable options for AVX and AVX512 for the XSAVE code.

I originally considered a generic argument that would disable any XSAVE feature, but it turns out you need special code to also disable all the CPUID bits, because otherwise kernel code may assume it exists, when it doesn't. MPX already has an own disable flag. Not clear it is useful for the others. So we only do it for AVX/AVX512 for now.

---
 Documentation/admin-guide/kernel-parameters.txt |  3 ++
 arch/x86/kernel/fpu/xstate.c                    | 61 +++++++++++++++++++++----
 2 files changed, 55 insertions(+), 9 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +3 -0
index 2ba45caabada..d9ae4a2c07ab 100644
@@ -785,6 +785,9 @@
 	dhash_entries=	[KNL]
 			Set number of hash buckets for dentry cache.
 
+	disable_avx	[X86] Disable support for AVX on Intel CPUs.
+	disable_avx512	[X86] Disable support for AVX512 on Intel CPUs.
+
 	disable_1tb_segments [PPC]
 			Disables the use of 1TB hash page table segments. This
 			causes the kernel to fall back to 256MB segments which
arch/x86/kernel/fpu/xstate.c +52 -9

96 lines not on this page · raw message

Signature · 1 line
2.9.3
#2 lore raw

Re: [PATCH 2/2] x86/fpu: Support disabling AVX and AVX512

Quoted 11 lines · Andi Kleen

From: Andi Kleen <[email protected]>

For performance testing it is useful to be able to disable AVX and AVX512. User programs check in XGETBV if AVX is supported by the OS. If we don't initialize the XSAVE state for AVX it will appear as if the OS is not supporting AVX. For kernel users we can also clear the internal cpu feature bits.

This patch implements disable options for AVX and AVX512 for the XSAVE code.

Quoted 6 lines

I originally considered a generic argument that would disable any XSAVE feature, but it turns out you need special code to also disable all the CPUID bits, because otherwise kernel code may assume it exists, when it doesn't. MPX already has an own disable flag. Not clear it is useful for the others. So we only do it for AVX/AVX512 for now.

Please read and follow Documentation/process/submitting-patches.rst.

Especially this paragraph:

Describe your changes in imperative mood, e.g. "make xyzzy do frotz" instead of "[This patch] makes xyzzy do frotz" or "[I] changed xyzzy to do frotz", as if you are giving orders to the codebase to change its behaviour.

Documentation/admin-guide/kernel-parameters.txt line 788

Quoted 1 line
Documentation/admin-guide/kernel-parameters.txt +1 -0
+	disable_avx	[X86] Disable support for AVX on Intel CPUs.

So that command line option fails on AMD CPUs, right?

Documentation/admin-guide/kernel-parameters.txt line 789

Quoted 1 line
Documentation/admin-guide/kernel-parameters.txt +1 -0
+	disable_avx512	[X86] Disable support for AVX512 on Intel CPUs.

Please drop the Intel stuff here as well. It's just a question of time until AMD gets that as well.

Quoted 14 lines
Documentation/admin-guide/kernel-parameters.txt +0 -0
 	disable_1tb_segments [PPC]
 			Disables the use of 1TB hash page table segments. This
 			causes the kernel to fall back to 256MB segments which
arch/x86/kernel/fpu/xstate.c +3 -0
index c24ac1efb12d..cf75638ec657 100644
@@ -16,6 +16,20 @@
 
 #include <asm/tlbflush.h>
 
+enum xsave_features {
+	XSAVE_X87,
+	XSAVE_SSE,

arch/x86/kernel/fpu/xstate.c line 31

Quoted 10 lines
arch/x86/kernel/fpu/xstate.c +10 -0
+	XSAVE_AVX,
+	XSAVE_MPX_BOUNDS,
+	XSAVE_MPX_CSR,
+	XSAVE_AVX512_OPMASK,
+	XSAVE_AVX512_HI256,
+	XSAVE_AVX512_ZMM_HI256,
+	XSAVE_PT,
+	XSAVE_PKU,
+	XSAVE_UNKNOWN
+};

What's that enum for? It's unused ....

arch/x86/kernel/fpu/xstate.c static const char *xfeature_names[] = · line 58

Quoted 8 lines
arch/x86/kernel/fpu/xstate.c +1 -0
 /*
  * Although we spell it out in here, the Processor Trace
  * xfeature is completely unused.  We use other mechanisms
@@ -41,6 +55,8 @@ static const char *xfeature_names[] =
  */
 u64 xfeatures_mask __read_mostly;
 
+u64 xfeatures_disabled __initdata;

Why is this global?

arch/x86/kernel/fpu/xstate.c void __init fpu__init_system_xstate(void) · line 762

Quoted 2 lines
arch/x86/kernel/fpu/xstate.c +1 -0
+	xfeatures_mask &= ~xfeatures_disabled;
 	xfeatures_mask &= fpu__get_supported_xfeatures_mask();

Thanks,

tglx

#3 lore raw

Re: [PATCH 2/2] x86/fpu: Support disabling AVX and AVX512

Quoted 7 lines · Thomas Gleixner
Quoted 7 lines
arch/x86/kernel/fpu/xstate.c +7 -0
+enum xsave_features {
+	XSAVE_X87,
+	XSAVE_SSE,
+	XSAVE_AVX,
+	XSAVE_MPX_BOUNDS,
+	XSAVE_MPX_CSR,
+	XSAVE_AVX512_OPMASK,
Quoted 8 lines

arch/x86/kernel/fpu/xstate.c line 31

Quoted 6 lines
arch/x86/kernel/fpu/xstate.c +6 -0
+	XSAVE_AVX512_HI256,
+	XSAVE_AVX512_ZMM_HI256,
+	XSAVE_PT,
+	XSAVE_PKU,
+	XSAVE_UNKNOWN
+};

What's that enum for? It's unused ....

It's not. The disable functions use it (although only the AVX parts)

-Andi

#4 lore raw

Re: [PATCH 2/2] x86/fpu: Support disabling AVX and AVX512

Quoted 17 lines · Andi Kleen
Quoted 7 lines · Thomas Gleixner
Quoted 7 lines
arch/x86/kernel/fpu/xstate.c +7 -0
+enum xsave_features {
+	XSAVE_X87,
+	XSAVE_SSE,
+	XSAVE_AVX,
+	XSAVE_MPX_BOUNDS,
+	XSAVE_MPX_CSR,
+	XSAVE_AVX512_OPMASK,
Quoted 8 lines

arch/x86/kernel/fpu/xstate.c line 31

Quoted 6 lines
arch/x86/kernel/fpu/xstate.c +6 -0
+	XSAVE_AVX512_HI256,
+	XSAVE_AVX512_ZMM_HI256,
+	XSAVE_PT,
+	XSAVE_PKU,
+	XSAVE_UNKNOWN
+};

What's that enum for? It's unused ....

Quoted 1 line

It's not. The disable functions use it (although only the AVX parts)

Sorry, misread it.

Thanks,

tglx