Skip to contentSkip to conversation

Updated cpuid-match patchkit v2

PATCH In review Andi Kleen 5 participants 22 messages Dec 19, 2011 – Dec 20, 2011

Subsystems (10)
x86 Kbuild ACPI Crypto API +6 more subsystems
lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

This patchkit allows automatic udev loading of drivers based on x86 CPUID models and feature ids.

This version addresses the review feedback, except:
- no multi match to keep things simple
- only matches boot cpu (also to make it simple)

I changed all the tables to const and did some other improvements.

More testing still appreciated, especially in a real distribution.

-Andi

author
#1 lore raw

PATCH 2/10 Add driver auto probing for x86 features

From: Andi Kleen <[email protected]>

Old patchkit, resurrect due to popular demand.

There's a growing number of drivers that support a specific x86 feature or CPU. Currently loading these drivers currently on a generic distribution requires various driver specific hacks and it often doesn't work.

This patch adds auto probing for drivers based on the x86 cpuid information, in particular based on vendor/family/model number and also based on CPUID feature bits.

For example a common issue is not loading the SSE 4.2 accelerated CRC module: this can significantly lower the performance of BTRFS which relies on fast CRC.

Another issue is loading the right CPUFREQ driver for the current CPU. Currently distributions often try all all possible driver until one sticks, which is not really a good way to do this.

It works with existing udev without any changes. The code exports the x86 information as a generic string in sysfs that can be matched by udev's pattern matching.

This scheme does not support numeric ranges, so if you want to handle e.g. ranges of model numbers they have to be encoded in ASCII or simply all models or families listed. Fixing that would require changing udev.

Another issue is that udev will happily load all drivers that match, there is currently no nice way to stop a specific driver from being loaded if it's not needed (e.g. if you don't need fast CRC) But there are not that many cpu specific drivers around and they're all not that bloated, so this isn't a particularly serious issue.

Originally this patch added the modalias to the normal cpu sysdevs. However sysdevs don't have all the infrastructure needed for udev, so it couldn't really autoload drivers. This patch instead adds the CPU modaliases to the cpuid devices, which are real devices with full support for udev. This implies that the cpuid driver has to be loaded to use this.

This patch just adds infrastructure, some driver conversions in followups.

Thanks to Kay for helping with some sysfs magic.

---
 arch/x86/include/asm/cpu_device_id.h |   13 +++++++
 arch/x86/kernel/cpu/Makefile         |    1 +
 arch/x86/kernel/cpu/match.c          |   48 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpuid.c              |   58 +++++++++++++++++++++++++++++++++-
 include/linux/mod_devicetable.h      |   20 ++++++++++++
 scripts/mod/file2alias.c             |   27 ++++++++++++++++
 6 files changed, 166 insertions(+), 1 deletions(-)
 create mode 100755 arch/x86/include/asm/cpu_device_id.h
 create mode 100644 arch/x86/kernel/cpu/match.c
arch/x86/include/asm/cpu_device_id.h new file +13 -0
new file mode 100755
index 0000000..ff501e5
@@ -0,0 +1,13 @@
+#ifndef _CPU_DEVICE_ID
+#define _CPU_DEVICE_ID 1
+
+/*
+ * Declare drivers belonging to specific x86 CPUs
+ * Similar in spirit to pci_device_id and related PCI functions
+ */
+
+#include <linux/mod_devicetable.h>
+
+extern const struct x86_cpu_id *x86_match_cpu(const struct x86_cpu_id *match);
+
+#endif
arch/x86/kernel/cpu/Makefile +1 -0
index 25f24dc..6ab6aa2 100644
@@ -16,6 +16,7 @@ obj-y			:= intel_cacheinfo.o scattered.o topology.o
 obj-y			+= proc.o capflags.o powerflags.o common.o
 obj-y			+= vmware.o hypervisor.o sched.o mshyperv.o
 obj-y			+= rdrand.o
+obj-y			+= match.o
 
 obj-$(CONFIG_X86_32)	+= bugs.o
 obj-$(CONFIG_X86_64)	+= bugs_64.o
arch/x86/kernel/cpu/match.c new file +48 -0

48 lines not on this page · raw message

arch/x86/kernel/cpuid.c +57 -1

88 lines not on this page · raw message

include/linux/mod_devicetable.h +20 -0
index 468819c..8971fec 100644
@@ -542,4 +542,24 @@ struct isapnp_device_id {
 	kernel_ulong_t driver_data;	/* data private to the driver */
 };
 
+/*
+ * Match x86 CPUs for CPU specific drivers.
+ * See documentation of "x86_match_cpu" for details.
+ */
+
+struct x86_cpu_id {
+	__u16 vendor;
+	__u16 family;
+	__u16 model;
+ 	__u16 feature;	/* bit index */
+	kernel_ulong_t driver_data;
+};
+
+#define X86_FEATURE_MATCH(x) { X86_VENDOR_ANY, X86_FAMILY_ANY, X86_MODEL_ANY, x }
+
+#define X86_VENDOR_ANY 0xffff
+#define X86_FAMILY_ANY 0
+#define X86_MODEL_ANY  0
+#define X86_FEATURE_ANY 0	/* Same as FPU, you can't test for that */
+
 #endif /* LINUX_MOD_DEVICETABLE_H */
scripts/mod/file2alias.c +27 -0
index f936d1f..e5dd089 100644
@@ -880,6 +880,29 @@ static int do_isapnp_entry(const char *filename,
 	return 1;
 }
 
+/* LOOKS like x86cpu:vendor:VVVV:family:FFFF:model:MMMM:feature:*,FEAT,*
+ * All fields are numbers. It would be nicer to use strings for vendor
+ * and feature, but getting those out of the build system here is too
+ * complicated.
+ */
+
+static int do_x86cpu_entry(const char *filename, struct x86_cpu_id *id, 
+			   char *alias)
+{
+	id->feature = TO_NATIVE(id->feature);
+	id->family = TO_NATIVE(id->family);
+	id->model = TO_NATIVE(id->model);
+	id->vendor = TO_NATIVE(id->vendor);
+
+	strcpy(alias, "x86cpu:");
+	ADD(alias, "vendor:",  id->vendor != X86_VENDOR_ANY, id->vendor);
+	ADD(alias, ":family:", id->family != X86_FAMILY_ANY, id->family);
+	ADD(alias, ":model:",  id->model  != X86_MODEL_ANY,  id->model);
+	ADD(alias, ":feature:*,", id->feature != X86_FEATURE_ANY, id->feature);
+	strcat(alias, ",*");
+	return 1;
+}
+
 /* Ignore any prefix, eg. some architectures prepend _ */
 static inline int sym_is(const char *symbol, const char *name)
 {
@@ -1047,6 +1070,10 @@ void handle_moddevtable(struct module *mod, struct elf_info *info,
 		do_table(symval, sym->st_size,
 			sizeof(struct isapnp_device_id), "isa",
 			do_isapnp_entry, mod);
+	else if (sym_is(symname, "__mod_x86cpu_device_table"))
+		do_table(symval, sym->st_size,
+			 sizeof(struct x86_cpu_id), "x86cpu",
+			 do_x86cpu_entry, mod);
 	free(zeros);
 }
 
Signature · 1 line
1.7.7.4
author
#2 lore raw

PATCH 2/10 Add driver auto probing for x86 features v2

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Old patchkit, resurrect due to popular demand.

There's a growing number of drivers that support a specific x86 feature or CPU. Currently loading these drivers currently on a generic distribution requires various driver specific hacks and it often doesn't work.

This patch adds auto probing for drivers based on the x86 cpuid information, in particular based on vendor/family/model number and also based on CPUID feature bits.

For example a common issue is not loading the SSE 4.2 accelerated CRC module: this can significantly lower the performance of BTRFS which relies on fast CRC.

Another issue is loading the right CPUFREQ driver for the current CPU. Currently distributions often try all all possible driver until one sticks, which is not really a good way to do this.

It works with existing udev without any changes. The code exports the x86 information as a generic string in sysfs that can be matched by udev's pattern matching.

This scheme does not support numeric ranges, so if you want to handle e.g. ranges of model numbers they have to be encoded in ASCII or simply all models or families listed. Fixing that would require changing udev.

Another issue is that udev will happily load all drivers that match, there is currently no nice way to stop a specific driver from being loaded if it's not needed (e.g. if you don't need fast CRC) But there are not that many cpu specific drivers around and they're all not that bloated, so this isn't a particularly serious issue.

Originally this patch added the modalias to the normal cpu sysdevs. However sysdevs don't have all the infrastructure needed for udev, so it couldn't really autoload drivers. This patch instead adds the CPU modaliases to the cpuid devices, which are real devices with full support for udev. This implies that the cpuid driver has to be loaded to use this.

This patch just adds infrastructure, some driver conversions in followups.

Thanks to Kay for helping with some sysfs magic.

v2: Constifcation, some updates

---
 arch/x86/include/asm/cpu_device_id.h |   13 +++++++
 arch/x86/kernel/cpu/Makefile         |    1 +
 arch/x86/kernel/cpu/match.c          |   48 ++++++++++++++++++++++++++++
 arch/x86/kernel/cpuid.c              |   58 +++++++++++++++++++++++++++++++++-
 include/linux/mod_devicetable.h      |   20 ++++++++++++
 scripts/mod/file2alias.c             |   27 ++++++++++++++++
 6 files changed, 166 insertions(+), 1 deletions(-)
 create mode 100755 arch/x86/include/asm/cpu_device_id.h
 create mode 100644 arch/x86/kernel/cpu/match.c
arch/x86/include/asm/cpu_device_id.h new file +13 -0
new file mode 100755
index 0000000..ff501e5
@@ -0,0 +1,13 @@
+#ifndef _CPU_DEVICE_ID
+#define _CPU_DEVICE_ID 1
+
+/*
+ * Declare drivers belonging to specific x86 CPUs
+ * Similar in spirit to pci_device_id and related PCI functions
+ */
+
+#include <linux/mod_devicetable.h>
+
+extern const struct x86_cpu_id *x86_match_cpu(const struct x86_cpu_id *match);
+
+#endif
arch/x86/kernel/cpu/Makefile +1 -0
index 25f24dc..6ab6aa2 100644
@@ -16,6 +16,7 @@ obj-y			:= intel_cacheinfo.o scattered.o topology.o
 obj-y			+= proc.o capflags.o powerflags.o common.o
 obj-y			+= vmware.o hypervisor.o sched.o mshyperv.o
 obj-y			+= rdrand.o
+obj-y			+= match.o
 
 obj-$(CONFIG_X86_32)	+= bugs.o
 obj-$(CONFIG_X86_64)	+= bugs_64.o
arch/x86/kernel/cpu/match.c new file +48 -0

48 lines not on this page · raw message

arch/x86/kernel/cpuid.c +57 -1

88 lines not on this page · raw message

include/linux/mod_devicetable.h +20 -0
index 468819c..8971fec 100644
@@ -542,4 +542,24 @@ struct isapnp_device_id {
 	kernel_ulong_t driver_data;	/* data private to the driver */
 };
 
+/*
+ * Match x86 CPUs for CPU specific drivers.
+ * See documentation of "x86_match_cpu" for details.
+ */
+
+struct x86_cpu_id {
+	__u16 vendor;
+	__u16 family;
+	__u16 model;
+ 	__u16 feature;	/* bit index */
+	kernel_ulong_t driver_data;
+};
+
+#define X86_FEATURE_MATCH(x) { X86_VENDOR_ANY, X86_FAMILY_ANY, X86_MODEL_ANY, x }
+
+#define X86_VENDOR_ANY 0xffff
+#define X86_FAMILY_ANY 0
+#define X86_MODEL_ANY  0
+#define X86_FEATURE_ANY 0	/* Same as FPU, you can't test for that */
+
 #endif /* LINUX_MOD_DEVICETABLE_H */
scripts/mod/file2alias.c +27 -0
index f936d1f..e5dd089 100644
@@ -880,6 +880,29 @@ static int do_isapnp_entry(const char *filename,
 	return 1;
 }
 
+/* LOOKS like x86cpu:vendor:VVVV:family:FFFF:model:MMMM:feature:*,FEAT,*
+ * All fields are numbers. It would be nicer to use strings for vendor
+ * and feature, but getting those out of the build system here is too
+ * complicated.
+ */
+
+static int do_x86cpu_entry(const char *filename, struct x86_cpu_id *id, 
+			   char *alias)
+{
+	id->feature = TO_NATIVE(id->feature);
+	id->family = TO_NATIVE(id->family);
+	id->model = TO_NATIVE(id->model);
+	id->vendor = TO_NATIVE(id->vendor);
+
+	strcpy(alias, "x86cpu:");
+	ADD(alias, "vendor:",  id->vendor != X86_VENDOR_ANY, id->vendor);
+	ADD(alias, ":family:", id->family != X86_FAMILY_ANY, id->family);
+	ADD(alias, ":model:",  id->model  != X86_MODEL_ANY,  id->model);
+	ADD(alias, ":feature:*,", id->feature != X86_FEATURE_ANY, id->feature);
+	strcat(alias, ",*");
+	return 1;
+}
+
 /* Ignore any prefix, eg. some architectures prepend _ */
 static inline int sym_is(const char *symbol, const char *name)
 {
@@ -1047,6 +1070,10 @@ void handle_moddevtable(struct module *mod, struct elf_info *info,
 		do_table(symval, sym->st_size,
 			sizeof(struct isapnp_device_id), "isa",
 			do_isapnp_entry, mod);
+	else if (sym_is(symname, "__mod_x86cpu_device_table"))
+		do_table(symval, sym->st_size,
+			 sizeof(struct x86_cpu_id), "x86cpu",
+			 do_x86cpu_entry, mod);
 	free(zeros);
 }
 
Signature · 1 line
1.7.7.4
author
#3 lore raw

PATCH 3/10 crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Add support for auto-loading of crypto drivers based on cpuid features. This enables auto-loading of the VIA and Intel specific drivers for AES, hashing and CRCs.

Requires the earlier infrastructure patch to add x86 modinfo. I kept it all in a single patch for now.

I dropped the printks when the driver cpuid doesn't match (imho drivers never should print anything in such a case)

One drawback is that udev doesn't know if the drivers are used or not, so they will be unconditionally loaded at boot up. That's better than not loading them at all, like it often happens.

---
 arch/x86/crypto/aesni-intel_glue.c         |   12 +++++++++---
 arch/x86/crypto/crc32c-intel.c             |   11 ++++++++---
 arch/x86/crypto/ghash-clmulni-intel_glue.c |   12 ++++++++----
 drivers/crypto/padlock-aes.c               |    9 ++++++++-
 drivers/crypto/padlock-sha.c               |   16 ++++++++--------
 5 files changed, 41 insertions(+), 19 deletions(-)
arch/x86/crypto/aesni-intel_glue.c +9 -3
index 545d0ce..b3350bd 100644
@@ -28,6 +28,7 @@
 #include <crypto/aes.h>
 #include <crypto/cryptd.h>
 #include <crypto/ctr.h>
+#include <asm/cpu_device_id.h>
 #include <asm/i387.h>
 #include <asm/aes.h>
 #include <crypto/scatterwalk.h>
@@ -1253,14 +1254,19 @@ static struct crypto_alg __rfc4106_alg = {
 };
 #endif
 
+
+static const struct x86_cpu_id aesni_cpu_id[] = {
+	X86_FEATURE_MATCH(X86_FEATURE_AES),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, aesni_cpu_id);
+
 static int __init aesni_init(void)
 {
 	int err;
 
-	if (!cpu_has_aes) {
-		printk(KERN_INFO "Intel AES-NI instructions are not detected.\n");
+	if (!x86_match_cpu(aesni_cpu_id))
 		return -ENODEV;
-	}
 
 	if ((err = crypto_fpu_init()))
 		goto fpu_err;
arch/x86/crypto/crc32c-intel.c +8 -3
index b9d0026..493f959 100644
@@ -31,6 +31,7 @@
 #include <crypto/internal/hash.h>
 
 #include <asm/cpufeature.h>
+#include <asm/cpu_device_id.h>
 
 #define CHKSUM_BLOCK_SIZE	1
 #define CHKSUM_DIGEST_SIZE	4
@@ -173,13 +174,17 @@ static struct shash_alg alg = {
 	}
 };
 
+static const struct x86_cpu_id crc32c_cpu_id[] = {
+	X86_FEATURE_MATCH(X86_FEATURE_XMM4_2),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, crc32c_cpu_id);
 
 static int __init crc32c_intel_mod_init(void)
 {
-	if (cpu_has_xmm4_2)
-		return crypto_register_shash(&alg);
-	else
+	if (!x86_match_cpu(crc32c_cpu_id))
 		return -ENODEV;
+	return crypto_register_shash(&alg);
 }
 
 static void __exit crc32c_intel_mod_fini(void)
arch/x86/crypto/ghash-clmulni-intel_glue.c +8 -4
index 976aa64..b4bf0a6 100644
@@ -20,6 +20,7 @@
 #include <crypto/gf128mul.h>
 #include <crypto/internal/hash.h>
 #include <asm/i387.h>
+#include <asm/cpu_device_id.h>
 
 #define GHASH_BLOCK_SIZE	16
 #define GHASH_DIGEST_SIZE	16
@@ -294,15 +295,18 @@ static struct ahash_alg ghash_async_alg = {
 	},
 };
 
+static const struct x86_cpu_id pcmul_cpu_id[] = {
+	X86_FEATURE_MATCH(X86_FEATURE_PCLMULQDQ), /* Pickle-Mickle-Duck */
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, pcmul_cpu_id);
+
 static int __init ghash_pclmulqdqni_mod_init(void)
 {
 	int err;
 
-	if (!cpu_has_pclmulqdq) {
-		printk(KERN_INFO "Intel PCLMULQDQ-NI instructions are not"
-		       " detected.\n");
+	if (!x86_match_cpu(pcmul_cpu_id))
 		return -ENODEV;
-	}
 
 	err = crypto_register_shash(&ghash_alg);
 	if (err)
drivers/crypto/padlock-aes.c +8 -1
index 29b9469..37b2e94 100644
@@ -19,6 +19,7 @@
 #include <linux/percpu.h>
 #include <linux/smp.h>
 #include <linux/slab.h>
+#include <asm/cpu_device_id.h>
 #include <asm/byteorder.h>
 #include <asm/processor.h>
 #include <asm/i387.h>
@@ -503,12 +504,18 @@ static struct crypto_alg cbc_aes_alg = {
 	}
 };
 
+static struct x86_cpu_id padlock_cpu_id[] = {
+	X86_FEATURE_MATCH(X86_FEATURE_XCRYPT),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, padlock_cpu_id);
+
 static int __init padlock_init(void)
 {
 	int ret;
 	struct cpuinfo_x86 *c = &cpu_data(0);
 
-	if (!cpu_has_xcrypt)
+	if (!x86_match_cpu(padlock_cpu_id))
 		return -ENODEV;
 
 	if (!cpu_has_xcrypt_enabled) {
drivers/crypto/padlock-sha.c +8 -8
index 06bdb4b..9266c0e 100644
@@ -22,6 +22,7 @@
 #include <linux/interrupt.h>
 #include <linux/kernel.h>
 #include <linux/scatterlist.h>
+#include <asm/cpu_device_id.h>
 #include <asm/i387.h>
 
 struct padlock_sha_desc {
@@ -526,6 +527,12 @@ static struct shash_alg sha256_alg_nano = {
 	}
 };
 
+static struct x86_cpu_id padlock_sha_ids[] = {
+	X86_FEATURE_MATCH(X86_FEATURE_PHE),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, padlock_sha_ids);
+
 static int __init padlock_init(void)
 {
 	int rc = -ENODEV;
@@ -533,15 +540,8 @@ static int __init padlock_init(void)
 	struct shash_alg *sha1;
 	struct shash_alg *sha256;
 
-	if (!cpu_has_phe) {
-		printk(KERN_NOTICE PFX "VIA PadLock Hash Engine not detected.\n");
-		return -ENODEV;
-	}
-
-	if (!cpu_has_phe_enabled) {
-		printk(KERN_NOTICE PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n");
+	if (!x86_match_cpu(padlock_sha_ids) || !cpu_has_phe_enabled)
 		return -ENODEV;
-	}
 
 	/* Register the newly added algorithm module if on *
 	* VIA Nano processor, or else just do as before */
Signature · 1 line
1.7.7.4
#4 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 7 lines · Andi Kleen

static int __init padlock_init(void)
{

drivers/crypto/padlock-sha.c +0 -0
 	int rc = -ENODEV;
@@ -533,15 +540,8 @@ static int __init padlock_init(void)
 	struct shash_alg *sha1;
 	struct shash_alg *sha256;
 

drivers/crypto/padlock-sha.c static int __init padlock_init(void) · line 544

Quoted 10 lines
drivers/crypto/padlock-sha.c +1 -8
-	if (!cpu_has_phe) {
-		printk(KERN_NOTICE PFX "VIA PadLock Hash Engine not detected.\n");
-		return -ENODEV;
-	}
-
-	if (!cpu_has_phe_enabled) {
-		printk(KERN_NOTICE PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n");
+	if (!x86_match_cpu(padlock_sha_ids) || !cpu_has_phe_enabled)
 		return -ENODEV;
-	}

don't we want to keep the printk ?

Seems that if it's been disabled by the BIOS for some reason, it will now silently fail instead of printing a nice warning.

Dave

maintainer
#5 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 22 lines · Dave Jones
Quoted 7 lines · Andi Kleen

static int __init padlock_init(void)
{

drivers/crypto/padlock-sha.c +0 -0
 	int rc = -ENODEV;
@@ -533,15 +540,8 @@ static int __init padlock_init(void)
 	struct shash_alg *sha1;
 	struct shash_alg *sha256;
 

drivers/crypto/padlock-sha.c static int __init padlock_init(void) · line 544

Quoted 10 lines
drivers/crypto/padlock-sha.c +1 -8
-	if (!cpu_has_phe) {
-		printk(KERN_NOTICE PFX "VIA PadLock Hash Engine not detected.\n");
-		return -ENODEV;
-	}
-
-	if (!cpu_has_phe_enabled) {
-		printk(KERN_NOTICE PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n");
+	if (!x86_match_cpu(padlock_sha_ids) || !cpu_has_phe_enabled)
 		return -ENODEV;
-	}

don't we want to keep the printk ?

Quoted 3 lines

Seems that if it's been disabled by the BIOS for some reason, it will now silently fail instead of printing a nice warning.

The first printk is insanely annoying, because the end result is that we get a printk of all the crypto engines that are *not* enabled... arguably it would be somewhat nice to get a printk when a crypto engine *is* enabled.

-hpa

#6 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 13 lines · H. Peter Anvin
Quoted 13 lines

drivers/crypto/padlock-sha.c static int __init padlock_init(void) · line 544

Quoted 10 lines
drivers/crypto/padlock-sha.c +1 -8
-	if (!cpu_has_phe) {
-		printk(KERN_NOTICE PFX "VIA PadLock Hash Engine not detected.\n");
-		return -ENODEV;
-	}
-
-	if (!cpu_has_phe_enabled) {
-		printk(KERN_NOTICE PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n");
+	if (!x86_match_cpu(padlock_sha_ids) || !cpu_has_phe_enabled)
 		return -ENODEV;
-	}

don't we want to keep the printk ?

Quoted 7 lines
Quoted 2 lines

Seems that if it's been disabled by the BIOS for some reason, it will now silently fail instead of printing a nice warning.

The first printk is insanely annoying, because the end result is that we get a printk of all the crypto engines that are *not* enabled... arguably it would be somewhat nice to get a printk when a crypto engine *is* enabled.

Sorry, I just meant the second printk, which should hopefully not get printed by default.

Dave

author
#7 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

In reply to H. Peter Anvin
Quoted 4 lines

The first printk is insanely annoying, because the end result is that we get a printk of all the crypto engines that are *not* enabled... arguably it would be somewhat nice to get a printk when a crypto engine *is* enabled.

Yes that was my rationale. I have some Intel system complaining that some VIA crypto is not available. That's very annoying and just has to go.

-Andi

maintainer
#8 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 8 lines · Andi Kleen
Quoted 4 lines

The first printk is insanely annoying, because the end result is that we get a printk of all the crypto engines that are *not* enabled... arguably it would be somewhat nice to get a printk when a crypto engine *is* enabled.

Yes that was my rationale. I have some Intel system complaining that some VIA crypto is not available. That's very annoying and just has to go.

What is really bad is that it doesn't tell you anywhere what it *is* using.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#9 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 2 lines · H. Peter Anvin

What is really bad is that it doesn't tell you anywhere what it *is* using.

You can always check /proc/crypto to see what is currently registered. The system will always choose the one with the highest priority out of those as the default.

Cheers,

Signature · 3 lines
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
maintainer
#10 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 8 lines · Herbert Xu
Quoted 2 lines · H. Peter Anvin

What is really bad is that it doesn't tell you anywhere what it *is* using.

You can always check /proc/crypto to see what is currently registered. The system will always choose the one with the highest priority out of those as the default.

Yes, but it's ridiculous for it to list all the things I'm *not* using, but not what I *am* using...

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#11 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 11 lines · H. Peter Anvin
Quoted 7 lines · Herbert Xu
Quoted 2 lines · H. Peter Anvin

What is really bad is that it doesn't tell you anywhere what it *is* using.

You can always check /proc/crypto to see what is currently registered. The system will always choose the one with the highest priority out of those as the default.

Yes, but it's ridiculous for it to list all the things I'm *not* using, but not what I *am* using...

If your system doesn't have the VIA Padlock it won't even show up in /proc/crypto since it'll fail to register.

Cheers,

Signature · 3 lines
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
maintainer
#12 lore raw

Re: [PATCH 03/10] crypto: Add support for x86 cpuid auto loading for x86 crypto drivers

Quoted 7 lines · Herbert Xu
Quoted 3 lines

Yes, but it's ridiculous for it to list all the things I'm *not* using, but not what I *am* using...

If your system doesn't have the VIA Padlock it won't even show up in /proc/crypto since it'll fail to register.

Wow. /proc/crypto is hard to read.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
author
#13 lore raw

PATCH 4/10 intel-idle: convert to x86_cpu_id auto probing

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

With this it should be automatically loaded on suitable systems by udev.

The old switch () is replaced with a table based approach, this also cleans up the code.

---
 drivers/idle/intel_idle.c |  116 +++++++++++++++++++++++++-------------------
 1 files changed, 66 insertions(+), 50 deletions(-)
drivers/idle/intel_idle.c +66 -50
index 5d2f8e1..0920cc5 100644
@@ -62,6 +62,7 @@
 #include <linux/notifier.h>
 #include <linux/cpu.h>
 #include <linux/module.h>
+#include <asm/cpu_device_id.h>
 #include <asm/mwait.h>
 #include <asm/msr.h>
 
@@ -81,6 +82,17 @@ static unsigned int mwait_substates;
 /* Reliable LAPIC Timer States, bit 1 for C1 etc.  */
 static unsigned int lapic_timer_reliable_states = (1 << 1);	 /* Default to only C1 */
 
+struct idle_cpu {
+	struct cpuidle_state *state_table;
+
+	/*
+	 * Hardware C-state auto-demotion may not always be optimal.
+	 * Indicate which enable bits to clear here.
+	 */
+	unsigned long auto_demotion_disable_flags;
+};
+
+static const struct idle_cpu *icpu;
 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
 static int intel_idle(struct cpuidle_device *dev,
 			struct cpuidle_driver *drv, int index);
@@ -88,12 +100,6 @@ static int intel_idle(struct cpuidle_device *dev,
 static struct cpuidle_state *cpuidle_state_table;
 
 /*
- * Hardware C-state auto-demotion may not always be optimal.
- * Indicate which enable bits to clear here.
- */
-static unsigned long long auto_demotion_disable_flags;
-
-/*
  * Set this flag for states where the HW flushes the TLB for us
  * and so we don't need cross-calls to keep it consistent.
  * If this flag is set, SW flushes the TLB, so even if the
@@ -320,27 +326,72 @@ static void auto_demotion_disable(void *dummy)
 	unsigned long long msr_bits;
 
 	rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
-	msr_bits &= ~auto_demotion_disable_flags;
+	msr_bits &= ~(icpu->auto_demotion_disable_flags);
 	wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
 }
 
+static const struct idle_cpu idle_cpu_nehalem = {
+	.state_table = nehalem_cstates,
+};
+
+static const struct idle_cpu idle_cpu_westmere = {
+	.state_table = nehalem_cstates,
+	.auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
+};
+
+static const struct idle_cpu idle_cpu_atom = {
+	.state_table = atom_cstates,
+};
+
+static const struct idle_cpu idle_cpu_lincroft = {
+	.state_table = atom_cstates,
+	.auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
+};
+
+static const struct idle_cpu idle_cpu_snb = {
+	.state_table = snb_cstates,
+};
+
+#define ICPU(model, cpu) \
+	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
+
+static const struct x86_cpu_id intel_idle_ids[] = {
+	ICPU(0x1a, idle_cpu_nehalem),
+	ICPU(0x1e, idle_cpu_nehalem),
+	ICPU(0x1f, idle_cpu_nehalem),
+	ICPU(0x25, idle_cpu_westmere),
+	ICPU(0x2c, idle_cpu_westmere),
+	ICPU(0x2f, idle_cpu_westmere),
+	ICPU(0x1c, idle_cpu_atom),
+	ICPU(0x26, idle_cpu_lincroft),
+	ICPU(0x2f, idle_cpu_westmere),
+	ICPU(0x2a, idle_cpu_snb),
+	ICPU(0x2d, idle_cpu_snb),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
+
 /*
  * intel_idle_probe()
  */
 static int intel_idle_probe(void)
 {
 	unsigned int eax, ebx, ecx;
+	const struct x86_cpu_id *id;
 
 	if (max_cstate == 0) {
 		pr_debug(PREFIX "disabled\n");
 		return -EPERM;
 	}
 
-	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
-		return -ENODEV;
-
-	if (!boot_cpu_has(X86_FEATURE_MWAIT))
+	id = x86_match_cpu(intel_idle_ids);
+	if (!id) {
+		if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
+		    boot_cpu_data.x86 == 6)
+			pr_debug(PREFIX "does not run on family %d model %d\n",
+				boot_cpu_data.x86, boot_cpu_data.x86_model);
 		return -ENODEV;
+	}
 
 	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
 		return -ENODEV;
@@ -353,44 +404,9 @@ static int intel_idle_probe(void)
 
 	pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
 
-
-	if (boot_cpu_data.x86 != 6)	/* family 6 */
-		return -ENODEV;
-
-	switch (boot_cpu_data.x86_model) {
-
-	case 0x1A:	/* Core i7, Xeon 5500 series */
-	case 0x1E:	/* Core i7 and i5 Processor - Lynnfield Jasper Forest */
-	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
-	case 0x2E:	/* Nehalem-EX Xeon */
-	case 0x2F:	/* Westmere-EX Xeon */
-	case 0x25:	/* Westmere */
-	case 0x2C:	/* Westmere */
-		cpuidle_state_table = nehalem_cstates;
-		auto_demotion_disable_flags =
-			(NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE);
-		break;
-
-	case 0x1C:	/* 28 - Atom Processor */
-		cpuidle_state_table = atom_cstates;
-		break;
-
-	case 0x26:	/* 38 - Lincroft Atom Processor */
-		cpuidle_state_table = atom_cstates;
-		auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE;
-		break;
-
-	case 0x2A:	/* SNB */
-	case 0x2D:	/* SNB Xeon */
-		cpuidle_state_table = snb_cstates;
-		break;
-
-	default:
-		pr_debug(PREFIX "does not run on family %d model %d\n",
-			boot_cpu_data.x86, boot_cpu_data.x86_model);
-		return -ENODEV;
-	}
-
+	icpu = (const struct idle_cpu *)id->driver_data;
+	cpuidle_state_table = icpu->state_table;
+	
 	if (boot_cpu_has(X86_FEATURE_ARAT))	/* Always Reliable APIC Timer */
 		lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
 	else {
@@ -470,7 +486,7 @@ static int intel_idle_cpuidle_driver_init(void)
 		drv->state_count += 1;
 	}
 
-	if (auto_demotion_disable_flags)
+	if (icpu->auto_demotion_disable_flags)
 		smp_call_function(auto_demotion_disable, NULL, 1);
 
 	return 0;
Signature · 1 line
1.7.7.4
author
#14 lore raw

PATCH 5/10 ACPI: Load acpi-cpufreq from processor driver automatically

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

The only left over hole in automatic cpufreq driver loading was the loading of ACPI cpufreq. This driver should be loaded when ACPI supports a _PDC method and the CPU vendor wants to use acpi cpufreq.

Simply add a request module call to the acpi processor core driver when this is true. This seems like the simplest solution for this.

---
 drivers/acpi/processor_driver.c  |    1 +
 drivers/acpi/processor_perflib.c |   22 ++++++++++++++++++++++
 include/acpi/processor.h         |    1 +
 3 files changed, 24 insertions(+), 0 deletions(-)
drivers/acpi/processor_driver.c +1 -0
index 9d7bc9f..9e933ed 100644
@@ -499,6 +499,7 @@ static int __cpuinit acpi_processor_add(struct acpi_device *device)
 
 #ifdef CONFIG_CPU_FREQ
 	acpi_processor_ppc_has_changed(pr, 0);
+	acpi_processor_load_module(pr);
 #endif
 	acpi_processor_get_throttling_info(pr);
 	acpi_processor_get_limit_info(pr);
drivers/acpi/processor_perflib.c +22 -0
index 85b3237..a533512 100644
@@ -240,6 +240,28 @@ void acpi_processor_ppc_exit(void)
 	acpi_processor_ppc_status &= ~PPC_REGISTERED;
 }
 
+/* 
+ * Do a quick check if the systems looks like it should use ACPI 
+ * cpufreq. We look at a _PCT method being available, but don't
+ * do a whole lot of sanity checks. 
+ */
+void acpi_processor_load_module(struct acpi_processor *pr)
+{
+	static int requested;
+	acpi_status status = 0;
+	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
+
+	if (!arch_has_acpi_pdc() || requested)
+		return;
+	status = acpi_evaluate_object(pr->handle, "_PCT", NULL, &buffer);
+	if (!ACPI_FAILURE(status)) {
+		printk(KERN_INFO PREFIX "Requesting acpi_cpufreq\n");
+		request_module_nowait("acpi_cpufreq");
+		requested = 1;
+	}
+	kfree(buffer.pointer);
+}
+
 static int acpi_processor_get_performance_control(struct acpi_processor *pr)
 {
 	int result = 0;
include/acpi/processor.h +1 -0
index 610f6fb..da57fdc 100644
@@ -224,6 +224,7 @@ struct acpi_processor_errata {
 	} piix4;
 };
 
+extern void acpi_processor_load_module(struct acpi_processor *pr);
 extern int acpi_processor_preregister_performance(struct
 						  acpi_processor_performance
 						  __percpu *performance);
Signature · 1 line
1.7.7.4
author
#15 lore raw

PATCH 6/10 HWMON: Convert via-cputemp to x86 cpuid autoprobing

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Use the new x86 cpuid autoprobe interface.

---
 drivers/hwmon/via-cputemp.c |   16 +++++++++++-----
 1 files changed, 11 insertions(+), 5 deletions(-)
drivers/hwmon/via-cputemp.c +11 -5
index 8eac67d..8689664 100644
@@ -37,6 +37,7 @@
 #include <linux/cpu.h>
 #include <asm/msr.h>
 #include <asm/processor.h>
+#include <asm/cpu_device_id.h>
 
 #define DRVNAME	"via_cputemp"
 
@@ -308,15 +309,20 @@ static struct notifier_block via_cputemp_cpu_notifier __refdata = {
 	.notifier_call = via_cputemp_cpu_callback,
 };
 
+static const struct x86_cpu_id cputemp_ids[] = {
+	{ X86_VENDOR_CENTAUR, 6, 0xa, }, /* C7 A */
+	{ X86_VENDOR_CENTAUR, 6, 0xd, }, /* C7 D */
+	{ X86_VENDOR_CENTAUR, 6, 0xf, }, /* Nano */
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, cputemp_ids);
+
 static int __init via_cputemp_init(void)
 {
 	int i, err;
 
-	if (cpu_data(0).x86_vendor != X86_VENDOR_CENTAUR) {
-		printk(KERN_DEBUG DRVNAME ": Not a VIA CPU\n");
-		err = -ENODEV;
-		goto exit;
-	}
+	if (!x86_match_cpu(cputemp_ids))
+		return -ENODEV;
 
 	err = platform_driver_register(&via_cputemp_driver);
 	if (err)
Signature · 1 line
1.7.7.4
author
#16 lore raw

PATCH 7/10 HWMON: Convert coretemp to x86 cpuid autoprobing

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Use the new x86 cpuid autoprobe interface for the Intel coretemp driver.

---
 drivers/hwmon/coretemp.c |   17 ++++++++++++++---
 1 files changed, 14 insertions(+), 3 deletions(-)
drivers/hwmon/coretemp.c +14 -3
index 104b376..b965f80 100644
@@ -39,6 +39,7 @@
 #include <linux/moduleparam.h>
 #include <asm/msr.h>
 #include <asm/processor.h>
+#include <asm/cpu_device_id.h>
 
 #define DRVNAME	"coretemp"
 
@@ -756,13 +757,23 @@ static struct notifier_block coretemp_cpu_notifier __refdata = {
 	.notifier_call = coretemp_cpu_callback,
 };
 
+static const struct x86_cpu_id coretemp_ids[] = {
+	{ X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTS },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
+
 static int __init coretemp_init(void)
 {
 	int i, err = -ENODEV;
 
-	/* quick check if we run Intel */
-	if (cpu_data(0).x86_vendor != X86_VENDOR_INTEL)
-		goto exit;
+	/*
+	 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
+	 * sensors. We check this bit only, all the early CPUs
+	 * without thermal sensors will be filtered out.
+	 */
+	if (!x86_match_cpu(coretemp_ids))
+		return -ENODEV;
 
 	err = platform_driver_register(&coretemp_driver);
 	if (err)
Signature · 1 line
1.7.7.4
author
#17 lore raw

PATCH 8/10 cpufreq: Add support for x86 cpuinfo auto loading

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

This marks all the x86 cpuinfo tables to the CPU specific device drivers, to allow auto loading by udev. This should simplify the distribution startup scripts for this greatly.

I didn't add MODULE_DEVICE_IDs to the centrino and p4-clockmod drivers, because those probably shouldn't be auto loaded and the acpi driver be used instead (not fully sure on that, would appreciate feedback)

The old nforce drivers autoload based on the PCI ID.

ACPI cpufreq is autoloaded in another patch.

---
 drivers/cpufreq/cpufreq-nforce2.c    |    8 ++++++++
 drivers/cpufreq/e_powersaver.c       |   20 +++++++++++---------
 drivers/cpufreq/elanfreq.c           |   14 +++++++-------
 drivers/cpufreq/gx-suspmod.c         |   15 ++++++++++-----
 drivers/cpufreq/longhaul.c           |    8 +++++++-
 drivers/cpufreq/longrun.c            |   13 ++++++++-----
 drivers/cpufreq/p4-clockmod.c        |   17 +++++++++++------
 drivers/cpufreq/powernow-k6.c        |   12 ++++++++----
 drivers/cpufreq/powernow-k7.c        |   14 ++++++++------
 drivers/cpufreq/powernow-k8.c        |   22 ++++++++++++++++------
 drivers/cpufreq/sc520_freq.c         |   14 ++++++++------
 drivers/cpufreq/speedstep-centrino.c |   24 ++++++++++++++++++++----
 drivers/cpufreq/speedstep-ich.c      |   15 +++++++++++++++
 drivers/cpufreq/speedstep-lib.c      |    1 +
 drivers/cpufreq/speedstep-smi.c      |   15 +++++++++++++++
 15 files changed, 153 insertions(+), 59 deletions(-)
drivers/cpufreq/cpufreq-nforce2.c +8 -0
index 7bac808..661251b 100644
@@ -385,6 +385,14 @@ static struct cpufreq_driver nforce2_driver = {
 	.owner = THIS_MODULE,
 };
 
+#ifdef MODULE
+static struct pci_device_id nforce2_ids[] = {
+	{ PCI_VENDOR_ID_NVIDIA, PCI_DEVICE_ID_NVIDIA_NFORCE2 },
+	{}
+};
+MODULE_DEVICE_TABLE(pci, nforce2_ids);
+#endif
+
 /**
  * nforce2_detect_chipset - detect the Southbridge which contains FSB PLL logic
  *
drivers/cpufreq/e_powersaver.c +11 -9
index 4bd6815..3fffbe6 100644
@@ -16,6 +16,7 @@
 #include <linux/io.h>
 #include <linux/delay.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/msr.h>
 #include <asm/tsc.h>
 
@@ -437,18 +438,19 @@ static struct cpufreq_driver eps_driver = {
 	.attr		= eps_attr,
 };
 
+
+/* This driver will work only on Centaur C7 processors with
+ * Enhanced SpeedStep/PowerSaver registers */
+static const struct x86_cpu_id eps_cpu_id[] = {
+	{ X86_VENDOR_CENTAUR, 6, X86_MODEL_ANY, X86_FEATURE_EST },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, eps_cpu_id);
+
 static int __init eps_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
-
-	/* This driver will work only on Centaur C7 processors with
-	 * Enhanced SpeedStep/PowerSaver registers */
-	if (c->x86_vendor != X86_VENDOR_CENTAUR
-	    || c->x86 != 6 || c->x86_model < 10)
-		return -ENODEV;
-	if (!cpu_has(c, X86_FEATURE_EST))
+	if (!x86_match_cpu(eps_cpu_id) || boot_cpu_data.x86_model < 10)
 		return -ENODEV;
-
 	if (cpufreq_register_driver(&eps_driver))
 		return -EINVAL;
 	return 0;
drivers/cpufreq/elanfreq.c +7 -7
index c587db4..960671f 100644
@@ -23,6 +23,7 @@
 #include <linux/delay.h>
 #include <linux/cpufreq.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/msr.h>
 #include <linux/timex.h>
 #include <linux/io.h>
@@ -277,17 +278,16 @@ static struct cpufreq_driver elanfreq_driver = {
 	.attr		= elanfreq_attr,
 };
 
+static const struct x86_cpu_id elan_id[] = {
+	{ X86_VENDOR_AMD, 4, 10, },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, elan_id);
 
 static int __init elanfreq_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
-
-	/* Test if we have the right hardware */
-	if ((c->x86_vendor != X86_VENDOR_AMD) ||
-		(c->x86 != 4) || (c->x86_model != 10)) {
-		printk(KERN_INFO "elanfreq: error: no Elan processor found!\n");
+	if (!x86_match_cpu(elan_id))
 		return -ENODEV;
-	}
 	return cpufreq_register_driver(&elanfreq_driver);
 }
 
drivers/cpufreq/gx-suspmod.c +10 -5
index ffe1f2c..05d0f7f 100644
@@ -82,6 +82,7 @@
 #include <linux/errno.h>
 #include <linux/slab.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/processor-cyrix.h>
 
 /* PCI config registers, all at F0 */
@@ -171,12 +172,20 @@ static struct pci_device_id gx_chipset_tbl[] __initdata = {
 	{ PCI_VDEVICE(CYRIX, PCI_DEVICE_ID_CYRIX_5510), },
 	{ 0, },
 };
+/* MODULE_DEVICE_TABLE here too? But the CPU should be enough. */
 
 static void gx_write_byte(int reg, int value)
 {
 	pci_write_config_byte(gx_params->cs55x0, reg, value);
 }
 
+static const struct x86_cpu_id gx_ids[] = {
+	{ X86_VENDOR_NSC, },
+	{ X86_VENDOR_CYRIX },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, gx_ids);
+
 /**
  * gx_detect_chipset:
  *
@@ -185,12 +194,8 @@ static __init struct pci_dev *gx_detect_chipset(void)
 {
 	struct pci_dev *gx_pci = NULL;
 
-	/* check if CPU is a MediaGX or a Geode. */
-	if ((boot_cpu_data.x86_vendor != X86_VENDOR_NSC) &&
-	    (boot_cpu_data.x86_vendor != X86_VENDOR_CYRIX)) {
-		pr_debug("error: no MediaGX/Geode processor found!\n");
+	if (!x86_match_cpu(gx_ids))
 		return NULL;
-	}
 
 	/* detect which companion chip is used */
 	for_each_pci_dev(gx_pci) {
drivers/cpufreq/longhaul.c +7 -1
index f47d26e..53ddbc7 100644
@@ -35,6 +35,7 @@
 #include <linux/acpi.h>
 
 #include <asm/msr.h>
+#include <asm/cpu_device_id.h>
 #include <acpi/processor.h>
 
 #include "longhaul.h"
@@ -951,12 +952,17 @@ static struct cpufreq_driver longhaul_driver = {
 	.attr	= longhaul_attr,
 };
 
+static const struct x86_cpu_id longhaul_id[] = {
+	{ X86_VENDOR_CENTAUR, 6 },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, longhaul_id);
 
 static int __init longhaul_init(void)
 {
 	struct cpuinfo_x86 *c = &cpu_data(0);
 
-	if (c->x86_vendor != X86_VENDOR_CENTAUR || c->x86 != 6)
+	if (!x86_match_cpu(longhaul_id))
 		return -ENODEV;
 
 #ifdef CONFIG_SMP
drivers/cpufreq/longrun.c +8 -5
index 34ea359..8bc9f5f 100644
@@ -14,6 +14,7 @@
 
 #include <asm/msr.h>
 #include <asm/processor.h>
+#include <asm/cpu_device_id.h>
 
 static struct cpufreq_driver	longrun_driver;
 
@@ -288,6 +289,12 @@ static struct cpufreq_driver longrun_driver = {
 	.owner		= THIS_MODULE,
 };
 
+static const struct x86_cpu_id longrun_ids[] = {
+	{ X86_VENDOR_TRANSMETA, X86_FAMILY_ANY, X86_MODEL_ANY,
+	  X86_FEATURE_LONGRUN },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, longrun_ids);
 
 /**
  * longrun_init - initializes the Transmeta Crusoe LongRun CPUFreq driver
@@ -296,12 +303,8 @@ static struct cpufreq_driver longrun_driver = {
  */
 static int __init longrun_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
-
-	if (c->x86_vendor != X86_VENDOR_TRANSMETA ||
-	    !cpu_has(c, X86_FEATURE_LONGRUN))
+	if (!x86_match_cpu(longrun_ids))
 		return -ENODEV;
-
 	return cpufreq_register_driver(&longrun_driver);
 }
 
drivers/cpufreq/p4-clockmod.c +11 -6
index 6be3e07..827629c9 100644
@@ -31,6 +31,7 @@
 #include <asm/processor.h>
 #include <asm/msr.h>
 #include <asm/timer.h>
+#include <asm/cpu_device_id.h>
 
 #include "speedstep-lib.h"
 
@@ -289,21 +290,25 @@ static struct cpufreq_driver p4clockmod_driver = {
 	.attr		= p4clockmod_attr,
 };
 
+static const struct x86_cpu_id cpufreq_p4_id[] = {
+	{ X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_ACC },
+	{}
+};
+
+/*
+ * Intentionally no MODULE_DEVICE_TABLE here: this driver should not
+ * be auto loaded.  Please don't add one.
+ */
 
 static int __init cpufreq_p4_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
 	int ret;
 
 	/*
 	 * THERM_CONTROL is architectural for IA32 now, so
 	 * we can rely on the capability checks
 	 */
-	if (c->x86_vendor != X86_VENDOR_INTEL)
-		return -ENODEV;
-
-	if (!test_cpu_cap(c, X86_FEATURE_ACPI) ||
-				!test_cpu_cap(c, X86_FEATURE_ACC))
+	if (!x86_match_cpu(cpufreq_p4_id) || !boot_cpu_has(X86_FEATURE_ACPI))
 		return -ENODEV;
 
 	ret = cpufreq_register_driver(&p4clockmod_driver);
drivers/cpufreq/powernow-k6.c +8 -4
index b3379d6..54dd031 100644
@@ -16,6 +16,7 @@
 #include <linux/timex.h>
 #include <linux/io.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/msr.h>
 
 #define POWERNOW_IOPORT 0xfff0          /* it doesn't matter where, as long
@@ -210,6 +211,12 @@ static struct cpufreq_driver powernow_k6_driver = {
 	.attr		= powernow_k6_attr,
 };
 
+static const struct x86_cpu_id powernow_k6_ids[] = {
+	{ X86_VENDOR_AMD, 5, 12 },
+	{ X86_VENDOR_AMD, 5, 13 },
+	{}
+};
+
 
 /**
  * powernow_k6_init - initializes the k6 PowerNow! CPUFreq driver
@@ -220,10 +227,7 @@ static struct cpufreq_driver powernow_k6_driver = {
  */
 static int __init powernow_k6_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
-
-	if ((c->x86_vendor != X86_VENDOR_AMD) || (c->x86 != 5) ||
-		((c->x86_model != 12) && (c->x86_model != 13)))
+	if (!x86_match_cpu(powernow_k6_ids))
 		return -ENODEV;
 
 	if (!request_region(POWERNOW_IOPORT, 16, "PowerNow!")) {
drivers/cpufreq/powernow-k7.c +8 -6
index d71d9f3..501d167 100644
@@ -28,6 +28,7 @@
 #include <asm/timer.h>		/* Needed for recalibrate_cpu_khz() */
 #include <asm/msr.h>
 #include <asm/system.h>
+#include <asm/cpu_device_id.h>
 
 #ifdef CONFIG_X86_POWERNOW_K7_ACPI
 #include <linux/acpi.h>
@@ -110,18 +111,19 @@ static int check_fsb(unsigned int fsbspeed)
 	return delta < 5;
 }
 
+static const struct x86_cpu_id powernow_k7_cpuids[] = {
+	{ X86_VENDOR_AMD, 7, },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, powernow_k7_cpuids);
+
 static int check_powernow(void)
 {
 	struct cpuinfo_x86 *c = &cpu_data(0);
 	unsigned int maxei, eax, ebx, ecx, edx;
 
-	if ((c->x86_vendor != X86_VENDOR_AMD) || (c->x86 != 6)) {
-#ifdef MODULE
-		printk(KERN_INFO PFX "This module only works with "
-				"AMD K7 CPUs\n");
-#endif
+	if (!x86_match_cpu(powernow_k7_cpuids))
 		return 0;
-	}
 
 	/* Get maximum capabilities */
 	maxei = cpuid_eax(0x80000000);
drivers/cpufreq/powernow-k8.c +16 -6
index bce576d..c87fabb 100644
@@ -37,6 +37,7 @@
 #include <linux/delay.h>
 
 #include <asm/msr.h>
+#include <asm/cpu_device_id.h>
 
 #include <linux/acpi.h>
 #include <linux/mutex.h>
@@ -514,6 +515,18 @@ static int core_voltage_post_transition(struct powernow_k8_data *data,
 	return 0;
 }
 
+static const struct x86_cpu_id powernow_k8_ids[] = {
+	{ X86_VENDOR_AMD, 0xf, },
+	{ X86_VENDOR_AMD, 0x10, },
+	{ X86_VENDOR_AMD, 0x11, },
+	{ X86_VENDOR_AMD, 0x12, },
+	{ X86_VENDOR_AMD, 0x13, },
+	{ X86_VENDOR_AMD, 0x14, },
+	/* Add more here? */
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, powernow_k8_ids);
+
 static void check_supported_cpu(void *_rc)
 {
 	u32 eax, ebx, ecx, edx;
@@ -521,13 +534,7 @@ static void check_supported_cpu(void *_rc)
 
 	*rc = -ENODEV;
 
-	if (__this_cpu_read(cpu_info.x86_vendor) != X86_VENDOR_AMD)
-		return;
-
 	eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
-	if (((eax & CPUID_XFAM) != CPUID_XFAM_K8) &&
-	    ((eax & CPUID_XFAM) < CPUID_XFAM_10H))
-		return;
 
 	if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) {
 		if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) ||
@@ -1542,6 +1549,9 @@ static int __cpuinit powernowk8_init(void)
 	unsigned int i, supported_cpus = 0, cpu;
 	int rv;
 
+	if (!x86_match_cpu(powernow_k8_ids))
+		return -ENODEV;
+
 	for_each_online_cpu(i) {
 		int rc;
 		smp_call_function_single(i, check_supported_cpu, &rc, 1);
drivers/cpufreq/sc520_freq.c +8 -6
index 1e205e6..e42e073 100644
@@ -22,6 +22,7 @@
 #include <linux/timex.h>
 #include <linux/io.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/msr.h>
 
 #define MMCR_BASE	0xfffef000	/* The default base address */
@@ -150,18 +151,19 @@ static struct cpufreq_driver sc520_freq_driver = {
 	.attr	= sc520_freq_attr,
 };
 
+static const struct x86_cpu_id sc520_ids[] = {
+	{ X86_VENDOR_AMD, 4, 9 },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, sc520_ids);
 
 static int __init sc520_freq_init(void)
 {
-	struct cpuinfo_x86 *c = &cpu_data(0);
 	int err;
 
-	/* Test if we have the right hardware */
-	if (c->x86_vendor != X86_VENDOR_AMD ||
-	    c->x86 != 4 || c->x86_model != 9) {
-		pr_debug("no Elan SC520 processor found!\n");
+	if (!x86_match_cpu(sc520_ids))
 		return -ENODEV;
-	}
+
 	cpuctl = ioremap((unsigned long)(MMCR_BASE + OFFS_CPUCTL), 1);
 	if (!cpuctl) {
 		printk(KERN_ERR "sc520_freq: error: failed to remap memory\n");
drivers/cpufreq/speedstep-centrino.c +20 -4
index 6ea3455..3a953d5 100644
@@ -25,6 +25,7 @@
 #include <asm/msr.h>
 #include <asm/processor.h>
 #include <asm/cpufeature.h>
+#include <asm/cpu_device_id.h>
 
 #define PFX		"speedstep-centrino: "
 #define MAINTAINER	"[email protected]"
@@ -595,6 +596,24 @@ static struct cpufreq_driver centrino_driver = {
 	.owner		= THIS_MODULE,
 };
 
+/*
+ * This doesn't replace the detailed checks above because
+ * the generic CPU IDs don't have a way to match for steppings
+ * or ASCII model IDs.
+ */
+static const struct x86_cpu_id centrino_ids[] = {
+	{ X86_VENDOR_INTEL, 6, 9, X86_FEATURE_EST },
+	{ X86_VENDOR_INTEL, 6, 13, X86_FEATURE_EST },
+	{ X86_VENDOR_INTEL, 6, 13, X86_FEATURE_EST },
+	{ X86_VENDOR_INTEL, 6, 13, X86_FEATURE_EST },
+	{ X86_VENDOR_INTEL, 15, 3, X86_FEATURE_EST },
+	{ X86_VENDOR_INTEL, 15, 4, X86_FEATURE_EST },
+	{}
+};
+#if 0
+/* Autoload or not? Do not for now. */
+MODULE_DEVICE_TABLE(x86cpu, centrino_ids);
+#endif
 
 /**
  * centrino_init - initializes the Enhanced SpeedStep CPUFreq driver
@@ -612,11 +631,8 @@ static struct cpufreq_driver centrino_driver = {
  */
 static int __init centrino_init(void)
 {
-	struct cpuinfo_x86 *cpu = &cpu_data(0);
-
-	if (!cpu_has(cpu, X86_FEATURE_EST))
+	if (!x86_match_cpu(centrino_ids))
 		return -ENODEV;
-
 	return cpufreq_register_driver(&centrino_driver);
 }
 
drivers/cpufreq/speedstep-ich.c +15 -0
index a748ce7..7432b3a 100644
@@ -25,6 +25,8 @@
 #include <linux/pci.h>
 #include <linux/sched.h>
 
+#include <asm/cpu_device_id.h>
+
 #include "speedstep-lib.h"
 
 
@@ -388,6 +390,16 @@ static struct cpufreq_driver speedstep_driver = {
 	.attr	= speedstep_attr,
 };
 
+static const struct x86_cpu_id ss_smi_ids[] = {
+	{ X86_VENDOR_INTEL, 6, 0xb, },
+	{ X86_VENDOR_INTEL, 6, 0x8, },
+	{ X86_VENDOR_INTEL, 15, 2 },
+	{}
+};
+#if 0
+/* Autoload or not? Do not for now. */
+MODULE_DEVICE_TABLE(x86cpu, ss_smi_ids);
+#endif
 
 /**
  * speedstep_init - initializes the SpeedStep CPUFreq driver
@@ -398,6 +410,9 @@ static struct cpufreq_driver speedstep_driver = {
  */
 static int __init speedstep_init(void)
 {
+	if (!x86_match_cpu(ss_smi_ids))
+		return -ENODEV;
+
 	/* detect processor */
 	speedstep_processor = speedstep_detect_processor();
 	if (!speedstep_processor) {
drivers/cpufreq/speedstep-lib.c +1 -0
index 8af2d2f..7047821 100644
@@ -249,6 +249,7 @@ EXPORT_SYMBOL_GPL(speedstep_get_frequency);
  *                 DETECT SPEEDSTEP-CAPABLE PROCESSOR                *
  *********************************************************************/
 
+/* Keep in sync with the x86_cpu_id tables in the different modules */
 unsigned int speedstep_detect_processor(void)
 {
 	struct cpuinfo_x86 *c = &cpu_data(0);
drivers/cpufreq/speedstep-smi.c +15 -0
index c76ead3..6a457fc 100644
@@ -20,6 +20,7 @@
 #include <linux/delay.h>
 #include <linux/io.h>
 #include <asm/ist.h>
+#include <asm/cpu_device_id.h>
 
 #include "speedstep-lib.h"
 
@@ -379,6 +380,17 @@ static struct cpufreq_driver speedstep_driver = {
 	.attr		= speedstep_attr,
 };
 
+static const struct x86_cpu_id ss_smi_ids[] = {
+	{ X86_VENDOR_INTEL, 6, 0xb, },
+	{ X86_VENDOR_INTEL, 6, 0x8, },
+	{ X86_VENDOR_INTEL, 15, 2 },
+	{}
+};
+#if 0
+/* Not auto loaded currently */
+MODULE_DEVICE_TABLE(x86cpu, ss_smi_ids);
+#endif
+
 /**
  * speedstep_init - initializes the SpeedStep CPUFreq driver
  *
@@ -388,6 +400,9 @@ static struct cpufreq_driver speedstep_driver = {
  */
 static int __init speedstep_init(void)
 {
+	if (!x86_match_cpu(ss_smi_ids))
+		return -ENODEV;
+
 	speedstep_processor = speedstep_detect_processor();
 
 	switch (speedstep_processor) {
Signature · 1 line
1.7.7.4
#18 lore raw

Re: [PATCH 08/10] cpufreq: Add support for x86 cpuinfo auto loading

Quoted 37 lines · Andi Kleen
drivers/cpufreq/gx-suspmod.c +9 -0
index ffe1f2c..05d0f7f 100644
@@ -82,6 +82,7 @@
 #include <linux/errno.h>
 #include <linux/slab.h>
 
+#include <asm/cpu_device_id.h>
 #include <asm/processor-cyrix.h>
 
 /* PCI config registers, all at F0 */
@@ -171,12 +172,20 @@ static struct pci_device_id gx_chipset_tbl[] __initdata = {
 	{ PCI_VDEVICE(CYRIX, PCI_DEVICE_ID_CYRIX_5510), },
 	{ 0, },
 };
+/* MODULE_DEVICE_TABLE here too? But the CPU should be enough. */
 
 static void gx_write_byte(int reg, int value)
 {
 	pci_write_config_byte(gx_params->cs55x0, reg, value);
 }
 
+static const struct x86_cpu_id gx_ids[] = {
+	{ X86_VENDOR_NSC, },
+	{ X86_VENDOR_CYRIX },
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, gx_ids);
+
 /**
  * gx_detect_chipset:
  *
@@ -185,12 +194,8 @@ static __init struct pci_dev *gx_detect_chipset(void)
 {
 	struct pci_dev *gx_pci = NULL;
 

drivers/cpufreq/gx-suspmod.c static __init struct pci_dev *gx_detect_chipset(void) · line 201

Quoted 10 lines
drivers/cpufreq/gx-suspmod.c +1 -5
-	/* check if CPU is a MediaGX or a Geode. */
-	if ((boot_cpu_data.x86_vendor != X86_VENDOR_NSC) &&
-	    (boot_cpu_data.x86_vendor != X86_VENDOR_CYRIX)) {
-		pr_debug("error: no MediaGX/Geode processor found!\n");
+	if (!x86_match_cpu(gx_ids))
 		return NULL;
-	}
 
 	/* detect which companion chip is used */
 	for_each_pci_dev(gx_pci) {

Looking closer at this one, I think we could possibly just autoload based on the PCI IDs in gx_chipset_tbl, and remove the cpu vendor checking entirely.
Pretty sure those Cyrix PCI devices won't show up anywhere else.

Otherwise, we could narrow the range this gets loaded on to Cyrix Fam 4 model 5, and NatSemi Fam 5, model 4. (afaik, that covers all the possible cpus this driver works on)

Dave

#19 lore raw

Re: [PATCH 08/10] cpufreq: Add support for x86 cpuinfo auto loading

Quoted 3 lines

Looking closer at this one, I think we could possibly just autoload based on the PCI IDs in gx_chipset_tbl, and remove the cpu vendor checking entirely.
Pretty sure those Cyrix PCI devices won't show up anywhere else.

Ok. Will do that.

Quoted 2 lines

Otherwise, we could narrow the range this gets loaded on to Cyrix Fam 4 model 5, and NatSemi Fam 5, model 4. (afaik, that covers all the possible cpus this driver works on)

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
maintainer
#20 lore raw

Re: [PATCH 08/10] cpufreq: Add support for x86 cpuinfo auto loading

Quoted 6 lines · Andi Kleen
Quoted 3 lines

Looking closer at this one, I think we could possibly just autoload based on the PCI IDs in gx_chipset_tbl, and remove the cpu vendor checking entirely.
Pretty sure those Cyrix PCI devices won't show up anywhere else.

Ok. Will do that.

I'll drop this subpatch then... the PCI device one should be orthogonal.

-hpa

author
#21 lore raw

PATCH 9/10 x86: autoload microcode driver on Intel and AMD systems

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Don't try to describe the actual models for now.

---
 arch/x86/kernel/microcode_core.c |   15 +++++++++++++++
 1 files changed, 15 insertions(+), 0 deletions(-)
arch/x86/kernel/microcode_core.c +15 -0
index 9d46f5e..6911a41 100644
@@ -86,6 +86,7 @@
 
 #include <asm/microcode.h>
 #include <asm/processor.h>
+#include <asm/cpu_device_id.h>
 
 MODULE_DESCRIPTION("Microcode Update Driver");
 MODULE_AUTHOR("Tigran Aivazian <[email protected]>");
@@ -502,6 +503,20 @@ static struct notifier_block __refdata mc_cpu_notifier = {
 	.notifier_call	= mc_cpu_callback,
 };
 
+#ifdef MODULE
+/* Autoload on Intel and AMD systems */
+static const struct x86_cpu_id microcode_id[] = {
+#ifdef CONFIG_MICROCODE_INTEL
+       { X86_VENDOR_INTEL, X86_VENDOR_ANY, X86_MODEL_ANY, },
+#endif
+#ifdef CONFIG_MICROCODE_AMD
+       { X86_VENDOR_AMD, X86_VENDOR_ANY, X86_MODEL_ANY, },
+#endif
+       {}
+};
+MODULE_DEVICE_TABLE(x86cpu, microcode_id);
+#endif
+
 static int __init microcode_init(void)
 {
 	struct cpuinfo_x86 *c = &cpu_data(0);
Signature · 1 line
1.7.7.4