Skip to contentSkip to conversation

Add support for Hygon Dhyana Family 18h processor

PATCH Superseded Pu Wen 3 participants 51 messages Feb 21, 2019 – Mar 19, 2019

v2 of 8 Superseded Latest is v8 →

lore
Share & tools
lore

Pu Wen

author
#0 lore raw

v2 cover Add support for Hygon Dhyana Family 18h processor

As a new x86 CPU Vendor, Chengdu Haiguang IC Design Co., Ltd (Hygon) is a Joint Venture between AMD and Haiguang Information Technology Co., Ltd., and aims at providing high performance x86 processor for China server market.

The first generation Hygon's processor(Dhyana) originates from AMD technology and shares most of the architecture with AMD's family 17h, but with different CPU vendor ID("HygonGenuine") and family series
number: 18h (Hygon will negotiate with AMD to make sure that only Hygon will use family 18h).

To enable support of Xen to Hygon Dhyana CPU, we add a new vendor type (X86_VENDOR_HYGON, with value of 5), and share most of the codes with AMD family 17h.

This patch series have been applied and tested successfully on Hygon Dhyana processor. Also tested on AMD EPYC (family 17h) processor, it works fine and makes no harm to the existing codes.

v1->v2:
- Rebased on 4.20.0-rc3 and tested against it.
- Move opt_cpuid_mask_l7s0_(eax/ebx) to common.c.
- Insert Hygon cases after AMD ones instead of above.
- Remove (rd/wr)msr_hygon_safe and use (rd/wr)msr_safe instead.
- Remove wrmsr_hygon and use wrmsrl instead.
- Remove the unnecessary change to xstate.
- Refine some codes and comments.
- Add Acked-by from Jan Beulich for x86/traps.
- Add Acked-by from Wei Liu for tools/libxc.

Pu Wen (14):
x86/cpu: Create Hygon Dhyana architecture support file x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2 x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU x86/cpu/mce: Add Hygon Dhyana support to the MCA infrastructure x86/spec_ctrl: Add Hygon Dhyana to the respective mitigation machinery x86/apic: Add Hygon Dhyana support
x86/acpi: Add Hygon Dhyana support
x86/iommu: Add Hygon Dhyana support
x86/pv: Add Hygon Dhyana support to emulate MSRs access x86/domain: Add Hygon Dhyana support
x86/domctl: Add Hygon Dhyana support
x86/traps: Add Hygon Dhyana support
x86/cpuid: Add Hygon Dhyana support
tools/libxc: Add Hygon Dhyana support

 tools/libxc/xc_cpuid_x86.c             |  16 ++-
 xen/arch/x86/acpi/cpu_idle.c           |   3 +-
 xen/arch/x86/acpi/cpufreq/cpufreq.c    |   8 +-
 xen/arch/x86/acpi/cpufreq/powernow.c   |   3 +-
 xen/arch/x86/apic.c                    |   5 +
 xen/arch/x86/cpu/Makefile              |   1 +
 xen/arch/x86/cpu/amd.c                 |   5 -
 xen/arch/x86/cpu/common.c              |   9 +-
 xen/arch/x86/cpu/cpu.h                 |   2 +
 xen/arch/x86/cpu/hygon.c               | 248 +++++++++++++++++++++++++++++++++
 xen/arch/x86/cpu/mcheck/amd_nonfatal.c |   5 +-
 xen/arch/x86/cpu/mcheck/mce.c          |   6 +-
 xen/arch/x86/cpu/mcheck/mce_amd.c      |   5 +-
 xen/arch/x86/cpu/mcheck/non-fatal.c    |   3 +-
 xen/arch/x86/cpu/mcheck/vmce.c         |   2 +
 xen/arch/x86/cpu/mtrr/generic.c        |   5 +-
 xen/arch/x86/cpu/vpmu.c                |   2 +
 xen/arch/x86/cpu/vpmu_amd.c            |   2 +
 xen/arch/x86/cpuid.c                   |  10 +-
 xen/arch/x86/dom0_build.c              |   3 +-
 xen/arch/x86/domain.c                  |   9 +-
 xen/arch/x86/domctl.c                  |  13 +-
 xen/arch/x86/pv/emul-priv-op.c         |  19 ++-
 xen/arch/x86/spec_ctrl.c               |   6 +-
 xen/arch/x86/traps.c                   |   3 +
 xen/include/asm-x86/iommu.h            |   1 +
 xen/include/asm-x86/x86-vendors.h      |   3 +-
 27 files changed, 354 insertions(+), 43 deletions(-)
 create mode 100644 xen/arch/x86/cpu/hygon.c
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#1 lore raw

PATCH v2 1/14 x86/cpu: Create Hygon Dhyana architecture support file

Add x86 architecture support for a new processor: Hygon Dhyana Family 18h. Carve out initialization codes from amd.c needed by Dhyana into a separate file hygon.c by removing unnecessary codes and make Hygon initialization codes more clear.

To identify Hygon Dhyana CPU, add a new vendor type X86_VENDOR_HYGON for system recognition.

As opt_cpuid_mask_l7s0_eax and opt_cpuid_mask_l7s0_ebx are used by both AMD and Hygon, so move them to common.c.

---
 xen/arch/x86/cpu/Makefile         |   1 +
 xen/arch/x86/cpu/amd.c            |   5 -
 xen/arch/x86/cpu/common.c         |   6 +
 xen/arch/x86/cpu/cpu.h            |   2 +
 xen/arch/x86/cpu/hygon.c          | 248 ++++++++++++++++++++++++++++++++++++++
 xen/include/asm-x86/x86-vendors.h |   3 +-
 6 files changed, 259 insertions(+), 6 deletions(-)
 create mode 100644 xen/arch/x86/cpu/hygon.c
xen/arch/x86/cpu/Makefile +1 -0
index 34a01ca..1db7d88 100644
@@ -8,4 +8,5 @@ obj-y += intel.o
 obj-y += intel_cacheinfo.o
 obj-y += mwait-idle.o
 obj-y += shanghai.o
+obj-y += hygon.o
 obj-y += vpmu.o vpmu_amd.o vpmu_intel.o
xen/arch/x86/cpu/amd.c +0 -5
index c790416..4c595cf 100644
@@ -32,11 +32,6 @@
 static char __initdata opt_famrev[14];
 string_param("cpuid_mask_cpu", opt_famrev);
 
-static unsigned int __initdata opt_cpuid_mask_l7s0_eax = ~0u;
-integer_param("cpuid_mask_l7s0_eax", opt_cpuid_mask_l7s0_eax);
-static unsigned int __initdata opt_cpuid_mask_l7s0_ebx = ~0u;
-integer_param("cpuid_mask_l7s0_ebx", opt_cpuid_mask_l7s0_ebx);
-
 static unsigned int __initdata opt_cpuid_mask_thermal_ecx = ~0u;
 integer_param("cpuid_mask_thermal_ecx", opt_cpuid_mask_thermal_ecx);
 
xen/arch/x86/cpu/common.c +6 -0
index de6c5c9..5bab845 100644
@@ -36,6 +36,11 @@ integer_param("cpuid_mask_ext_ecx", opt_cpuid_mask_ext_ecx);
 unsigned int opt_cpuid_mask_ext_edx = ~0u;
 integer_param("cpuid_mask_ext_edx", opt_cpuid_mask_ext_edx);
 
+unsigned int  opt_cpuid_mask_l7s0_eax = ~0u;
+integer_param("cpuid_mask_l7s0_eax", opt_cpuid_mask_l7s0_eax);
+unsigned int  opt_cpuid_mask_l7s0_ebx = ~0u;
+integer_param("cpuid_mask_l7s0_ebx", opt_cpuid_mask_l7s0_ebx);
+
 unsigned int __initdata expected_levelling_cap;
 unsigned int __read_mostly levelling_caps;
 
@@ -704,6 +709,7 @@ void __init early_cpu_init(void)
 {
 	intel_cpu_init();
 	amd_init_cpu();
+	hygon_init_cpu();
 	centaur_init_cpu();
 	shanghai_init_cpu();
 	early_cpu_detect();
xen/arch/x86/cpu/cpu.h +2 -0
index 2fcb931..9ea53e5 100644
@@ -13,11 +13,13 @@ extern bool_t opt_arat;
 extern unsigned int opt_cpuid_mask_ecx, opt_cpuid_mask_edx;
 extern unsigned int opt_cpuid_mask_xsave_eax;
 extern unsigned int opt_cpuid_mask_ext_ecx, opt_cpuid_mask_ext_edx;
+extern unsigned int opt_cpuid_mask_l7s0_eax, opt_cpuid_mask_l7s0_ebx;
 
 extern int get_model_name(struct cpuinfo_x86 *c);
 extern void display_cacheinfo(struct cpuinfo_x86 *c);
 
 int intel_cpu_init(void);
 int amd_init_cpu(void);
+int hygon_init_cpu(void);
 int centaur_init_cpu(void);
 int shanghai_init_cpu(void);
xen/arch/x86/cpu/hygon.c new file +248 -0

248 lines not on this page · raw message

xen/include/asm-x86/x86-vendors.h +2 -1
index 38a81c3..fa1cbb4 100644
@@ -9,6 +9,7 @@
 #define X86_VENDOR_AMD 2
 #define X86_VENDOR_CENTAUR 3
 #define X86_VENDOR_SHANGHAI 4
-#define X86_VENDOR_NUM 5
+#define X86_VENDOR_HYGON 5
+#define X86_VENDOR_NUM 6
 
 #endif	/* __XEN_X86_VENDORS_H__ */
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#2 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 11 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

Add x86 architecture support for a new processor: Hygon Dhyana Family 18h. Carve out initialization codes from amd.c needed by Dhyana into a separate file hygon.c by removing unnecessary codes and make Hygon initialization codes more clear.

To identify Hygon Dhyana CPU, add a new vendor type X86_VENDOR_HYGON for system recognition.

As opt_cpuid_mask_l7s0_eax and opt_cpuid_mask_l7s0_ebx are used by both AMD and Hygon, so move them to common.c.

I'm not convinced of this - it'll get in the way of introducing something like Linux'es CONFIG_CPU_SUP_*. Our command line parsing logic handles multiple instances of an option quite fine I think, so I'd prefer if these were kept static to both files. Or did Andrew ask you to do this?

xen/arch/x86/cpu/Makefile obj-y += intel.o · line 12

Quoted 8 lines
xen/arch/x86/cpu/Makefile +1 -0
@@ -8,4 +8,5 @@ obj-y += intel.o
 obj-y += intel_cacheinfo.o
 obj-y += mwait-idle.o
 obj-y += shanghai.o
+obj-y += hygon.o
 obj-y += vpmu.o vpmu_amd.o vpmu_intel.o

Please insert at the alphabetically correct place.

Quoted 6 lines
xen/arch/x86/cpu/cpu.h +0 -0
@@ -13,11 +13,13 @@ extern bool_t opt_arat;
 extern unsigned int opt_cpuid_mask_ecx, opt_cpuid_mask_edx;
 extern unsigned int opt_cpuid_mask_xsave_eax;
 extern unsigned int opt_cpuid_mask_ext_ecx, opt_cpuid_mask_ext_edx;

xen/arch/x86/cpu/cpu.h extern bool_t opt_arat; · line 25

Quoted 10 lines
xen/arch/x86/cpu/cpu.h +2 -0
+extern unsigned int opt_cpuid_mask_l7s0_eax, opt_cpuid_mask_l7s0_ebx;
 
 extern int get_model_name(struct cpuinfo_x86 *c);
 extern void display_cacheinfo(struct cpuinfo_x86 *c);
 
 int intel_cpu_init(void);
 int amd_init_cpu(void);
+int hygon_init_cpu(void);
 int centaur_init_cpu(void);
 int shanghai_init_cpu(void);

Whereas here you should really put the new declaration at the end.

Quoted 17 lines
xen/arch/x86/cpu/hygon.c new file +14 -0
@@ -0,0 +1,248 @@
+#include <xen/init.h>
+#include <asm/processor.h>
+#include <asm/hvm/support.h>
+#include <asm/spec_ctrl.h>
+
+#include "cpu.h"
+
+/*
+ * Sets caps in expected_levelling_cap, probes for the specified mask MSR, and
+ * set caps in levelling_caps if it is found.  Returns the default value.
+ */
+static uint64_t __init _probe_mask_msr(unsigned int msr, uint64_t caps)
+{
+	uint64_t value;

xen/arch/x86/cpu/hygon.c line 24

Quoted 10 lines
xen/arch/x86/cpu/hygon.c new file +10 -0
+
+	expected_levelling_cap |= caps;
+
+	if ((rdmsr_safe(msr, value) == 0) && (wrmsr_safe(msr, value) == 0))
+		levelling_caps |= caps;
+
+	return value;
+}
+
+/* Probe for the existance of the expected masking MSRs. */

Please don't drop the other part of the original comment.

xen/arch/x86/cpu/hygon.c line 36

Quoted 12 lines
xen/arch/x86/cpu/hygon.c new file +12 -0
+static void __init noinline probe_masking_msrs(void)
+{
+	const struct cpuinfo_x86 *c = &boot_cpu_data;
+
+	/* Work out which masking MSRs we should have. */
+	cpuidmask_defaults._1cd =
+		_probe_mask_msr(MSR_K8_FEATURE_MASK, LCAP_1cd);
+	cpuidmask_defaults.e1cd =
+		_probe_mask_msr(MSR_K8_EXT_FEATURE_MASK, LCAP_e1cd);
+	if (c->cpuid_level >= 7)
+		cpuidmask_defaults._7ab0 =
+			_probe_mask_msr(MSR_AMD_L7S0_FEATURE_MASK, LCAP_7ab0);

There's more relevant code here in the original function.

Quoted 112 lines
xen/arch/x86/cpu/hygon.c new file +112 -0
+}
+
+/*
+ * Context switch CPUID masking state to the next domain.  Only called if
+ * CPUID Faulting isn't available, but masking MSRs have been detected.  A
+ * parameter of NULL is used to context switch to the default host state (by
+ * the cpu bringup-code, crash path, etc).
+ */
+static void hygon_ctxt_switch_masking(const struct vcpu *next)
+{
+	struct cpuidmasks *these_masks = &this_cpu(cpuidmasks);
+	const struct domain *nextd = next ? next->domain : NULL;
+	const struct cpuidmasks *masks =
+		(nextd && is_pv_domain(nextd) && nextd->arch.pv.cpuidmasks)
+		? nextd->arch.pv.cpuidmasks : &cpuidmask_defaults;
+
+	if ((levelling_caps & LCAP_1cd) == LCAP_1cd) {
+		uint64_t val = masks->_1cd;
+
+		/*
+		 * OSXSAVE defaults to 1, which causes fast-forwarding of
+		 * Xen's real setting.  Clobber it if disabled by the guest
+		 * kernel.
+		 */
+		if (next && is_pv_vcpu(next) && !is_idle_vcpu(next) &&
+		    !(next->arch.pv.ctrlreg[4] & X86_CR4_OSXSAVE))
+			val &= ~((uint64_t)cpufeat_mask(X86_FEATURE_OSXSAVE) << 32);
+
+		if (unlikely(these_masks->_1cd != val)) {
+			wrmsrl(MSR_K8_FEATURE_MASK, val);
+			these_masks->_1cd = val;
+		}
+	}
+
+#define LAZY(cap, msr, field)						\
+	({								\
+		if (unlikely(these_masks->field != masks->field) &&	\
+		    ((levelling_caps & cap) == cap)) {							\
+			wrmsrl(msr, masks->field);			\
+			these_masks->field = masks->field;		\
+		}							\
+	})
+
+	LAZY(LCAP_e1cd, MSR_K8_EXT_FEATURE_MASK,   e1cd);
+	LAZY(LCAP_7ab0, MSR_AMD_L7S0_FEATURE_MASK, _7ab0);
+#undef LAZY
+}
+
+/*
+ * Mask the features and extended features returned by CPUID.  Parameters are
+ * set from the boot line via user-defined masks.
+ */
+static void __init noinline hygon_init_levelling(void)
+{
+	probe_masking_msrs();
+
+	if ((levelling_caps & LCAP_1cd) == LCAP_1cd) {
+		uint32_t ecx, edx, tmp;
+
+		cpuid(0x00000001, &tmp, &tmp, &ecx, &edx);
+
+		if (~(opt_cpuid_mask_ecx & opt_cpuid_mask_edx)) {
+			ecx &= opt_cpuid_mask_ecx;
+			edx &= opt_cpuid_mask_edx;
+		}
+
+		/* Fast-forward bits - Must be set. */
+		if (ecx & cpufeat_mask(X86_FEATURE_XSAVE))
+			ecx |= cpufeat_mask(X86_FEATURE_OSXSAVE);
+		edx |= cpufeat_mask(X86_FEATURE_APIC);
+
+		/* Allow the HYPERVISOR bit to be set via guest policy. */
+		ecx |= cpufeat_mask(X86_FEATURE_HYPERVISOR);
+
+		cpuidmask_defaults._1cd = ((uint64_t)ecx << 32) | edx;
+	}
+
+	if ((levelling_caps & LCAP_e1cd) == LCAP_e1cd) {
+		uint32_t ecx, edx, tmp;
+
+		cpuid(0x80000001, &tmp, &tmp, &ecx, &edx);
+
+		if (~(opt_cpuid_mask_ext_ecx & opt_cpuid_mask_ext_edx)) {
+			ecx &= opt_cpuid_mask_ext_ecx;
+			edx &= opt_cpuid_mask_ext_edx;
+		}
+
+		/* Fast-forward bits - Must be set. */
+		edx |= cpufeat_mask(X86_FEATURE_APIC);
+
+		cpuidmask_defaults.e1cd = ((uint64_t)ecx << 32) | edx;
+	}
+
+	if ((levelling_caps & LCAP_7ab0) == LCAP_7ab0) {
+		uint32_t eax, ebx, tmp;
+
+		cpuid(0x00000007, &eax, &ebx, &tmp, &tmp);
+
+		if (~(opt_cpuid_mask_l7s0_eax & opt_cpuid_mask_l7s0_ebx)) {
+			eax &= opt_cpuid_mask_l7s0_eax;
+			ebx &= opt_cpuid_mask_l7s0_ebx;
+		}
+
+		cpuidmask_defaults._7ab0 &= ((uint64_t)eax << 32) | ebx;
+	}
+
+	if (opt_cpu_info) {
+		printk(XENLOG_INFO "Levelling caps: %#x\n", levelling_caps);
+		printk(XENLOG_INFO
+		       "MSR defaults: 1d 0x%08x, 1c 0x%08x, e1d 0x%08x, "
+		       "e1c 0x%08x, 7a0 0x%08x, 7b0 0x%08x\n",
+		       (uint32_t)cpuidmask_defaults._1cd,

xen/arch/x86/cpu/hygon.c line 158

Quoted 10 lines
xen/arch/x86/cpu/hygon.c new file +10 -0
+		       (uint32_t)(cpuidmask_defaults._1cd >> 32),
+		       (uint32_t)cpuidmask_defaults.e1cd,
+		       (uint32_t)(cpuidmask_defaults.e1cd >> 32),
+		       (uint32_t)(cpuidmask_defaults._7ab0 >> 32),
+		       (uint32_t)cpuidmask_defaults._7ab0);
+	}
+
+	if (levelling_caps)
+		ctxt_switch_masking = hygon_ctxt_switch_masking;
+}

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

xen/arch/x86/cpu/hygon.c line 200

Quoted 12 lines
xen/arch/x86/cpu/hygon.c new file +12 -0
+static void init_hygon(struct cpuinfo_x86 *c)
+{
+	u32 l, h;
+	unsigned long long value;
+
+	/* Attempt to set lfence to be Dispatch Serialising. */
+	if (rdmsr_safe(MSR_AMD64_DE_CFG, value))
+		/* Unable to read.  Assume the safer default. */
+		__clear_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);
+	else if (value & AMD64_DE_CFG_LFENCE_SERIALISE)
+		/* Already dispatch serialising. */
+		__set_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);

Didn't you cut off too much from the original code (which again may better be shared, by splitting out into a function)?

Quoted 16 lines
xen/arch/x86/cpu/hygon.c new file +16 -0
+	/*
+	 * If the user has explicitly chosen to disable Memory Disambiguation
+	 * to mitigiate Speculative Store Bypass, poke the appropriate MSR.
+	 */
+ 	if (opt_ssbd && !rdmsr_safe(MSR_AMD64_LS_CFG, value)) {
+		value |= 1ull << 10;
+		wrmsr_safe(MSR_AMD64_LS_CFG, value);
+	}
+
+	display_cacheinfo(c);
+
+	if (cpu_has(c, X86_FEATURE_ITSC)) {
+		__set_bit(X86_FEATURE_CONSTANT_TSC, c->x86_capability);
+		__set_bit(X86_FEATURE_NONSTOP_TSC, c->x86_capability);
+		__set_bit(X86_FEATURE_TSC_RELIABLE, c->x86_capability);
+	}

xen/arch/x86/cpu/hygon.c line 227

Quoted 10 lines
xen/arch/x86/cpu/hygon.c new file +10 -0
+
+	c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
+
+	hygon_get_topology(c);
+
+	/* Hygon CPUs do not support SYSENTER outside of legacy mode. */
+	__clear_bit(X86_FEATURE_SEP, c->x86_capability);
+
+	/* Hygon processors have APIC timer running in deep C states. */
+	if ( opt_arat )

Please don't copy the bad spaces inside the parentheses here.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#3 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 9 lines · Jan Beulich
Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

As opt_cpuid_mask_l7s0_eax and opt_cpuid_mask_l7s0_ebx are used by both AMD and Hygon, so move them to common.c.

I'm not convinced of this - it'll get in the way of introducing something like Linux'es CONFIG_CPU_SUP_*. Our command line parsing logic handles multiple instances of an option quite fine I think, so I'd prefer if these were kept static to both files. Or did Andrew ask you to do this?

Yes, Andrew asked that these should be moved from the AMD specific code into the common cpu code (alongside the other masks) rather than duplicated here.

Quoted 10 lines

xen/arch/x86/cpu/Makefile obj-y += intel.o · line 12

Quoted 8 lines
xen/arch/x86/cpu/Makefile +1 -0
@@ -8,4 +8,5 @@ obj-y += intel.o
  obj-y += intel_cacheinfo.o
  obj-y += mwait-idle.o
  obj-y += shanghai.o
+obj-y += hygon.o
  obj-y += vpmu.o vpmu_amd.o vpmu_intel.o

Please insert at the alphabetically correct place.

OK.

...

Quoted 7 lines
Quoted 5 lines

int intel_cpu_init(void);
int amd_init_cpu(void);
+int hygon_init_cpu(void);
int centaur_init_cpu(void);
int shanghai_init_cpu(void);

Whereas here you should really put the new declaration at the end.

OK.

...

Quoted 3 lines
Quoted 1 line

+/* Probe for the existance of the expected masking MSRs. */

Please don't drop the other part of the original comment.

OK. Will revert to the original one.

Quoted 6 lines
Quoted 6 lines
xen/arch/x86/cpu/Makefile +6 -0
+static void __init noinline probe_masking_msrs(void)
+{
+	const struct cpuinfo_x86 *c = &boot_cpu_data;
+
+	/* Work out which masking MSRs we should have. */
+	cpuidmask_defaults._1cd =
Quoted 8 lines

xen/arch/x86/cpu/Makefile

Quoted 6 lines
xen/arch/x86/cpu/Makefile +6 -0
+		_probe_mask_msr(MSR_K8_FEATURE_MASK, LCAP_1cd);
+	cpuidmask_defaults.e1cd =
+		_probe_mask_msr(MSR_K8_EXT_FEATURE_MASK, LCAP_e1cd);
+	if (c->cpuid_level >= 7)
+		cpuidmask_defaults._7ab0 =
+			_probe_mask_msr(MSR_AMD_L7S0_FEATURE_MASK, LCAP_7ab0);

There's more relevant code here in the original function.

The code is used for family 15h. Hygon CPU do not need it.

...

Quoted 11 lines
Quoted 11 lines
xen/arch/x86/cpu/Makefile +11 -0
+	if (opt_cpu_info) {
+		printk(XENLOG_INFO "Levelling caps: %#x\n", levelling_caps);
+		printk(XENLOG_INFO
+		       "MSR defaults: 1d 0x%08x, 1c 0x%08x, e1d 0x%08x, "
+		       "e1c 0x%08x, 7a0 0x%08x, 7b0 0x%08x\n",
+		       (uint32_t)cpuidmask_defaults._1cd,
+		       (uint32_t)(cpuidmask_defaults._1cd >> 32),
+		       (uint32_t)cpuidmask_defaults.e1cd,
+		       (uint32_t)(cpuidmask_defaults.e1cd >> 32),
+		       (uint32_t)(cpuidmask_defaults._7ab0 >> 32),
+		       (uint32_t)cpuidmask_defaults._7ab0);
Quoted 8 lines

xen/arch/x86/cpu/Makefile

Quoted 5 lines
xen/arch/x86/cpu/Makefile +5 -0
+	}
+
+	if (levelling_caps)
+		ctxt_switch_masking = hygon_ctxt_switch_masking;
+}

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance. There are many codes to support AMD's different families, the family/model/stepping checking are deeply embedded in codes. If we add Hygon family/model/stepping checking in future version, it will make code interleaved together and hard to maintain.

In fact in Linux we keep hygon.c separated from AMD even though there are some duplicated codes at the moment. Andrew also agreed the logic that following suit with Linux.

Quoted 7 lines
Quoted 7 lines
xen/arch/x86/cpu/Makefile +7 -0
+static void init_hygon(struct cpuinfo_x86 *c)
+{
+	u32 l, h;
+	unsigned long long value;
+
+	/* Attempt to set lfence to be Dispatch Serialising. */
+	if (rdmsr_safe(MSR_AMD64_DE_CFG, value))
Quoted 8 lines

xen/arch/x86/cpu/Makefile

Quoted 5 lines
xen/arch/x86/cpu/Makefile +5 -0
+		/* Unable to read.  Assume the safer default. */
+		__clear_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);
+	else if (value & AMD64_DE_CFG_LFENCE_SERIALISE)
+		/* Already dispatch serialising. */
+		__set_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);

Didn't you cut off too much from the original code (which again may better be shared, by splitting out into a function)?

The cut codes are used for other AMD families except family 17h. Hygon Dhyana is derived from AMD Zen, so we only need the family 17h parts. This also make the init flow minimal.

Quoted 4 lines

xen/arch/x86/cpu/Makefile

Quoted 2 lines
xen/arch/x86/cpu/Makefile +2 -0
+	/* Hygon processors have APIC timer running in deep C states. */
+	if ( opt_arat )

Please don't copy the bad spaces inside the parentheses here.

OK. Thanks for the reminding.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#4 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 18 lines
Quoted 1 line
Quoted 1 line · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 15 lines · Jan Beulich

xen/arch/x86/cpu/hygon.c line 36

Quoted 13 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:
xen/arch/x86/cpu/hygon.c new file +12 -0
+static void __init noinline probe_masking_msrs(void)
+{
+	const struct cpuinfo_x86 *c = &boot_cpu_data;
+
+	/* Work out which masking MSRs we should have. */
+	cpuidmask_defaults._1cd =
+		_probe_mask_msr(MSR_K8_FEATURE_MASK, LCAP_1cd);
+	cpuidmask_defaults.e1cd =
+		_probe_mask_msr(MSR_K8_EXT_FEATURE_MASK, LCAP_e1cd);
+	if (c->cpuid_level >= 7)
+		cpuidmask_defaults._7ab0 =
+			_probe_mask_msr(MSR_AMD_L7S0_FEATURE_MASK, LCAP_7ab0);

There's more relevant code here in the original function.

Quoted 1 line

The code is used for family 15h. Hygon CPU do not need it.

There's a single Fam15 conditional in the middle of the function. Everything beyond that is again family-independent.

Quoted 20 lines
Quoted 11 lines
Quoted 11 lines
xen/arch/x86/cpu/hygon.c new file +11 -0
+	if (opt_cpu_info) {
+		printk(XENLOG_INFO "Levelling caps: %#x\n", levelling_caps);
+		printk(XENLOG_INFO
+		       "MSR defaults: 1d 0x%08x, 1c 0x%08x, e1d 0x%08x, "
+		       "e1c 0x%08x, 7a0 0x%08x, 7b0 0x%08x\n",
+		       (uint32_t)cpuidmask_defaults._1cd,
+		       (uint32_t)(cpuidmask_defaults._1cd >> 32),
+		       (uint32_t)cpuidmask_defaults.e1cd,
+		       (uint32_t)(cpuidmask_defaults.e1cd >> 32),
+		       (uint32_t)(cpuidmask_defaults._7ab0 >> 32),
+		       (uint32_t)cpuidmask_defaults._7ab0);
Quoted 8 lines

xen/arch/x86/cpu/hygon.c line 158

Quoted 5 lines
xen/arch/x86/cpu/hygon.c new file +5 -0
+	}
+
+	if (levelling_caps)
+		ctxt_switch_masking = hygon_ctxt_switch_masking;
+}

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

Quoted 3 lines

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

Quoted 4 lines

There are many codes to support AMD's
different families, the family/model/stepping checking are deeply embedded in codes. If we add Hygon family/model/stepping checking in future version, it will make code interleaved together and hard to maintain.

We can think about adding the duplication when it becomes unwieldy to maintain the shared variant.

Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
xen/arch/x86/cpu/hygon.c new file +7 -0
+static void init_hygon(struct cpuinfo_x86 *c)
+{
+	u32 l, h;
+	unsigned long long value;
+
+	/* Attempt to set lfence to be Dispatch Serialising. */
+	if (rdmsr_safe(MSR_AMD64_DE_CFG, value))
Quoted 8 lines

xen/arch/x86/cpu/hygon.c line 200

Quoted 5 lines
xen/arch/x86/cpu/hygon.c new file +5 -0
+		/* Unable to read.  Assume the safer default. */
+		__clear_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);
+	else if (value & AMD64_DE_CFG_LFENCE_SERIALISE)
+		/* Already dispatch serialising. */
+		__set_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);

Didn't you cut off too much from the original code (which again may better be shared, by splitting out into a function)?

Quoted 3 lines

The cut codes are used for other AMD families except family 17h. Hygon Dhyana is derived from AMD Zen, so we only need the family 17h parts. This also make the init flow minimal.

How is

		else if (wrmsr_safe(MSR_AMD64_DE_CFG,
				    value | AMD64_DE_CFG_LFENCE_SERIALISE) ||
			 rdmsr_safe(MSR_AMD64_DE_CFG, value) ||
			 !(value & AMD64_DE_CFG_LFENCE_SERIALISE))
			/* Attempt to set failed.  Assume the safer default. */
			__clear_bit(X86_FEATURE_LFENCE_DISPATCH,
				    c->x86_capability);
		else
			/* Successfully enabled! */
			__set_bit(X86_FEATURE_LFENCE_DISPATCH,
				  c->x86_capability);

family dependent in any way?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#5 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 43 lines · Jan Beulich
Quoted 18 lines
Quoted 1 line
Quoted 1 line · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 15 lines · Jan Beulich

xen/arch/x86/cpu/hygon.c line 36

Quoted 13 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:
xen/arch/x86/cpu/hygon.c new file +12 -0
+static void __init noinline probe_masking_msrs(void)
+{
+	const struct cpuinfo_x86 *c = &boot_cpu_data;
+
+	/* Work out which masking MSRs we should have. */
+	cpuidmask_defaults._1cd =
+		_probe_mask_msr(MSR_K8_FEATURE_MASK, LCAP_1cd);
+	cpuidmask_defaults.e1cd =
+		_probe_mask_msr(MSR_K8_EXT_FEATURE_MASK, LCAP_e1cd);
+	if (c->cpuid_level >= 7)
+		cpuidmask_defaults._7ab0 =
+			_probe_mask_msr(MSR_AMD_L7S0_FEATURE_MASK, LCAP_7ab0);

There's more relevant code here in the original function.

Quoted 1 line

The code is used for family 15h. Hygon CPU do not need it.

There's a single Fam15 conditional in the middle of the function. Everything beyond that is again family-independent.

Quoted 20 lines
Quoted 19 lines
Quoted 6 lines
xen/arch/x86/cpu/hygon.c new file +6 -0
+	if (opt_cpu_info) {
+		printk(XENLOG_INFO "Levelling caps: %#x\n", levelling_caps);
+		printk(XENLOG_INFO
+		       "MSR defaults: 1d 0x%08x, 1c 0x%08x, e1d 0x%08x, "
+		       "e1c 0x%08x, 7a0 0x%08x, 7b0 0x%08x\n",
+		       (uint32_t)cpuidmask_defaults._1cd,

xen/arch/x86/cpu/hygon.c line 158

Quoted 10 lines
xen/arch/x86/cpu/hygon.c new file +10 -0
+		       (uint32_t)(cpuidmask_defaults._1cd >> 32),
+		       (uint32_t)cpuidmask_defaults.e1cd,
+		       (uint32_t)(cpuidmask_defaults.e1cd >> 32),
+		       (uint32_t)(cpuidmask_defaults._7ab0 >> 32),
+		       (uint32_t)cpuidmask_defaults._7ab0);
+	}
+
+	if (levelling_caps)
+		ctxt_switch_masking = hygon_ctxt_switch_masking;
+}

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

Quoted 6 lines
Quoted 3 lines

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

That's OK. With this method only init_levelling in amd.c should be added an amd_ prefix and called by hygon.c.

But I'm afraid Hygon should have its own init functions and not call the AMD ones. The current Hygon init functions have been heavily stripped from the original AMD's.

Quoted 42 lines
Quoted 4 lines

There are many codes to support AMD's
different families, the family/model/stepping checking are deeply embedded in codes. If we add Hygon family/model/stepping checking in future version, it will make code interleaved together and hard to maintain.

We can think about adding the duplication when it becomes unwieldy to maintain the shared variant.

Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
xen/arch/x86/cpu/hygon.c new file +7 -0
+static void init_hygon(struct cpuinfo_x86 *c)
+{
+	u32 l, h;
+	unsigned long long value;
+
+	/* Attempt to set lfence to be Dispatch Serialising. */
+	if (rdmsr_safe(MSR_AMD64_DE_CFG, value))
Quoted 8 lines

xen/arch/x86/cpu/hygon.c line 200

Quoted 5 lines
xen/arch/x86/cpu/hygon.c new file +5 -0
+		/* Unable to read.  Assume the safer default. */
+		__clear_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);
+	else if (value & AMD64_DE_CFG_LFENCE_SERIALISE)
+		/* Already dispatch serialising. */
+		__set_bit(X86_FEATURE_LFENCE_DISPATCH, c->x86_capability);

Didn't you cut off too much from the original code (which again may better be shared, by splitting out into a function)?

Quoted 3 lines

The cut codes are used for other AMD families except family 17h. Hygon Dhyana is derived from AMD Zen, so we only need the family 17h parts. This also make the init flow minimal.

How is

		else if (wrmsr_safe(MSR_AMD64_DE_CFG,
				    value | AMD64_DE_CFG_LFENCE_SERIALISE) ||
			 rdmsr_safe(MSR_AMD64_DE_CFG, value) ||
			 !(value & AMD64_DE_CFG_LFENCE_SERIALISE))
			/* Attempt to set failed.  Assume the safer default. */
			__clear_bit(X86_FEATURE_LFENCE_DISPATCH,
				    c->x86_capability);
		else
			/* Successfully enabled! */
			__set_bit(X86_FEATURE_LFENCE_DISPATCH,
				  c->x86_capability);
Quoted 1 line

family dependent in any way?

Independent here.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#6 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 52 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 10:57, <[email protected]> wrote:
Quoted 43 lines · Jan Beulich
Quoted 18 lines
Quoted 1 line
Quoted 1 line · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 15 lines · Jan Beulich

xen/arch/x86/cpu/hygon.c line 36

Quoted 13 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:
xen/arch/x86/cpu/hygon.c new file +12 -0
+static void __init noinline probe_masking_msrs(void)
+{
+	const struct cpuinfo_x86 *c = &boot_cpu_data;
+
+	/* Work out which masking MSRs we should have. */
+	cpuidmask_defaults._1cd =
+		_probe_mask_msr(MSR_K8_FEATURE_MASK, LCAP_1cd);
+	cpuidmask_defaults.e1cd =
+		_probe_mask_msr(MSR_K8_EXT_FEATURE_MASK, LCAP_e1cd);
+	if (c->cpuid_level >= 7)
+		cpuidmask_defaults._7ab0 =
+			_probe_mask_msr(MSR_AMD_L7S0_FEATURE_MASK, LCAP_7ab0);

There's more relevant code here in the original function.

Quoted 1 line

The code is used for family 15h. Hygon CPU do not need it.

There's a single Fam15 conditional in the middle of the function. Everything beyond that is again family-independent.

Quoted 20 lines
Quoted 19 lines
Quoted 6 lines
xen/arch/x86/cpu/hygon.c new file +6 -0
+	if (opt_cpu_info) {
+		printk(XENLOG_INFO "Levelling caps: %#x\n", levelling_caps);
+		printk(XENLOG_INFO
+		       "MSR defaults: 1d 0x%08x, 1c 0x%08x, e1d 0x%08x, "
+		       "e1c 0x%08x, 7a0 0x%08x, 7b0 0x%08x\n",
+		       (uint32_t)cpuidmask_defaults._1cd,

xen/arch/x86/cpu/hygon.c line 158

Quoted 10 lines
xen/arch/x86/cpu/hygon.c new file +10 -0
+		       (uint32_t)(cpuidmask_defaults._1cd >> 32),
+		       (uint32_t)cpuidmask_defaults.e1cd,
+		       (uint32_t)(cpuidmask_defaults.e1cd >> 32),
+		       (uint32_t)(cpuidmask_defaults._7ab0 >> 32),
+		       (uint32_t)cpuidmask_defaults._7ab0);
+	}
+
+	if (levelling_caps)
+		ctxt_switch_masking = hygon_ctxt_switch_masking;
+}

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

Quoted 6 lines
Quoted 3 lines

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

Quoted 6 lines

That's OK. With this method only init_levelling in amd.c should be added an amd_ prefix and called by hygon.c.

But I'm afraid Hygon should have its own init functions and not call the AMD ones. The current Hygon init functions have been heavily stripped from the original AMD's.

Let me give you this rule of thumb (subject to discussion): If you can safely re-use any non-trivial current AMD function with at most minor adjustments (and irrespective of certain code there being unreachable on Hygon), then I think it would be better to re-use it than to duplicate it.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#7 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 16 lines · Jan Beulich
Quoted 8 lines · On 16.03.19 at 10:57, <[email protected]> wrote:
Quoted 7 lines · Jan Beulich
Quoted 6 lines · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 2 lines · Jan Beulich

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Quoted 7 lines
Quoted 2 lines

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

That's OK. With this method only init_levelling in amd.c should be added an amd_ prefix and called by hygon.c.
But I'm afraid Hygon should have its own init functions and not call the AMD ones. The current Hygon init functions have been heavily stripped from the original AMD's.

Quoted 5 lines

Let me give you this rule of thumb (subject to discussion): If you can safely re-use any non-trivial current AMD function with at most minor adjustments (and irrespective of certain code there being unreachable on Hygon), then I think it would be better to re-use it than to duplicate it.

I tried, but it will add 0x18 to init_amd(). It's strange because AMD does not have family 18h now. So it becomes unwieldy to maintain init_amd() just at this time. The same situation for amd_get_topology().

So I think it's better to carve out init_hygon() and hygon_get_topology() (which also need Hygon's own adjustment for computing phys_proc_id).

I think it's time to develop a new patch series for your review.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#8 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 24 lines
Quoted 1 line
Quoted 1 line · On 19.03.19 at 13:33, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich
Quoted 8 lines · On 16.03.19 at 10:57, <[email protected]> wrote:
Quoted 7 lines · Jan Beulich
Quoted 6 lines · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 2 lines · Jan Beulich

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Quoted 7 lines
Quoted 2 lines

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

That's OK. With this method only init_levelling in amd.c should be added an amd_ prefix and called by hygon.c.
But I'm afraid Hygon should have its own init functions and not call the AMD ones. The current Hygon init functions have been heavily stripped from the original AMD's.

Quoted 5 lines

Let me give you this rule of thumb (subject to discussion): If you can safely re-use any non-trivial current AMD function with at most minor adjustments (and irrespective of certain code there being unreachable on Hygon), then I think it would be better to re-use it than to duplicate it.

Quoted 6 lines

I tried, but it will add 0x18 to init_amd(). It's strange because AMD does not have family 18h now. So it becomes unwieldy to maintain init_amd() just at this time. The same situation for amd_get_topology().

So I think it's better to carve out init_hygon() and hygon_get_topology() (which also need Hygon's own adjustment for computing phys_proc_id).

Just to be clear: I've never objected to a separate init_hygon(). For amd_get_topology() it's less clear: It looks to me as if you wouldn't need to add any Hygon conditionals to the function, as long as cpu_has(c, X86_FEATURE_TOPOEXT) is true in your case.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#9 lore raw

Re: [PATCH v2 01/14] x86/cpu: Create Hygon Dhyana architecture support file

Quoted 28 lines · Jan Beulich
Quoted 27 lines · On 19.03.19 at 13:33, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich
Quoted 8 lines · On 16.03.19 at 10:57, <[email protected]> wrote:
Quoted 7 lines · Jan Beulich
Quoted 6 lines · On 15.03.19 at 11:17, <[email protected]> wrote:
Quoted 2 lines · Jan Beulich

This is a lot of duplicated code with only minor differences. I think you would be better off calling into the AMD original functions.

These functions and AMD original ones are static. So Hygon cannot directly call into them. Or we can put them into the common cpu code, but I think it's not good for future maintenance.

Quoted 7 lines
Quoted 2 lines

Just make non-static what needs to be, add an amd_ prefix, and call it from your code.

That's OK. With this method only init_levelling in amd.c should be added an amd_ prefix and called by hygon.c.
But I'm afraid Hygon should have its own init functions and not call the AMD ones. The current Hygon init functions have been heavily stripped from the original AMD's.

Quoted 5 lines

Let me give you this rule of thumb (subject to discussion): If you can safely re-use any non-trivial current AMD function with at most minor adjustments (and irrespective of certain code there being unreachable on Hygon), then I think it would be better to re-use it than to duplicate it.

I tried, but it will add 0x18 to init_amd(). It's strange because AMD does not have family 18h now. So it becomes unwieldy to maintain init_amd() just at this time. The same situation for amd_get_topology().

Quoted 7 lines
Quoted 2 lines

So I think it's better to carve out init_hygon() and hygon_get_topology() (which also need Hygon's own adjustment for computing phys_proc_id).

Just to be clear: I've never objected to a separate init_hygon(). For amd_get_topology() it's less clear: It looks to me as if you wouldn't need to add any Hygon conditionals to the function, as long as cpu_has(c, X86_FEATURE_TOPOEXT) is true in your case.

Yes, it's no need to add Hygon conditional to amd_get_topology() as >=0x17 conditionals are default supported.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#10 lore raw

PATCH v2 2/14 x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

In reply to Pu Wen

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

---
 xen/arch/x86/cpu/mtrr/generic.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
xen/arch/x86/cpu/mtrr/generic.c +3 -2
index 8f9cf1b..94ee7d6 100644
@@ -217,8 +217,9 @@ static void __init print_mtrr_state(const char *level)
 			printk("%s  %u disabled\n", level, i);
 	}
 
-	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD
-	    && boot_cpu_data.x86 >= 0xf) {
+	if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
+	     boot_cpu_data.x86 >= 0xf) ||
+	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
 		uint64_t syscfg, tom2;
 
 		rdmsrl(MSR_K8_SYSCFG, syscfg);
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 2/14
#11 lore raw

Re: [PATCH v2 02/14] x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

I'm afraid I won't be able to offer Reviewed-by tags for any of this series without you pointing us at the documentation for the processor (a link should really be in the cover letter).

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#12 lore raw

Re: [PATCH v2 02/14] x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

Quoted 11 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

I'm afraid I won't be able to offer Reviewed-by tags for any of this series without you pointing us at the documentation for the processor (a link should really be in the cover letter).

Do you mean the specification for Hygon processor? We don't have any public spec now, but you can reference to the AMD Zen one[1].

Reference:
[1] https://www.amd.com/system/files/TechDocs/54945_PPR_Family_17h_Models_00h-0Fh.pdf

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#13 lore raw

Re: [PATCH v2 02/14] x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:06, <[email protected]> wrote:
Quoted 6 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

Quoted 8 lines
Quoted 5 lines

I'm afraid I won't be able to offer Reviewed-by tags for any of this series without you pointing us at the documentation for the processor (a link should really be in the cover letter).

Do you mean the specification for Hygon processor? We don't have any public spec now, but you can reference to the AMD Zen one[1].

Well, no, not really. There are undoubtedly differences (see your reply to my pci_cfg_ok() question as an example). I can ack changes on the basis that you know how your processors behave. I can give R-b only if I have Hygon-specific doc to check against.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#14 lore raw

Re: [PATCH v2 02/14] x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

Quoted 14 lines · Jan Beulich
Quoted 14 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:06, <[email protected]> wrote:
Quoted 11 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

I'm afraid I won't be able to offer Reviewed-by tags for any of this series without you pointing us at the documentation for the processor (a link should really be in the cover letter).

Quoted 7 lines
Quoted 2 lines

Do you mean the specification for Hygon processor? We don't have any public spec now, but you can reference to the AMD Zen one[1].

Well, no, not really. There are undoubtedly differences (see your reply to my pci_cfg_ok() question as an example). I can ack changes on the basis that you know how your processors behave. I can give R-b only if I have Hygon-specific doc to check against.

In fact there is no definition of MSR C001_001F(MSR_AMD64_NB_CFG) in the AMD family 17h models 00h-0Fh PPR, I don't know why there is 0x17 support in pci_cfg_ok().

Besides, the other MSRs and CPUIDs which are used by Hygon in this patch series are all defined in this PPR, and the meanings are fully the same. For example in current patch, bit 21 of MSR C001_0010(MSR_K8_TOP_MEM2) in print_mtrr_state().

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#15 lore raw

Re: [PATCH v2 02/14] x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2

Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 18.03.19 at 16:51, <[email protected]> wrote:
Quoted 17 lines · Jan Beulich
Quoted 16 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:06, <[email protected]> wrote:
Quoted 11 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

The Hygon Dhyana CPU supports the MSR way to get TOP_MEM2. So add Hygon Dhyana support to print the value of TOP_MEM2.

I'm afraid I won't be able to offer Reviewed-by tags for any of this series without you pointing us at the documentation for the processor (a link should really be in the cover letter).

Do you mean the specification for Hygon processor? We don't have any public spec now, but you can reference to the AMD Zen one[1].

Quoted 8 lines
Quoted 4 lines

Well, no, not really. There are undoubtedly differences (see your reply to my pci_cfg_ok() question as an example). I can ack changes on the basis that you know how your processors behave. I can give R-b only if I have Hygon-specific doc to check against.

In fact there is no definition of MSR C001_001F(MSR_AMD64_NB_CFG) in the AMD family 17h models 00h-0Fh PPR, I don't know why there is 0x17 support in pci_cfg_ok().

Hmm, interesting observation. The extended access mechanism appears to still exist, but is controlled by a different PCI config space bit now, and there doesn't seem to be any mirroring MSR anymore. Brian, could you confirm this? If there's indeed no NB_CFG MSR anymore, we may need to adjust some of our code elsewhere as well; the control logic of extended accesses looks to need adjustment in any event.

The use of 0x17 looks to date back to a 2010 commit (3157bb4e13), where decimal 17 was converted to 0x17, without there being any description to clarify whether this wasn't actually a mistake.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#16 lore raw

PATCH v2 3/14 x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

In reply to Pu Wen

The PMU architecture for the Hygon Dhyana CPU is similar to the AMD family 17h one. So add Hygon Dhyana support in vpmu_arch_initialise and vpmu_init by using the code path of AMD.

Since current Xen vPMU still not support Zen(0x17), so add both 0x17 and 0x18 support by using the 0x15's case in amd_vpmu_init, for 0x17 and 0x18 have the same performance event select and counter MSRs as 0x15 has.

---
 xen/arch/x86/cpu/vpmu.c     | 2 ++
 xen/arch/x86/cpu/vpmu_amd.c | 2 ++
 2 files changed, 4 insertions(+)
xen/arch/x86/cpu/vpmu.c +2 -0
index 13da7d0..f679e79 100644
@@ -459,6 +459,7 @@ static int vpmu_arch_initialise(struct vcpu *v)
     switch ( vendor )
     {
     case X86_VENDOR_AMD:
+    case X86_VENDOR_HYGON:
         ret = svm_vpmu_initialise(v);
         break;
 
@@ -876,6 +877,7 @@ static int __init vpmu_init(void)
     switch ( vendor )
     {
     case X86_VENDOR_AMD:
+    case X86_VENDOR_HYGON:
         if ( amd_vpmu_init() )
            vpmu_mode = XENPMU_MODE_OFF;
         break;
xen/arch/x86/cpu/vpmu_amd.c +2 -0
index 5efc39b..c9abe6e 100644
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
     switch ( current_cpu_data.x86 )
     {
     case 0x15:
+    case 0x17:
+    case 0x18:
         num_counters = F15H_NUM_COUNTERS;
         counters = AMD_F15H_COUNTERS;
         ctrls = AMD_F15H_CTRLS;
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#17 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
     switch ( current_cpu_data.x86 )
     {
     case 0x15:
+    case 0x17:
+    case 0x18:
         num_counters = F15H_NUM_COUNTERS;
         counters = AMD_F15H_COUNTERS;
         ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#18 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
      switch ( current_cpu_data.x86 )
      {
      case 0x15:
+    case 0x17:
+    case 0x18:
          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#19 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
      switch ( current_cpu_data.x86 )
      {
      case 0x15:
+    case 0x17:
+    case 0x18:
          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#20 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 19 lines · Jan Beulich
Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
       switch ( current_cpu_data.x86 )
       {
       case 0x15:
+    case 0x17:
+    case 0x18:
           num_counters = F15H_NUM_COUNTERS;
           counters = AMD_F15H_COUNTERS;
           ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 6 lines
Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

How to check vendor? Maybe like this:

    case 0x15:
    case 0x17:
    case 0x18:
        if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
            boot_cpu_data.x86 == 0x18)
            return -EINVAL;

        num_counters = F15H_NUM_COUNTERS;
        counters = AMD_F15H_COUNTERS;
        ctrls = AMD_F15H_CTRLS;

or just add Hygon support at beginning of amd_vpmu_init():

    if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
        num_counters = F15H_NUM_COUNTERS;
        counters = AMD_F15H_COUNTERS;
        ctrls = AMD_F15H_CTRLS;
        k7_counters_mirrored = 1;
    }
Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#21 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 40 lines
Quoted 1 line
Quoted 1 line · On 19.03.19 at 12:32, <[email protected]> wrote:
Quoted 19 lines · Jan Beulich
Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
       switch ( current_cpu_data.x86 )
       {
       case 0x15:
+    case 0x17:
+    case 0x18:
           num_counters = F15H_NUM_COUNTERS;
           counters = AMD_F15H_COUNTERS;
           ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 6 lines
Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

How to check vendor? Maybe like this:

     case 0x15:
     case 0x17:
     case 0x18:
         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
             boot_cpu_data.x86 == 0x18)
             return -EINVAL;

         num_counters = F15H_NUM_COUNTERS;
         counters = AMD_F15H_COUNTERS;
         ctrls = AMD_F15H_CTRLS;
Quoted 7 lines

or just add Hygon support at beginning of amd_vpmu_init():

     if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
         num_counters = F15H_NUM_COUNTERS;
         counters = AMD_F15H_COUNTERS;
         ctrls = AMD_F15H_CTRLS;
         k7_counters_mirrored = 1;
     }

A suitable variant of the latter or

int __init amd_vpmu_init(void)

{
    unsigned int i, fam = current_cpu_data.x86

    /* <suitable comment> */
    if ( current_cpu_data.x86_vendor == X86_VENDOR_HYGON && fam == 0x18 )
       fam = 0x17;

    switch ( fam )
    ...

or perhaps even better would be two separate switch()-es, one for AMD and one for Hygon. Possibly even a separate hygon_vpmu_init().

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#22 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 50 lines · Jan Beulich
Quoted 39 lines · On 19.03.19 at 12:32, <[email protected]> wrote:
Quoted 19 lines · Jan Beulich
Quoted 18 lines · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 11 lines · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
        switch ( current_cpu_data.x86 )
        {
        case 0x15:
+    case 0x17:
+    case 0x18:
            num_counters = F15H_NUM_COUNTERS;
            counters = AMD_F15H_COUNTERS;
            ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 6 lines
Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

How to check vendor? Maybe like this:

      case 0x15:
      case 0x17:
      case 0x18:
          if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
              boot_cpu_data.x86 == 0x18)
              return -EINVAL;

          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;
Quoted 7 lines

or just add Hygon support at beginning of amd_vpmu_init():

      if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;
          k7_counters_mirrored = 1;
      }

A suitable variant of the latter or

Quoted 7 lines

int __init amd_vpmu_init(void)

{
     unsigned int i, fam = current_cpu_data.x86

     /* <suitable comment> */
     if ( current_cpu_data.x86_vendor == X86_VENDOR_HYGON && fam == 0x18 )
        fam = 0x17;

This is the minimum change, I think it's better.

Quoted 6 lines

switch ( fam )
...

or perhaps even better would be two separate switch()-es, one for AMD and one for Hygon. Possibly even a separate hygon_vpmu_init().

A separate hygon_vpmu_init() is also fine except that the last part of the function can be shared.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#23 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 63 lines
Quoted 1 line
Quoted 1 line · On 19.03.19 at 14:47, <[email protected]> wrote:
Quoted 50 lines · Jan Beulich
Quoted 39 lines · On 19.03.19 at 12:32, <[email protected]> wrote:
Quoted 19 lines · Jan Beulich
Quoted 18 lines · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 11 lines · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
        switch ( current_cpu_data.x86 )
        {
        case 0x15:
+    case 0x17:
+    case 0x18:
            num_counters = F15H_NUM_COUNTERS;
            counters = AMD_F15H_COUNTERS;
            ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 6 lines
Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

How to check vendor? Maybe like this:

      case 0x15:
      case 0x17:
      case 0x18:
          if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
              boot_cpu_data.x86 == 0x18)
              return -EINVAL;

          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;
Quoted 7 lines

or just add Hygon support at beginning of amd_vpmu_init():

      if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
          num_counters = F15H_NUM_COUNTERS;
          counters = AMD_F15H_COUNTERS;
          ctrls = AMD_F15H_CTRLS;
          k7_counters_mirrored = 1;
      }

A suitable variant of the latter or

Quoted 7 lines

int __init amd_vpmu_init(void)

{
     unsigned int i, fam = current_cpu_data.x86

     /* <suitable comment> */
     if ( current_cpu_data.x86_vendor == X86_VENDOR_HYGON && fam == 0x18 )
        fam = 0x17;

This is the minimum change, I think it's better.

Quoted 1 line
Quoted 8 lines
Quoted 5 lines

switch ( fam )
...

or perhaps even better would be two separate switch()-es, one for AMD and one for Hygon. Possibly even a separate hygon_vpmu_init().

A separate hygon_vpmu_init() is also fine except that the last part of the function can be shared.

So perhaps split that part out into a static _vpmu_init() or common_init()?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#24 lore raw

Re: [PATCH v2 03/14] x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU

Quoted 36 lines · Jan Beulich
Quoted 35 lines · On 19.03.19 at 14:47, <[email protected]> wrote:
Quoted 34 lines · Jan Beulich
Quoted 33 lines · On 19.03.19 at 12:32, <[email protected]> wrote:
Quoted 19 lines · Jan Beulich
Quoted 18 lines · On 16.03.19 at 11:11, <[email protected]> wrote:
Quoted 16 lines · Jan Beulich

xen/arch/x86/cpu/vpmu_amd.c int __init amd_vpmu_init(void) · line 552

Quoted 11 lines · On 21.02.19 at 10:50, <[email protected]> wrote:
xen/arch/x86/cpu/vpmu_amd.c +2 -0
@@ -545,6 +545,8 @@ int __init amd_vpmu_init(void)
         switch ( current_cpu_data.x86 )
         {
         case 0x15:
+    case 0x17:
+    case 0x18:
             num_counters = F15H_NUM_COUNTERS;
             counters = AMD_F15H_COUNTERS;
             ctrls = AMD_F15H_CTRLS;

Unless you know what AMD Fam18 will look like, you can't do it like this. Fam18 really needs to be further qualified by a vendor check at this point in time.

Quoted 6 lines
Quoted 2 lines

Hygon will negotiate with AMD to make sure that only Hygon should use Fam18h.

In the success case of which please state this in the description. Until those negotiations have succeeded I'm afraid I'm going to insist to see the extra check added.

or just add Hygon support at beginning of amd_vpmu_init():

       if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
           num_counters = F15H_NUM_COUNTERS;
           counters = AMD_F15H_COUNTERS;
           ctrls = AMD_F15H_CTRLS;
           k7_counters_mirrored = 1;
       }
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

or perhaps even better would be two separate switch()-es, one for AMD and one for Hygon. Possibly even a separate hygon_vpmu_init().

A separate hygon_vpmu_init() is also fine except that the last part of the function can be shared.

So perhaps split that part out into a static _vpmu_init() or common_init()?

Okay, I'll try to split that part.

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#25 lore raw

PATCH v2 4/14 x86/cpu/mce: Add Hygon Dhyana support to the MCA infrastructure

In reply to Pu Wen

The machine check architecture for Hygon Dhyana CPU is similar to the AMD family 17h one. Add vendor checking for Hygon Dhyana to share the code path of AMD family 17h.

---
 xen/arch/x86/cpu/common.c              | 3 ++-
 xen/arch/x86/cpu/mcheck/amd_nonfatal.c | 5 +++--
 xen/arch/x86/cpu/mcheck/mce.c          | 6 ++++--
 xen/arch/x86/cpu/mcheck/mce_amd.c      | 5 ++++-
 xen/arch/x86/cpu/mcheck/non-fatal.c    | 3 ++-
 xen/arch/x86/cpu/mcheck/vmce.c         | 2 ++
 6 files changed, 17 insertions(+), 7 deletions(-)
xen/arch/x86/cpu/common.c +2 -1
index 5bab845..214bdb5 100644
@@ -350,7 +350,8 @@ static void __init early_cpu_detect(void)
 			hap_paddr_bits = PADDR_BITS;
 	}
 
-	if (c->x86_vendor != X86_VENDOR_AMD)
+	if (c->x86_vendor != X86_VENDOR_AMD &&
+	    c->x86_vendor != X86_VENDOR_HYGON)
 		park_offline_cpus = opt_mce;
 
 	initialize_cpu_data(0);
xen/arch/x86/cpu/mcheck/amd_nonfatal.c +3 -2
index 222f539..589dac5 100644
@@ -203,10 +203,11 @@ static void mce_amd_work_fn(void *data)
 
 void __init amd_nonfatal_mcheck_init(struct cpuinfo_x86 *c)
 {
-	if (c->x86_vendor != X86_VENDOR_AMD)
+	if (c->x86_vendor != X86_VENDOR_AMD &&
+	    c->x86_vendor != X86_VENDOR_HYGON)
 		return;
 
-	/* Assume we are on K8 or newer AMD CPU here */
+	/* Assume we are on K8 or newer AMD or Hygon CPU here */
 
 	/* The threshold bitfields in MSR_IA32_MC4_MISC has
 	 * been introduced along with the SVME feature bit. */
xen/arch/x86/cpu/mcheck/mce.c +4 -2
index 30cdb06..0798dea 100644
@@ -778,6 +778,7 @@ void mcheck_init(struct cpuinfo_x86 *c, bool bsp)
     switch ( c->x86_vendor )
     {
     case X86_VENDOR_AMD:
+    case X86_VENDOR_HYGON:
         inited = amd_mcheck_init(c);
         break;
 
@@ -1172,10 +1173,11 @@ static bool x86_mc_msrinject_verify(struct xen_mc_msrinject *mci)
 
             /* MSRs that the HV will take care of */
             case MSR_K8_HWCR:
-                if ( c->x86_vendor == X86_VENDOR_AMD )
+                if ( c->x86_vendor == X86_VENDOR_AMD ||
+                     c->x86_vendor == X86_VENDOR_HYGON )
                     reason = "HV will operate HWCR";
                 else
-                    reason = "only supported on AMD";
+                    reason = "only supported on AMD or Hygon";
                 break;
 
             default:
xen/arch/x86/cpu/mcheck/mce_amd.c +4 -1
index d125bc1..040e449 100644
@@ -274,7 +274,10 @@ enum mcheck_type
 amd_mcheck_init(struct cpuinfo_x86 *ci)
 {
     uint32_t i;
-    enum mcequirk_amd_flags quirkflag = mcequirk_lookup_amd_quirkdata(ci);
+    enum mcequirk_amd_flags quirkflag = 0;
+
+    if (ci->x86_vendor != X86_VENDOR_HYGON)
+        quirkflag = mcequirk_lookup_amd_quirkdata(ci);
 
     /* Assume that machine check support is available.
      * The minimum provided support is at least the K8. */
xen/arch/x86/cpu/mcheck/non-fatal.c +2 -1
index d12e8f2..77be418 100644
@@ -101,7 +101,8 @@ static int __init init_nonfatal_mce_checker(void)
 	 */
 	switch (c->x86_vendor) {
 	case X86_VENDOR_AMD:
-		/* Assume we are on K8 or newer AMD CPU here */
+	case X86_VENDOR_HYGON:
+		/* Assume we are on K8 or newer AMD or Hygon CPU here */
 		amd_nonfatal_mcheck_init(c);
 		break;
 
xen/arch/x86/cpu/mcheck/vmce.c +2 -0
index f15835e..4f5de07 100644
@@ -154,6 +154,7 @@ static int bank_mce_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
             break;
 
         case X86_VENDOR_AMD:
+        case X86_VENDOR_HYGON:
             ret = vmce_amd_rdmsr(v, msr, val);
             break;
 
@@ -284,6 +285,7 @@ static int bank_mce_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
             break;
 
         case X86_VENDOR_AMD:
+        case X86_VENDOR_HYGON:
             ret = vmce_amd_wrmsr(v, msr, val);
             break;
 
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 4/14
#26 lore raw

Re: [PATCH v2 04/14] x86/cpu/mce: Add Hygon Dhyana support to the MCA infrastructure

Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:51, <[email protected]> wrote:

The machine check architecture for Hygon Dhyana CPU is similar to the AMD family 17h one. Add vendor checking for Hygon Dhyana to share the code path of AMD family 17h.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#27 lore raw

PATCH v2 5/14 x86/spec_ctrl: Add Hygon Dhyana to the respective mitigation machinery

In reply to Pu Wen

The Hygon Dhyana CPU has the same speculative execution as AMD family 17h, so share AMD Retpoline and PTI mitigation code with Hygon Dhyana.

---
 xen/arch/x86/spec_ctrl.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
xen/arch/x86/spec_ctrl.c +4 -2
index ad72ecd..38a6ed2 100644
@@ -306,7 +306,8 @@ static bool __init retpoline_safe(uint64_t caps)
 {
     unsigned int ucode_rev = this_cpu(ucode_cpu_info).cpu_sig.rev;
 
-    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD )
+    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+         boot_cpu_data.x86_vendor == X86_VENDOR_HYGON )
         return true;
 
     if ( boot_cpu_data.x86_vendor != X86_VENDOR_INTEL ||
@@ -612,7 +613,8 @@ int8_t __read_mostly opt_xpti_domu = -1;
 
 static __init void xpti_init_default(uint64_t caps)
 {
-    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD )
+    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+         boot_cpu_data.x86_vendor == X86_VENDOR_HYGON )
         caps = ARCH_CAPABILITIES_RDCL_NO;
 
     if ( caps & ARCH_CAPABILITIES_RDCL_NO )
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 5/14
#28 lore raw

Re: [PATCH v2 05/14] x86/spec_ctrl: Add Hygon Dhyana to the respective mitigation machinery

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:51, <[email protected]> wrote:

The Hygon Dhyana CPU has the same speculative execution as AMD family 17h, so share AMD Retpoline and PTI mitigation code with Hygon Dhyana.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#29 lore raw

PATCH v2 6/14 x86/apic: Add Hygon Dhyana support

In reply to Pu Wen

Add Hygon Dhyana support to use modern APIC.

---
 xen/arch/x86/apic.c | 5 +++++
 1 file changed, 5 insertions(+)
xen/arch/x86/apic.c +5 -0
index 2a24326..004d685 100644
@@ -92,6 +92,11 @@ static int modern_apic(void)
     if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
         boot_cpu_data.x86 >= 0xf)
         return 1;
+
+    /* Hygon systems use modern APIC */
+    if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
+        return 1;
+
     lvr = apic_read(APIC_LVR);
     version = GET_APIC_VERSION(lvr);
     return version >= 0x14;
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#31 lore raw

PATCH v2 7/14 x86/acpi: Add Hygon Dhyana support

In reply to Pu Wen

Add Hygon Dhyana support to the acpi cpufreq and cpuidle subsystems by using the code path of AMD.

---
 xen/arch/x86/acpi/cpu_idle.c         | 3 ++-
 xen/arch/x86/acpi/cpufreq/cpufreq.c  | 8 +++++---
 xen/arch/x86/acpi/cpufreq/powernow.c | 3 ++-
 3 files changed, 9 insertions(+), 5 deletions(-)
xen/arch/x86/acpi/cpu_idle.c +2 -1
index 14b0278..f57825c 100644
@@ -795,7 +795,8 @@ void acpi_dead_idle(void)
             __mwait(cx->address, 0);
         }
     }
-    else if ( current_cpu_data.x86_vendor == X86_VENDOR_AMD &&
+    else if ( (current_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+               current_cpu_data.x86_vendor == X86_VENDOR_HYGON) &&
               cx->entry_method == ACPI_CSTATE_EM_SYSIO )
     {
         /* Intel prefers not to use SYSIO */
xen/arch/x86/acpi/cpufreq/cpufreq.c +5 -3
index 844ab85..14c18bd 100644
@@ -649,7 +649,8 @@ static int __init cpufreq_driver_init(void)
         (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL))
         ret = cpufreq_register_driver(&acpi_cpufreq_driver);
     else if ((cpufreq_controller == FREQCTL_xen) &&
-        (boot_cpu_data.x86_vendor == X86_VENDOR_AMD))
+        (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+         boot_cpu_data.x86_vendor == X86_VENDOR_HYGON))
         ret = powernow_register_driver();
 
     return ret;
@@ -660,9 +661,10 @@ int cpufreq_cpu_init(unsigned int cpuid)
 {
     int ret;
 
-    /* Currently we only handle Intel and AMD processor */
+    /* Currently we only handle Intel, AMD and Hygon processor */
     if ( (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL ) ||
-         (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ) )
+         (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ) ||
+         (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON ) )
         ret = cpufreq_add_cpu(cpuid);
     else
         ret = -EFAULT;
xen/arch/x86/acpi/cpufreq/powernow.c +2 -1
index 025b37d..f245908 100644
@@ -360,7 +360,8 @@ unsigned int __init powernow_register_driver()
 
     for_each_online_cpu(i) {
         struct cpuinfo_x86 *c = &cpu_data[i];
-        if (c->x86_vendor != X86_VENDOR_AMD)
+        if (c->x86_vendor != X86_VENDOR_AMD &&
+            c->x86_vendor != X86_VENDOR_HYGON)
             ret = -ENODEV;
         else
         {
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 7/14
#32 lore raw

Re: [PATCH v2 07/14] x86/acpi: Add Hygon Dhyana support

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:52, <[email protected]> wrote:

Add Hygon Dhyana support to the acpi cpufreq and cpuidle subsystems by using the code path of AMD.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#33 lore raw

PATCH v2 8/14 x86/iommu: Add Hygon Dhyana support

In reply to Pu Wen

The IOMMU architecture for the Hygon Dhyana CPU is similar to the AMD family 17h one. So add Hygon Dhyana support to it by sharing the code path of AMD.

---
 xen/include/asm-x86/iommu.h | 1 +
 1 file changed, 1 insertion(+)
xen/include/asm-x86/iommu.h +1 -0
index 8dc3924..699a8f7 100644
@@ -74,6 +74,7 @@ static inline int iommu_hardware_setup(void)
     case X86_VENDOR_INTEL:
         return intel_vtd_setup();
     case X86_VENDOR_AMD:
+    case X86_VENDOR_HYGON:
         return amd_iov_detect();
     }
 
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 8/14
#34 lore raw

Re: [PATCH v2 08/14] x86/iommu: Add Hygon Dhyana support

Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:52, <[email protected]> wrote:

The IOMMU architecture for the Hygon Dhyana CPU is similar to the AMD family 17h one. So add Hygon Dhyana support to it by sharing the code path of AMD.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#35 lore raw

PATCH v2 9/14 x86/pv: Add Hygon Dhyana support to emulate MSRs access

In reply to Pu Wen

The Hygon Dhyana CPU supports lots of MSRs(such as perf event select and counter MSRs, hardware configuration MSR, MMIO configuration base address MSR, MPERF/APERF MSRs) as AMD CPU does, so add Hygon Dhyana support to the PV emulation infrastructure by using the code path of AMD.

---
 xen/arch/x86/pv/emul-priv-op.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
xen/arch/x86/pv/emul-priv-op.c +13 -6
index 942ece2..1521041 100644
@@ -911,7 +911,9 @@ static int read_msr(unsigned int reg, uint64_t *val,
             /* fall through */
     case MSR_AMD_FAM15H_EVNTSEL0 ... MSR_AMD_FAM15H_PERFCTR5:
     case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3:
-            if ( vpmu_msr || (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) )
+            if ( vpmu_msr ||
+                (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) ||
+                (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) )
             {
                 if ( vpmu_do_rdmsr(reg, val) )
                     break;
@@ -993,7 +995,8 @@ static int write_msr(unsigned int reg, uint64_t val,
     case MSR_K8_PSTATE6:
     case MSR_K8_PSTATE7:
     case MSR_K8_HWCR:
-        if ( boot_cpu_data.x86_vendor != X86_VENDOR_AMD )
+        if ( boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
+             boot_cpu_data.x86_vendor != X86_VENDOR_HYGON )
             break;
         if ( likely(!is_cpufreq_controller(currd)) ||
              wrmsr_safe(reg, val) == 0 )
@@ -1014,8 +1017,9 @@ static int write_msr(unsigned int reg, uint64_t val,
         break;
 
     case MSR_FAM10H_MMIO_CONF_BASE:
-        if ( boot_cpu_data.x86_vendor != X86_VENDOR_AMD ||
-             boot_cpu_data.x86 < 0x10 || boot_cpu_data.x86 > 0x17 )
+        if ( (boot_cpu_data.x86_vendor != X86_VENDOR_AMD ||
+              boot_cpu_data.x86 < 0x10 || boot_cpu_data.x86 > 0x17) &&
+              boot_cpu_data.x86_vendor != X86_VENDOR_HYGON )
             break;
         if ( !is_hardware_domain(currd) || !is_pinned_vcpu(curr) )
             return X86EMUL_OKAY;
@@ -1054,7 +1058,8 @@ static int write_msr(unsigned int reg, uint64_t val,
     case MSR_IA32_MPERF:
     case MSR_IA32_APERF:
         if ( (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) &&
-             (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) )
+             (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) &&
+             (boot_cpu_data.x86_vendor != X86_VENDOR_HYGON) )
             break;
         if ( likely(!is_cpufreq_controller(currd)) ||
              wrmsr_safe(reg, val) == 0 )
@@ -1087,7 +1092,9 @@ static int write_msr(unsigned int reg, uint64_t val,
             vpmu_msr = true;
     case MSR_AMD_FAM15H_EVNTSEL0 ... MSR_AMD_FAM15H_PERFCTR5:
     case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3:
-            if ( vpmu_msr || (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) )
+            if ( vpmu_msr ||
+                (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) ||
+                (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) )
             {
                 if ( (vpmu_mode & XENPMU_MODE_ALL) &&
                      !is_hardware_domain(currd) )
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 9/14
#36 lore raw

Re: [PATCH v2 09/14] x86/pv: Add Hygon Dhyana support to emulate MSRs access

Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:52, <[email protected]> wrote:

The Hygon Dhyana CPU supports lots of MSRs(such as perf event select and counter MSRs, hardware configuration MSR, MMIO configuration base address MSR, MPERF/APERF MSRs) as AMD CPU does, so add Hygon Dhyana support to the PV emulation infrastructure by using the code path of AMD.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#37 lore raw

PATCH v2 10/14 x86/domain: Add Hygon Dhyana support

In reply to Pu Wen

Add Hygon Dhyana support to handle HyperTransport range.

Also loading a nul selector does not clear bases and limits on Hygon CPUs, so add Hygon Dhyana support to the function preload_segment.

---
 xen/arch/x86/dom0_build.c | 3 ++-
 xen/arch/x86/domain.c     | 9 +++++----
 2 files changed, 7 insertions(+), 5 deletions(-)
xen/arch/x86/dom0_build.c +2 -1
index 2b4d9e9..c0657ee 100644
@@ -532,7 +532,8 @@ int __init dom0_setup_permissions(struct domain *d)
                             paddr_to_pfn(MSI_ADDR_BASE_LO +
                                          MSI_ADDR_DEST_ID_MASK));
     /* HyperTransport range. */
-    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD )
+    if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+         boot_cpu_data.x86_vendor == X86_VENDOR_HYGON )
         rc |= iomem_deny_access(d, paddr_to_pfn(0xfdULL << 32),
                                 paddr_to_pfn((1ULL << 40) - 1));
 
xen/arch/x86/domain.c +5 -4
index 32dc4253..319eea1 100644
@@ -1300,13 +1300,14 @@ arch_do_vcpu_op(
 }
 
 /*
- * Loading a nul selector does not clear bases and limits on AMD CPUs. Be on
- * the safe side and re-initialize both to flat segment values before loading
- * a nul selector.
+ * Loading a nul selector does not clear bases and limits on AMD or Hygon
+ * CPUs. Be on the safe side and re-initialize both to flat segment values
+ * before loading a nul selector.
  */
 #define preload_segment(seg, value) do {              \
     if ( !((value) & ~3) &&                           \
-         boot_cpu_data.x86_vendor == X86_VENDOR_AMD ) \
+        (boot_cpu_data.x86_vendor == X86_VENDOR_AMD || \
+         boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) ) \
         asm volatile ( "movl %k0, %%" #seg            \
                        :: "r" (FLAT_USER_DS32) );     \
 } while ( false )
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 10/14
#38 lore raw

Re: [PATCH v2 10/14] x86/domain: Add Hygon Dhyana support

Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:53, <[email protected]> wrote:

Add Hygon Dhyana support to handle HyperTransport range.

Also loading a nul selector does not clear bases and limits on Hygon CPUs, so add Hygon Dhyana support to the function preload_segment.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#39 lore raw

PATCH v2 11/14 x86/domctl: Add Hygon Dhyana support

In reply to Pu Wen

Add Hygon Dhyana support to update cpuid info for creating PV guest.

---
 xen/arch/x86/domctl.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
xen/arch/x86/domctl.c +10 -3
index 9bf2d08..3a9e290 100644
@@ -176,6 +176,7 @@ static int update_domain_cpuid_info(struct domain *d,
                 break;
 
             case X86_VENDOR_AMD:
+            case X86_VENDOR_HYGON:
                 mask &= ((uint64_t)ecx << 32) | edx;
 
                 /*
@@ -220,7 +221,8 @@ static int update_domain_cpuid_info(struct domain *d,
             uint32_t eax = ctl->eax;
             uint32_t ebx = p->feat._7b0;
 
-            if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD )
+            if ( boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+                 boot_cpu_data.x86_vendor == X86_VENDOR_HYGON )
                 mask &= ((uint64_t)eax << 32) | ebx;
 
             d->arch.pv.cpuidmasks->_7ab0 = mask;
@@ -281,8 +283,12 @@ static int update_domain_cpuid_info(struct domain *d,
             if ( cpu_has_cmp_legacy )
                 ecx |= cpufeat_mask(X86_FEATURE_CMP_LEGACY);
 
-            /* If not emulating AMD, clear the duplicated features in e1d. */
-            if ( p->x86_vendor != X86_VENDOR_AMD )
+            /*
+             * If not emulating AMD and Hygon, clear the duplicated features
+             * in e1d.
+             */
+            if ( p->x86_vendor != X86_VENDOR_AMD &&
+                 p->x86_vendor != X86_VENDOR_HYGON )
                 edx &= ~CPUID_COMMON_1D_FEATURES;
 
             switch ( boot_cpu_data.x86_vendor )
@@ -292,6 +298,7 @@ static int update_domain_cpuid_info(struct domain *d,
                 break;
 
             case X86_VENDOR_AMD:
+            case X86_VENDOR_HYGON:
                 mask &= ((uint64_t)ecx << 32) | edx;
 
                 /*
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 11/14
#40 lore raw

Re: [PATCH v2 11/14] x86/domctl: Add Hygon Dhyana support

Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:53, <[email protected]> wrote:

Add Hygon Dhyana support to update cpuid info for creating PV guest.

with one remark:

xen/arch/x86/domctl.c static int update_domain_cpuid_info(struct domain *d, · line 288

Quoted 9 lines
xen/arch/x86/domctl.c +3 -2
@@ -281,8 +283,12 @@ static int update_domain_cpuid_info(struct domain *d,
             if ( cpu_has_cmp_legacy )
                 ecx |= cpufeat_mask(X86_FEATURE_CMP_LEGACY);
 
-            /* If not emulating AMD, clear the duplicated features in e1d. */
-            if ( p->x86_vendor != X86_VENDOR_AMD )
+            /*
+             * If not emulating AMD and Hygon, clear the duplicated features
+             * in e1d.

I think this wants to be "or" instead of "and".

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#41 lore raw

Re: [PATCH v2 11/14] x86/domctl: Add Hygon Dhyana support

Quoted 12 lines · Jan Beulich
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:53, <[email protected]> wrote:

Add Hygon Dhyana support to update cpuid info for creating PV guest.

with one remark:

Quoted 4 lines
xen/arch/x86/domctl.c +0 -0
@@ -281,8 +283,12 @@ static int update_domain_cpuid_info(struct domain *d,
              if ( cpu_has_cmp_legacy )
                  ecx |= cpufeat_mask(X86_FEATURE_CMP_LEGACY);
 
Quoted 7 lines

xen/arch/x86/domctl.c static int update_domain_cpuid_info(struct domain *d, · line 288

Quoted 5 lines
xen/arch/x86/domctl.c +3 -2
-            /* If not emulating AMD, clear the duplicated features in e1d. */
-            if ( p->x86_vendor != X86_VENDOR_AMD )
+            /*
+             * If not emulating AMD and Hygon, clear the duplicated features
+             * in e1d.

I think this wants to be "or" instead of "and".

Yes, will change to "or".

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#42 lore raw

PATCH v2 12/14 x86/traps: Add Hygon Dhyana support

In reply to Pu Wen

The Hygon Dhyana processor has the methold to get the last exception source IP from MSR0000_01DD. So add support for it if the boot param ler is true.

---
 xen/arch/x86/traps.c | 3 +++
 1 file changed, 3 insertions(+)
xen/arch/x86/traps.c +3 -0
index 05ddc39..97bf9e2 100644
@@ -1973,6 +1973,9 @@ static unsigned int calc_ler_msr(void)
             return MSR_IA32_LASTINTFROMIP;
         }
         break;
+
+    case X86_VENDOR_HYGON:
+        return MSR_IA32_LASTINTFROMIP;
     }
 
     return 0;
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#43 lore raw

PATCH v2 13/14 x86/cpuid: Add Hygon Dhyana support

In reply to Pu Wen

The Hygon Dhyana family 18h processor shares the same cpuid leaves as the AMD family 17h one. So add Hygon Dhyana support to caculate the cpuid policies as the AMD CPU does.

---
 xen/arch/x86/cpuid.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/x86/cpuid.c +7 -3
index ab0aab6..f760594 100644
@@ -240,6 +240,7 @@ static void recalculate_misc(struct cpuid_policy *p)
         break;
 
     case X86_VENDOR_AMD:
+    case X86_VENDOR_HYGON:
         zero_leaves(p->basic.raw, 0x2, 0x3);
         memset(p->cache.raw, 0, sizeof(p->cache.raw));
         zero_leaves(p->basic.raw, 0x9, 0xa);
@@ -390,7 +391,8 @@ static void __init calculate_hvm_max_policy(void)
      * long mode (and init_amd() has cleared it out of host capabilities), but
      * HVM guests are able if running in protected mode.
      */
-    if ( (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) &&
+    if ( (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
+          boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) &&
          raw_cpuid_policy.basic.sep )
         __set_bit(X86_FEATURE_SEP, hvm_featureset);
 
@@ -465,7 +467,8 @@ void recalculate_cpuid_policy(struct domain *d)
     p->basic.max_leaf   = min(p->basic.max_leaf,   max->basic.max_leaf);
     p->feat.max_subleaf = min(p->feat.max_subleaf, max->feat.max_subleaf);
     p->extd.max_leaf    = 0x80000000 | min(p->extd.max_leaf & 0xffff,
-                                           (p->x86_vendor == X86_VENDOR_AMD
+                                          ((p->x86_vendor == X86_VENDOR_AMD ||
+                                            p->x86_vendor == X86_VENDOR_HYGON)
                                             ? CPUID_GUEST_NR_EXTD_AMD
                                             : CPUID_GUEST_NR_EXTD_INTEL) - 1);
 
@@ -507,7 +510,8 @@ void recalculate_cpuid_policy(struct domain *d)
     if ( is_pv_32bit_domain(d) )
     {
         __clear_bit(X86_FEATURE_LM, max_fs);
-        if ( boot_cpu_data.x86_vendor != X86_VENDOR_AMD )
+        if ( boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
+             boot_cpu_data.x86_vendor != X86_VENDOR_HYGON )
             __clear_bit(X86_FEATURE_SYSCALL, max_fs);
     }
 
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 13/14
#44 lore raw

Re: [PATCH v2 13/14] x86/cpuid: Add Hygon Dhyana support

Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:54, <[email protected]> wrote:

The Hygon Dhyana family 18h processor shares the same cpuid leaves as the AMD family 17h one. So add Hygon Dhyana support to caculate the cpuid policies as the AMD CPU does.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#45 lore raw

PATCH v2 14/14 tools/libxc: Add Hygon Dhyana support

In reply to Pu Wen

Add Hygon Dhyana support to caculate the cpuid policies for creating PV or HVM guest by using the code path of AMD.

---
 tools/libxc/xc_cpuid_x86.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
tools/libxc/xc_cpuid_x86.c +12 -4
index 098affe..d0cb9ae 100644
@@ -234,6 +234,7 @@ struct cpuid_domain_info
         VENDOR_UNKNOWN,
         VENDOR_INTEL,
         VENDOR_AMD,
+        VENDOR_HYGON,
     } vendor;
 
     bool hvm;
@@ -304,6 +305,10 @@ static int get_cpuid_domain_info(xc_interface *xch, uint32_t domid,
               regs[2] == 0x444d4163U &&
               regs[3] == 0x69746e65U )
         info->vendor = VENDOR_AMD;
+    else if ( regs[1] == 0x6f677948U && /* "HygonGenuine" */
+              regs[2] == 0x656e6975U &&
+              regs[3] == 0x6e65476eU )
+        info->vendor = VENDOR_HYGON;
     else
         info->vendor = VENDOR_UNKNOWN;
 
@@ -568,7 +573,8 @@ static void xc_cpuid_hvm_policy(const struct cpuid_domain_info *info,
         break;
     }
 
-    if ( info->vendor == VENDOR_AMD )
+    if ( info->vendor == VENDOR_AMD ||
+         info->vendor == VENDOR_HYGON )
         amd_xc_cpuid_policy(info, input, regs);
     else
         intel_xc_cpuid_policy(info, input, regs);
@@ -630,7 +636,8 @@ static void xc_cpuid_pv_policy(const struct cpuid_domain_info *info,
 
     case 0x80000000:
     {
-        unsigned int max = info->vendor == VENDOR_AMD
+        unsigned int max = (info->vendor == VENDOR_AMD||
+                            info->vendor == VENDOR_HYGON)
             ? DEF_MAX_AMDEXT : DEF_MAX_INTELEXT;
 
         if ( regs[0] > max )
@@ -736,7 +743,8 @@ static void sanitise_featureset(struct cpuid_domain_info *info)
         if ( !info->pv64 )
         {
             clear_bit(X86_FEATURE_LM, info->featureset);
-            if ( info->vendor != VENDOR_AMD )
+            if ( info->vendor != VENDOR_AMD &&
+                 info->vendor != VENDOR_HYGON )
                 clear_bit(X86_FEATURE_SYSCALL, info->featureset);
         }
 
@@ -787,7 +795,7 @@ int xc_cpuid_apply_policy(xc_interface *xch, uint32_t domid,
     input[0] = 0x80000000;
     cpuid(input, regs);
 
-    if ( info.vendor == VENDOR_AMD )
+    if ( info.vendor == VENDOR_AMD || info.vendor == VENDOR_HYGON )
         ext_max = (regs[0] <= DEF_MAX_AMDEXT) ? regs[0] : DEF_MAX_AMDEXT;
     else
         ext_max = (regs[0] <= DEF_MAX_INTELEXT) ? regs[0] : DEF_MAX_INTELEXT;
Signature · 7 lines
2.7.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#47 lore raw
Quoted 1 line · Wei Liu

I think the version should have been v5?

Aha. This is the second revision of the patch series. So why should it have been v5?

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#48 lore raw
Quoted 5 lines · Pu Wen
Quoted 1 line · Wei Liu

I think the version should have been v5?

Aha. This is the second revision of the patch series. So why should it have been v5?

I have seen several previous postings from last year. But I just realised they were for Linux, not Xen.

Sorry for the noise.

Wei.

Quoted 4 lines
Signature · 3 lines
-- 
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#49 lore raw
In reply to Pu Wen
Quoted 16 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

Pu Wen (14):
x86/cpu: Create Hygon Dhyana architecture support file x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2 x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU x86/cpu/mce: Add Hygon Dhyana support to the MCA infrastructure x86/spec_ctrl: Add Hygon Dhyana to the respective mitigation machinery x86/apic: Add Hygon Dhyana support
x86/acpi: Add Hygon Dhyana support
x86/iommu: Add Hygon Dhyana support
x86/pv: Add Hygon Dhyana support to emulate MSRs access x86/domain: Add Hygon Dhyana support
x86/domctl: Add Hygon Dhyana support
x86/traps: Add Hygon Dhyana support
x86/cpuid: Add Hygon Dhyana support
tools/libxc: Add Hygon Dhyana support

In addition to these changes, what about pci_cfg_ok()? I don't think I've seen any of the patches touch it.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

author
#50 lore raw
Quoted 17 lines · Jan Beulich
Quoted 16 lines
Quoted 1 line
Quoted 1 line · On 21.02.19 at 10:48, <[email protected]> wrote:

Pu Wen (14):
x86/cpu: Create Hygon Dhyana architecture support file x86/cpu/mtrr: Add Hygon Dhyana support to get TOP_MEM2 x86/cpu/vpmu: Add Hygon Dhyana and AMD Zen support for vPMU x86/cpu/mce: Add Hygon Dhyana support to the MCA infrastructure x86/spec_ctrl: Add Hygon Dhyana to the respective mitigation machinery x86/apic: Add Hygon Dhyana support
x86/acpi: Add Hygon Dhyana support
x86/iommu: Add Hygon Dhyana support
x86/pv: Add Hygon Dhyana support to emulate MSRs access x86/domain: Add Hygon Dhyana support
x86/domctl: Add Hygon Dhyana support
x86/traps: Add Hygon Dhyana support
x86/cpuid: Add Hygon Dhyana support
tools/libxc: Add Hygon Dhyana support

Quoted 2 lines

In addition to these changes, what about pci_cfg_ok()? I don't think I've seen any of the patches touch it.

Because Hygon does not access the extended NB configuration space right now, so there is no change to pci_cfg_ok().

Signature · 7 lines
Regards,
Pu Wen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel