vDSO support for Hyper-V guest on ARM64
Subsystems (6)
cover vDSO support for Hyper-V guest on ARM64
Hi,
This is the RFC patchset for vDSO support in ARM64 Hyper-V guest. To test it, Michael's ARM64 support patchset:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
is needed.
Similar as x86, Hyper-V on ARM64 use a TSC page for guests to read the virtualized hardware timer, this TSC page is read-only for the guests, so could be used for vDSO data page. And the vDSO (userspace) code could use the same code for timer reading as kernel, since they read the same TSC page.
This patchset therefore extends ARM64's __vsdo_init() to allow multiple data pages and introduces the vclock_mode concept similar to x86 to allow different platforms (bare-metal, Hyper-V, etc.) to switch to different __arch_get_hw_counter() implementations. The rest of this patchset does the necessary setup for Hyper-V guests: mapping tsc page, enabling userspace to read cntvct, etc. to enable vDSO.
This patchset consists of 6 patches:
patch #1 allows hv_get_raw_timer() definition to be overridden for userspace and kernel to share the same hv_read_tsc_page() definition.
patch #2 extends ARM64 to support multiple vDSO data pages.
patch #3 introduces vclock_mode similiar to x86 to allow different __arch_get_hw_counter() implementations for different clocksources.
patch #4 maps Hyper-V TSC page into vDSO data page.
patch #5 allows userspace to read cntvct, so that userspace can efficiently read the clocksource.
patch #6 enables the vDSO for ARM64 Hyper-V guest.
The whole patchset is based on v5.5-rc1 plus Michael's ARM64 support patchset, and I've done a few tests with:
https://github.com/nlynch-mentor/vdsotest
Comments and suggestions are welcome!
Regards,
Boqun
RFC 1/6 arm64: hyperv: Allow hv_get_raw_timer() definition to be overridden
In order to support vDSO, hv_read_tsc_page() should be able to be called from userspace if tsc page mapped. As a result, hv_get_raw_timer(), called by hv_read_tsc_page() requires to be called by both kernel and vDSO. Currently, it's defined as arch_timer_read_counter(), which is a function pointer initialized (using a kernel address) by the arch timer driver, therefore not usable in vDSO.
Fix this by allowing a previous definition to override the default one, so that in vDSO code, we can define it as a function callable in userspace.
--- arch/arm64/include/asm/mshyperv.h | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
arch/arm64/include/asm/mshyperv.h +8 -1
@@ -97,8 +97,15 @@ extern void hv_get_vpreg_128(u32 reg, struct hv_get_vp_register_output *result);#define hv_disable_stimer0_percpu_irq (irq) disable_percpu_irq(irq)#endif -/* ARM64 specific code to read the hardware clock */ +/* +* ARM64 specific code to read the hardware clock. +* +* This could be used in both kernel space and userspace (vDSO), so make it +* possible for a previous definition to override the default one. +*/ +#ifndef hv_get_raw_timer#define hv_get_raw_timer () arch_timer_read_counter() +#endif #include <asm-generic/mshyperv.h>
Signature · 1 line
2.24.0
RFC 2/6 arm64: vdso: Add support for multiple vDSO data pages
In reply to Boqun FengSplit __vdso_abi::vdso_pages into nr_vdso_{data,code}_pages, so that __setup_additional_pages() could work with multiple vDSO data pages with the setup from __vdso_init().
Multiple vDSO data pages are required when running in a virtualized environment, where the cycles read from cntvct at userspace need to be adjusted with some data from a page maintained by the hypervisor. For example, the TSC page in Hyper-V.
This is a prerequisite for vDSO support in ARM64 on Hyper-V.
--- arch/arm64/kernel/vdso.c | 43 ++++++++++++++++++++++++---------------- 1 file changed, 26 insertions(+), 17 deletions(-)
arch/arm64/kernel/vdso.c +26 -17
@@ -50,7 +50,8 @@ struct __vdso_abi {const char *name;const char *vdso_code_start;const char *vdso_code_end; -unsigned long vdso_pages; +unsigned long nr_vdso_data_pages; +unsigned long nr_vdso_code_pages;/* Data Mapping */ struct vm_special_mapping *dm;/* Code Mapping */
@@ -101,6 +102,8 @@ static int __vdso_init(enum arch_vdso_type arch_index) {int i;struct page **vdso_pagelist; +struct page **vdso_code_pagelist; +unsigned long nr_vdso_pages;unsigned long pfn;if (memcmp (vdso_lookup[arch_index].vdso_code_start,"\177ELF" ,4 )) {
@@ -108,14 +111,18 @@ static int __vdso_init(enum arch_vdso_type arch_index)return -EINVAL ; } - vdso_lookup[arch_index].vdso_pages = ( + vdso_lookup[arch_index].nr_vdso_data_pages =1 ; + + vdso_lookup[arch_index].nr_vdso_code_pages = ( vdso_lookup[arch_index].vdso_code_end - vdso_lookup[arch_index].vdso_code_start) >>PAGE_SHIFT ; -/* Allocate the vDSO pagelist, plus a page for the data. */ - vdso_pagelist =kcalloc (vdso_lookup[arch_index].vdso_pages +1 , -sizeof (struct page *), + nr_vdso_pages = vdso_lookup[arch_index].nr_vdso_data_pages + + vdso_lookup[arch_index].nr_vdso_code_pages; + +/* Allocate the vDSO pagelist. */ + vdso_pagelist =kcalloc (nr_vdso_pages,sizeof (struct page *),GFP_KERNEL );if (vdso_pagelist ==NULL )return -ENOMEM ;
@@ -123,15 +130,17 @@ static int __vdso_init(enum arch_vdso_type arch_index)/* Grab the vDSO data page. */ vdso_pagelist[0 ] =phys_to_page (__pa_symbol (vdso_data)); -/* Grab the vDSO code pages. */ pfn =sym_to_pfn (vdso_lookup[arch_index].vdso_code_start); -for (i =0 ; i < vdso_lookup[arch_index].vdso_pages; i++) - vdso_pagelist[i +1 ] =pfn_to_page (pfn + i); + vdso_code_pagelist = vdso_pagelist + + vdso_lookup[arch_index].nr_vdso_data_pages; + +for (i =0 ; i < vdso_lookup[arch_index].nr_vdso_code_pages; i++) + vdso_code_pagelist[i] =pfn_to_page (pfn + i); - vdso_lookup[arch_index].dm->pages = &vdso_pagelist[0 ]; - vdso_lookup[arch_index].cm->pages = &vdso_pagelist[1 ]; + vdso_lookup[arch_index].dm->pages = vdso_pagelist; + vdso_lookup[arch_index].cm->pages = vdso_code_pagelist;return 0 ; }
@@ -141,26 +150,26 @@ static int __setup_additional_pages(enum arch_vdso_type arch_index, struct linux_binprm *bprm, int uses_interp) { -unsigned long vdso_base, vdso_text_len, vdso_mapping_len; +unsigned long vdso_base, vdso_text_len, vdso_data_len;void *ret; - vdso_text_len = vdso_lookup[arch_index].vdso_pages <<PAGE_SHIFT ; -/* Be sure to map the data page */ - vdso_mapping_len = vdso_text_len +PAGE_SIZE ; + vdso_data_len = vdso_lookup[arch_index].nr_vdso_data_pages <<PAGE_SHIFT ; + vdso_text_len = vdso_lookup[arch_index].nr_vdso_code_pages <<PAGE_SHIFT ; - vdso_base =get_unmapped_area (NULL ,0 , vdso_mapping_len,0 ,0 ); + vdso_base =get_unmapped_area (NULL ,0 , + vdso_data_len + vdso_text_len,0 ,0 );if (IS_ERR_VALUE (vdso_base)) { ret =ERR_PTR (vdso_base); goto up_fail; } - ret =_install_special_mapping (mm, vdso_base,PAGE_SIZE , + ret =_install_special_mapping (mm, vdso_base, vdso_data_len,VM_READ |VM_MAYREAD , vdso_lookup[arch_index].dm);if (IS_ERR (ret)) goto up_fail; - vdso_base +=PAGE_SIZE ; + vdso_base += vdso_data_len; mm->context.vdso = (void *)vdso_base; ret = _install_special_mapping(mm, vdso_base, vdso_text_len, VM_READ|VM_EXEC|
Signature · 1 line
2.24.0
RFC 3/6 arm/arm64: clocksource: Introduce vclock_mode
In reply to Boqun FengSimilar to x86, use a vclock_mode in arch_clocksource_data to differ clocksoures use different read function in vDSO.
No functional changes, only preparation for support vDSO in ARM64 on Hyper-V.
Note: the changes for arm are only because arm and arm64 share the same code in the arch timer driver and require arch_clocksource_data having the same field.
--- arch/arm/include/asm/clocksource.h | 6 +++++- arch/arm/kernel/vdso.c | 1 - arch/arm64/include/asm/clocksource.h | 6 +++++- arch/arm64/include/asm/mshyperv.h | 2 +- arch/arm64/include/asm/vdso/compat_gettimeofday.h | 5 +++-- arch/arm64/include/asm/vdso/gettimeofday.h | 5 +++-- arch/arm64/include/asm/vdso/vsyscall.h | 4 +--- drivers/clocksource/arm_arch_timer.c | 8 ++++---- 8 files changed, 22 insertions(+), 15 deletions(-)
arch/arm/include/asm/clocksource.h +5 -1
@@ -1,8 +1,12 @@#ifndef _ASM_CLOCKSOURCE_H#define _ASM_CLOCKSOURCE_H +#define VCLOCK_NONE 0/* No vDSO clock available. */ +#define VCLOCK_CNTVCT 1/* vDSO should use cntvcnt */ +#define VCLOCK_MAX 1 +struct arch_clocksource_data { -bool vdso_direct;/* Usable for direct VDSO access? */ +int vclock_mode; };#endif
arch/arm/kernel/vdso.c +0 -1
@@ -263,4 +263,3 @@ void arm_install_vdso(struct mm_struct *mm, unsigned long addr)if (!IS_ERR (vma)) mm->context.vdso = addr; } -
arch/arm64/include/asm/clocksource.h +5 -1
@@ -2,8 +2,12 @@#ifndef _ASM_CLOCKSOURCE_H#define _ASM_CLOCKSOURCE_H +#define VCLOCK_NONE 0/* No vDSO clock available. */ +#define VCLOCK_CNTVCT 1/* vDSO should use cntvcnt */ +#define VCLOCK_MAX 1 +struct arch_clocksource_data { -bool vdso_direct;/* Usable for direct VDSO access? */ +int vclock_mode; };#endif
arch/arm64/include/asm/mshyperv.h +1 -1
@@ -90,7 +90,7 @@ extern void hv_get_vpreg_128(u32 reg, struct hv_get_vp_register_output *result);#define hv_set_reference_tsc (val) \ hv_set_vpreg(HV_REGISTER_REFERENCE_TSC, val)#define hv_set_clocksource_vdso (val) \ - ((val).archdata.vdso_direct = false) + ((val).archdata.vclock_mode = VCLOCK_NONE)#if IS_ENABLED (CONFIG_HYPERV )#define hv_enable_stimer0_percpu_irq (irq) enable_percpu_irq(irq, 0)
arch/arm64/include/asm/vdso/compat_gettimeofday.h +3 -2
@@ -8,6 +8,7 @@#ifndef __ASSEMBLY__#include <asm/unistd.h> +#include <asm/clocksource.h> #include <uapi/linux/time.h> #include <asm/vdso/compat_barrier.h>
@@ -117,10 +118,10 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)u64 res;/* -* clock_mode == 0 implies that vDSO are enabled otherwise +* clock_mode == VCLOCK_NONE implies that vDSO are disabled so * fallback on syscall. */ -if (clock_mode) +if (clock_mode ==VCLOCK_NONE )return __VDSO_USE_SYSCALL;/*
arch/arm64/include/asm/vdso/gettimeofday.h +3 -2
@@ -8,6 +8,7 @@#ifndef __ASSEMBLY__#include <asm/unistd.h> +#include <asm/clocksource.h> #include <uapi/linux/time.h> #define __VDSO_USE_SYSCALL ULLONG_MAX
@@ -71,10 +72,10 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)u64 res;/* -* clock_mode == 0 implies that vDSO are enabled otherwise +* clock_mode == VCLOCK_NONE implies that vDSO are disabled so * fallback on syscall. */ -if (clock_mode) +if (clock_mode ==VCLOCK_NONE )return __VDSO_USE_SYSCALL;/*
arch/arm64/include/asm/vdso/vsyscall.h +1 -3
@@ -24,9 +24,7 @@ struct vdso_data *__arm64_get_k_vdso_data(void)static __always_inline int __arm64_get_clock_mode(struct timekeeper *tk) { -u32 use_syscall = !tk->tkr_mono.clock->archdata.vdso_direct; - -return use_syscall; +return tk->tkr_mono.clock->archdata.vclock_mode; }#define __arch_get_clock_mode __arm64_get_clock_mode
drivers/clocksource/arm_arch_timer.c +4 -4
@@ -69,7 +69,7 @@ static enum arch_timer_ppi_nr arch_timer_uses_ppi = ARCH_TIMER_VIRT_PPI;static bool arch_timer_c3stop;static bool arch_timer_mem_use_virtual;static bool arch_counter_suspend_stop; -static bool vdso_default = true; +static int vdso_default =VCLOCK_CNTVCT ;static cpumask_t evtstrm_available =CPU_MASK_NONE ;static bool evtstrm_enable =IS_ENABLED (CONFIG_ARM_ARCH_TIMER_EVTSTREAM );
@@ -560,8 +560,8 @@ void arch_timer_enable_workaround(const struct arch_timer_erratum_workaround *wa* change both the default value and the vdso itself. */ if (wa->read_cntvct_el0) { - clocksource_counter.archdata.vdso_direct = false; - vdso_default = false; + clocksource_counter.archdata.vclock_mode =VCLOCK_NONE ; + vdso_default =VCLOCK_NONE ; } }
@@ -979,7 +979,7 @@ static void __init arch_counter_register(unsigned type) } arch_timer_read_counter = rd; - clocksource_counter.archdata.vdso_direct = vdso_default; + clocksource_counter.archdata.vclock_mode = vdso_default; }else { arch_timer_read_counter = arch_counter_get_cntvct_mem; }
Signature · 1 line
2.24.0
RFC 4/6 arm64: vdso: hyperv: Map tsc page into vDSO if enabled
In reply to Boqun FengOn Hyper-V, a tsc page has the data for adjusting cntvct numbers to clocksource cycles, and that's how Hyper-V guest kernel reads the clocksource. In order to allow userspace to read the same clocksource directly, the tsc page has to been mapped into userspace via vDSO.
Use the framework for vDSO set-up in __vdso_init() to do this.
Note: if HYPERV_TIMER=y but the kernel is using other clocksource or doesn't have the hyperv timer clocksource, tsc page will still be mapped into userspace.
--- arch/arm64/kernel/vdso.c | 12 ++++++++++++ arch/arm64/kernel/vdso/vdso.lds.S | 12 +++++++++++- 2 files changed, 23 insertions(+), 1 deletion(-)
arch/arm64/kernel/vdso.c +12 -0
@@ -9,6 +9,7 @@#include <linux/cache.h> #include <linux/clocksource.h> +#include <clocksource/hyperv_timer.h> #include <linux/elf.h> #include <linux/err.h> #include <linux/errno.h>
@@ -105,14 +106,22 @@ static int __vdso_init(enum arch_vdso_type arch_index)struct page **vdso_code_pagelist;unsigned long nr_vdso_pages;unsigned long pfn; +struct ms_hyperv_tsc_page *tsc_page; +int tsc_page_idx;if (memcmp (vdso_lookup[arch_index].vdso_code_start,"\177ELF" ,4 )) {pr_err ("vDSO is not a valid ELF object!\n" );return -EINVAL ; } +/* One vDSO data page */ vdso_lookup[arch_index].nr_vdso_data_pages =1 ; +/* Grab the Hyper-V tsc page, if enabled, add one more page */ + tsc_page =hv_get_tsc_page (); +if (tsc_page) + tsc_page_idx = vdso_lookup[arch_index].nr_vdso_data_pages++; + vdso_lookup[arch_index].nr_vdso_code_pages = ( vdso_lookup[arch_index].vdso_code_end - vdso_lookup[arch_index].vdso_code_start) >>
@@ -130,6 +139,9 @@ static int __vdso_init(enum arch_vdso_type arch_index)/* Grab the vDSO data page. */ vdso_pagelist[0 ] =phys_to_page (__pa_symbol (vdso_data)); +if (tsc_page) + vdso_pagelist[tsc_page_idx] =phys_to_page (__pa (tsc_page)); +/* Grab the vDSO code pages. */ pfn =sym_to_pfn (vdso_lookup[arch_index].vdso_code_start);
arch/arm64/kernel/vdso/vdso.lds.S +11 -1
@@ -17,7 +17,17 @@ OUTPUT_ARCH(aarch64) SECTIONS { - PROVIDE(_vdso_data = . - PAGE_SIZE); + /* + * vdso data pages: + * vdso data (1 page) + * hv tsc page (1 page if enabled) + */ + PROVIDE(_vdso_data = _hvclock_page - PAGE_SIZE); +#ifdef CONFIG_HYPERV_TIMER + PROVIDE(_hvclock_page = . - PAGE_SIZE); +#else + PROVIDE(_hvclock_page = .); +#endif . = VDSO_LBASE + SIZEOF_HEADERS; .hash : { *(.hash) } :text
Signature · 1 line
2.24.0
RFC 5/6 arm64: hyperv: Enable userspace to read cntvct
In reply to Boqun FengSince reading hyperv-timer clocksource requires reading cntvct, userspace should be allowed to read it, otherwise reading cntvct will result in traps, which makes vsyscall's cost similar compared to syscall's.
So enable it on every cpu when a Hyper-V guest booting up.
--- arch/arm64/hyperv/hv_init.c | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm64/hyperv/hv_init.c +7 -0
@@ -27,6 +27,7 @@#include <linux/sched_clock.h> #include <asm-generic/bug.h> #include <asm/hyperv-tlfs.h> +#include <asm/arch_timer.h> #include <asm/mshyperv.h> #include <asm/sysreg.h> #include <clocksource/hyperv_timer.h>
@@ -45,6 +46,7 @@ EXPORT_SYMBOL_GPL(hv_max_vp_index);static int hv_cpu_init (unsigned int cpu) {u64 msr_vp_index; +u32 cntkctl;hv_get_vp_index (msr_vp_index);
@@ -53,6 +55,11 @@ static int hv_cpu_init(unsigned int cpu)if (msr_vp_index > hv_max_vp_index) hv_max_vp_index = msr_vp_index; +/* Enable EL0 to access cntvct */ + cntkctl =arch_timer_get_cntkctl (); + cntkctl |=ARCH_TIMER_USR_VCT_ACCESS_EN ; +arch_timer_set_cntkctl (cntkctl); +return 0 ; }
Signature · 1 line
2.24.0
RFC 6/6 arm64: hyperv: Enable vDSO
In reply to Boqun FengSimilar to x86, add a new vclock_mode VCLOCK_HVCLOCK, and reuse the hv_read_tsc_page() for userspace to read tsc page clocksource.
--- arch/arm64/include/asm/clocksource.h | 3 ++- arch/arm64/include/asm/mshyperv.h | 2 +- arch/arm64/include/asm/vdso/gettimeofday.h | 19 +++++++++++++++++++ 3 files changed, 22 insertions(+), 2 deletions(-)
arch/arm64/include/asm/clocksource.h +2 -1
@@ -4,7 +4,8 @@#define VCLOCK_NONE 0/* No vDSO clock available. */ #define VCLOCK_CNTVCT 1/* vDSO should use cntvcnt */ -#define VCLOCK_MAX 1 +#define VCLOCK_HVCLOCK 2/* vDSO should use vread_hvclock() */ +#define VCLOCK_MAX 2struct arch_clocksource_data {int vclock_mode;
arch/arm64/include/asm/mshyperv.h +1 -1
@@ -90,7 +90,7 @@ extern void hv_get_vpreg_128(u32 reg, struct hv_get_vp_register_output *result);#define hv_set_reference_tsc (val) \ hv_set_vpreg(HV_REGISTER_REFERENCE_TSC, val)#define hv_set_clocksource_vdso (val) \ - ((val).archdata.vclock_mode = VCLOCK_NONE) + ((val).archdata.vclock_mode = VCLOCK_HVCLOCK)#if IS_ENABLED (CONFIG_HYPERV )#define hv_enable_stimer0_percpu_irq (irq) enable_percpu_irq(irq, 0)
arch/arm64/include/asm/vdso/gettimeofday.h +19 -0
@@ -67,6 +67,20 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)return ret; } +#ifdef CONFIG_HYPERV_TIMER +/* This will override the default hv_get_raw_timer() */ +#define hv_get_raw_timer () __arch_counter_get_cntvct() +#include <clocksource/hyperv_timer.h> + +extern struct ms_hyperv_tsc_page +_hvclock_page __attribute__((visibility("hidden"))); + +static u64 vread_hvclock (void ) +{ +return hv_read_tsc_page (&_hvclock_page); +} +#endif + static __always_inline u64 __arch_get_hw_counter(s32 clock_mode) {u64 res;
@@ -78,6 +92,11 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)if (clock_mode ==VCLOCK_NONE )return __VDSO_USE_SYSCALL; +#ifdef CONFIG_HYPERV_TIMER +if (likely (clock_mode ==VCLOCK_HVCLOCK )) +return vread_hvclock (); +#endif +/* * This isb() is required to prevent that the counter value * is speculated.
Signature · 1 line
2.24.0
Re: [RFC 6/6] arm64: hyperv: Enable vDSO
Quoted 48 lines · Boqun Feng
Similar to x86, add a new
vclock_modeVCLOCK_HVCLOCK, and reuse thehv_read_tsc_page()for userspace to read tsc page clocksource.Signed-off-by Boqun Feng (Microsoft) <[email protected]>--- arch/arm64/include/asm/clocksource.h | 3 ++- arch/arm64/include/asm/mshyperv.h | 2 +- arch/arm64/include/asm/vdso/gettimeofday.h | 19 +++++++++++++++++++ 3 files changed, 22 insertions(+), 2 deletions(-)arch/arm64/include/asm/clocksource.h +2
-1@@ -4,7 +4,8 @@#define VCLOCK_NONE 0/* No vDSO clock available. */ #define VCLOCK_CNTVCT 1/* vDSO should use cntvcnt */ -#define VCLOCK_MAX 1 +#define VCLOCK_HVCLOCK 2/* vDSO should use vread_hvclock() */ +#define VCLOCK_MAX 2struct arch_clocksource_data {int vclock_mode;arch/arm64/include/asm/mshyperv.h +1
-1@@ -90,7 +90,7 @@ extern void hv_get_vpreg_128(u32 reg, struct hv_get_vp_register_output *result);#define hv_set_reference_tsc (val) \ hv_set_vpreg(HV_REGISTER_REFERENCE_TSC, val)#define hv_set_clocksource_vdso (val) \ - ((val).archdata.vclock_mode = VCLOCK_NONE) + ((val).archdata.vclock_mode = VCLOCK_HVCLOCK)#if IS_ENABLED (CONFIG_HYPERV )#define hv_enable_stimer0_percpu_irq (irq) enable_percpu_irq(irq, 0)arch/arm64/include/asm/vdso/gettimeofday.h +3
-0@@ -67,6 +67,20 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)return ret; } +#ifdef CONFIG_HYPERV_TIMER +/* This will override the default hv_get_raw_timer() */ +#define hv_get_raw_timer () __arch_counter_get_cntvct()
arch/arm64/include/asm/vdso/gettimeofday.h int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts) · line 82
Quoted 10 lines
arch/arm64/include/asm/vdso/gettimeofday.h +10
-0+#include <clocksource/hyperv_timer.h> + +extern struct ms_hyperv_tsc_page +_hvclock_page __attribute__((visibility("hidden"))); + +static u64 vread_hvclock (void ) +{ +return hv_read_tsc_page (&_hvclock_page); +} +#endif
The function is almost the same on x86 (&_hvclock_page -> &hvclock_page), would it maybe make sense to move this to arch neutral clocksource/hyperv_timer.h?
arch/arm64/include/asm/vdso/gettimeofday.h static __always_inline u64 __arch_get_hw_counter(s32 clock_mode) · line 97
Quoted 11 lines
arch/arm64/include/asm/vdso/gettimeofday.h +4
-0+ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode) {u64 res;@@ -78,6 +92,11 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)if (clock_mode ==VCLOCK_NONE )return __VDSO_USE_SYSCALL; +#ifdef CONFIG_HYPERV_TIMER +if (likely (clock_mode ==VCLOCK_HVCLOCK )) +return vread_hvclock ();
I'm not sure likely() is justified here: it'll make ALL builds which enable CONFIG_HYPERV_TIMER (e.g. distro kernels) to prefer VCLOCK_HVCLOCK, even if the kernel is not running on Hyper-V.
Quoted 5 lines
arch/arm64/include/asm/vdso/gettimeofday.h +2
-0+#endif +/* * This isb() is required to prevent that the counter value * is speculated.
Signature · 1 line
Vitaly
Re: [RFC 6/6] arm64: hyperv: Enable vDSO
Quoted 56 lines · Vitaly Kuznetsov
Quoted 54 lines · Boqun Feng
Similar to x86, add a new
vclock_modeVCLOCK_HVCLOCK, and reuse thehv_read_tsc_page()for userspace to read tsc page clocksource.Signed-off-by Boqun Feng (Microsoft) <[email protected]>--- arch/arm64/include/asm/clocksource.h | 3 ++- arch/arm64/include/asm/mshyperv.h | 2 +- arch/arm64/include/asm/vdso/gettimeofday.h | 19 +++++++++++++++++++ 3 files changed, 22 insertions(+), 2 deletions(-)arch/arm64/include/asm/clocksource.h +2
-1@@ -4,7 +4,8 @@#define VCLOCK_NONE 0/* No vDSO clock available. */ #define VCLOCK_CNTVCT 1/* vDSO should use cntvcnt */ -#define VCLOCK_MAX 1 +#define VCLOCK_HVCLOCK 2/* vDSO should use vread_hvclock() */ +#define VCLOCK_MAX 2struct arch_clocksource_data {int vclock_mode;arch/arm64/include/asm/mshyperv.h +1
-1@@ -90,7 +90,7 @@ extern void hv_get_vpreg_128(u32 reg, struct hv_get_vp_register_output *result);#define hv_set_reference_tsc (val) \ hv_set_vpreg(HV_REGISTER_REFERENCE_TSC, val)#define hv_set_clocksource_vdso (val) \ - ((val).archdata.vclock_mode = VCLOCK_NONE) + ((val).archdata.vclock_mode = VCLOCK_HVCLOCK)#if IS_ENABLED (CONFIG_HYPERV )#define hv_enable_stimer0_percpu_irq (irq) enable_percpu_irq(irq, 0)arch/arm64/include/asm/vdso/gettimeofday.h +9
-0@@ -67,6 +67,20 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)return ret; } +#ifdef CONFIG_HYPERV_TIMER +/* This will override the default hv_get_raw_timer() */ +#define hv_get_raw_timer () __arch_counter_get_cntvct() +#include <clocksource/hyperv_timer.h> + +extern struct ms_hyperv_tsc_page +_hvclock_page __attribute__((visibility("hidden"))); + +static u64 vread_hvclock (void )
Quoted 9 lines
arch/arm64/include/asm/vdso/gettimeofday.h int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts) · line 82
Quoted 4 lines
arch/arm64/include/asm/vdso/gettimeofday.h +4
-0+{ +return hv_read_tsc_page (&_hvclock_page); +} +#endif The function is almost the same on x86 (
&_hvclock_page->&hvclock_page), would it maybe make sense to move this to arch neutralclocksource/hyperv_timer.h?
I'm not sure whether the underscore matters in the vDSO data symbol, so I follow the architectural name convention. If the leading underscore doesn't have special purpose I'm happy to move this to arch neutral header file.
Quoted 8 lines
Quoted 8 lines
arch/arm64/include/asm/vdso/gettimeofday.h +1
-0+ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode) {u64 res;@@ -78,6 +92,11 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)if (clock_mode ==VCLOCK_NONE )return __VDSO_USE_SYSCALL;
Quoted 8 lines
arch/arm64/include/asm/vdso/gettimeofday.h static __always_inline u64 __arch_get_hw_counter(s32 clock_mode) · line 97
Quoted 3 lines
arch/arm64/include/asm/vdso/gettimeofday.h +3
-0+#ifdef CONFIG_HYPERV_TIMER +if (likely (clock_mode ==VCLOCK_HVCLOCK )) +return vread_hvclock ();I'm not sure
likely()is justified here: it'll make ALL builds which enableCONFIG_HYPERV_TIMER(e.g. distro kernels) to preferVCLOCK_HVCLOCK, even if the kernel is not running on Hyper-V.
Make sense. Thanks for pointing this out! I will change it in the next version.
Regards,
Boqun
Quoted 9 lines
Quoted 5 lines
arch/arm64/include/asm/vdso/gettimeofday.h +2
-0+#endif +/* * This isb() is required to prevent that the counter value * is speculated. Signature · 1 line
Vitaly
Hi Boqun Feng,
sorry for the late reply.
Quoted 7 lines · Boqun Feng
Hi,
This is the RFC patchset for vDSO support in ARM64 Hyper-V guest. To test it, Michael's ARM64 support patchset:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
Quoted 8 lines
is needed.
Similar as x86, Hyper-V on ARM64 use a TSC page for guests to read the virtualized hardware timer, this TSC page is read-only for the guests, so could be used for vDSO data page. And the vDSO (userspace) code could use the same code for timer reading as kernel, since they read the same TSC page.
I had a look to your patches and overall, I could not understand why we can't use the arch_timer to do the same things you are doing with the one you introduced in this series. What confuses me is that KVM works just fine with the arch_timer which was designed with virtualization in mind. Why do we need another one? Could you please explain?
Quoted 39 lines
This patchset therefore extends ARM64's
__vsdo_init()to allow multiple data pages and introduces thevclock_modeconcept similar to x86 to allow different platforms (bare-metal, Hyper-V, etc.) to switch to different__arch_get_hw_counter()implementations. The rest of this patchset does the necessary setup for Hyper-V guests: mapping tsc page, enabling userspace to read cntvct, etc. to enable vDSO.This patchset consists of 6 patches:
patch #1 allows
hv_get_raw_timer()definition to be overridden for userspace and kernel to share the samehv_read_tsc_page()definition.patch #2 extends ARM64 to support multiple vDSO data pages.
patch #3 introduces
vclock_modesimiliar to x86 to allow different__arch_get_hw_counter()implementations for different clocksources.patch #4 maps Hyper-V TSC page into vDSO data page.
patch #5 allows userspace to read cntvct, so that userspace can efficiently read the clocksource.
patch #6 enables the vDSO for ARM64 Hyper-V guest.
The whole patchset is based on v5.5-rc1 plus Michael's ARM64 support patchset, and I've done a few tests with:
https://github.com/nlynch-mentor/vdsotest
Comments and suggestions are welcome!
Regards,
Boqun_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Signature · 2 lines
Regards, Vincenzo
Hi Vincenzo,
Quoted 4 lines · Vincenzo Frascino
Hi Boqun Feng,
sorry for the late reply.
That's OK, thanks for your review ;-)
Quoted 17 lines · Boqun Feng
Quoted 7 lines
Hi,
This is the RFC patchset for vDSO support in ARM64 Hyper-V guest. To test it, Michael's ARM64 support patchset:
https://lore.kernel.org/linux-arm-kernel/[email protected]/Quoted 8 lines
is needed.
Similar as x86, Hyper-V on ARM64 use a TSC page for guests to read the virtualized hardware timer, this TSC page is read-only for the guests, so could be used for vDSO data page. And the vDSO (userspace) code could use the same code for timer reading as kernel, since they read the same TSC page.
Quoted 6 lines
I had a look to your patches and overall, I could not understand why we can't use the
arch_timerto do the same things you are doing with the one you introduced in this series. What confuses me is that KVM works just fine with thearch_timerwhich was designed with virtualization in mind. Why do we need another one? Could you please explain?
Please note that the guest VM on Hyper-V for ARM64 doesn't use arch_timer as the clocksource. See:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
, ACPI_SIG_GTDT is used for setting up Hyper-V synthetic clocksource and other initialization work.
So just to be clear, your suggestion is
1) Hyper-V guest on ARM64 should use arch_timer as clocksource and vDSO will just work.
or
2) Even though arch_timer is not used as the clocksource, we can still use it for vDSO.
?
Regards,
Boqun
Quoted 43 lines
Quoted 39 lines
This patchset therefore extends ARM64's
__vsdo_init()to allow multiple data pages and introduces thevclock_modeconcept similar to x86 to allow different platforms (bare-metal, Hyper-V, etc.) to switch to different__arch_get_hw_counter()implementations. The rest of this patchset does the necessary setup for Hyper-V guests: mapping tsc page, enabling userspace to read cntvct, etc. to enable vDSO.This patchset consists of 6 patches:
patch #1 allows
hv_get_raw_timer()definition to be overridden for userspace and kernel to share the samehv_read_tsc_page()definition.patch #2 extends ARM64 to support multiple vDSO data pages.
patch #3 introduces
vclock_modesimiliar to x86 to allow different__arch_get_hw_counter()implementations for different clocksources.patch #4 maps Hyper-V TSC page into vDSO data page.
patch #5 allows userspace to read cntvct, so that userspace can efficiently read the clocksource.
patch #6 enables the vDSO for ARM64 Hyper-V guest.
The whole patchset is based on v5.5-rc1 plus Michael's ARM64 support patchset, and I've done a few tests with:
https://github.com/nlynch-mentor/vdsotest
Comments and suggestions are welcome!
Regards,
Boqun_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernelSignature · 2 lines
Regards, Vincenzo
Hi Boqun Feng,
Quoted 2 lines · Boqun Feng
Hi Vincenzo,
[...]
Quoted 8 lines
Quoted 7 lines
I had a look to your patches and overall, I could not understand why we can't use the
arch_timerto do the same things you are doing with the one you introduced in this series. What confuses me is that KVM works just fine with thearch_timerwhich was designed with virtualization in mind. Why do we need another one? Could you please explain?
Quoted 8 lines
Please note that the guest VM on Hyper-V for ARM64 doesn't use
arch_timeras the clocksource. See:https://lore.kernel.org/linux-arm-kernel/[email protected]/,
ACPI_SIG_GTDTis used for setting up Hyper-V synthetic clocksource and other initialization work.
I had a look a look at it and my question stands, why do we need another timer on arm64?
Quoted 5 lines
So just to be clear, your suggestion is
1) Hyper-V guest on ARM64 should use
arch_timeras clocksource and vDSO will just work.
Quoted 7 lines
or
2) Even though
arch_timeris not used as the clocksource, we can still use it for vDSO.?
Option #1 would be the preferred solution, unless there is a good reason against.
Quoted 3 lines
Regards,
Boqun
Signature · 2 lines
Regards, Vincenzo
Quoted 19 lines · Vincenzo Frascino
Hi Boqun Feng,
Quoted 2 lines · Boqun Feng
Hi Vincenzo,
[...]
Quoted 11 lines
Quoted 7 lines
I had a look to your patches and overall, I could not understand why we can't use the
arch_timerto do the same things you are doing with the one you introduced in this series. What confuses me is that KVM works just fine with thearch_timerwhich was designed with virtualization in mind. Why do we need another one? Could you please explain?Please note that the guest VM on Hyper-V for ARM64 doesn't use
arch_timeras the clocksource. See:
Quoted 9 lines
Quoted 5 lines
https://lore.kernel.org/linux-arm-kernel/[email protected]/,
ACPI_SIG_GTDTis used for setting up Hyper-V synthetic clocksource and other initialization work.I had a look a look at it and my question stands, why do we need another timer on arm64?
Sorry for the late response. It's weekend and Chinese New Year, so I got to spend some time making (and mostly eating) dumplings ;-)
After discussion with Michael, here is some explanation why we need another timer:
The synthetic clocks that Hyper-V presents in a guest VM were originally created for the x86 architecture. They provide a level of abstraction that solves problems like continuity across live migrations where the hardware clock (i.e., TSC in the case x86) frequency may be different across the migration. When Hyper-V was brought to ARM64, this abstraction was maintained to provide consistency across the x86 and
ARM64 architectures, and for both Windows and Linux guest VMs. The
core Linux code for the Hyper-V clocks (in drivers/clocksource/hyperv_timer.c) is architecture neutral and works on both x86 and ARM64. As you can see, this part is done in Michael's patchset.
Arguably, Hyper-V for ARM64 should have optimized for consistency with the ARM64 community rather with the existing x86 implementation and existing guest code in Windows. But at this point, it is what it is, and the Hyper-V clocks do solve problems like migration that aren’t addressed in ARM64 until v8.4 of the architecture with the addition of the counter hardware scaling feature. Hyper-V doesn’t currently map the ARM arch timer interrupts into guest VMs, so we need to use the existing Hyper-V clocks and the common code that already exists.
Does the above answer your question?
Regards,
Boqun
Quoted 22 lines
Quoted 5 lines
So just to be clear, your suggestion is
1) Hyper-V guest on ARM64 should use
arch_timeras clocksource and vDSO will just work.Quoted 7 lines
or
2) Even though
arch_timeris not used as the clocksource, we can still use it for vDSO.?
Option #1 would be the preferred solution, unless there is a good reason against.
Quoted 3 lines
Regards,
BoqunSignature · 2 lines
Regards, Vincenzo
Quoted 26 lines · Boqun Feng
Quoted 25 lines · Vincenzo Frascino
Hi Boqun Feng,
Quoted 2 lines · Boqun Feng
Hi Vincenzo,
[...]
Quoted 16 lines
Quoted 7 lines
I had a look to your patches and overall, I could not understand why we can't use the
arch_timerto do the same things you are doing with the one you introduced in this series. What confuses me is that KVM works just fine with thearch_timerwhich was designed with virtualization in mind. Why do we need another one? Could you please explain?Please note that the guest VM on Hyper-V for ARM64 doesn't use
arch_timeras the clocksource. See:https://lore.kernel.org/linux-arm-kernel/[email protected]/,
ACPI_SIG_GTDTis used for setting up Hyper-V synthetic clocksource and other initialization work.
Quoted 6 lines
Quoted 3 lines
I had a look a look at it and my question stands, why do we need another timer on arm64?
Sorry for the late response. It's weekend and Chinese New Year, so I got to spend some time making (and mostly eating) dumplings ;-)
And you haven't been sharing! ;-)
Quoted 15 lines
After discussion with Michael, here is some explanation why we need another timer:
The synthetic clocks that Hyper-V presents in a guest VM were originally created for the x86 architecture. They provide a level of abstraction that solves problems like continuity across live migrations where the hardware clock (i.e., TSC in the case x86) frequency may be different across the migration. When Hyper-V was brought to ARM64, this abstraction was maintained to provide consistency across the x86 and
ARM64 architectures, and for both Windows and Linux guest VMs. The
core Linux code for the Hyper-V clocks (indrivers/clocksource/hyperv_timer.c) is architecture neutral and works on both x86 and ARM64. As you can see, this part is done in Michael's patchset.
Quoted 8 lines
Arguably, Hyper-V for ARM64 should have optimized for consistency with the ARM64 community rather with the existing x86 implementation and existing guest code in Windows. But at this point, it is what it is, and the Hyper-V clocks do solve problems like migration that aren’t addressed in ARM64 until v8.4 of the architecture with the addition of the counter hardware scaling feature. Hyper-V doesn’t currently map the ARM arch timer interrupts into guest VMs, so we need to use the existing Hyper-V clocks and the common code that already exists.
The migration thing is a bit of a red herring. Do you really anticipate VM migration across systems that have their timers running at different frequencies *today*? And even if you did, there are ways to deal with it with the arch timers (patches to that effect were posted on the list, and there was even a bit of an ARM spec for it).
I find it odd to try and make arm64 "just another x86", while the architecture gives you most of what you need already. I guess I'm tainted.
Thanks,
M.
Signature · 6 lines
Who you jivin' with that Cosmik Debris? _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel