PV extension of HVM (Hybrid) for Xen
Subsystems (8)
v7 cover PV extension of HVM (Hybrid) for Xen
Hi, Jeremy
Here is the seventh version of patchset to enable PV extension of HVM support in Linux kernel of Xen.
The patch based on upstream 2.6.33 with pv_ops domU support.
The PV extension of HVM is started from real mode like HVM guest, but also with a a range of PV features(e.g. PV timer, event channel, and PV
drivers in the future). So guest with this feature can takes the advantages of both H/W virtualization and Para-Virtualization.
The first two of the patchset imported several header file from Jeremy's tree and Xen tree, respect to Jeremy and Keir's works.
The whole patchset based on Linux upstream.
You need a line like:
cpuid = [ '0x40000002:edx=0x7' ]
in HVM configuration file to expose all PV feature to guest, and
CONFIG_XEN
in the guest kernel configuration file to enable the PV extension support. It would implicit enable CONFIG_XEN_HVM_PV if CONFIG_X86_64 is selected as well.
And the compiled image can be used as native/pv domU/hvm guest/pv feature hvm kernel.
Change from v6:
1. Make PV clocksource as a separate feature rather than default one. Notice PV evtchn still depends on it.
2. Some comments update.
Change from v5:
Update the comments from Jeremy.
Change from v4:
1. Add a new CONFIG_XEN_HVM_PV to enable the feature in kernel
2. Separate the related code form enlighted.c to hvmpv.c
3. Separate the feature "PV clocksource" from evtchn. Now we can support HVM guest with PV clocksource. This would be enabled by default.
4. Drop PV halt and pv drivers in this edition. We can work on that later.
5. Update the patchset following Jeremy's comments.
Change from v3:
1. Rebase to Linux 2.6.33 release.
2. change the name to "PV extension of HVM"
3. Some minor coding polishing.
Change from v2:
1. change the name "hybrid" to "PV featured HVM".
2. Unified the PV driver's judgement of xen_domain() to xen_evtchn_enabled().
3. Move the function(evtchn) initialize hypercall near the real enabling place, rather than a unified place before function enabled.
4. Remove the reserved E820 region for grant table. Use QEmu Xen platform device's MMIO instead.
The major change from v1:
1. SMP support.
2. Modify the entrance point to avoid most of genernic kernel modification.
3. Binding PV timer with event channel mechanism.
--
regards
Yang, Sheng
arch/x86/include/asm/xen/cpuid.h | 75 +++++++++++ arch/x86/include/asm/xen/hypercall.h | 6 + arch/x86/include/asm/xen/hypervisor.h | 6 + arch/x86/kernel/setup.c | 4 + arch/x86/xen/Kconfig | 4 + arch/x86/xen/Makefile | 1 + arch/x86/xen/enlighten.c | 6 +- arch/x86/xen/hvmpv.c | 229 +++++++++++++++++++++++++++++++++ arch/x86/xen/irq.c | 28 ++++ arch/x86/xen/smp.c | 76 ++++++++++- arch/x86/xen/time.c | 12 ++- arch/x86/xen/xen-ops.h | 17 +++ drivers/block/xen-blkfront.c | 2 +- drivers/input/xen-kbdfront.c | 2 +- drivers/net/xen-netfront.c | 2 +- drivers/video/xen-fbfront.c | 2 +- drivers/xen/events.c | 74 ++++++++++- drivers/xen/grant-table.c | 2 +- drivers/xen/xenbus/xenbus_probe.c | 4 +- include/xen/events.h | 4 + include/xen/hvm.h | 28 ++++ include/xen/interface/hvm/hvm_op.h | 82 ++++++++++++ include/xen/interface/hvm/params.h | 111 ++++++++++++++++ include/xen/interface/xen.h | 6 +- include/xen/xen.h | 15 ++ 25 files changed, 776 insertions(+), 22 deletions(-)
PATCH v7 5/7 x86/xen: The entrance for PV extension of HVM
xen_guest_init() would setup the environment.
--- arch/x86/include/asm/xen/hypervisor.h | 6 ++++++ arch/x86/kernel/setup.c | 4 ++++ 2 files changed, 10 insertions(+), 0 deletions(-)
arch/x86/include/asm/xen/hypervisor.h +6 -0
@@ -37,4 +37,10 @@extern struct shared_info *HYPERVISOR_shared_info;extern struct start_info *xen_start_info; +#ifdef CONFIG_XEN_HVM_PV +void __init xen_guest_init (void ); +#else +static inline void xen_guest_init (void ) {}; +#endif +#endif /* _ASM_X86_XEN_HYPERVISOR_H */
arch/x86/kernel/setup.c +4 -0
@@ -113,6 +113,8 @@#endif #include <asm/mce.h> +#include <asm/xen/hypervisor.h> +/* * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. * The direct mapping extends to max_pfn_mapped, so that we can directly access
@@ -740,6 +742,8 @@ void __init setup_arch(char **cmdline_p) x86_init.oem.arch_setup (); +xen_guest_init (); +setup_memory_map ();parse_setup_data ();/* update the e820_saved too */
Signature · 1 line
1.5.4.5
PATCH v7 6/7 xen: Enable PV clocksource for HVM
In reply to Sheng YangPV clocksource can provide a reliable clocksource for HVM running on Xen.
To enable it, put following line to the HVM configure file:
cpuid = [ '0x40000002:edx=0x3' ]
It would set bit 0 and bit 1 in 0x40000002:edx, which enable PV extension framework and PV clocksource.
--- arch/x86/xen/hvmpv.c | 24 ++++++++++++++++++++++++ arch/x86/xen/time.c | 12 +++++++++++- arch/x86/xen/xen-ops.h | 1 + 3 files changed, 36 insertions(+), 1 deletions(-)
arch/x86/xen/hvmpv.c +24 -0
@@ -43,6 +43,8 @@ static void __init xen_hvm_pv_banner(void) pv_info.name);printk (KERN_INFO "Xen version: %d.%d%s\n" , version >>16 , version &0xffff , extra.extraversion); +if (xen_hvm_pv_clock_enabled ()) +printk (KERN_INFO "PV feature: PV clocksource enabled\n" ); }static int __init xen_para_available (void )
@@ -82,6 +84,9 @@ static int __init init_hvm_pv_info(void)if (!(edx &XEN_CPUID_FEAT2_HVM_PV ))return -ENODEV ; +if (edx &XEN_CPUID_FEAT2_HVM_PV_CLOCK ) + xen_hvm_pv_features |=XEN_HVM_PV_CLOCK_ENABLED ; +if (pages <1 )return -ENODEV ;
@@ -112,6 +117,23 @@ static int __init init_shared_info(void)return 0 ; } +static void __init init_pv_clocksource (void ) +{ +if (!xen_hvm_pv_clock_enabled ()) +return ; + +if (enable_hvm_pv (HVM_PV_CLOCK )) +BUG (); + + pv_time_ops.sched_clock = xen_sched_clock; + + x86_platform.calibrate_tsc = xen_tsc_khz; + x86_platform.get_wallclock = xen_get_wallclock; + x86_platform.set_wallclock = xen_set_wallclock; + +xen_register_clocksource (); +} +void __init xen_guest_init (void ) {int r;
@@ -134,4 +156,6 @@ void __init xen_guest_init(void) pv_info = xen_hvm_pv_info; xen_domain_type =XEN_HVM_DOMAIN ; + +init_pv_clocksource (); }
arch/x86/xen/time.c +11 -1
@@ -472,11 +472,21 @@ void xen_timer_resume(void) } } +static bool xen_clocksource_enabled; + +void xen_register_clocksource (void ) +{ +if (!xen_clocksource_enabled) { +clocksource_register (&xen_clocksource); + xen_clocksource_enabled =1 ; + } +} +__init void xen_time_init (void ) {int cpu =smp_processor_id (); -clocksource_register (&xen_clocksource); +xen_register_clocksource ();if (HYPERVISOR_vcpu_op (VCPUOP_stop_periodic_timer, cpu,NULL ) ==0 ) {/* Successfully turned off 100Hz tick, so we have the
arch/x86/xen/xen-ops.h +1 -0
@@ -48,6 +48,7 @@ cycle_t xen_clocksource_read(void);void xen_setup_cpu_clockevents (void );unsigned long xen_tsc_khz (void );void __init xen_time_init (void ); +void xen_register_clocksource (void );unsigned long xen_get_wallclock (void );int xen_set_wallclock (unsigned long time);unsigned long long xen_sched_clock (void );
Signature · 1 line
1.5.4.5
PATCH v7 3/7 xen: Make pv drivers only work with xen_pv_domain()
In reply to Sheng YangOtherwise they would still try to enable with HVM domain type.
--- drivers/block/xen-blkfront.c | 2 +- drivers/input/xen-kbdfront.c | 2 +- drivers/net/xen-netfront.c | 2 +- drivers/video/xen-fbfront.c | 2 +- drivers/xen/grant-table.c | 2 +- drivers/xen/xenbus/xenbus_probe.c | 4 ++-- 6 files changed, 7 insertions(+), 7 deletions(-)
drivers/block/xen-blkfront.c +1 -1
@@ -1068,7 +1068,7 @@ static struct xenbus_driver blkfront = {static int __init xlblk_init (void ) { -if (!xen_domain ()) +if (!xen_pv_domain ())return -ENODEV ;if (register_blkdev (XENVBD_MAJOR ,DEV_NAME )) {
drivers/input/xen-kbdfront.c +1 -1
@@ -338,7 +338,7 @@ static struct xenbus_driver xenkbd_driver = {static int __init xenkbd_init (void ) { -if (!xen_domain ()) +if (!xen_pv_domain ())return -ENODEV ;/* Nothing to do if running in dom0. */
drivers/net/xen-netfront.c +1 -1
@@ -1804,7 +1804,7 @@ static struct xenbus_driver netfront_driver = {static int __init netif_init (void ) { -if (!xen_domain ()) +if (!xen_pv_domain ())return -ENODEV ;if (xen_initial_domain ())
drivers/video/xen-fbfront.c +1 -1
@@ -683,7 +683,7 @@ static struct xenbus_driver xenfb_driver = {static int __init xenfb_init (void ) { -if (!xen_domain ()) +if (!xen_pv_domain ())return -ENODEV ;/* Nothing to do if running in dom0. */
drivers/xen/grant-table.c +1 -1
@@ -510,7 +510,7 @@ static int __devinit gnttab_init(void)unsigned int max_nr_glist_frames, nr_glist_frames;unsigned int nr_init_grefs; -if (!xen_domain ()) +if (!xen_pv_domain ())return -ENODEV ; nr_grant_frames =1 ;
drivers/xen/xenbus/xenbus_probe.c +2 -2
@@ -786,7 +786,7 @@ static int __init xenbus_probe_init(void)DPRINTK ("" ); err = -ENODEV ; -if (!xen_domain ()) +if (!xen_pv_domain ()) goto out_error;/* Register ourselves with the kernel bus subsystem */
@@ -922,7 +922,7 @@ static void wait_for_devices(struct xenbus_driver *xendrv)struct device_driver *drv = xendrv ? &xendrv->driver :NULL ;unsigned int seconds_waited =0 ; -if (!ready_to_wait_for_devices || !xen_domain ()) +if (!ready_to_wait_for_devices || !xen_pv_domain ())return ;while (exists_connecting_device (drv)) {
Signature · 1 line
1.5.4.5
PATCH v7 1/7 xen: add support for hvm_op
In reply to Sheng YangFrom: Jeremy Fitzhardinge <[email protected]>
Add support for hvm_op hypercall.
--- arch/x86/include/asm/xen/hypercall.h | 6 ++ include/xen/hvm.h | 23 +++++++ include/xen/interface/hvm/hvm_op.h | 72 ++++++++++++++++++++++ include/xen/interface/hvm/params.h | 111 ++++++++++++++++++++++++++++++++++ 4 files changed, 212 insertions(+), 0 deletions(-) create mode 100644 include/xen/hvm.h create mode 100644 include/xen/interface/hvm/hvm_op.h create mode 100644 include/xen/interface/hvm/params.h
arch/x86/include/asm/xen/hypercall.h +6 -0
@@ -417,6 +417,12 @@ HYPERVISOR_nmi_op(unsigned long op, unsigned long arg)return _hypercall2 (int, nmi_op, op, arg); } +static inline unsigned long __must_check +HYPERVISOR_hvm_op (int op,void *arg) +{ + return _hypercall2(unsigned long, hvm_op, op, arg); +} +static inline void MULTI_fpu_taskswitch (struct multicall_entry *mcl,int set) {
include/xen/hvm.h new file +23 -0
@@ -0,0 +1,23 @@ +/* Simple wrappers around HVM functions */ +#ifndef XEN_HVM_H__ +#define XEN_HVM_H__ + +#include <xen/interface/hvm/params.h> + +static inline unsigned long hvm_get_parameter (int idx) +{ +struct xen_hvm_param xhv; +int r; + + xhv.domid =DOMID_SELF ; + xhv.index = idx; + r =HYPERVISOR_hvm_op (HVMOP_get_param, &xhv); +if (r <0 ) { +printk (KERN_ERR "cannot get hvm parameter %d: %d.\n" , + idx, r); +return 0 ; + } +return xhv.value; +} + +#endif /* XEN_HVM_H__ */
include/xen/interface/hvm/hvm_op.h new file +72 -0
72 lines not on this page · raw message
include/xen/interface/hvm/params.h new file +111 -0
111 lines not on this page · raw message
Signature · 1 line
1.5.4.5
PATCH v7 7/7 xen: Enable event channel of PV extension of HVM
In reply to Sheng YangWe mapped each IOAPIC pin to a VIRQ, so that we can deliver interrupt through these VIRQs.
We used X86_PLATFORM_IPI_VECTOR as the notification vector for hypervisor to notify guest about the event.
The patch also enabled SMP support, then we can support IPI through evtchn as well.
When this feature is enabled, we would relay on Xen PV timer for clockevent, rather than other hardware emulated ones.
Then we don't use IOAPIC/LAPIC, eliminated the overhead brought by unnecessary VMExit caused by LAPIC.
PV evtchn depends on PV clocksource. To enable it, put following line in the HVM configure file:
cpuid = [ '0x40000002:edx=0x7' ]
It would enable PV extension framework(bit 0), PV clocksource(bit 1), as well as PV evtchn(bit 2).
Notice if you try to enable PV evtchn without other two bits set, the setting would be ignored.
--- arch/x86/xen/enlighten.c | 6 ++-- arch/x86/xen/hvmpv.c | 70 +++++++++++++++++++++++++++++++++++++++- arch/x86/xen/irq.c | 28 ++++++++++++++++ arch/x86/xen/smp.c | 76 ++++++++++++++++++++++++++++++++++++++++-- arch/x86/xen/xen-ops.h | 16 +++++++++ drivers/xen/events.c | 74 ++++++++++++++++++++++++++++++++++++++--- include/xen/events.h | 4 ++ include/xen/hvm.h | 5 +++ include/xen/interface/xen.h | 6 +++- 9 files changed, 270 insertions(+), 15 deletions(-)
arch/x86/xen/enlighten.c +3 -3
@@ -717,7 +717,7 @@ static u32 xen_safe_apic_wait_icr_idle(void)return 0 ; } -static void set_xen_basic_apic_ops (void ) +void xen_set_basic_apic_ops (void ) { apic->read = xen_apic_read; apic->write = xen_apic_write;
@@ -1026,7 +1026,7 @@ static void xen_crash_shutdown(struct pt_regs *regs)xen_reboot (SHUTDOWN_crash); } -static const struct machine_ops __initdata xen_machine_ops = { +const struct machine_ops __initdata xen_machine_ops = { .restart = xen_restart, .halt = xen_machine_halt, .power_off = xen_machine_halt,
@@ -1116,7 +1116,7 @@ asmlinkage void __init xen_start_kernel(void)/* * set up the basic apic ops. */ -set_xen_basic_apic_ops (); +xen_set_basic_apic_ops ();#endif if (xen_feature (XENFEAT_mmu_pt_update_preserve_ad)) {
arch/x86/xen/hvmpv.c +69 -1
100 lines not on this page · raw message
arch/x86/xen/irq.c +28 -0
@@ -2,6 +2,7 @@#include <asm/x86_init.h> +#include <xen/xen.h> #include <xen/interface/xen.h> #include <xen/interface/sched.h> #include <xen/interface/vcpu.h>
@@ -131,3 +132,30 @@ void __init xen_init_irq_ops() pv_irq_ops = xen_irq_ops; x86_init.irqs.intr_init = xen_init_IRQ; } + +#ifdef CONFIG_XEN_HVM_PV +static void xen_hvm_pv_evtchn_disable (void ) +{ +native_irq_disable (); +xen_irq_disable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_disable); + +static void xen_hvm_pv_evtchn_enable (void ) +{ +native_irq_enable (); +xen_irq_enable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_enable); + +void __init xen_hvm_pv_init_irq_ops (void ) +{ +if (xen_hvm_pv_evtchn_enabled ()) { + pv_irq_ops.irq_disable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_disable); + pv_irq_ops.irq_enable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_enable); + x86_init.irqs.intr_init = xen_hvm_pv_evtchn_init_IRQ; + } +} +#endif
arch/x86/xen/smp.c +72 -4
127 lines not on this page · raw message
arch/x86/xen/xen-ops.h +16 -0
@@ -52,6 +52,12 @@ void xen_register_clocksource(void);unsigned long xen_get_wallclock (void );int xen_set_wallclock (unsigned long time);unsigned long long xen_sched_clock (void ); +void xen_set_basic_apic_ops (void ); + +#ifdef CONFIG_XEN_HVM_PV +void __init xen_hvm_pv_init_irq_ops (void ); +void __init xen_hvm_pv_evtchn_init_IRQ (void ); +#endif /* CONFIG_XEN_HVM_PV */ irqreturn_t xen_debug_interrupt (int irq,void *dev_id);
@@ -62,9 +68,17 @@ void xen_setup_vcpu_info_placement(void);#ifdef CONFIG_SMP void xen_smp_init (void ); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_pv_smp_init (void ); +#endif /* CONFIG_XEN_HVM_PV */ +extern cpumask_var_t xen_cpu_initialized_map;#else static inline void xen_smp_init (void ) {} +#ifdef CONFIG_XEN_HVM_PV +static inline void xen_hvm_pv_smp_init (void ) {} +#endif /* CONFIG_XEN_HVM_PV */ +#endif #ifdef CONFIG_PARAVIRT_SPINLOCKS
@@ -102,4 +116,6 @@ void xen_sysret32(void);void xen_sysret64 (void );void xen_adjust_exception_frame (void ); +extern const struct machine_ops xen_machine_ops; +#endif /* XEN_OPS_H */
drivers/xen/events.c +68 -6
112 lines not on this page · raw message
include/xen/events.h +4 -0
@@ -56,4 +56,8 @@ void xen_poll_irq(int irq);/* Determine the IRQ which is bound to an event channel */ unsigned irq_from_evtchn (unsigned int evtchn); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_evtchn_do_upcall (struct pt_regs *regs); +#endif +#endif /* _XEN_EVENTS_H */
include/xen/hvm.h +5 -0
@@ -20,4 +20,9 @@ static inline unsigned long hvm_get_parameter(int idx) return xhv.value; } +#define HVM_CALLBACK_VIA_TYPE_VECTOR 0x2 +#define HVM_CALLBACK_VIA_TYPE_SHIFT 56 +#define HVM_CALLBACK_VECTOR (x) (((uint64_t)HVM_CALLBACK_VIA_TYPE_VECTOR)<<\ + HVM_CALLBACK_VIA_TYPE_SHIFT | (x)) +#endif /* XEN_HVM_H__ */
include/xen/interface/xen.h +5 -1
@@ -90,7 +90,11 @@#define VIRQ_ARCH_6 22#define VIRQ_ARCH_7 23 -#define NR_VIRQS 24 +#define VIRQ_EMUL_PIN_START 24 +#define VIRQ_EMUL_PIN_NUM 16 + +#define NR_VIRQS (VIRQ_EMUL_PIN_START + VIRQ_EMUL_PIN_NUM) +/* * MMU-UPDATE REQUESTS *
Signature · 1 line
1.5.4.5
PATCH v7 2/7 xen: Import cpuid.h from Xen
In reply to Sheng YangFrom: Keir Fraser <[email protected]>
Which would be used by CPUID detection later
--- arch/x86/include/asm/xen/cpuid.h | 68 ++++++++++++++++++++++++++++++++++++++ 1 files changed, 68 insertions(+), 0 deletions(-) create mode 100644 arch/x86/include/asm/xen/cpuid.h
arch/x86/include/asm/xen/cpuid.h new file +68 -0
@@ -0,0 +1,68 @@ +/****************************************************************************** +* arch/include/asm/xen/cpuid.h +* +* CPUID interface to Xen. +* +* Permission is hereby granted, free of charge, to any person obtaining a copy +* of this software and associated documentation files (the "Software"), to +* deal in the Software without restriction, including without limitation the +* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +* sell copies of the Software, and to permit persons to whom the Software is +* furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice shall be included in +* all copies or substantial portions of the Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER +* DEALINGS IN THE SOFTWARE. +* +* Copyright (c) 2007 Citrix Systems, Inc. +* +* Authors: +* Keir Fraser <[email protected]> +*/ + +#ifndef __ASM_X86_XEN_CPUID_H__ +#define __ASM_X86_XEN_CPUID_H__ + +/* Xen identification leaves start at 0x40000000. */ +#define XEN_CPUID_FIRST_LEAF 0x40000000 +#define XEN_CPUID_LEAF (i) (XEN_CPUID_FIRST_LEAF + (i)) + +/* +* Leaf 1 (0x40000000) +* EAX: Largest Xen-information leaf. All leaves up to an including @EAX +* are supported by the Xen host. +* EBX-EDX: "XenVMMXenVMM" signature, allowing positive identification +* of a Xen host. +*/ +#define XEN_CPUID_SIGNATURE_EBX 0x566e6558/* "XenV" */ +#define XEN_CPUID_SIGNATURE_ECX 0x65584d4d/* "MMXe" */ +#define XEN_CPUID_SIGNATURE_EDX 0x4d4d566e/* "nVMM" */ + +/* +* Leaf 2 (0x40000001) +* EAX[31:16]: Xen major version. +* EAX[15: 0]: Xen minor version. +* EBX-EDX: Reserved (currently all zeroes). +*/ + +/* +* Leaf 3 (0x40000002) +* EAX: Number of hypercall transfer pages. This register is always guaranteed +* to specify one hypercall page. +* EBX: Base address of Xen-specific MSRs. +* ECX: Features 1. Unused bits are set to zero. +* EDX: Features 2. Unused bits are set to zero. +*/ + +/* Does the host support MMU_PT_UPDATE_PRESERVE_AD for this guest? */ +#define _XEN_CPUID_FEAT1_MMU_PT_UPDATE_PRESERVE_AD 0 +#define XEN_CPUID_FEAT1_MMU_PT_UPDATE_PRESERVE_AD (1u<<0) + +#endif /* __XEN_PUBLIC_ARCH_X86_CPUID_H__ */
Signature · 1 line
1.5.4.5
PATCH v7 4/7 xen/hvm: Xen PV extension of HVM initialization
In reply to Sheng YangThe PV extended HVM(once known as Hybrid) is started from real mode like HVM guest, but also with a component based PV feature selection(e.g. PV halt, PV timer, event channel, then PV drivers). So guest can takes the advantages of both H/W virtualization and Para-Virtualization.
This patch introduced the PV extension of HVM guest initialization.
Guest would detect the capability using CPUID 0x40000002.edx, then call HVMOP_enable_pv hypercall to enable pv support in hypervisor.
--- arch/x86/include/asm/xen/cpuid.h | 7 ++ arch/x86/xen/Kconfig | 4 + arch/x86/xen/Makefile | 1 + arch/x86/xen/hvmpv.c | 137 ++++++++++++++++++++++++++++++++++++ include/xen/interface/hvm/hvm_op.h | 10 +++ include/xen/xen.h | 15 ++++ 6 files changed, 174 insertions(+), 0 deletions(-) create mode 100644 arch/x86/xen/hvmpv.c
arch/x86/include/asm/xen/cpuid.h +7 -0
@@ -65,4 +65,11 @@#define _XEN_CPUID_FEAT1_MMU_PT_UPDATE_PRESERVE_AD 0#define XEN_CPUID_FEAT1_MMU_PT_UPDATE_PRESERVE_AD (1u<<0) +#define _XEN_CPUID_FEAT2_HVM_PV 0 +#define XEN_CPUID_FEAT2_HVM_PV (1u<<0) +#define _XEN_CPUID_FEAT2_HVM_PV_CLOCK 1 +#define XEN_CPUID_FEAT2_HVM_PV_CLOCK (1u<<1) +#define _XEN_CPUID_FEAT2_HVM_PV_EVTCHN 1 +#define XEN_CPUID_FEAT2_HVM_PV_EVTCHN (1u<<2) +#endif /* __XEN_PUBLIC_ARCH_X86_CPUID_H__ */
arch/x86/xen/Kconfig +4 -0
@@ -36,3 +36,7 @@ config XEN_DEBUG_FS help Enable statistics output and various tuning options in debugfs. Enabling this option may incur a significant performance overhead. + +config XEN_HVM_PV + def_bool y + depends on XEN && X86_64
arch/x86/xen/Makefile +1 -0
@@ -17,4 +17,5 @@ obj-y := enlighten.o setup.o multicalls.o mmu.o irq.o \ obj-$(CONFIG_SMP) += smp.o obj-$(CONFIG_PARAVIRT_SPINLOCKS)+= spinlock.o obj-$(CONFIG_XEN_DEBUG_FS) += debugfs.o +obj-$(CONFIG_XEN_HVM_PV) += hvmpv.o
arch/x86/xen/hvmpv.c new file +137 -0
137 lines not on this page · raw message
include/xen/interface/hvm/hvm_op.h +10 -0
@@ -69,4 +69,14 @@ DEFINE_GUEST_HANDLE_STRUCT(xen_hvm_set_pci_link_route);/* Flushes all VCPU TLBs: @arg must be NULL. */ #define HVMOP_flush_tlbs 5 +/* Enable PV extended HVM mode. Should called by BSP */ +#define HVMOP_enable_pv 9 +struct xen_hvm_pv_type { +/* Should be DOMID_SELF so far */ +domid_t domid; +/* The features want to be enabled */ +uint32_t flags; +#define HVM_PV_CLOCK (1ull<<0) +}; +#endif /* __XEN_PUBLIC_HVM_HVM_OP_H__ */
include/xen/xen.h +15 -0
@@ -19,6 +19,21 @@ extern enum xen_domain_type xen_domain_type;#define xen_hvm_domain () (xen_domain() && \ xen_domain_type == XEN_HVM_DOMAIN) +#ifdef CONFIG_XEN_HVM_PV + +#define XEN_HVM_PV_CLOCK_ENABLED (1u << 0) +#define XEN_HVM_PV_EVTCHN_ENABLED (1u << 1) +extern u32 xen_hvm_pv_features; + +#define xen_hvm_pv_clock_enabled () \ + (xen_hvm_pv_features & XEN_HVM_PV_CLOCK_ENABLED) +#define xen_hvm_pv_evtchn_enabled () \ + (xen_hvm_pv_features & XEN_HVM_PV_EVTCHN_ENABLED) +#else +#define xen_hvm_pv_clock_enabled () 0 +#define xen_hvm_pv_evtchn_enabled () 0 +#endif /* CONFIG_XEN_HVM_PV */ +#ifdef CONFIG_XEN_DOM0 #include <xen/interface/xen.h> #include <asm/xen/hypervisor.h>
Signature · 1 line
1.5.4.5
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Sheng YangHave you actually tried booting a guest with 2 vcpus? Are you sure it works for you?
Quoted 574 lines · Sheng Yang
We mapped each IOAPIC pin to a VIRQ, so that we can deliver interrupt through these VIRQs.
We used
X86_PLATFORM_IPI_VECTORas the notification vector for hypervisor to notify guest about the event.The patch also enabled SMP support, then we can support IPI through evtchn as well.
When this feature is enabled, we would relay on Xen PV timer for clockevent, rather than other hardware emulated ones.
Then we don't use IOAPIC/LAPIC, eliminated the overhead brought by unnecessary VMExit caused by LAPIC.
PV evtchn depends on PV clocksource. To enable it, put following line in the HVM configure file:
cpuid = [ '0x40000002:edx=0x7' ]
It would enable PV extension framework(bit 0), PV clocksource(bit 1), as well as PV evtchn(bit 2).
Notice if you try to enable PV evtchn without other two bits set, the setting would be ignored.
Signed-off-by Sheng Yang <[email protected]>--- arch/x86/xen/enlighten.c | 6 ++-- arch/x86/xen/hvmpv.c | 70 +++++++++++++++++++++++++++++++++++++++- arch/x86/xen/irq.c | 28 ++++++++++++++++ arch/x86/xen/smp.c | 76 ++++++++++++++++++++++++++++++++++++++++-- arch/x86/xen/xen-ops.h | 16 +++++++++ drivers/xen/events.c | 74 ++++++++++++++++++++++++++++++++++++++--- include/xen/events.h | 4 ++ include/xen/hvm.h | 5 +++ include/xen/interface/xen.h | 6 +++- 9 files changed, 270 insertions(+), 15 deletions(-)arch/x86/xen/enlighten.c +3
-3@@ -717,7 +717,7 @@ static u32 xen_safe_apic_wait_icr_idle(void)return 0 ; } -static void set_xen_basic_apic_ops (void ) +void xen_set_basic_apic_ops (void ) { apic->read = xen_apic_read; apic->write = xen_apic_write;@@ -1026,7 +1026,7 @@ static void xen_crash_shutdown(struct pt_regs *regs)xen_reboot (SHUTDOWN_crash); } -static const struct machine_ops __initdata xen_machine_ops = { +const struct machine_ops __initdata xen_machine_ops = { .restart = xen_restart, .halt = xen_machine_halt, .power_off = xen_machine_halt,@@ -1116,7 +1116,7 @@ asmlinkage void __init xen_start_kernel(void)/* * set up the basic apic ops. */ -set_xen_basic_apic_ops (); +xen_set_basic_apic_ops ();#endif if (xen_feature (XENFEAT_mmu_pt_update_preserve_ad)) {arch/x86/xen/hvmpv.c +69
-1@@ -17,6 +17,7 @@#include <xen/interface/version.h> #include <xen/interface/memory.h> +#include <asm/reboot.h> #include <asm/xen/cpuid.h> #include <asm/xen/hypercall.h> #include <asm/xen/hypervisor.h> @@ -45,6 +46,8 @@ static void __init xen_hvm_pv_banner(void) version >> 16, version & 0xffff, extra.extraversion);if (xen_hvm_pv_clock_enabled ())printk (KERN_INFO "PV feature: PV clocksource enabled\n" ); +if (xen_hvm_pv_evtchn_enabled ()) +printk (KERN_INFO "PV feature: Event channel enabled\n" ); }static int __init xen_para_available (void )@@ -84,9 +87,14 @@ static int __init init_hvm_pv_info(void)if (!(edx &XEN_CPUID_FEAT2_HVM_PV ))return -ENODEV ; -if (edx &XEN_CPUID_FEAT2_HVM_PV_CLOCK ) +if (edx &XEN_CPUID_FEAT2_HVM_PV_CLOCK ) { xen_hvm_pv_features |=XEN_HVM_PV_CLOCK_ENABLED ; +/* Evtchn depends on PV clocksource */ +if (edx &XEN_CPUID_FEAT2_HVM_PV_EVTCHN ) + xen_hvm_pv_features |=XEN_HVM_PV_EVTCHN_ENABLED ; + } +if (pages <1 )return -ENODEV ;@@ -134,6 +142,64 @@ static void __init init_pv_clocksource(void)xen_register_clocksource (); } +static int set_callback_via (uint64_t via) +{ +struct xen_hvm_param a; + + a.domid =DOMID_SELF ; + a.index =HVM_PARAM_CALLBACK_IRQ ; + a.value = via; +return HYPERVISOR_hvm_op (HVMOP_set_param, &a); +} + +void do_hvm_pv_evtchn_intr (void ) +{ +per_cpu (irq_count,smp_processor_id ())++; +xen_hvm_evtchn_do_upcall (get_irq_regs ()); +per_cpu (irq_count,smp_processor_id ())--; +} + +#ifdef CONFIG_X86_LOCAL_APIC +static void xen_hvm_pv_evtchn_apic_write (u32 reg,u32 val) +{ +/* The only one reached here should be EOI */ +WARN_ON (reg !=APIC_EOI ); +} +#endif + +static void __init init_pv_evtchn (void ) +{ +uint64_t callback_via; + +if (!xen_hvm_pv_evtchn_enabled ()) +return ; + +xen_hvm_pv_init_irq_ops (); + + x86_init.timers.timer_init = xen_time_init; + x86_init.timers.setup_percpu_clockev = x86_init_noop; + x86_cpuinit.setup_percpu_clockev = x86_init_noop; + + pv_apic_ops.startup_ipi_hook = paravirt_nop; +#ifdef CONFIG_X86_LOCAL_APIC +/* +* set up the basic apic ops. +*/ +xen_set_basic_apic_ops (); + apic->write = xen_hvm_pv_evtchn_apic_write; +#endif + + callback_via =HVM_CALLBACK_VECTOR (X86_PLATFORM_IPI_VECTOR ); +set_callback_via (callback_via); + + x86_platform_ipi_callback = do_hvm_pv_evtchn_intr; + +disable_acpi (); + +xen_hvm_pv_smp_init (); + machine_ops = xen_machine_ops; +} +void __init xen_guest_init (void ) {int r;@@ -158,4 +224,6 @@ void __init xen_guest_init(void) xen_domain_type =XEN_HVM_DOMAIN ;init_pv_clocksource (); + +init_pv_evtchn (); }arch/x86/xen/irq.c +28
-0@@ -2,6 +2,7 @@#include <asm/x86_init.h> +#include <xen/xen.h> #include <xen/interface/xen.h> #include <xen/interface/sched.h> #include <xen/interface/vcpu.h> @@ -131,3 +132,30 @@ void __init xen_init_irq_ops() pv_irq_ops = xen_irq_ops; x86_init.irqs.intr_init = xen_init_IRQ; } + +#ifdef CONFIG_XEN_HVM_PV +static void xen_hvm_pv_evtchn_disable (void ) +{ +native_irq_disable (); +xen_irq_disable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_disable); + +static void xen_hvm_pv_evtchn_enable (void ) +{ +native_irq_enable (); +xen_irq_enable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_enable); + +void __init xen_hvm_pv_init_irq_ops (void ) +{ +if (xen_hvm_pv_evtchn_enabled ()) { + pv_irq_ops.irq_disable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_disable); + pv_irq_ops.irq_enable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_enable); + x86_init.irqs.intr_init = xen_hvm_pv_evtchn_init_IRQ; + } +} +#endif arch/x86/xen/smp.c +72
-4@@ -15,18 +15,24 @@#include <linux/sched.h> #include <linux/err.h> #include <linux/smp.h> +#include <linux/nmi.h> #include <asm/paravirt.h> #include <asm/desc.h> #include <asm/pgtable.h> #include <asm/cpu.h> +#include <asm/trampoline.h> +#include <asm/tlbflush.h> +#include <asm/mtrr.h> #include <xen/interface/xen.h> #include <xen/interface/vcpu.h> #include <asm/xen/interface.h> #include <asm/xen/hypercall.h> +#include <asm/xen/hypervisor.h> +#include <xen/xen.h> #include <xen/page.h> #include <xen/events.h> @@ -63,8 +69,12 @@ static __cpuinit void cpu_bringup(void)touch_softlockup_watchdog ();preempt_disable (); -xen_enable_sysenter (); -xen_enable_syscall (); +if (xen_pv_domain ()) { +xen_enable_sysenter (); +xen_enable_syscall (); + } + +set_mtrr_aps_delayed_init (); cpu =smp_processor_id ();smp_store_cpu_info (cpu);@@ -171,7 +181,8 @@ static void __init xen_smp_prepare_boot_cpu(void)/* We've switched to the "real" per-cpu gdt, so make sure the old memory can be recycled */ -make_lowmem_page_readwrite (xen_initial_gdt); +if (xen_feature (XENFEAT_writable_descriptor_tables)) +make_lowmem_page_readwrite (xen_initial_gdt);xen_setup_vcpu_info_placement (); }@@ -282,6 +293,39 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle)return 0 ; } +static __cpuinit int +hvm_pv_cpu_initialize_context (unsigned int cpu,struct task_struct *idle) +{ +struct vcpu_guest_context *ctxt; +unsigned long start_ip; + +if (cpumask_test_and_set_cpu (cpu, xen_cpu_initialized_map)) +return 0 ; + + ctxt =kzalloc (sizeof (*ctxt),GFP_KERNEL ); +if (ctxt ==NULL ) +return -ENOMEM ; + + early_gdt_descr.address = (unsigned long )get_cpu_gdt_table (cpu); + initial_code = (unsigned long )cpu_bringup_and_idle; + stack_start.sp = (void *) idle->thread.sp; + +/* start_ip had better be page-aligned! */ + start_ip =setup_trampoline (); + +/* only start_ip is what we want */ + ctxt->flags =VGCF_HVM_GUEST ; + ctxt->user_regs.eip = start_ip; + +printk (KERN_INFO "Booting processor %d ip 0x%lx\n" , cpu, start_ip); + +if (HYPERVISOR_vcpu_op (VCPUOP_initialise, cpu, ctxt)) +BUG (); + +kfree (ctxt); +return 0 ; +} +static int __cpuinit xen_cpu_up (unsigned int cpu) {struct task_struct *idle =idle_task (cpu);@@ -292,6 +336,7 @@ static int __cpuinit xen_cpu_up(unsigned int cpu)irq_ctx_init (cpu);#else clear_tsk_thread_flag (idle,TIF_FORK ); + initial_gs =per_cpu_offset (cpu);per_cpu (kernel_stack, cpu) = (unsigned long )task_stack_page (idle) -KERNEL_STACK_OFFSET +THREAD_SIZE ;@@ -305,7 +350,13 @@ static int __cpuinit xen_cpu_up(unsigned int cpu)/* make sure interrupts start blocked */ per_cpu (xen_vcpu, cpu)->evtchn_upcall_mask =1 ; - rc =cpu_initialize_context (cpu, idle); +if (xen_pv_domain ()) + rc =cpu_initialize_context (cpu, idle); +else if (xen_hvm_pv_evtchn_enabled ()) + rc =hvm_pv_cpu_initialize_context (cpu, idle); +else +BUG (); +if (rc)return rc;@@ -480,3 +531,20 @@ void __init xen_smp_init(void)xen_fill_possible_map ();xen_init_spinlocks (); } + +#ifdef CONFIG_XEN_HVM_PV +static void xen_hvm_pv_flush_tlb_others (const struct cpumask *cpumask, +struct mm_struct *mm,unsigned long va) +{ +/* TODO Make it more specific */ +flush_tlb_all (); +} + +void __init xen_hvm_pv_smp_init (void ) +{ +if (xen_hvm_pv_evtchn_enabled ()) { + smp_ops = xen_smp_ops; + pv_mmu_ops.flush_tlb_others = xen_hvm_pv_flush_tlb_others; + } +} +#endif arch/x86/xen/xen-ops.h +16
-0@@ -52,6 +52,12 @@ void xen_register_clocksource(void);unsigned long xen_get_wallclock (void );int xen_set_wallclock (unsigned long time);unsigned long long xen_sched_clock (void ); +void xen_set_basic_apic_ops (void ); + +#ifdef CONFIG_XEN_HVM_PV +void __init xen_hvm_pv_init_irq_ops (void ); +void __init xen_hvm_pv_evtchn_init_IRQ (void ); +#endif /* CONFIG_XEN_HVM_PV */ irqreturn_t xen_debug_interrupt (int irq,void *dev_id);@@ -62,9 +68,17 @@ void xen_setup_vcpu_info_placement(void);#ifdef CONFIG_SMP void xen_smp_init (void ); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_pv_smp_init (void ); +#endif /* CONFIG_XEN_HVM_PV */ +extern cpumask_var_t xen_cpu_initialized_map;#else static inline void xen_smp_init (void ) {} +#ifdef CONFIG_XEN_HVM_PV +static inline void xen_hvm_pv_smp_init (void ) {} +#endif /* CONFIG_XEN_HVM_PV */ +#endif #ifdef CONFIG_PARAVIRT_SPINLOCKS @@ -102,4 +116,6 @@ void xen_sysret32(void);void xen_sysret64 (void );void xen_adjust_exception_frame (void ); +extern const struct machine_ops xen_machine_ops; +#endif /* XEN_OPS_H */ drivers/xen/events.c +68
-6@@ -32,14 +32,17 @@#include <asm/irq.h> #include <asm/idle.h> #include <asm/sync_bitops.h> +#include <asm/desc.h> #include <asm/xen/hypercall.h> #include <asm/xen/hypervisor.h> +#include <xen/xen.h> #include <xen/xen-ops.h> #include <xen/events.h> #include <xen/interface/xen.h> #include <xen/interface/event_channel.h> +/* * This lock protects updates to the following mapping and reference-count * arrays. The lock does not need to be acquired to read the mapping tables. @@ -616,17 +619,13 @@ static DEFINE_PER_CPU(unsigned, xed_nesting_count);* a bitset of words which contain pending event bits. The second * level is a bitset of pending events themselves. */ -void xen_evtchn_do_upcall (struct pt_regs *regs) +void __xen_evtchn_do_upcall (struct pt_regs *regs) {int cpu =get_cpu (); -struct pt_regs *old_regs =set_irq_regs (regs);struct shared_info *s = HYPERVISOR_shared_info;struct vcpu_info *vcpu_info =__get_cpu_var (xen_vcpu);unsigned count; -exit_idle (); -irq_enter (); - do { unsigned long pending_words;@@ -662,10 +661,25 @@ void xen_evtchn_do_upcall(struct pt_regs *regs) }while (count !=1 ); out: +put_cpu (); +} + +void xen_evtchn_do_upcall (struct pt_regs *regs) +{ +struct pt_regs *old_regs =set_irq_regs (regs); + +exit_idle (); +irq_enter (); + +__xen_evtchn_do_upcall (regs); +irq_exit ();set_irq_regs (old_regs); +} -put_cpu (); +void xen_hvm_evtchn_do_upcall (struct pt_regs *regs) +{ +__xen_evtchn_do_upcall (regs); }/* Rebind a new event channel to an existing irq. */ @@ -944,3 +958,51 @@ void __init xen_init_IRQ(void)irq_ctx_init (smp_processor_id ()); } + +void __init xen_hvm_pv_evtchn_init_IRQ (void ) +{ +int i; + +xen_init_IRQ (); +for (i =0 ; i <NR_IRQS_LEGACY ; i++) { +struct evtchn_bind_virq bind_virq; +struct irq_desc *desc =irq_to_desc (i); +int virq, evtchn; + + virq = i +VIRQ_EMUL_PIN_START ; + bind_virq.virq = virq; + bind_virq.vcpu =0 ; + +if (HYPERVISOR_event_channel_op (EVTCHNOP_bind_virq, + &bind_virq) !=0 ) +BUG (); + + evtchn = bind_virq.port; + evtchn_to_irq[evtchn] = i; + irq_info[i] =mk_virq_info (evtchn, virq); + + desc->status =IRQ_DISABLED ; + desc->action =NULL ; + desc->depth =1 ; + +/* +* 16 old-style INTA-cycle interrupts: +*/ +set_irq_chip_and_handler_name (i, &xen_dynamic_chip, + handle_level_irq,"event" ); + } + +/* +* Cover the whole vector space, no vector can escape +* us. (some of these will be overridden and become +* 'special' SMP interrupts) +*/ +for (i =0 ; i < (NR_VECTORS -FIRST_EXTERNAL_VECTOR ); i++) { +int vector =FIRST_EXTERNAL_VECTOR + i; +if (vector !=IA32_SYSCALL_VECTOR ) +set_intr_gate (vector, interrupt[i]); + } + +/* generic IPI for platform specific use, now used for HVM evtchn */ +alloc_intr_gate (X86_PLATFORM_IPI_VECTOR , x86_platform_ipi); +}include/xen/events.h +4
-0@@ -56,4 +56,8 @@ void xen_poll_irq(int irq);/* Determine the IRQ which is bound to an event channel */ unsigned irq_from_evtchn (unsigned int evtchn); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_evtchn_do_upcall (struct pt_regs *regs); +#endif +#endif /* _XEN_EVENTS_H */ include/xen/hvm.h +5
-0@@ -20,4 +20,9 @@ static inline unsigned long hvm_get_parameter(int idx) return xhv.value; } +#define HVM_CALLBACK_VIA_TYPE_VECTOR 0x2 +#define HVM_CALLBACK_VIA_TYPE_SHIFT 56 +#define HVM_CALLBACK_VECTOR (x) (((uint64_t)HVM_CALLBACK_VIA_TYPE_VECTOR)<<\ + HVM_CALLBACK_VIA_TYPE_SHIFT | (x)) +#endif /* XEN_HVM_H__ */ include/xen/interface/xen.h +5
-2@@ -90,7 +90,11 @@#define VIRQ_ARCH_6 22#define VIRQ_ARCH_7 23 -#define NR_VIRQS 24 +#define VIRQ_EMUL_PIN_START 24 +#define VIRQ_EMUL_PIN_NUM 16 + +#define NR_VIRQS (VIRQ_EMUL_PIN_START + VIRQ_EMUL_PIN_NUM) +/* * MMU-UPDATE REQUESTS * -- 1.5.4.5
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Sheng YangI think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change there.
Look at xen_setup_pirqs() to get an idea about how it works, in my patch series enabling pirq\interrupt mappings is basically a one line change. Besides pirqs and interrupts can coexist without problems and even make pv on hvm work out of the box.
Even though you don't seem interested, I am still open on collaborating with you to complete this patch series.
Quoted 52 lines · Sheng Yang
drivers/xen/events.c +48
-0@@ -944,3 +958,51 @@ void __init xen_init_IRQ(void)irq_ctx_init (smp_processor_id ()); } + +void __init xen_hvm_pv_evtchn_init_IRQ (void ) +{ +int i; + +xen_init_IRQ (); +for (i =0 ; i <NR_IRQS_LEGACY ; i++) { +struct evtchn_bind_virq bind_virq; +struct irq_desc *desc =irq_to_desc (i); +int virq, evtchn; + + virq = i +VIRQ_EMUL_PIN_START ; + bind_virq.virq = virq; + bind_virq.vcpu =0 ; + +if (HYPERVISOR_event_channel_op (EVTCHNOP_bind_virq, + &bind_virq) !=0 ) +BUG (); + + evtchn = bind_virq.port; + evtchn_to_irq[evtchn] = i; + irq_info[i] =mk_virq_info (evtchn, virq); + + desc->status =IRQ_DISABLED ; + desc->action =NULL ; + desc->depth =1 ; + +/* +* 16 old-style INTA-cycle interrupts: +*/ +set_irq_chip_and_handler_name (i, &xen_dynamic_chip, + handle_level_irq,"event" ); + } + +/* +* Cover the whole vector space, no vector can escape +* us. (some of these will be overridden and become +* 'special' SMP interrupts) +*/ +for (i =0 ; i < (NR_VECTORS -FIRST_EXTERNAL_VECTOR ); i++) { +int vector =FIRST_EXTERNAL_VECTOR + i; +if (vector !=IA32_SYSCALL_VECTOR ) +set_intr_gate (vector, interrupt[i]); + } + +/* generic IPI for platform specific use, now used for HVM evtchn */ +alloc_intr_gate (X86_PLATFORM_IPI_VECTOR , x86_platform_ipi); +}
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Stefano StabelliniQuoted 2 lines · On Tuesday 09 March 2010 01:00:52 Stefano Stabellini wrote:
Have you actually tried booting a guest with 2 vcpus? Are you sure it works for you?
I mostly tested my patches with 4 vcpus, so that I can know if the SMP code works well.
Signature · 2 lines
regards Yang, Sheng
Quoted 579 lines · Sheng Yang
Quoted 576 lines
We mapped each IOAPIC pin to a VIRQ, so that we can deliver interrupt through these VIRQs.
We used
X86_PLATFORM_IPI_VECTORas the notification vector for hypervisor to notify guest about the event.The patch also enabled SMP support, then we can support IPI through evtchn as well.
When this feature is enabled, we would relay on Xen PV timer for clockevent, rather than other hardware emulated ones.
Then we don't use IOAPIC/LAPIC, eliminated the overhead brought by unnecessary VMExit caused by LAPIC.
PV evtchn depends on PV clocksource. To enable it, put following line in the HVM configure file:
cpuid = [ '0x40000002:edx=0x7' ]
It would enable PV extension framework(bit 0), PV clocksource(bit 1), as well as PV evtchn(bit 2).
Notice if you try to enable PV evtchn without other two bits set, the setting would be ignored.
Signed-off-by Sheng Yang <[email protected]>--- arch/x86/xen/enlighten.c | 6 ++-- arch/x86/xen/hvmpv.c | 70 +++++++++++++++++++++++++++++++++++++++- arch/x86/xen/irq.c | 28 ++++++++++++++++ arch/x86/xen/smp.c | 76 ++++++++++++++++++++++++++++++++++++++++-- arch/x86/xen/xen-ops.h | 16 +++++++++ drivers/xen/events.c | 74 ++++++++++++++++++++++++++++++++++++++--- include/xen/events.h | 4 ++ include/xen/hvm.h | 5 +++ include/xen/interface/xen.h | 6 +++- 9 files changed, 270 insertions(+), 15 deletions(-)arch/x86/xen/enlighten.c +3
-3@@ -717,7 +717,7 @@ static u32 xen_safe_apic_wait_icr_idle(void)return 0 ; } -static void set_xen_basic_apic_ops (void ) +void xen_set_basic_apic_ops (void ) { apic->read = xen_apic_read; apic->write = xen_apic_write;@@ -1026,7 +1026,7 @@ static void xen_crash_shutdown(struct pt_regs *regs) xen_reboot(SHUTDOWN_crash); } -static const struct machine_ops __initdata xen_machine_ops = { +const struct machine_ops __initdata xen_machine_ops = { .restart = xen_restart, .halt = xen_machine_halt, .power_off = xen_machine_halt,@@ -1116,7 +1116,7 @@ asmlinkage void __init xen_start_kernel(void)/* * set up the basic apic ops. */ -set_xen_basic_apic_ops (); +xen_set_basic_apic_ops ();#endif if (xen_feature (XENFEAT_mmu_pt_update_preserve_ad)) {arch/x86/xen/hvmpv.c +69
-1@@ -17,6 +17,7 @@#include <xen/interface/version.h> #include <xen/interface/memory.h> +#include <asm/reboot.h> #include <asm/xen/cpuid.h> #include <asm/xen/hypercall.h> #include <asm/xen/hypervisor.h> @@ -45,6 +46,8 @@ static void __init xen_hvm_pv_banner(void) version >> 16, version & 0xffff, extra.extraversion);if (xen_hvm_pv_clock_enabled ())printk (KERN_INFO "PV feature: PV clocksource enabled\n" ); +if (xen_hvm_pv_evtchn_enabled ()) +printk (KERN_INFO "PV feature: Event channel enabled\n" ); }static int __init xen_para_available (void )@@ -84,9 +87,14 @@ static int __init init_hvm_pv_info(void)if (!(edx &XEN_CPUID_FEAT2_HVM_PV ))return -ENODEV ; -if (edx &XEN_CPUID_FEAT2_HVM_PV_CLOCK ) +if (edx &XEN_CPUID_FEAT2_HVM_PV_CLOCK ) { xen_hvm_pv_features |=XEN_HVM_PV_CLOCK_ENABLED ; +/* Evtchn depends on PV clocksource */ +if (edx &XEN_CPUID_FEAT2_HVM_PV_EVTCHN ) + xen_hvm_pv_features |=XEN_HVM_PV_EVTCHN_ENABLED ; + } +if (pages <1 )return -ENODEV ;@@ -134,6 +142,64 @@ static void __init init_pv_clocksource(void)xen_register_clocksource (); } +static int set_callback_via (uint64_t via) +{ +struct xen_hvm_param a; + + a.domid =DOMID_SELF ; + a.index =HVM_PARAM_CALLBACK_IRQ ; + a.value = via; +return HYPERVISOR_hvm_op (HVMOP_set_param, &a); +} + +void do_hvm_pv_evtchn_intr (void ) +{ +per_cpu (irq_count,smp_processor_id ())++; +xen_hvm_evtchn_do_upcall (get_irq_regs ()); +per_cpu (irq_count,smp_processor_id ())--; +} + +#ifdef CONFIG_X86_LOCAL_APIC +static void xen_hvm_pv_evtchn_apic_write (u32 reg,u32 val) +{ +/* The only one reached here should be EOI */ +WARN_ON (reg !=APIC_EOI ); +} +#endif + +static void __init init_pv_evtchn (void ) +{ +uint64_t callback_via; + +if (!xen_hvm_pv_evtchn_enabled ()) +return ; + +xen_hvm_pv_init_irq_ops (); + + x86_init.timers.timer_init = xen_time_init; + x86_init.timers.setup_percpu_clockev = x86_init_noop; + x86_cpuinit.setup_percpu_clockev = x86_init_noop; + + pv_apic_ops.startup_ipi_hook = paravirt_nop; +#ifdef CONFIG_X86_LOCAL_APIC +/* +* set up the basic apic ops. +*/ +xen_set_basic_apic_ops (); + apic->write = xen_hvm_pv_evtchn_apic_write; +#endif + + callback_via =HVM_CALLBACK_VECTOR (X86_PLATFORM_IPI_VECTOR ); +set_callback_via (callback_via); + + x86_platform_ipi_callback = do_hvm_pv_evtchn_intr; + +disable_acpi (); + +xen_hvm_pv_smp_init (); + machine_ops = xen_machine_ops; +} +void __init xen_guest_init (void ) {int r;@@ -158,4 +224,6 @@ void __init xen_guest_init(void) xen_domain_type =XEN_HVM_DOMAIN ;init_pv_clocksource (); + +init_pv_evtchn (); }arch/x86/xen/irq.c +28
-0@@ -2,6 +2,7 @@#include <asm/x86_init.h> +#include <xen/xen.h> #include <xen/interface/xen.h> #include <xen/interface/sched.h> #include <xen/interface/vcpu.h> @@ -131,3 +132,30 @@ void __init xen_init_irq_ops() pv_irq_ops = xen_irq_ops; x86_init.irqs.intr_init = xen_init_IRQ; } + +#ifdef CONFIG_XEN_HVM_PV +static void xen_hvm_pv_evtchn_disable (void ) +{ +native_irq_disable (); +xen_irq_disable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_disable); + +static void xen_hvm_pv_evtchn_enable (void ) +{ +native_irq_enable (); +xen_irq_enable (); +} +PV_CALLEE_SAVE_REGS_THUNK (xen_hvm_pv_evtchn_enable); + +void __init xen_hvm_pv_init_irq_ops (void ) +{ +if (xen_hvm_pv_evtchn_enabled ()) { + pv_irq_ops.irq_disable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_disable); + pv_irq_ops.irq_enable = +PV_CALLEE_SAVE (xen_hvm_pv_evtchn_enable); + x86_init.irqs.intr_init = xen_hvm_pv_evtchn_init_IRQ; + } +} +#endif arch/x86/xen/smp.c +38
-3@@ -15,18 +15,24 @@#include <linux/sched.h> #include <linux/err.h> #include <linux/smp.h> +#include <linux/nmi.h> #include <asm/paravirt.h> #include <asm/desc.h> #include <asm/pgtable.h> #include <asm/cpu.h> +#include <asm/trampoline.h> +#include <asm/tlbflush.h> +#include <asm/mtrr.h> #include <xen/interface/xen.h> #include <xen/interface/vcpu.h> #include <asm/xen/interface.h> #include <asm/xen/hypercall.h> +#include <asm/xen/hypervisor.h> +#include <xen/xen.h> #include <xen/page.h> #include <xen/events.h> @@ -63,8 +69,12 @@ static __cpuinit void cpu_bringup(void)touch_softlockup_watchdog ();preempt_disable (); -xen_enable_sysenter (); -xen_enable_syscall (); +if (xen_pv_domain ()) { +xen_enable_sysenter (); +xen_enable_syscall (); + } + +set_mtrr_aps_delayed_init (); cpu =smp_processor_id ();smp_store_cpu_info (cpu);@@ -171,7 +181,8 @@ static void __init xen_smp_prepare_boot_cpu(void)/* We've switched to the "real" per-cpu gdt, so make sure the old memory can be recycled */ -make_lowmem_page_readwrite (xen_initial_gdt); +if (xen_feature (XENFEAT_writable_descriptor_tables)) +make_lowmem_page_readwrite (xen_initial_gdt);xen_setup_vcpu_info_placement (); }@@ -282,6 +293,39 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle) return 0; } +static __cpuinit int +hvm_pv_cpu_initialize_context (unsigned int cpu,struct task_struct *idle) +{ +struct vcpu_guest_context *ctxt; +unsigned long start_ip; + +if (cpumask_test_and_set_cpu (cpu, xen_cpu_initialized_map)) +return 0 ; + + ctxt =kzalloc (sizeof (*ctxt),GFP_KERNEL ); +if (ctxt ==NULL ) +return -ENOMEM ; + + early_gdt_descr.address = (unsigned long )get_cpu_gdt_table (cpu); + initial_code = (unsigned long )cpu_bringup_and_idle; + stack_start.sp = (void *) idle->thread.sp; + +/* start_ip had better be page-aligned! */ + start_ip =setup_trampoline (); + +/* only start_ip is what we want */ + ctxt->flags =VGCF_HVM_GUEST ; + ctxt->user_regs.eip = start_ip; + + printk(KERN_INFO "Booting processor %d ip 0x%lx\n", cpu,
start_ip); +arch/x86/xen/smp.c +19
-1+if (HYPERVISOR_vcpu_op (VCPUOP_initialise, cpu, ctxt)) +BUG (); + +kfree (ctxt); +return 0 ; +} +static int __cpuinit xen_cpu_up (unsigned int cpu) {struct task_struct *idle =idle_task (cpu);@@ -292,6 +336,7 @@ static int __cpuinit xen_cpu_up(unsigned int cpu)irq_ctx_init (cpu);#else clear_tsk_thread_flag (idle,TIF_FORK ); + initial_gs =per_cpu_offset (cpu);per_cpu (kernel_stack, cpu) = (unsigned long )task_stack_page (idle) -KERNEL_STACK_OFFSET +THREAD_SIZE ;@@ -305,7 +350,13 @@ static int __cpuinit xen_cpu_up(unsigned int cpu)/* make sure interrupts start blocked */ per_cpu (xen_vcpu, cpu)->evtchn_upcall_mask =1 ; - rc =cpu_initialize_context (cpu, idle); +if (xen_pv_domain ()) + rc =cpu_initialize_context (cpu, idle); +else if (xen_hvm_pv_evtchn_enabled ()) + rc =hvm_pv_cpu_initialize_context (cpu, idle); +else +BUG (); +if (rc)return rc;@@ -480,3 +531,20 @@ void __init xen_smp_init(void)xen_fill_possible_map ();xen_init_spinlocks (); } + +#ifdef CONFIG_XEN_HVM_PV +static void xen_hvm_pv_flush_tlb_others (const struct cpumask *cpumask, + struct mm_struct *mm, unsignedlong va) +{
arch/x86/xen/smp.c +9
-0+/* TODO Make it more specific */ +flush_tlb_all (); +} + +void __init xen_hvm_pv_smp_init (void ) +{ +if (xen_hvm_pv_evtchn_enabled ()) { + smp_ops = xen_smp_ops; + pv_mmu_ops.flush_tlb_others =
xen_hvm_pv_flush_tlb_others; + }arch/x86/xen/smp.c +2
-0+} +#endif arch/x86/xen/xen-ops.h +16
-0@@ -52,6 +52,12 @@ void xen_register_clocksource(void);unsigned long xen_get_wallclock (void );int xen_set_wallclock (unsigned long time);unsigned long long xen_sched_clock (void ); +void xen_set_basic_apic_ops (void ); + +#ifdef CONFIG_XEN_HVM_PV +void __init xen_hvm_pv_init_irq_ops (void ); +void __init xen_hvm_pv_evtchn_init_IRQ (void ); +#endif /* CONFIG_XEN_HVM_PV */ irqreturn_t xen_debug_interrupt (int irq,void *dev_id);@@ -62,9 +68,17 @@ void xen_setup_vcpu_info_placement(void);#ifdef CONFIG_SMP void xen_smp_init (void ); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_pv_smp_init (void ); +#endif /* CONFIG_XEN_HVM_PV */ +extern cpumask_var_t xen_cpu_initialized_map;#else static inline void xen_smp_init (void ) {} +#ifdef CONFIG_XEN_HVM_PV +static inline void xen_hvm_pv_smp_init (void ) {} +#endif /* CONFIG_XEN_HVM_PV */ +#endif #ifdef CONFIG_PARAVIRT_SPINLOCKS @@ -102,4 +116,6 @@ void xen_sysret32(void);void xen_sysret64 (void );void xen_adjust_exception_frame (void ); +extern const struct machine_ops xen_machine_ops; +#endif /* XEN_OPS_H */ drivers/xen/events.c +3
-0@@ -32,14 +32,17 @@#include <asm/irq.h> #include <asm/idle.h> #include <asm/sync_bitops.h> +#include <asm/desc.h> #include <asm/xen/hypercall.h> #include <asm/xen/hypervisor.h> +#include <xen/xen.h> #include <xen/xen-ops.h> #include <xen/events.h> #include <xen/interface/xen.h> #include <xen/interface/event_channel.h> +/* * This lock protects updates to the following mapping and reference-count * arrays. The lock does not need to be acquired to read the mapping tables. @@ -616,17 +619,13 @@ static
DEFINE_PER_CPU(unsigned, xed_nesting_count); * a bitset of words which contain pending event bits.
The second * level is a bitset of pending events themselves.
*/drivers/xen/events.c +63
-6-void xen_evtchn_do_upcall (struct pt_regs *regs) +void __xen_evtchn_do_upcall (struct pt_regs *regs) {int cpu =get_cpu (); -struct pt_regs *old_regs =set_irq_regs (regs);struct shared_info *s = HYPERVISOR_shared_info;struct vcpu_info *vcpu_info =__get_cpu_var (xen_vcpu);unsigned count; -exit_idle (); -irq_enter (); - do { unsigned long pending_words;@@ -662,10 +661,25 @@ void xen_evtchn_do_upcall(struct pt_regs *regs) }while (count !=1 ); out: +put_cpu (); +} + +void xen_evtchn_do_upcall (struct pt_regs *regs) +{ +struct pt_regs *old_regs =set_irq_regs (regs); + +exit_idle (); +irq_enter (); + +__xen_evtchn_do_upcall (regs); +irq_exit ();set_irq_regs (old_regs); +} -put_cpu (); +void xen_hvm_evtchn_do_upcall (struct pt_regs *regs) +{ +__xen_evtchn_do_upcall (regs); }/* Rebind a new event channel to an existing irq. */ @@ -944,3 +958,51 @@ void __init xen_init_IRQ(void)irq_ctx_init (smp_processor_id ()); } + +void __init xen_hvm_pv_evtchn_init_IRQ (void ) +{ +int i; + +xen_init_IRQ (); +for (i =0 ; i <NR_IRQS_LEGACY ; i++) { +struct evtchn_bind_virq bind_virq; +struct irq_desc *desc =irq_to_desc (i); +int virq, evtchn; + + virq = i +VIRQ_EMUL_PIN_START ; + bind_virq.virq = virq; + bind_virq.vcpu =0 ; + +if (HYPERVISOR_event_channel_op (EVTCHNOP_bind_virq, + &bind_virq) !=0 ) +BUG (); + + evtchn = bind_virq.port; + evtchn_to_irq[evtchn] = i; + irq_info[i] =mk_virq_info (evtchn, virq); + + desc->status =IRQ_DISABLED ; + desc->action =NULL ; + desc->depth =1 ; + +/* +* 16 old-style INTA-cycle interrupts: +*/ +set_irq_chip_and_handler_name (i, &xen_dynamic_chip, + handle_level_irq,"event" ); + } + +/* +* Cover the whole vector space, no vector can escape +* us. (some of these will be overridden and become +* 'special' SMP interrupts) +*/ +for (i =0 ; i < (NR_VECTORS -FIRST_EXTERNAL_VECTOR ); i++) { +int vector =FIRST_EXTERNAL_VECTOR + i; +if (vector !=IA32_SYSCALL_VECTOR ) +set_intr_gate (vector, interrupt[i]); + } + +/* generic IPI for platform specific use, now used for HVM evtchn */ +
alloc_intr_gate(X86_PLATFORM_IPI_VECTOR, x86_platform_ipi); +}include/xen/events.h +4
-0@@ -56,4 +56,8 @@ void xen_poll_irq(int irq);/* Determine the IRQ which is bound to an event channel */ unsigned irq_from_evtchn (unsigned int evtchn); +#ifdef CONFIG_XEN_HVM_PV +void xen_hvm_evtchn_do_upcall (struct pt_regs *regs); +#endif +#endif /* _XEN_EVENTS_H */ include/xen/hvm.h +3
-0@@ -20,4 +20,9 @@ static inline unsigned long hvm_get_parameter(int idx)return xhv.value; } +#define HVM_CALLBACK_VIA_TYPE_VECTOR 0x2 +#define HVM_CALLBACK_VIA_TYPE_SHIFT 56 +#define HVM_CALLBACK_VECTOR (x)(((uint64_t)HVM_CALLBACK_VIA_TYPE_VECTOR)<<\ +
HVM_CALLBACK_VIA_TYPE_SHIFT| (x)) +
#endif /*XEN_HVM_H__*/include/xen/interface/xen.h +5
-2@@ -90,7 +90,11 @@#define VIRQ_ARCH_6 22#define VIRQ_ARCH_7 23 -#define NR_VIRQS 24 +#define VIRQ_EMUL_PIN_START 24 +#define VIRQ_EMUL_PIN_NUM 16 + +#define NR_VIRQS (VIRQ_EMUL_PIN_START + VIRQ_EMUL_PIN_NUM) +/* * MMU-UPDATE REQUESTS * -- 1.5.4.5
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Stefano StabelliniQuoted 3 lines · On Tuesday 09 March 2010 01:10:56 Stefano Stabellini wrote:
I think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change ther.
The code existed in the pv_ops dom0 side, but not in the upstream Linux. The latter is our target. We want this work to be accepted by upstream Linux soon.
Thanks.
Signature · 2 lines
regards Yang, Sheng
Quoted 61 lines
Look at
xen_setup_pirqs()to get an idea about how it works, in my patch series enabling pirq\interrupt mappings is basically a one line change. Besides pirqs and interrupts can coexist without problems and even make pv on hvm work out of the box.Even though you don't seem interested, I am still open on collaborating with you to complete this patch series.
Quoted 51 lines · Sheng Yang
drivers/xen/events.c +46
-0@@ -944,3 +958,51 @@ void __init xen_init_IRQ(void)irq_ctx_init (smp_processor_id ()); } + +void __init xen_hvm_pv_evtchn_init_IRQ (void ) +{ +int i; + +xen_init_IRQ (); +for (i =0 ; i <NR_IRQS_LEGACY ; i++) { +struct evtchn_bind_virq bind_virq; +struct irq_desc *desc =irq_to_desc (i); +int virq, evtchn; + + virq = i +VIRQ_EMUL_PIN_START ; + bind_virq.virq = virq; + bind_virq.vcpu =0 ; + +if (HYPERVISOR_event_channel_op (EVTCHNOP_bind_virq, + &bind_virq) !=0 ) +BUG (); + + evtchn = bind_virq.port; + evtchn_to_irq[evtchn] = i; + irq_info[i] =mk_virq_info (evtchn, virq); + + desc->status =IRQ_DISABLED ; + desc->action =NULL ; + desc->depth =1 ; + +/* +* 16 old-style INTA-cycle interrupts: +*/ +set_irq_chip_and_handler_name (i, &xen_dynamic_chip, + handle_level_irq,"event" ); + } + +/* +* Cover the whole vector space, no vector can escape +* us. (some of these will be overridden and become +* 'special' SMP interrupts) +*/ +for (i =0 ; i < (NR_VECTORS -FIRST_EXTERNAL_VECTOR ); i++) { +int vector =FIRST_EXTERNAL_VECTOR + i; +if (vector !=IA32_SYSCALL_VECTOR ) +set_intr_gate (vector, interrupt[i]); + } + +/* generic IPI for platform specific use, now used for HVM evtchn */ +
alloc_intr_gate(X86_PLATFORM_IPI_VECTOR, x86_platform_ipi); +}
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 8 lines · Sheng Yang
Quoted 4 lines · On Tuesday 09 March 2010 01:10:56 Stefano Stabellini wrote:
I think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change ther.
The code existed in the
pv_opsdom0 side, but not in the upstream Linux. The latter is our target. We want this work to be accepted by upstream Linux soon.
I don't think its a sound idea to rush this stuff upstream, for several reasons:
1. It is currently based on 2.6.33, which means it will not get into any distros in the near future. If it were based on 2.6.32 then it has a much greater chance of being accepted into distros. (They may decide to backport the work, but it would be helpful if that were done in advance.)
2. It has significant overlaps with the current xen.git development which is also targeted for upstream. There's no point in creating an unnecessary duplicate mechanism when the infrastructure will be in place anyway.
3. The code has had very little exposure within the Xen community, especially since the Xen-side patches have not been accepted into xen-unstable yet, let alone a released hypervisor. On the kernel side, it would help if the patches were based on 2.6.32 (or even .31) so they can be merged into the xen.git kernels people are actually using.
4. The kernel Xen code is already complicated with a number of different code paths according to architecture, options, features etc. I would like to see some strong concrete evidence that these changes make a worthwhile improvement on a real workload before adding yet another thing which needs to be tested and maintained indefinitely.
J
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 7 lines · Jeremy Fitzhardinge
Quoted 6 lines · Sheng Yang
Quoted 4 lines · On Tuesday 09 March 2010 01:10:56 Stefano Stabellini wrote:
I think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change ther.
Quoted 7 lines
Quoted 2 lines
The code existed in the
pv_opsdom0 side, but not in the upstream Linux. The latter is our target. We want this work to be accepted by upstream Linux soon.2. It has significant overlaps with the current xen.git development which is also targeted for upstream. There's no point in creating an unnecessary duplicate mechanism when the infrastructure will be in place anyway.
There's also nothing stopping us from upstreaming portions of the "dom0" patchset ahead of full dom0 support if it is useful for some domU feature.
Ian.
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 11 lines · Ian Campbell
Quoted 9 lines · Jeremy Fitzhardinge
Quoted 8 lines · Sheng Yang
Quoted 4 lines · On Tuesday 09 March 2010 01:10:56 Stefano Stabellini wrote:
I think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change ther.
The code existed in the
pv_opsdom0 side, but not in the upstream Linux. The latter is our target. We want this work to be accepted by upstream Linux soon.
Quoted 9 lines
Quoted 4 lines
2. It has significant overlaps with the current xen.git development which is also targeted for upstream. There's no point in creating an unnecessary duplicate mechanism when the infrastructure will be in place anyway.
There's also nothing stopping us from upstreaming portions of the "dom0" patchset ahead of full dom0 support if it is useful for some domU feature.
Indeed.
You just need the pirq mappings and few other things, it shouldn't be too hard.
At this point I am going to do that myself after this patch series is completed.
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 6 lines · Stefano Stabellini
Indeed.
You just need the pirq mappings and few other things, it shouldn't be too hard.
At this point I am going to do that myself after this patch series is completed.
Thanks. You might want to coordinate with Konrad who's doing the same thing to get pcifront into an upstreamable form; I think there would be some overlap.
J
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Stefano StabelliniQuoted 21 lines · On Tuesday 09 March 2010 18:22:31 Stefano Stabellini wrote:
Quoted 11 lines · Ian Campbell
Quoted 10 lines · Jeremy Fitzhardinge
Quoted 8 lines · Sheng Yang
Quoted 3 lines · On Tuesday 09 March 2010 01:10:56 Stefano Stabellini wrote:
I think that mapping interrupts into VIRQs is a bad idea: you should map interrupts into pirqs instead, the code exists already on the kernel side so we don't need to do any (ugly) change ther.
The code existed in the
pv_opsdom0 side, but not in the upstream Linux. The latter is our target. We want this work to be accepted by upstream Linux soon.Quoted 8 lines
Quoted 4 lines
2. It has significant overlaps with the current xen.git development which is also targeted for upstream. There's no point in
creating an unnecessary duplicate mechanism when the infrastructure will be in place anyway.There's also nothing stopping us from upstreaming portions of the "dom0" patchset ahead of full dom0 support if it is useful for some domU feature.
Quoted 5 lines
Indeed.
You just need the pirq mappings and few other things, it shouldn't be too hard.
At this point I am going to do that myself after this patch series is completed.
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
But I think the former 6 patches can be taken as the base. It won't overlapped with others. And PV clocksource is definitely important for HVM guest. We would like to get the first 6 ones checked in Linux upstream first, then we can work on others. And it would the work easier for us later.
So Jeremy, how about a even more simplified version only contained framework and pv clocksource? I think it's pretty elegant and solid first step into Linux upstream.
If you agree, I would send the updated patchset soon, only framework and pv clocksource.
Signature · 2 lines
regards Yang, Sheng
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Jeremy FitzhardingeQuoted 6 lines · On Tuesday 09 March 2010 14:26:12 Jeremy Fitzhardinge wrote:
Quoted 5 lines · Stefano Stabellini
Indeed.
You just need the pirq mappings and few other things, it shouldn't be too hard.
At this point I am going to do that myself after this patch series is completed.
Stefano,
Take a look atgit://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen.git
pv/pcifront-2.6.32
It has a Linus 2.6.32 as a base and all of the patches necessary for PCI front to work. That includes a lot of patches in the events and the pirq/event channel to make it work. Also tie ins to the MSI/MSI-X and IRQ interrupts to have a PCI device work in PV guest.
Quoted 6 lines
Thanks. You might want to coordinate with Konrad who's doing the same thing to get pcifront into an upstreamable form; I think there would be some overlap.
J
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
In reply to Sheng YangQuoted 5 lines · Sheng Yang
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
Most of my patch series can be upstreamed right now, the only thing that needs some extra work is the pirq remapping.
So yes, we can upstream the rest and that would also make PV on HVM work on linux upstream ASAP.
I think that is a worthy goal by itself, therefore I am going to send another reduced patch series, ready to be upstreamed, without pirq remappings, with support for PV on HVM.
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 6 lines · On Wednesday 10 March 2010 23:29:30 Stefano Stabellini wrote:
Quoted 4 lines · Sheng Yang
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
Quoted 8 lines
Most of my patch series can be upstreamed right now, the only thing that needs some extra work is the pirq remapping.
So yes, we can upstream the rest and that would also make PV on HVM work on linux upstream ASAP.
I think that is a worthy goal by itself, therefore I am going to send another reduced patch series, ready to be upstreamed, without pirq remappings, with support for PV on HVM.
I think my first 6 patches can be used without modification. And I've already spent lots of efforts on them with Jeremy.
We can work on evtchn and platform pci later.
Thanks.
Signature · 2 lines
regards Yang, Sheng
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 15 lines · Sheng Yang
Quoted 6 lines · On Wednesday 10 March 2010 23:29:30 Stefano Stabellini wrote:
Quoted 4 lines · Sheng Yang
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
Quoted 8 lines
Most of my patch series can be upstreamed right now, the only thing that needs some extra work is the pirq remapping.
So yes, we can upstream the rest and that would also make PV on HVM work on linux upstream ASAP.
I think that is a worthy goal by itself, therefore I am going to send another reduced patch series, ready to be upstreamed, without pirq remappings, with support for PV on HVM.
Quoted 5 lines
I think my first 6 patches can be used without modification. And I've already spent lots of efforts on them with Jeremy.
We can work on evtchn and platform pci later.
Agreed on evtchn but platform pci should be present even in the basic version.
Please read the series "basic PV on HVM support" I posted yesterday because I believe has already everything needed to be accepted upstream right now; at the same time it introduces the pv clocksource and the vector based callback mechanism that is going to be needed for the interrupt remapping we are going to work on later on.
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 21 lines · On Thursday 11 March 2010 19:43:00 Stefano Stabellini wrote:
Quoted 15 lines · Sheng Yang
Quoted 6 lines · On Wednesday 10 March 2010 23:29:30 Stefano Stabellini wrote:
Quoted 4 lines · Sheng Yang
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
Quoted 7 lines
Most of my patch series can be upstreamed right now, the only thing that needs some extra work is the pirq remapping.
So yes, we can upstream the rest and that would also make PV on HVM work on linux upstream ASAP.
I think that is a worthy goal by itself, therefore I am going to send another reduced patch series, ready to be upstreamed, without pirq remappings, with support for PV on HVM.Quoted 4 lines
I think my first 6 patches can be used without modification. And I've already spent lots of efforts on them with Jeremy.
We can work on evtchn and platform pci later.
Quoted 8 lines
Agreed on evtchn but platform pci should be present even in the basic version.
Please read the series "basic PV on HVM support" I posted yesterday because I believe has already everything needed to be accepted upstream right now; at the same time it introduces the pv clocksource and the vector based callback mechanism that is going to be needed for the interrupt remapping we are going to work on later on.
IIRC, I think your platform PCI driver also depends on evtchn to inject interrupt?
Signature · 2 lines
regards Yang, Sheng
Re: [Xen-devel] [PATCH 7/7] xen: Enable event channel of PV extension of HVM
Quoted 30 lines · Sheng Yang
Quoted 21 lines · On Thursday 11 March 2010 19:43:00 Stefano Stabellini wrote:
Quoted 15 lines · Sheng Yang
Quoted 6 lines · On Wednesday 10 March 2010 23:29:30 Stefano Stabellini wrote:
Quoted 4 lines · Sheng Yang
I think we can leave the controversial thing later. At least, we want a framework for PV extension of HVM. We can work together to determine what is the better way for evtchn, as well as porting pirqs. (And the later MSI work may also depends on it)
Quoted 7 lines
Most of my patch series can be upstreamed right now, the only thing that needs some extra work is the pirq remapping.
So yes, we can upstream the rest and that would also make PV on HVM work on linux upstream ASAP.
I think that is a worthy goal by itself, therefore I am going to send another reduced patch series, ready to be upstreamed, without pirq remappings, with support for PV on HVM.Quoted 4 lines
I think my first 6 patches can be used without modification. And I've already spent lots of efforts on them with Jeremy.
We can work on evtchn and platform pci later.
Quoted 8 lines
Agreed on evtchn but platform pci should be present even in the basic version.
Please read the series "basic PV on HVM support" I posted yesterday because I believe has already everything needed to be accepted upstream right now; at the same time it introduces the pv clocksource and the vector based callback mechanism that is going to be needed for the interrupt remapping we are going to work on later on.
Quoted 3 lines
IIRC, I think your platform PCI driver also depends on evtchn to inject interrupt?
The original platform PCI driver used to register an interrupt based callback mechanism but in the last patch series I sent I removed it in favor of the new vector based callback you are familiar with. However I am currently working on providing the former GSI based method as a fallback, otherwise PV on HVM won't work on xen < 4.