Paravirt ops is currently only capable of either replacing a lot of Linux internal code or none at all. The are users that don't need all of the possibilities pv-ops delivers though. On KVM for example we're perfectly fine not using the PV MMU, thus not touching any MMU code. That way we don't have to improve pv-ops to become fast, we just don't compile the MMU parts in! This patchset splits pv-ops into several smaller config options split by feature category and then converts the KVM pv-ops code to use only the bits that are required, lowering overhead. Alexander Graf (3): Split paravirt ops by functionality Only export selected pv-ops feature structs Split the KVM pv-ops support by feature arch/x86/Kconfig | 72 +++++++++++++++++++++++++- arch/x86/include/asm/apic.h | 2 +- arch/x86/include/asm/desc.h | 4 +- arch/x86/include/asm/fixmap.h | 2 +- arch/x86/include/asm/highmem.h | 2 +- arch/x86/include/asm/io_32.h | 4 +- arch/x86/include/asm/io_64.h | 2 +- arch/x86/include/asm/irqflags.h | 21 ++++++-- arch/x86/include/asm/mmu_context.h | 4 +- arch/x86/include/asm/msr.h | 4 +- arch/x86/include/asm/paravirt.h | 44 ++++++++++++++++- arch/x86/include/asm/paravirt_types.h | 12 +++++ arch/x86/include/asm/pgalloc.h | 2 +- arch/x86/include/asm/pgtable-3level_types.h | 2 +- arch/x86/include/asm/pgtable.h | 2 +- arch/x86/include/asm/processor.h | 2 +- arch/x86/include/asm/required-features.h | 2 +- arch/x86/include/asm/smp.h | 2 +- arch/x86/include/asm/system.h | 13 +++-- arch/x86/include/asm/tlbflush.h | 4 +- arch/x86/kernel/head_64.S | 2 +- arch/x86/kernel/kvm.c | 22 ++++++--- arch/x86/kernel/paravirt.c | 37 +++++++++++-- arch/x86/kernel/tsc.c | 2 +- arch/x86/kernel/vsmp_64.c | 2 +- arch/x86/xen/Kconfig | 2 +- 26 files changed, 219 insertions(+), 50 deletions(-)
Currently when using paravirt ops it's an all-or-nothing option. We can
either
use pv-ops for CPU, MMU, timing, etc. or not at all.
Now there are some use cases where we don't need the full feature set, but
only
a small chunk of it. KVM is a pretty prominent example for this.
So let's make everything a bit more fine-grained. We already have a
splitting
by function groups, namely "cpu", "mmu", "time",
"irq", "apic" and "spinlock".
Taking that existing splitting and extending it to only compile in the PV
capable bits sounded like a natural fit. That way we don't get performance
hits
in MMU code from using the KVM PV clock which only needs the TIME parts of
pv-ops.
We define a new CONFIG_PARAVIRT_ALL option that basically does the same thing
the CONFIG_PARAVIRT did before this splitting. We move all users of
CONFIG_PARAVIRT to CONFIG_PARAVIRT_ALL, so they behave the same way they did
before.
So here it is - the splitting! I would have made the patch smaller, but this
was the closest I could get to atomic (for bisect) while staying sane.
Signed-off-by: Alexander Graf <agraf at suse.de>
---
arch/x86/Kconfig | 47 ++++++++++++++++++++++++--
arch/x86/include/asm/apic.h | 2 +-
arch/x86/include/asm/desc.h | 4 +-
arch/x86/include/asm/fixmap.h | 2 +-
arch/x86/include/asm/highmem.h | 2 +-
arch/x86/include/asm/io_32.h | 4 ++-
arch/x86/include/asm/io_64.h | 2 +-
arch/x86/include/asm/irqflags.h | 21 +++++++++---
arch/x86/include/asm/mmu_context.h | 4 +-
arch/x86/include/asm/msr.h | 4 +-
arch/x86/include/asm/paravirt.h | 44 ++++++++++++++++++++++++-
arch/x86/include/asm/paravirt_types.h | 12 +++++++
arch/x86/include/asm/pgalloc.h | 2 +-
arch/x86/include/asm/pgtable-3level_types.h | 2 +-
arch/x86/include/asm/pgtable.h | 2 +-
arch/x86/include/asm/processor.h | 2 +-
arch/x86/include/asm/required-features.h | 2 +-
arch/x86/include/asm/smp.h | 2 +-
arch/x86/include/asm/system.h | 13 +++++--
arch/x86/include/asm/tlbflush.h | 4 +-
arch/x86/kernel/head_64.S | 2 +-
arch/x86/kernel/paravirt.c | 2 +
arch/x86/kernel/tsc.c | 2 +-
arch/x86/kernel/vsmp_64.c | 2 +-
arch/x86/xen/Kconfig | 2 +-
25 files changed, 149 insertions(+), 38 deletions(-)
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 0c7b699..8c150b6 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -350,7 +350,7 @@ endif
config X86_VSMP
bool "ScaleMP vSMP"
- select PARAVIRT
+ select PARAVIRT_ALL
depends on X86_64 && PCI
depends on X86_EXTENDED_PLATFORM
---help---
@@ -493,7 +493,7 @@ source "arch/x86/xen/Kconfig"
config VMI
bool "VMI Guest support (DEPRECATED)"
- select PARAVIRT
+ select PARAVIRT_ALL
depends on X86_32
---help---
VMI provides a paravirtualized interface to the VMware ESX server
@@ -512,7 +512,6 @@ config VMI
config KVM_CLOCK
bool "KVM paravirtualized clock"
- select PARAVIRT
select PARAVIRT_CLOCK
---help---
Turning on this option will allow you to run a paravirtualized clock
@@ -523,7 +522,7 @@ config KVM_CLOCK
config KVM_GUEST
bool "KVM Guest support"
- select PARAVIRT
+ select PARAVIRT_ALL
---help---
This option enables various optimizations for running under the KVM
hypervisor.
@@ -551,8 +550,48 @@ config PARAVIRT_SPINLOCKS
If you are unsure how to answer this question, answer N.
+config PARAVIRT_CPU
+ bool
+ select PARAVIRT
+ default n
+
+config PARAVIRT_TIME
+ bool
+ select PARAVIRT
+ default n
+
+config PARAVIRT_IRQ
+ bool
+ select PARAVIRT
+ default n
+
+config PARAVIRT_APIC
+ bool
+ select PARAVIRT
+ default n
+
+config PARAVIRT_MMU
+ bool
+ select PARAVIRT
+ default n
+
+#
+# This is a placeholder to activate the old "include all pv-ops
functionality"
+# behavior. If you're using this I'd recommend looking through your
code to see
+# if you can be more specific. It probably saves you a few cycles!
+#
+config PARAVIRT_ALL
+ bool
+ select PARAVIRT_CPU
+ select PARAVIRT_TIME
+ select PARAVIRT_IRQ
+ select PARAVIRT_APIC
+ select PARAVIRT_MMU
+ default n
+
config PARAVIRT_CLOCK
bool
+ select PARAVIRT_TIME
default n
endif
diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
index 474d80d..b54c24a 100644
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -81,7 +81,7 @@ static inline bool apic_from_smp_config(void)
/*
* Basic functions accessing APICs.
*/
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_APIC
#include <asm/paravirt.h>
#endif
diff --git a/arch/x86/include/asm/desc.h b/arch/x86/include/asm/desc.h
index e8de2f6..cf65891 100644
--- a/arch/x86/include/asm/desc.h
+++ b/arch/x86/include/asm/desc.h
@@ -78,7 +78,7 @@ static inline int desc_empty(const void *ptr)
return !(desc[0] | desc[1]);
}
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_CPU
#include <asm/paravirt.h>
#else
#define load_TR_desc() native_load_tr_desc()
@@ -108,7 +108,7 @@ static inline void paravirt_alloc_ldt(struct desc_struct
*ldt, unsigned entries)
static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
{
}
-#endif /* CONFIG_PARAVIRT */
+#endif /* CONFIG_PARAVIRT_CPU */
#define store_ldt(ldt) asm("sldt %0" : "=m"(ldt))
diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
index 14f9890..5f29317 100644
--- a/arch/x86/include/asm/fixmap.h
+++ b/arch/x86/include/asm/fixmap.h
@@ -156,7 +156,7 @@ void __native_set_fixmap(enum fixed_addresses idx, pte_t
pte);
void native_set_fixmap(enum fixed_addresses idx,
phys_addr_t phys, pgprot_t flags);
-#ifndef CONFIG_PARAVIRT
+#ifndef CONFIG_PARAVIRT_MMU
static inline void __set_fixmap(enum fixed_addresses idx,
phys_addr_t phys, pgprot_t flags)
{
diff --git a/arch/x86/include/asm/highmem.h b/arch/x86/include/asm/highmem.h
index 014c2b8..458d785 100644
--- a/arch/x86/include/asm/highmem.h
+++ b/arch/x86/include/asm/highmem.h
@@ -66,7 +66,7 @@ void *kmap_atomic_pfn(unsigned long pfn, enum km_type type);
void *kmap_atomic_prot_pfn(unsigned long pfn, enum km_type type, pgprot_t
prot);
struct page *kmap_atomic_to_page(void *ptr);
-#ifndef CONFIG_PARAVIRT
+#ifndef CONFIG_PARAVIRT_MMU
#define kmap_atomic_pte(page, type) kmap_atomic(page, type)
#endif
diff --git a/arch/x86/include/asm/io_32.h b/arch/x86/include/asm/io_32.h
index a299900..a263c6f 100644
--- a/arch/x86/include/asm/io_32.h
+++ b/arch/x86/include/asm/io_32.h
@@ -109,7 +109,9 @@ extern void io_delay_init(void);
#if defined(CONFIG_PARAVIRT)
#include <asm/paravirt.h>
-#else
+#endif
+
+#ifndef CONFIG_PARAVIRT_CPU
static inline void slow_down_io(void)
{
diff --git a/arch/x86/include/asm/io_64.h b/arch/x86/include/asm/io_64.h
index 2440678..82c6eae 100644
--- a/arch/x86/include/asm/io_64.h
+++ b/arch/x86/include/asm/io_64.h
@@ -40,7 +40,7 @@ extern void native_io_delay(void);
extern int io_delay_type;
extern void io_delay_init(void);
-#if defined(CONFIG_PARAVIRT)
+#if defined(CONFIG_PARAVIRT_CPU)
#include <asm/paravirt.h>
#else
diff --git a/arch/x86/include/asm/irqflags.h b/arch/x86/include/asm/irqflags.h
index 9e2b952..b8d8f4c 100644
--- a/arch/x86/include/asm/irqflags.h
+++ b/arch/x86/include/asm/irqflags.h
@@ -58,9 +58,11 @@ static inline void native_halt(void)
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
-#else
+#endif
+
#ifndef __ASSEMBLY__
+#ifndef CONFIG_PARAVIRT_IRQ
static inline unsigned long __raw_local_save_flags(void)
{
return native_save_fl();
@@ -110,12 +112,17 @@ static inline unsigned long __raw_local_irq_save(void)
return flags;
}
-#else
+#endif /* CONFIG_PARAVIRT_IRQ */
+
+#else /* __ASSEMBLY__ */
+#ifndef CONFIG_PARAVIRT_IRQ
#define ENABLE_INTERRUPTS(x) sti
#define DISABLE_INTERRUPTS(x) cli
+#endif /* !CONFIG_PARAVIRT_IRQ */
#ifdef CONFIG_X86_64
+#ifndef CONFIG_PARAVIRT_CPU
#define SWAPGS swapgs
/*
* Currently paravirt can't handle swapgs nicely when we
@@ -128,8 +135,6 @@ static inline unsigned long __raw_local_irq_save(void)
*/
#define SWAPGS_UNSAFE_STACK swapgs
-#define PARAVIRT_ADJUST_EXCEPTION_FRAME /* */
-
#define INTERRUPT_RETURN iretq
#define USERGS_SYSRET64 \
swapgs; \
@@ -141,16 +146,22 @@ static inline unsigned long __raw_local_irq_save(void)
swapgs; \
sti; \
sysexit
+#endif /* !CONFIG_PARAVIRT_CPU */
+
+#ifndef CONFIG_PARAVIRT_IRQ
+#define PARAVIRT_ADJUST_EXCEPTION_FRAME /* */
+#endif /* !CONFIG_PARAVIRT_IRQ */
#else
+#ifndef CONFIG_PARAVIRT_CPU
#define INTERRUPT_RETURN iret
#define ENABLE_INTERRUPTS_SYSEXIT sti; sysexit
#define GET_CR0_INTO_EAX movl %cr0, %eax
+#endif /* !CONFIG_PARAVIRT_CPU */
#endif
#endif /* __ASSEMBLY__ */
-#endif /* CONFIG_PARAVIRT */
#ifndef __ASSEMBLY__
#define raw_local_save_flags(flags) \
diff --git a/arch/x86/include/asm/mmu_context.h
b/arch/x86/include/asm/mmu_context.h
index 4a2d4e0..a209e67 100644
--- a/arch/x86/include/asm/mmu_context.h
+++ b/arch/x86/include/asm/mmu_context.h
@@ -6,14 +6,14 @@
#include <asm/pgalloc.h>
#include <asm/tlbflush.h>
#include <asm/paravirt.h>
-#ifndef CONFIG_PARAVIRT
+#ifndef CONFIG_PARAVIRT_MMU
#include <asm-generic/mm_hooks.h>
static inline void paravirt_activate_mm(struct mm_struct *prev,
struct mm_struct *next)
{
}
-#endif /* !CONFIG_PARAVIRT */
+#endif /* !CONFIG_PARAVIRT_MMU */
/*
* Used for LDT copy/destruction.
diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 7e2b6ba..80ec5a5 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -123,7 +123,7 @@ static inline unsigned long long native_read_pmc(int
counter)
return EAX_EDX_VAL(val, low, high);
}
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_CPU
#include <asm/paravirt.h>
#else
#include <linux/errno.h>
@@ -234,7 +234,7 @@ do {
\
#define rdtscpll(val, aux) (val) = native_read_tscp(&(aux))
-#endif /* !CONFIG_PARAVIRT */
+#endif /* !CONFIG_PARAVIRT_CPU */
#define checking_wrmsrl(msr, val) wrmsr_safe((msr), (u32)(val), \
diff --git a/arch/x86/include/asm/paravirt.h b/arch/x86/include/asm/paravirt.h
index efb3899..e543098 100644
--- a/arch/x86/include/asm/paravirt.h
+++ b/arch/x86/include/asm/paravirt.h
@@ -18,6 +18,7 @@ static inline int paravirt_enabled(void)
return pv_info.paravirt_enabled;
}
+#ifdef CONFIG_PARAVIRT_CPU
static inline void load_sp0(struct tss_struct *tss,
struct thread_struct *thread)
{
@@ -58,7 +59,9 @@ static inline void write_cr0(unsigned long x)
{
PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
}
+#endif /* CONFIG_PARAVIRT_CPU */
+#ifdef CONFIG_PARAVIRT_MMU
static inline unsigned long read_cr2(void)
{
return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
@@ -78,7 +81,9 @@ static inline void write_cr3(unsigned long x)
{
PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
}
+#endif /* CONFIG_PARAVIRT_MMU */
+#ifdef CONFIG_PARAVIRT_CPU
static inline unsigned long read_cr4(void)
{
return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
@@ -92,8 +97,9 @@ static inline void write_cr4(unsigned long x)
{
PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
}
+#endif /* CONFIG_PARAVIRT_CPU */
-#ifdef CONFIG_X86_64
+#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_CPU)
static inline unsigned long read_cr8(void)
{
return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
@@ -105,6 +111,7 @@ static inline void write_cr8(unsigned long x)
}
#endif
+#ifdef CONFIG_PARAVIRT_IRQ
static inline void raw_safe_halt(void)
{
PVOP_VCALL0(pv_irq_ops.safe_halt);
@@ -114,14 +121,18 @@ static inline void halt(void)
{
PVOP_VCALL0(pv_irq_ops.safe_halt);
}
+#endif /* CONFIG_PARAVIRT_IRQ */
+#ifdef CONFIG_PARAVIRT_CPU
static inline void wbinvd(void)
{
PVOP_VCALL0(pv_cpu_ops.wbinvd);
}
+#endif
#define get_kernel_rpl() (pv_info.kernel_rpl)
+#ifdef CONFIG_PARAVIRT_CPU
static inline u64 paravirt_read_msr(unsigned msr, int *err)
{
return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
@@ -224,12 +235,16 @@ do { \
} while (0)
#define rdtscll(val) (val = paravirt_read_tsc())
+#endif /* CONFIG_PARAVIRT_CPU */
+#ifdef CONFIG_PARAVIRT_TIME
static inline unsigned long long paravirt_sched_clock(void)
{
return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
}
+#endif /* CONFIG_PARAVIRT_TIME */
+#ifdef CONFIG_PARAVIRT_CPU
static inline unsigned long long paravirt_read_pmc(int counter)
{
return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
@@ -345,8 +360,9 @@ static inline void slow_down_io(void)
pv_cpu_ops.io_delay();
#endif
}
+#endif /* CONFIG_PARAVIRT_CPU */
-#ifdef CONFIG_SMP
+#if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_APIC)
static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
unsigned long start_esp)
{
@@ -355,6 +371,7 @@ static inline void startup_ipi_hook(int phys_apicid,
unsigned long start_eip,
}
#endif
+#ifdef CONFIG_PARAVIRT_MMU
static inline void paravirt_activate_mm(struct mm_struct *prev,
struct mm_struct *next)
{
@@ -698,7 +715,9 @@ static inline void pmd_clear(pmd_t *pmdp)
set_pmd(pmdp, __pmd(0));
}
#endif /* CONFIG_X86_PAE */
+#endif /* CONFIG_PARAVIRT_MMU */
+#ifdef CONFIG_PARAVIRT_CPU
#define __HAVE_ARCH_START_CONTEXT_SWITCH
static inline void arch_start_context_switch(struct task_struct *prev)
{
@@ -709,7 +728,9 @@ static inline void arch_end_context_switch(struct
task_struct *next)
{
PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
}
+#endif /* CONFIG_PARAVIRT_CPU */
+#ifdef CONFIG_PARAVIRT_MMU
#define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
static inline void arch_enter_lazy_mmu_mode(void)
{
@@ -728,6 +749,7 @@ static inline void __set_fixmap(unsigned /* enum
fixed_addresses */ idx,
{
pv_mmu_ops.set_fixmap(idx, phys, flags);
}
+#endif /* CONFIG_PARAVIRT_MMU */
#if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
@@ -838,6 +860,7 @@ static __always_inline void __raw_spin_unlock(struct
raw_spinlock *lock)
#define __PV_IS_CALLEE_SAVE(func) \
((struct paravirt_callee_save) { func })
+#ifdef CONFIG_PARAVIRT_IRQ
static inline unsigned long __raw_local_save_flags(void)
{
return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
@@ -866,6 +889,7 @@ static inline unsigned long __raw_local_irq_save(void)
raw_local_irq_disable();
return f;
}
+#endif /* CONFIG_PARAVIRT_IRQ */
/* Make sure as little as possible of this mess escapes. */
@@ -948,10 +972,13 @@ extern void default_banner(void);
#define PARA_INDIRECT(addr) *%cs:addr
#endif
+#ifdef CONFIG_PARAVIRT_CPU
#define INTERRUPT_RETURN \
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
+#endif /* CONFIG_PARAVIRT_CPU */
+#ifdef CONFIG_PARAVIRT_IRQ
#define DISABLE_INTERRUPTS(clobbers) \
PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
@@ -963,13 +990,17 @@ extern void default_banner(void);
PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable); \
PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
+#endif /* CONFIG_PARAVIRT_IRQ */
+#ifdef CONFIG_PARAVIRT_CPU
#define USERGS_SYSRET32 \
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32), \
CLBR_NONE, \
jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
+#endif /* CONFIG_PARAVIRT_CPU */
#ifdef CONFIG_X86_32
+#ifdef CONFIG_PARAVIRT_CPU
#define GET_CR0_INTO_EAX \
push %ecx; push %edx; \
call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
@@ -979,10 +1010,12 @@ extern void default_banner(void);
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
CLBR_NONE, \
jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
+#endif /* CONFIG_PARAVIRT_CPU */
#else /* !CONFIG_X86_32 */
+#ifdef CONFIG_PARAVIRT_CPU
/*
* If swapgs is used while the userspace stack is still current,
* there's no way to call a pvop. The PV replacement *must* be
@@ -1002,17 +1035,23 @@ extern void default_banner(void);
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs) \
)
+#endif /* CONFIG_PARAVIRT_CPU */
+#ifdef CONFIG_PARAVIRT_MMU
#define GET_CR2_INTO_RCX \
call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
movq %rax, %rcx; \
xorq %rax, %rax;
+#endif /* CONFIG_PARAVIRT_MMU */
+#ifdef CONFIG_PARAVIRT_IRQ
#define PARAVIRT_ADJUST_EXCEPTION_FRAME \
PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
CLBR_NONE, \
call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
+#endif /* CONFIG_PARAVIRT_IRQ */
+#ifdef CONFIG_PARAVIRT_CPU
#define USERGS_SYSRET64 \
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64), \
CLBR_NONE, \
@@ -1022,6 +1061,7 @@ extern void default_banner(void);
PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
CLBR_NONE, \
jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
+#endif /* CONFIG_PARAVIRT_CPU */
#endif /* CONFIG_X86_32 */
#endif /* __ASSEMBLY__ */
diff --git a/arch/x86/include/asm/paravirt_types.h
b/arch/x86/include/asm/paravirt_types.h
index 9357473..e190450 100644
--- a/arch/x86/include/asm/paravirt_types.h
+++ b/arch/x86/include/asm/paravirt_types.h
@@ -343,12 +343,24 @@ struct paravirt_patch_template {
extern struct pv_info pv_info;
extern struct pv_init_ops pv_init_ops;
+#ifdef CONFIG_PARAVIRT_TIME
extern struct pv_time_ops pv_time_ops;
+#endif
+#ifdef CONFIG_PARAVIRT_CPU
extern struct pv_cpu_ops pv_cpu_ops;
+#endif
+#ifdef CONFIG_PARAVIRT_IRQ
extern struct pv_irq_ops pv_irq_ops;
+#endif
+#ifdef CONFIG_PARAVIRT_APIC
extern struct pv_apic_ops pv_apic_ops;
+#endif
+#ifdef CONFIG_PARAVIRT_MMU
extern struct pv_mmu_ops pv_mmu_ops;
+#endif
+#ifdef CONFIG_PARAVIRT_SPINLOCKS
extern struct pv_lock_ops pv_lock_ops;
+#endif
#define PARAVIRT_PATCH(x) \
(offsetof(struct paravirt_patch_template, x) / sizeof(void *))
diff --git a/arch/x86/include/asm/pgalloc.h b/arch/x86/include/asm/pgalloc.h
index 0e8c2a0..94cce3d 100644
--- a/arch/x86/include/asm/pgalloc.h
+++ b/arch/x86/include/asm/pgalloc.h
@@ -7,7 +7,7 @@
static inline int __paravirt_pgd_alloc(struct mm_struct *mm) { return 0; }
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
#include <asm/paravirt.h>
#else
#define paravirt_pgd_alloc(mm) __paravirt_pgd_alloc(mm)
diff --git a/arch/x86/include/asm/pgtable-3level_types.h
b/arch/x86/include/asm/pgtable-3level_types.h
index 1bd5876..be58e74 100644
--- a/arch/x86/include/asm/pgtable-3level_types.h
+++ b/arch/x86/include/asm/pgtable-3level_types.h
@@ -18,7 +18,7 @@ typedef union {
} pte_t;
#endif /* !__ASSEMBLY__ */
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
#define SHARED_KERNEL_PMD (pv_info.shared_kernel_pmd)
#else
#define SHARED_KERNEL_PMD 1
diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index af6fd36..b68edfc 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -26,7 +26,7 @@ extern unsigned long empty_zero_page[PAGE_SIZE /
sizeof(unsigned long)];
extern spinlock_t pgd_lock;
extern struct list_head pgd_list;
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
#include <asm/paravirt.h>
#else /* !CONFIG_PARAVIRT */
#define set_pte(ptep, pte) native_set_pte(ptep, pte)
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index c3429e8..a42a807 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -571,7 +571,7 @@ static inline void native_swapgs(void)
#endif
}
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_CPU
#include <asm/paravirt.h>
#else
#define __cpuid native_cpuid
diff --git a/arch/x86/include/asm/required-features.h
b/arch/x86/include/asm/required-features.h
index 64cf2d2..f68edf2 100644
--- a/arch/x86/include/asm/required-features.h
+++ b/arch/x86/include/asm/required-features.h
@@ -48,7 +48,7 @@
#endif
#ifdef CONFIG_X86_64
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
/* Paravirtualized systems may not have PSE or PGE available */
#define NEED_PSE 0
#define NEED_PGE 0
diff --git a/arch/x86/include/asm/smp.h b/arch/x86/include/asm/smp.h
index 1e79678..fdd889a 100644
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -66,7 +66,7 @@ struct smp_ops {
extern void set_cpu_sibling_map(int cpu);
#ifdef CONFIG_SMP
-#ifndef CONFIG_PARAVIRT
+#ifndef CONFIG_PARAVIRT_APIC
#define startup_ipi_hook(phys_apicid, start_eip, start_esp) do { } while (0)
#endif
extern struct smp_ops smp_ops;
diff --git a/arch/x86/include/asm/system.h b/arch/x86/include/asm/system.h
index f08f973..63ca93c 100644
--- a/arch/x86/include/asm/system.h
+++ b/arch/x86/include/asm/system.h
@@ -302,13 +302,18 @@ static inline void native_wbinvd(void)
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
-#else
-#define read_cr0() (native_read_cr0())
-#define write_cr0(x) (native_write_cr0(x))
+#endif/* CONFIG_PARAVIRT */
+
+#ifndef CONFIG_PARAVIRT_MMU
#define read_cr2() (native_read_cr2())
#define write_cr2(x) (native_write_cr2(x))
#define read_cr3() (native_read_cr3())
#define write_cr3(x) (native_write_cr3(x))
+#endif /* CONFIG_PARAVIRT_MMU */
+
+#ifndef CONFIG_PARAVIRT_CPU
+#define read_cr0() (native_read_cr0())
+#define write_cr0(x) (native_write_cr0(x))
#define read_cr4() (native_read_cr4())
#define read_cr4_safe() (native_read_cr4_safe())
#define write_cr4(x) (native_write_cr4(x))
@@ -322,7 +327,7 @@ static inline void native_wbinvd(void)
/* Clear the 'TS' bit */
#define clts() (native_clts())
-#endif/* CONFIG_PARAVIRT */
+#endif /* CONFIG_PARAVIRT_CPU */
#define stts() write_cr0(read_cr0() | X86_CR0_TS)
diff --git a/arch/x86/include/asm/tlbflush.h b/arch/x86/include/asm/tlbflush.h
index 7f3eba0..89e055c 100644
--- a/arch/x86/include/asm/tlbflush.h
+++ b/arch/x86/include/asm/tlbflush.h
@@ -7,7 +7,7 @@
#include <asm/processor.h>
#include <asm/system.h>
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
#include <asm/paravirt.h>
#else
#define __flush_tlb() __native_flush_tlb()
@@ -162,7 +162,7 @@ static inline void reset_lazy_tlbstate(void)
#endif /* SMP */
-#ifndef CONFIG_PARAVIRT
+#ifndef CONFIG_PARAVIRT_MMU
#define flush_tlb_others(mask, mm, va) native_flush_tlb_others(mask, mm, va)
#endif
diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index 780cd92..1284d8d 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -20,7 +20,7 @@
#include <asm/processor-flags.h>
#include <asm/percpu.h>
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_MMU
#include <asm/asm-offsets.h>
#include <asm/paravirt.h>
#else
diff --git a/arch/x86/kernel/paravirt.c b/arch/x86/kernel/paravirt.c
index 1b1739d..c8530bd 100644
--- a/arch/x86/kernel/paravirt.c
+++ b/arch/x86/kernel/paravirt.c
@@ -155,12 +155,14 @@ unsigned paravirt_patch_default(u8 type, u16 clobbers,
void *insnbuf,
else if (opfunc == _paravirt_ident_64)
ret = paravirt_patch_ident_64(insnbuf, len);
+#ifdef CONFIG_PARAVIRT_CPU
else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
/* If operation requires a jmp, then jmp */
ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
+#endif
else
/* Otherwise call the function; assume target could
clobber any caller-save reg */
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index cd982f4..96aad98 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -66,7 +66,7 @@ u64 native_sched_clock(void)
/* We need to define a real function for sched_clock, to override the
weak default version */
-#ifdef CONFIG_PARAVIRT
+#ifdef CONFIG_PARAVIRT_TIME
unsigned long long sched_clock(void)
{
return paravirt_sched_clock();
diff --git a/arch/x86/kernel/vsmp_64.c b/arch/x86/kernel/vsmp_64.c
index a1d804b..23f4612 100644
--- a/arch/x86/kernel/vsmp_64.c
+++ b/arch/x86/kernel/vsmp_64.c
@@ -22,7 +22,7 @@
#include <asm/paravirt.h>
#include <asm/setup.h>
-#if defined CONFIG_PCI && defined CONFIG_PARAVIRT
+#if defined CONFIG_PCI && defined CONFIG_PARAVIRT_IRQ
/*
* Interrupt control on vSMPowered systems:
* ~AC is a shadow of IF. If IF is 'on' AC should be 'off'
diff --git a/arch/x86/xen/Kconfig b/arch/x86/xen/Kconfig
index b83e119..eef41bd 100644
--- a/arch/x86/xen/Kconfig
+++ b/arch/x86/xen/Kconfig
@@ -4,7 +4,7 @@
config XEN
bool "Xen guest support"
- select PARAVIRT
+ select PARAVIRT_ALL
select PARAVIRT_CLOCK
depends on X86_64 || (X86_32 && X86_PAE && !X86_VISWS)
depends on X86_CMPXCHG && X86_TSC
--
1.6.0.2
Alexander Graf
2009-Nov-18 00:13 UTC
[PATCH 2/3] Only export selected pv-ops feature structs
To really check for sure that we're not using any pv-ops code by accident,
we should make sure that we don't even export the structures used to access
pv-ops exported functions.
So let's surround the pv-ops structs by #ifdefs.
Signed-off-by: Alexander Graf <agraf at suse.de>
---
arch/x86/kernel/paravirt.c | 35 +++++++++++++++++++++++++++++------
1 files changed, 29 insertions(+), 6 deletions(-)
diff --git a/arch/x86/kernel/paravirt.c b/arch/x86/kernel/paravirt.c
index c8530bd..0619e7c 100644
--- a/arch/x86/kernel/paravirt.c
+++ b/arch/x86/kernel/paravirt.c
@@ -124,11 +124,21 @@ static void *get_call_destination(u8 type)
{
struct paravirt_patch_template tmpl = {
.pv_init_ops = pv_init_ops,
+#ifdef CONFIG_PARAVIRT_TIME
.pv_time_ops = pv_time_ops,
+#endif
+#ifdef CONFIG_PARAVIRT_CPU
.pv_cpu_ops = pv_cpu_ops,
+#endif
+#ifdef CONFIG_PARAVIRT_IRQ
.pv_irq_ops = pv_irq_ops,
+#endif
+#ifdef CONFIG_PARAVIRT_APIC
.pv_apic_ops = pv_apic_ops,
+#endif
+#ifdef CONFIG_PARAVIRT_MMU
.pv_mmu_ops = pv_mmu_ops,
+#endif
#ifdef CONFIG_PARAVIRT_SPINLOCKS
.pv_lock_ops = pv_lock_ops,
#endif
@@ -185,6 +195,7 @@ unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
return insn_len;
}
+#ifdef CONFIG_PARAVIRT_MMU
static void native_flush_tlb(void)
{
__native_flush_tlb();
@@ -203,6 +214,7 @@ static void native_flush_tlb_single(unsigned long addr)
{
__native_flush_tlb_single(addr);
}
+#endif /* CONFIG_PARAVIRT_MMU */
/* These are in entry.S */
extern void native_iret(void);
@@ -284,6 +296,7 @@ enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
return percpu_read(paravirt_lazy_mode);
}
+#ifdef CONFIG_PARAVIRT_MMU
void arch_flush_lazy_mmu_mode(void)
{
preempt_disable();
@@ -295,6 +308,7 @@ void arch_flush_lazy_mmu_mode(void)
preempt_enable();
}
+#endif /* CONFIG_PARAVIRT_MMU */
struct pv_info pv_info = {
.name = "bare hardware",
@@ -306,11 +320,16 @@ struct pv_info pv_info = {
struct pv_init_ops pv_init_ops = {
.patch = native_patch,
};
+EXPORT_SYMBOL_GPL(pv_info);
+#ifdef CONFIG_PARAVIRT_TIME
struct pv_time_ops pv_time_ops = {
.sched_clock = native_sched_clock,
};
+EXPORT_SYMBOL_GPL(pv_time_ops);
+#endif
+#ifdef CONFIG_PARAVIRT_IRQ
struct pv_irq_ops pv_irq_ops = {
.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
@@ -322,7 +341,10 @@ struct pv_irq_ops pv_irq_ops = {
.adjust_exception_frame = paravirt_nop,
#endif
};
+EXPORT_SYMBOL (pv_irq_ops);
+#endif
+#ifdef CONFIG_PARAVIRT_CPU
struct pv_cpu_ops pv_cpu_ops = {
.cpuid = native_cpuid,
.get_debugreg = native_get_debugreg,
@@ -383,12 +405,17 @@ struct pv_cpu_ops pv_cpu_ops = {
.start_context_switch = paravirt_nop,
.end_context_switch = paravirt_nop,
};
+EXPORT_SYMBOL (pv_cpu_ops);
+#endif
+#ifdef CONFIG_PARAVIRT_APIC
struct pv_apic_ops pv_apic_ops = {
#ifdef CONFIG_X86_LOCAL_APIC
.startup_ipi_hook = paravirt_nop,
#endif
};
+EXPORT_SYMBOL_GPL(pv_apic_ops);
+#endif
#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
/* 32-bit pagetable entries */
@@ -398,6 +425,7 @@ struct pv_apic_ops pv_apic_ops = {
#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
#endif
+#ifdef CONFIG_PARAVIRT_MMU
struct pv_mmu_ops pv_mmu_ops = {
.read_cr2 = native_read_cr2,
@@ -470,10 +498,5 @@ struct pv_mmu_ops pv_mmu_ops = {
.set_fixmap = native_set_fixmap,
};
-
-EXPORT_SYMBOL_GPL(pv_time_ops);
-EXPORT_SYMBOL (pv_cpu_ops);
EXPORT_SYMBOL (pv_mmu_ops);
-EXPORT_SYMBOL_GPL(pv_apic_ops);
-EXPORT_SYMBOL_GPL(pv_info);
-EXPORT_SYMBOL (pv_irq_ops);
+#endif
--
1.6.0.2
Currently selecting KVM guest support enabled multiple features at once that
not everyone necessarily wants to have, namely:
- PV MMU
- zero io delay
- apic detection workaround
Let's split them off so we don't drag in the full pv-ops framework just
to
detect we're running on KVM. That gives us more chances to tweak
performance!
Signed-off-by: Alexander Graf <agraf at suse.de>
---
arch/x86/Kconfig | 29 ++++++++++++++++++++++++++++-
arch/x86/kernel/kvm.c | 22 +++++++++++++++-------
2 files changed, 43 insertions(+), 8 deletions(-)
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 8c150b6..97d4f92 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -522,11 +522,38 @@ config KVM_CLOCK
config KVM_GUEST
bool "KVM Guest support"
- select PARAVIRT_ALL
+ select PARAVIRT
---help---
This option enables various optimizations for running under the KVM
hypervisor.
+config KVM_IODELAY
+ bool "KVM IO-delay support"
+ depends on KVM_GUEST
+ select PARAVIRT_CPU
+ ---help---
+ Usually we wait for PIO access to complete. When inside KVM there's
+ no need to do that, as we know that we're not going through a bus,
+ but process PIO requests instantly.
+
+ This option disables PIO waits, but drags in CPU-bound pv-ops. Thus
+ you will probably get more speed loss than speedup using this option.
+
+ If in doubt, say N.
+
+config KVM_MMU
+ bool "KVM PV MMU support"
+ depends on KVM_GUEST
+ select PARAVIRT_MMU
+ ---help---
+ This option enables the paravirtualized MMU for KVM. In most cases
+ it's pretty useless and shouldn't be used.
+
+ It will only cost you performance, because it drags in pv-ops for
+ memory management.
+
+ If in doubt, say N.
+
source "arch/x86/lguest/Kconfig"
config PARAVIRT
diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
index 63b0ec8..7e0207f 100644
--- a/arch/x86/kernel/kvm.c
+++ b/arch/x86/kernel/kvm.c
@@ -29,6 +29,16 @@
#include <linux/hardirq.h>
#include <asm/timer.h>
+#ifdef CONFIG_KVM_IODELAY
+/*
+ * No need for any "IO delay" on KVM
+ */
+static void kvm_io_delay(void)
+{
+}
+#endif /* CONFIG_KVM_IODELAY */
+
+#ifdef CONFIG_KVM_MMU
#define MMU_QUEUE_SIZE 1024
struct kvm_para_state {
@@ -43,13 +53,6 @@ static struct kvm_para_state *kvm_para_state(void)
return &per_cpu(para_state, raw_smp_processor_id());
}
-/*
- * No need for any "IO delay" on KVM
- */
-static void kvm_io_delay(void)
-{
-}
-
static void kvm_mmu_op(void *buffer, unsigned len)
{
int r;
@@ -194,15 +197,19 @@ static void kvm_leave_lazy_mmu(void)
mmu_queue_flush(state);
paravirt_leave_lazy_mmu();
}
+#endif /* CONFIG_KVM_MMU */
static void __init paravirt_ops_setup(void)
{
pv_info.name = "KVM";
pv_info.paravirt_enabled = 1;
+#ifdef CONFIG_KVM_IODELAY
if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
pv_cpu_ops.io_delay = kvm_io_delay;
+#endif
+#ifdef CONFIG_KVM_MMU
if (kvm_para_has_feature(KVM_FEATURE_MMU_OP)) {
pv_mmu_ops.set_pte = kvm_set_pte;
pv_mmu_ops.set_pte_at = kvm_set_pte_at;
@@ -226,6 +233,7 @@ static void __init paravirt_ops_setup(void)
pv_mmu_ops.lazy_mode.enter = kvm_enter_lazy_mmu;
pv_mmu_ops.lazy_mode.leave = kvm_leave_lazy_mmu;
}
+#endif /* CONFIG_KVM_MMU */
#ifdef CONFIG_X86_IO_APIC
no_timer_check = 1;
#endif
--
1.6.0.2
On 11/18/2009 02:13 AM, Alexander Graf wrote:> Paravirt ops is currently only capable of either replacing a lot of Linux > internal code or none at all. The are users that don't need all of the > possibilities pv-ops delivers though. > > On KVM for example we're perfectly fine not using the PV MMU, thus not > touching any MMU code. That way we don't have to improve pv-ops to become > fast, we just don't compile the MMU parts in! > > This patchset splits pv-ops into several smaller config options split by > feature category and then converts the KVM pv-ops code to use only the > bits that are required, lowering overhead. > > Alexander Graf (3): > Split paravirt ops by functionality > Only export selected pv-ops feature structs > Split the KVM pv-ops support by feature > >The whole thing looks good to me. Let's wait for Jeremy to ack though. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic.
Alexander Graf wrote:> Paravirt ops is currently only capable of either replacing a lot of Linux > internal code or none at all. The are users that don't need all of the > possibilities pv-ops delivers though. > > On KVM for example we're perfectly fine not using the PV MMU, thus not > touching any MMU code. That way we don't have to improve pv-ops to become > fast, we just don't compile the MMU parts in! > > This patchset splits pv-ops into several smaller config options split by > feature category and then converts the KVM pv-ops code to use only the > bits that are required, lowering overhead. >So has this ended up in some tree yet?
Possibly Parallel Threads
- [PATCH 0/3] Split up pv-ops
- [PATCH 00/10] x86/paravirt: several cleanups
- [PATCH v2 00/11] x86/paravirt: several cleanups
- [PATCH 00/13] x86/paravirt: Make pv ops code generation more closely match reality
- [PATCH 00/13] x86/paravirt: Make pv ops code generation more closely match reality