x86/xen: PVH fixes
Subsystems (2)
v4 cover x86/xen: PVH fixes
Two patches for PVH Linux guests.
This has been tested on an Intel processor (but not AMD).
Changes in v4 (David):
- cpu == 0 => boot CPU
- Reduce #ifdefs.
- Add patch for XEN_PVH docs.
Changes in v3 (David):
- Use common xen_pvh_cpu_early_init() function for boot and secondary VCPUs.
Changes in v2: (Mukesh):
- Use assembly macro to unify code for boot and secondary VCPUs.
PATCH v4 1/2 x86/xen: Set EFER.NX and EFER.SCE in PVH guests
From: Mukesh Rathor <[email protected]>
This fixes two bugs in PVH guests:
- Not setting EFER.NX means the NX bit in page table entries is ignored on Intel processors and causes reserved bit page faults on AMD processors.
- After the Xen commit 7645640d6ff1 (x86/PVH: don't set EFER_SCE for pvh guest) PVH guests are required to set EFER.SCE to enable the SYSCALL instruction.
Secondary VCPUs are started with pagetables with the NX bit set so EFER.NX must be set before using any stack or data segment. xen_pvh_cpu_early_init() is the new secondary VCPU entry point that sets EFER before jumping to cpu_bringup_and_idle().
--- arch/x86/xen/enlighten.c | 2 ++ arch/x86/xen/smp.c | 28 +++++++++++++++++----------- arch/x86/xen/smp.h | 8 ++++++++ arch/x86/xen/xen-head.S | 30 ++++++++++++++++++++++++++++++ 4 files changed, 57 insertions(+), 11 deletions(-)
arch/x86/xen/enlighten.c +2 -0
@@ -1508,6 +1508,8 @@ static void __init xen_pvh_early_guest_init(void)return ; xen_have_vector_callback =1 ; + +xen_pvh_cpu_early_init (0 );xen_pvh_set_cr_flags (0 );#ifdef CONFIG_X86_32
arch/x86/xen/smp.c +17 -11
@@ -37,6 +37,7 @@#include <xen/hvc-console.h> #include "xen-ops.h" #include "mmu.h" +#include "smp.h" cpumask_var_t xen_cpu_initialized_map;
@@ -99,8 +100,12 @@ static void cpu_bringup(void)wmb ();/* make sure everything is out */ } -/* Note: cpu parameter is only relevant for PVH */ -static void cpu_bringup_and_idle (int cpu) +/* +* Note: cpu parameter is only relevant for PVH. The reason for passing it +* is we can't do smp_processor_id until the percpu segments are loaded, for +* which we need the cpu number! So we pass it in rdi as first parameter. +*/ +asmlinkage __visible void cpu_bringup_and_idle(int cpu) {#ifdef CONFIG_X86_64 if (xen_feature(XENFEAT_auto_translated_physmap) &&
@@ -374,11 +379,10 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle) ctxt->user_regs.fs = __KERNEL_PERCPU; ctxt->user_regs.gs = __KERNEL_STACK_CANARY;#endif - ctxt->user_regs.eip = (unsigned long )cpu_bringup_and_idle; -memset (&ctxt->fpu_ctxt,0 ,sizeof (ctxt->fpu_ctxt));if (!xen_feature (XENFEAT_auto_translated_physmap)) { + ctxt->user_regs.eip = (unsigned long )cpu_bringup_and_idle; ctxt->flags =VGCF_IN_KERNEL ; ctxt->user_regs.eflags =0x1000 ;/* IOPL_RING1 */ ctxt->user_regs.ds = __USER_DS;
@@ -413,15 +417,17 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle) (unsigned long )xen_failsafe_callback; ctxt->user_regs.cs = __KERNEL_CS;per_cpu (xen_cr3, cpu) =__pa (swapper_pg_dir); -#ifdef CONFIG_X86_32 - } -#else - }else -/* N.B. The user_regs.eip (cpu_bringup_and_idle) is called with -* %rdi having the cpu number - which means are passing in -* as the first parameter the cpu. Subtle! + } +#ifdef CONFIG_XEN_PVH + else { +/* +* The vcpu comes on kernel page tables which have the NX pte +* bit set. This means before DS/SS is touched, NX in +* EFER must be set. Hence the following assembly glue code. */ + ctxt->user_regs.eip = (unsigned long )xen_pvh_cpu_early_init; ctxt->user_regs.rdi = cpu; + }#endif ctxt->user_regs.esp = idle->thread.sp0 -sizeof (struct pt_regs ); ctxt->ctrlreg[3 ] =xen_pfn_to_cr3 (virt_to_mfn (swapper_pg_dir));
arch/x86/xen/smp.h +8 -0
@@ -8,4 +8,12 @@ extern void xen_send_IPI_allbutself(int vector);extern void xen_send_IPI_all (int vector);extern void xen_send_IPI_self (int vector); +#ifdef CONFIG_XEN_PVH +extern void xen_pvh_cpu_early_init (int cpu); +#else +static inline void xen_pvh_cpu_early_init (int cpu) +{ +} +#endif +#endif
arch/x86/xen/xen-head.S +30 -0
@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi + + mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax + + bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */ + jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */ + .pushsection .text .balign PAGE_SIZE ENTRY(hypercall_page)
Signature · 1 line
1.7.10.4
Re: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
Quoted 22 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:11, <[email protected]> wrote:
arch/x86/xen/xen-head.S +15
-0@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi +
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 74
Quoted 10 lines
arch/x86/xen/xen-head.S +10
-0+ mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax + + bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */
Isn't Linux is specifically moving away from this assumption?
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 78
Quoted 4 lines
arch/x86/xen/xen-head.S +4
-0+ jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */
Also, does all this really need to be done in assembly?
Jan
Re: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
Quoted 26 lines · Jan Beulich
Quoted 26 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:11, <[email protected]> wrote:
arch/x86/xen/xen-head.S +19
-0@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi + + mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax +
Quoted 8 lines
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 74
Quoted 6 lines
arch/x86/xen/xen-head.S +6
-0+ bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */Isn't Linux is specifically moving away from this assumption?
I don't think this is relevant. Is Xen ever going boot with a non-zero VCPU?
Quoted 6 lines
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 78
Quoted 4 lines
arch/x86/xen/xen-head.S +4
-0+ jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */Also, does all this really need to be done in assembly?
There's no usable stack until EFER.NX is set. I couldn't think of a way to write this in C that would guarantee no stack use, so I didn't suggest it.
Can you think of a way?
David
Re: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
Quoted 37 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:46, <[email protected]> wrote:
Quoted 26 lines · Jan Beulich
Quoted 26 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:11, <[email protected]> wrote:
arch/x86/xen/xen-head.S +19
-0@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi + + mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax +Quoted 8 lines
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 74
Quoted 6 lines
arch/x86/xen/xen-head.S +6
-0+ bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */Isn't Linux is specifically moving away from this assumption?
Quoted 2 lines
I don't think this is relevant. Is Xen ever going boot with a non-zero VCPU?
That's not the question here. The question is - will CPU 0 ever be possible to be brought down and back up? And I hope the answer to this isn't "No, never".
Quoted 7 lines
Quoted 6 lines
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 78
Quoted 4 lines
arch/x86/xen/xen-head.S +4
-0+ jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */Also, does all this really need to be done in assembly?
Quoted 5 lines
There's no usable stack until EFER.NX is set. I couldn't think of a way to write this in C that would guarantee no stack use, so I didn't suggest it.
Can you think of a way?
No stack use is a very convincing argument for using assembly? I must have overlooked it if that was said somewhere in a comment.
Jan
Re: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
Quoted 37 lines · Jan Beulich
Quoted 37 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:46, <[email protected]> wrote:
Quoted 34 lines · Jan Beulich
Quoted 22 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:11, <[email protected]> wrote:
arch/x86/xen/xen-head.S +15
-0@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi +arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 74
Quoted 10 lines
arch/x86/xen/xen-head.S +10
-0+ mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax + + bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */Isn't Linux is specifically moving away from this assumption?
Quoted 6 lines
Quoted 2 lines
I don't think this is relevant. Is Xen ever going boot with a non-zero VCPU?
That's not the question here. The question is - will CPU 0 ever be possible to be brought down and back up? And I hope the answer to this isn't "No, never".
Oh, that's a good point. I'll add the flag back in.
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
arch/x86/xen/xen-head.S ENTRY(startup_xen) · line 78
Quoted 4 lines
arch/x86/xen/xen-head.S +4
-0+ jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */Also, does all this really need to be done in assembly?
Quoted 7 lines
Quoted 5 lines
There's no usable stack until EFER.NX is set. I couldn't think of a way to write this in C that would guarantee no stack use, so I didn't suggest it.
Can you think of a way?
No stack use is a very convincing argument for using assembly?
Is that question mark a typo? If not, I'm not really sure how to answer.
Quoted 3 lines
I must have overlooked it if that was said somewhere in a comment.
From the commit message:
"Secondary VCPUs are started with pagetables with the NX bit set so EFER.NX must be set before using any stack or data segment. xen_pvh_cpu_early_init() is the new secondary VCPU entry point that sets EFER before jumping to cpu_bringup_and_idle()."
And a comment:
+ * The vcpu comes on kernel page tables which have the NX pte
+ * bit set. This means before DS/SS is touched, NX in
+ * EFER must be set. Hence the following assembly glue code.
If this isn't sufficiently clear, can you suggest some improved wording?
David
Re: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 18:21, <[email protected]> wrote:
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 05.09.14 at 16:46, <[email protected]> wrote:
There's no usable stack until EFER.NX is set. I couldn't think of a way to write this in C that would guarantee no stack use, so I didn't suggest it.
Quoted 5 lines
Quoted 3 lines
Quoted 1 line
Can you think of a way?
No stack use is a very convincing argument for using assembly?
Is that question mark a typo? If not, I'm not really sure how to answer.
Yes, it is. I'm sorry for the confusion.
Jan
Boris Ostrovsky
reviewerRe: [PATCH 1/2] x86/xen: Set EFER.NX and EFER.SCE in PVH guests
In reply to David VrabelQuoted 31 lines · David Vrabel
From: Mukesh Rathor <[email protected]>
This fixes two bugs in PVH guests:
- Not setting EFER.NX means the NX bit in page table entries is ignored on Intel processors and causes reserved bit page faults on AMD processors.
- After the Xen commit
7645640d6ff1(x86/PVH: don't setEFER_SCEfor pvh guest) PVH guests are required to set EFER.SCE to enable the SYSCALL instruction.Secondary VCPUs are started with pagetables with the NX bit set so EFER.NX must be set before using any stack or data segment.
xen_pvh_cpu_early_init()is the new secondary VCPU entry point that sets EFER before jumping tocpu_bringup_and_idle().Signed-off-by Mukesh Rathor <[email protected]>Signed-off-by David Vrabel <[email protected]>--- arch/x86/xen/enlighten.c | 2 ++ arch/x86/xen/smp.c | 28 +++++++++++++++++----------- arch/x86/xen/smp.h | 8 ++++++++ arch/x86/xen/xen-head.S | 30 ++++++++++++++++++++++++++++++ 4 files changed, 57 insertions(+), 11 deletions(-)arch/x86/xen/enlighten.c +0
-0
arch/x86/xen/enlighten.c static void __init xen_pvh_early_guest_init(void) · line 1512
Quoted 5 lines
arch/x86/xen/enlighten.c +2
-0@@ -1508,6 +1508,8 @@ static void __init xen_pvh_early_guest_init(void)return ; xen_have_vector_callback =1 ; + +xen_pvh_cpu_early_init (0 );
Nit: for consistency may be better to call this xen_pvh_early_cpu_init().
Quoted 46 lines
arch/x86/xen/enlighten.c +0
-0xen_pvh_set_cr_flags (0 );#ifdef CONFIG_X86_32 arch/x86/xen/smp.c +8
-4@@ -37,6 +37,7 @@#include <xen/hvc-console.h> #include "xen-ops.h" #include "mmu.h" +#include "smp.h" cpumask_var_t xen_cpu_initialized_map;@@ -99,8 +100,12 @@ static void cpu_bringup(void)wmb ();/* make sure everything is out */ } -/* Note: cpu parameter is only relevant for PVH */ -static void cpu_bringup_and_idle (int cpu) +/* +* Note: cpu parameter is only relevant for PVH. The reason for passing it +* is we can't do smp_processor_id until the percpu segments are loaded, for +* which we need the cpu number! So we pass it in rdi as first parameter. +*/ +asmlinkage __visible void cpu_bringup_and_idle(int cpu) {#ifdef CONFIG_X86_64 if (xen_feature(XENFEAT_auto_translated_physmap) &&@@ -374,11 +379,10 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle) ctxt->user_regs.fs = __KERNEL_PERCPU; ctxt->user_regs.gs = __KERNEL_STACK_CANARY;#endif - ctxt->user_regs.eip = (unsigned long )cpu_bringup_and_idle; -memset (&ctxt->fpu_ctxt,0 ,sizeof (ctxt->fpu_ctxt));if (!xen_feature (XENFEAT_auto_translated_physmap)) { + ctxt->user_regs.eip = (unsigned long )cpu_bringup_and_idle; ctxt->flags =VGCF_IN_KERNEL ; ctxt->user_regs.eflags =0x1000 ;/* IOPL_RING1 */ ctxt->user_regs.ds = __USER_DS;@@ -413,15 +417,17 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle) (unsigned long )xen_failsafe_callback; ctxt->user_regs.cs = __KERNEL_CS;
arch/x86/xen/smp.c cpu_initialize_context(unsigned int cpu, struct task_struct *idle) · line 421
Quoted 10 lines
arch/x86/xen/smp.c +2
-7per_cpu (xen_cr3, cpu) =__pa (swapper_pg_dir); -#ifdef CONFIG_X86_32 - } -#else - }else -/* N.B. The user_regs.eip (cpu_bringup_and_idle) is called with -* %rdi having the cpu number - which means are passing in -* as the first parameter the cpu. Subtle! + } +#ifdef CONFIG_XEN_PVH
Since you are wrapping this in ifdef here we should probably also do the same for xen_pvh_early_guest_init() xen_cpu_secondary_cpu_init().
-boris
Quoted 70 lines
arch/x86/xen/smp.c +7
-0+ else { +/* +* The vcpu comes on kernel page tables which have the NX pte +* bit set. This means before DS/SS is touched, NX in +* EFER must be set. Hence the following assembly glue code. */ + ctxt->user_regs.eip = (unsigned long )xen_pvh_cpu_early_init; ctxt->user_regs.rdi = cpu; + }#endif ctxt->user_regs.esp = idle->thread.sp0 -sizeof (struct pt_regs ); ctxt->ctrlreg[3 ] =xen_pfn_to_cr3 (virt_to_mfn (swapper_pg_dir));arch/x86/xen/smp.h +8
-0@@ -8,4 +8,12 @@ extern void xen_send_IPI_allbutself(int vector);extern void xen_send_IPI_all (int vector);extern void xen_send_IPI_self (int vector); +#ifdef CONFIG_XEN_PVH +extern void xen_pvh_cpu_early_init (int cpu); +#else +static inline void xen_pvh_cpu_early_init (int cpu) +{ +} +#endif +#endif arch/x86/xen/xen-head.S +30
-0@@ -47,6 +47,36 @@ ENTRY(startup_xen) __FINIT +#ifdef CONFIG_XEN_PVH + +/** + * xen_pvh_cpu_early_init() - early PVH VCPU initialization + * @cpu: this cpu number (%rdi) + * + * Note: This is called as a function on the boot CPU and as the secondary + * CPU entry point. + */ +ENTRY(xen_pvh_cpu_early_init) + /* Gather features to see if NX implemented. */ + mov $0x80000001, %eax + cpuid + mov %edx,%esi + + mov $MSR_EFER, %ecx + rdmsr + bts $_EFER_SCE, %eax + + bt $20,%esi + jnc 1f /* No NX, skip setting it */ + bts $_EFER_NX, %eax +1: wrmsr + + cmp $0,%rdi /* non-zero => secondary cpu */ + jne cpu_bringup_and_idle + ret + +#endif /* CONFIG_XEN_PVH */ + .pushsection .text .balign PAGE_SIZE ENTRY(hypercall_page)
PATCH v4 2/2 x86/xen: document CONFIG_XEN_PVH option
In reply to David Vrabel--- arch/x86/xen/Kconfig | 8 ++++++++ 1 file changed, 8 insertions(+)
arch/x86/xen/Kconfig +8 -0
@@ -50,3 +50,11 @@ config XEN_PVH bool "Support for running as a PVH guest" depends on X86_64 && XEN && XEN_PVHVM def_bool n + help + PVH is an experimental guest mode that is a cross between + full PV and HVM. The hypervisor ABI for PVH guests is not + yet stable and this kernel may not boot in PVH mode on later + Xen hypervisors. + + Do not enable this option unless you are a developer working + on PVH guest support.
Signature · 1 line
1.7.10.4