x86: Add Support for Paging-Write Feature
PATCH v2 1/2 x86: Rename _rsvd field to pw and move it to the bit 58
In reply to Petr BenešFrom: Petr Beneš <[email protected]>
The EPT Paging-write feature (when enabled by the TERTIARY_EXEC_EPT_PAGING_WRITE bit) uses bit 58 of the EPT entry to indicate that guest paging may update the page, even if the W access is not set.
This patch is a preparation for the EPT Paging-write feature.
--- xen/arch/x86/include/asm/hvm/vmx/vmx.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
xen/arch/x86/include/asm/hvm/vmx/vmx.h +2 -2
@@ -34,8 +34,8 @@ typedef union {EPT/VT-d usage */ mfn : 40, /* bits 51:12 - Machine physical frame number */ sa_p2mt : 6, /* bits 57:52 - Software available 2 */ - access : 4, /* bits 61:58 - p2m_access_t */ - _rsvd : 1, /* bit 62 - reserved */ + pw : 1, /* bit 58 - Paging-write access */ + access : 4, /* bits 62:59 - p2m_access_t */ suppress_ve : 1; /* bit 63 - suppress #VE */ };u64 epte;
Signature · 1 line
2.34.1
v2 cover x86: Add Support for Paging-Write Feature
From: Petr Beneš <[email protected]>
Changes since v1:
- Added signed-off-by tags
This patch introduces a new XENMEM_access_r_pw permission. Functionally, it is similar to XENMEM_access_r, but for processors with TERTIARY_EXEC_EPT_PAGING_WRITE support (Intel 12th Gen/Alder Lake and later), it also permits the CPU to write to the page during guest page-table walks (e.g., updating A/D bits) without triggering an EPT violation.
This behavior works by both enabling the EPT paging-write feature and setting the EPT paging-write flag in the EPT leaf entry.
This feature provides a significant performance boost for introspection tools that monitor guest page-table updates. Previously, every page-table modification by the guest—including routine updates like setting A/D bits—triggered an EPT violation, adding unnecessary overhead. The new XENMEM_access_r_pw permission allows these "uninteresting" updates to occur without EPT violations, improving efficiency.
Additionally, this feature simplifies the handling of race conditions in scenarios where an introspection tool:
- Sets an "invisible breakpoint" in the altp2m view for a function F
- Monitors guest page-table updates to track whether the page containing F is paged out
- Encounters a cleared Access (A) bit on the page containing F while the guest is about to execute the breakpoint
In the current implementation:
- If xc_monitor_inguest_pagefault() is enabled, the introspection tool must emulate both the breakpoint and the setting of the Access bit.
- If xc_monitor_inguest_pagefault() is disabled, Xen handles the EPT violation without notifying the introspection tool, setting the Access bit and emulating the instruction. However, Xen fetches the instruction from the default view instead of the altp2m view, potentially causing the breakpoint to be missed.
With this patch, setting XENMEM_access_r_pw for monitored guest page-tables prevents EPT violations in these cases. This change enhances performance and reduces complexity for introspection tools, ensuring seamless breakpoint handling while tracking guest page-table updates.
Petr Beneš (2):
x86: Rename _rsvd field to pw and move it to the bit 58
x86: Add Support for Paging-Write Feature
xen/arch/arm/mem_access.c | 4 ++++ xen/arch/arm/mmu/p2m.c | 1 + xen/arch/x86/hvm/hvm.c | 1 + xen/arch/x86/hvm/monitor.c | 1 + xen/arch/x86/hvm/vmx/vmcs.c | 4 +++- xen/arch/x86/include/asm/hvm/vmx/vmcs.h | 3 +++ xen/arch/x86/include/asm/hvm/vmx/vmx.h | 4 ++-- xen/arch/x86/include/asm/p2m.h | 1 + xen/arch/x86/mm/hap/nested_hap.c | 3 +++ xen/arch/x86/mm/mem_access.c | 3 +++ xen/arch/x86/mm/p2m-ept.c | 4 ++++ xen/include/public/memory.h | 9 +++++++++ xen/include/xen/mem_access.h | 6 ++++++ 13 files changed, 41 insertions(+), 3 deletions(-)
Signature · 1 line
2.34.1
PATCH v2 2/2 x86: Add Support for Paging-Write Feature
From: Petr Beneš <[email protected]>
This patch introduces a new XENMEM_access_r_pw permission. Functionally, it is similar to XENMEM_access_r, but for processors with TERTIARY_EXEC_EPT_PAGING_WRITE support (Intel 12th Gen/Alder Lake and later), it also permits the CPU to write to the page during guest page-table walks (e.g., updating A/D bits) without triggering an EPT violation.
This behavior works by both enabling the EPT paging-write feature and setting the EPT paging-write flag in the EPT leaf entry.
This feature provides a significant performance boost for introspection tools that monitor guest page-table updates. Previously, every page-table modification by the guest—including routine updates like setting A/D bits—triggered an EPT violation, adding unnecessary overhead. The new XENMEM_access_r_pw permission allows these "uninteresting" updates to occur without EPT violations, improving efficiency.
Additionally, this feature simplifies the handling of race conditions in scenarios where an introspection tool:
- Sets an "invisible breakpoint" in the altp2m view for a function F
- Monitors guest page-table updates to track whether the page containing F is paged out
- Encounters a cleared Access (A) bit on the page containing F while the guest is about to execute the breakpoint
In the current implementation:
- If xc_monitor_inguest_pagefault() is enabled, the introspection tool must emulate both the breakpoint and the setting of the Access bit.
- If xc_monitor_inguest_pagefault() is disabled, Xen handles the EPT violation without notifying the introspection tool, setting the Access bit and emulating the instruction. However, Xen fetches the instruction from the default view instead of the altp2m view, potentially causing the breakpoint to be missed.
With this patch, setting XENMEM_access_r_pw for monitored guest page-tables prevents EPT violations in these cases. This change enhances performance and reduces complexity for introspection tools, ensuring seamless breakpoint handling while tracking guest page-table updates.
--- xen/arch/arm/mem_access.c | 4 ++++ xen/arch/arm/mmu/p2m.c | 1 + xen/arch/x86/hvm/hvm.c | 1 + xen/arch/x86/hvm/monitor.c | 1 + xen/arch/x86/hvm/vmx/vmcs.c | 4 +++- xen/arch/x86/include/asm/hvm/vmx/vmcs.h | 3 +++ xen/arch/x86/include/asm/p2m.h | 1 + xen/arch/x86/mm/hap/nested_hap.c | 3 +++ xen/arch/x86/mm/mem_access.c | 3 +++ xen/arch/x86/mm/p2m-ept.c | 4 ++++ xen/include/public/memory.h | 9 +++++++++ xen/include/xen/mem_access.h | 6 ++++++ 12 files changed, 39 insertions(+), 1 deletion(-)
xen/arch/arm/mem_access.c +4 -0
@@ -32,6 +32,7 @@ static int __p2m_get_mem_access(struct domain *d, gfn_t gfn,ACCESS (rwx),ACCESS (rx2rw),ACCESS (n2rwx), +ACCESS (r_pw), #undef ACCESS };
@@ -172,6 +173,7 @@ p2m_mem_access_check_and_get_page(vaddr_t gva, unsigned long flag,break ; else goto err; +case XENMEM_access_r_pw:case XENMEM_access_rx2rw:case XENMEM_access_rx:case XENMEM_access_r:
@@ -253,6 +255,7 @@ bool p2m_mem_access_check(paddr_t gpa, vaddr_t gla, const struct npfec npfec) violation = npfec.read_access || npfec.insn_fetch;break ;case XENMEM_access_r: +case XENMEM_access_r_pw: violation = npfec.write_access || npfec.insn_fetch;break ;default :
@@ -361,6 +364,7 @@ long p2m_set_mem_access(struct domain *d, gfn_t gfn, uint32_t nr,ACCESS (rwx),ACCESS (rx2rw),ACCESS (n2rwx), +ACCESS (r_pw), #undef ACCESS };
xen/arch/arm/mmu/p2m.c +1 -0
@@ -597,6 +597,7 @@ static void p2m_set_permission(lpae_t *e, p2m_type_t t, p2m_access_t a) e->p2m.read =0 ;break ;case p2m_access_r: +case p2m_access_r_pw: e->p2m.write =0 ; e->p2m.xn =1 ;break ;
xen/arch/x86/hvm/hvm.c +1 -0
@@ -1897,6 +1897,7 @@ int hvm_hap_nested_page_fault(paddr_t gpa, unsigned long gla, violation = npfec.read_access || npfec.write_access || npfec.insn_fetch;break ;case p2m_access_r: +case p2m_access_r_pw: violation = npfec.write_access || npfec.insn_fetch;break ;case p2m_access_w:
xen/arch/x86/hvm/monitor.c +1 -0
@@ -295,6 +295,7 @@ bool hvm_monitor_check_p2m(unsigned long gla, gfn_t gfn, uint32_t pfec,case XENMEM_access_r:case XENMEM_access_n: +case XENMEM_access_r_pw:if ( pfec & PFEC_write_access ) req.u.mem_access.flags |=MEM_ACCESS_R |MEM_ACCESS_W ;if ( pfec & PFEC_insn_fetch )
xen/arch/x86/hvm/vmx/vmcs.c +3 -1
@@ -203,6 +203,7 @@ static void __init vmx_display_features(void)P (cpu_has_vmx_bus_lock_detection,"Bus Lock Detection" );P (cpu_has_vmx_notify_vm_exiting,"Notify VM Exit" );P (cpu_has_vmx_virt_spec_ctrl,"Virtualize SPEC_CTRL" ); +P (cpu_has_vmx_ept_paging_write,"EPT Paging-Write" );#undef Pif ( !printed )
@@ -366,7 +367,8 @@ static int vmx_init_vmcs_config(bool bsp)if ( _vmx_cpu_based_exec_control &CPU_BASED_ACTIVATE_TERTIARY_CONTROLS ) { -uint64_t opt =TERTIARY_EXEC_VIRT_SPEC_CTRL ; +uint64_t opt = (TERTIARY_EXEC_VIRT_SPEC_CTRL | +TERTIARY_EXEC_EPT_PAGING_WRITE ); _vmx_tertiary_exec_control = adjust_vmx_controls2( "Tertiary Exec Control", 0, opt,
xen/arch/x86/include/asm/hvm/vmx/vmcs.h +3 -0
@@ -273,6 +273,9 @@ extern uint64_t vmx_tertiary_exec_control;#define cpu_has_vmx_virt_spec_ctrl \ (vmx_tertiary_exec_control & TERTIARY_EXEC_VIRT_SPEC_CTRL) +#define cpu_has_vmx_ept_paging_write \ + (vmx_tertiary_exec_control & TERTIARY_EXEC_EPT_PAGING_WRITE) +#define VMX_EPT_EXEC_ONLY_SUPPORTED 0x00000001#define VMX_EPT_WALK_LENGTH_4_SUPPORTED 0x00000040#define VMX_EPT_MEMORY_TYPE_UC 0x00000100
xen/arch/x86/include/asm/p2m.h +1 -0
@@ -980,6 +980,7 @@ static inline unsigned int p2m_access_to_iommu_flags(p2m_access_t p2ma)case p2m_access_r:case p2m_access_rx:case p2m_access_rx2rw: +case p2m_access_r_pw:return IOMMUF_readable;case p2m_access_w:
xen/arch/x86/mm/hap/nested_hap.c +3 -0
@@ -213,6 +213,9 @@ int nestedhvm_hap_nested_page_fault(case p2m_access_n2rwx: p2ma_10 = p2m_access_n;break ; +case p2m_access_r_pw: + p2ma_10 = p2m_access_r; +break ;default : p2ma_10 = p2m_access_n;/* For safety, remove all permissions. */
xen/arch/x86/mm/mem_access.c +3 -0
@@ -45,6 +45,7 @@ static int _p2m_get_mem_access(struct p2m_domain *p2m, gfn_t gfn,ACCESS (rwx),ACCESS (rx2rw),ACCESS (n2rwx), +ACCESS (r_pw), #undef ACCESS };
@@ -94,6 +95,7 @@ bool p2m_mem_access_emulate_check(struct vcpu *v,break ;case XENMEM_access_r: +case XENMEM_access_r_pw: violation = data->flags &MEM_ACCESS_WX ;break ;
@@ -312,6 +314,7 @@ bool xenmem_access_to_p2m_access(const struct p2m_domain *p2m,ACCESS (rwx),ACCESS (rx2rw),ACCESS (n2rwx), +ACCESS (r_pw), #undef ACCESS };
xen/arch/x86/mm/p2m-ept.c +4 -0
@@ -176,6 +176,10 @@ static void ept_p2m_type_to_flags(const struct p2m_domain *p2m,break ;case p2m_access_rwx:break ; +case p2m_access_r_pw: + entry->w = entry->x =0 ; + entry->pw = !!cpu_has_vmx_ept_paging_write; +break ; }/*
xen/include/public/memory.h +9 -0
@@ -426,6 +426,15 @@ typedef enum {* pausing the vcpu */ XENMEM_access_n2rwx, + +/* +* Same as XENMEM_access_r, but on processors with +* the TERTIARY_EXEC_EPT_PAGING_WRITE support, +* CPU-initiated page-table walks can still +* write to it (e.g., update A/D bits) +*/ + XENMEM_access_r_pw, +/* Take the domain default */ XENMEM_access_default } xenmem_access_t;
xen/include/xen/mem_access.h +6 -0
@@ -64,6 +64,12 @@ typedef enum {* generates an event but does not pause the * vcpu */ + p2m_access_r_pw = 10, /* Special: same as R, but on processors with +* the TERTIARY_EXEC_EPT_PAGING_WRITE support, +* CPU-initiated page-table walks can still +* write to it (e.g., update A/D bits) +*/ +/* NOTE: Assumed to be only 4 bits right now on x86. */ } p2m_access_t;
Signature · 1 line
2.34.1
xen/arch/x86/mm/p2m-ept.c static void ept_p2m_type_to_flags(const struct p2m_domain *p2m, · line 183
Quoted 11 lines · Petr Beneš
xen/arch/x86/mm/p2m-ept.c +4
-0@@ -176,6 +176,10 @@ static void ept_p2m_type_to_flags(const struct p2m_domain *p2m,break ;case p2m_access_rwx:break ; +case p2m_access_r_pw: + entry->w = entry->x =0 ; + entry->pw = !!cpu_has_vmx_ept_paging_write; +break ; }
To avoid the flag becoming "sticky", I think you need to clear it ahead of this switch() (or right at the start of the function).
Jan
Quoted 5 lines · Jan Beulich
To avoid the flag becoming "sticky", I think you need to clear it ahead of this
switch()(or right at the start of the function).Jan
When thinking about it now, wouldn't it make sense to only reset it in the p2m_ram_rw case above? I mean, where else would be guest page-tables located.
P.
Quoted 8 lines · Petr Beneš
Quoted 3 lines · Jan Beulich
To avoid the flag becoming "sticky", I think you need to clear it ahead of this
switch()(or right at the start of the function).When thinking about it now, wouldn't it make sense to only reset it in the
p2m_ram_rwcase above? I mean, where else would be guest page-tables located.
I'm afraid that unless you can prove otherwise, my general take on such is "you never know".
Jan