x86/vmx: Partially revert "x86/vmx: implement Notify VM Exit"
PATCH 1/1 x86/vmx: Partially revert "x86/vmx: implement Notify VM Exit"
The original patch tried to do two things - implement VMNotify, and re-optimise VT-x to not intercept #DB/#AC by default.
The second part is buggy in multiple ways. Both GDBSX and Introspection need to conditionally intercept #DB, which was not accounted for. Also, #DB interception has nothing at all to do with cpu_has_monitor_trap_flag.
Revert the second half, leaving #DB/#AC intercepted unilaterally, but with VMNotify active by default when available.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Kevin Tian <[email protected]>
#DB/#AC are not fastpaths in the slightest. This perf optimisation can be
reworked at some point later with rather more care and testing.
It's *really* not as urgent as getting VMNotify active by default.
--- xen/arch/x86/hvm/vmx/vmcs.c | 11 ++--------- xen/arch/x86/hvm/vmx/vmx.c | 16 ++-------------- 2 files changed, 4 insertions(+), 23 deletions(-)
xen/arch/x86/hvm/vmx/vmcs.c +2 -9
@@ -1296,17 +1296,10 @@ static int construct_vmcs(struct vcpu *v) v->arch.hvm.vmx.exception_bitmap =HVM_TRAP_MASK | (paging_mode_hap (d) ?0 : (1U << TRAP_page_fault)) | (v->arch.fully_eager_fpu ?0 : (1U << TRAP_no_device)); +if ( cpu_has_vmx_notify_vm_exiting ) - {__vmwrite (NOTIFY_WINDOW , vm_notify_window); -/* -* Disable #AC and #DB interception: by using VM Notify Xen is -* guaranteed to get a VM exit even if the guest manages to lock the -* CPU. -*/ - v->arch.hvm.vmx.exception_bitmap &= ~((1U << TRAP_debug) | - (1U << TRAP_alignment_check)); - } +vmx_update_exception_bitmap (v); v->arch.hvm.guest_cr[0 ] =X86_CR0_PE |X86_CR0_ET ;
xen/arch/x86/hvm/vmx/vmx.c +2 -14
@@ -1564,19 +1564,10 @@ static void cf_check vmx_update_host_cr3(struct vcpu *v)void vmx_update_debug_state (struct vcpu *v) { -unsigned int mask =1u << TRAP_int3; - -if ( !cpu_has_monitor_trap_flag && cpu_has_vmx_notify_vm_exiting ) -/* -* Only allow toggling TRAP_debug if notify VM exit is enabled, as -* unconditionally setting TRAP_debug is part of the XSA-156 fix. -*/ - mask |=1u << TRAP_debug; -if ( v->arch.hvm.debug_state_latch ) - v->arch.hvm.vmx.exception_bitmap |= mask; + v->arch.hvm.vmx.exception_bitmap |=1U << TRAP_int3;else - v->arch.hvm.vmx.exception_bitmap &= ~mask; + v->arch.hvm.vmx.exception_bitmap &= ~(1U << TRAP_int3);vmx_vmcs_enter (v);vmx_update_exception_bitmap (v);
@@ -4192,9 +4183,6 @@ void vmx_vmexit_handler(struct cpu_user_regs *regs)switch ( vector ) {case TRAP_debug: -if ( cpu_has_monitor_trap_flag && cpu_has_vmx_notify_vm_exiting ) - goto exit_and_crash; -/* * Updates DR6 where debugger can peek (See 3B 23.2.1, * Table 23-1, "Exit Qualification for Debug Exceptions").
Signature · 1 line
2.11.0
Quoted 11 lines
From: Andrew Cooper <[email protected]>
Sent: Thursday, January 19, 2023 3:37 AMThe original patch tried to do two things - implement VMNotify, and re-optimise VT-x to not intercept #DB/#AC by default.
The second part is buggy in multiple ways. Both GDBSX and Introspection need
to conditionally intercept #DB, which was not accounted for. Also, #DB interception has nothing at all to do withcpu_has_monitor_trap_flag.
Quoted 5 lines
Revert the second half, leaving #DB/#AC intercepted unilaterally, but with VMNotify active by default when available.
Fixes 573279cde1c4 ("x86/vmx: implement Notify VM Exit")Signed-off-by Andrew Cooper <[email protected]>