Message ID | 20240712235701.1458888-9-seanjc@google.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [GIT,PULL,(sort,of)] KVM: x86: Static call changes for 6.11 | expand |
On 7/13/24 01:56, Sean Christopherson wrote: > Here's a massage pull request for the static_call() changes, just in case you > want to go this route instead of applying patches directly after merging > everything else for 6.11 (it was easy to generate this). If you want to go the > patches route, I'll post 'em next week. > > The following changes since commit c1c8a908a5f4c372f8a8dca0501b56ffc8d260fe: > > Merge branch 'vmx' (2024-06-28 22:22:53 +0000) > > are available in the Git repository at: > > https://github.com/kvm-x86/linux.git tags/kvm-x86-static_calls-6.11 > > for you to fetch changes up to b528de209c858f61953023b405a4abbf9a9933da: > > KVM: x86/pmu: Add kvm_pmu_call() to simplify static calls of kvm_pmu_ops (2024-06-28 15:23:49 -0700) Thanks, indeed there was no straggler static_call() after applying this. However, there might be a problem: static_call_cond() is equal to static_call() only if CONFIG_HAVE_STATIC_CALL_INLINE, and arch/x86 has this: select HAVE_STATIC_CALL_INLINE if HAVE_OBJTOOL select HAVE_OBJTOOL if X86_64 And indeed if I apply diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig index 80e5afde69f4..d20159d4a37a 100644 --- a/arch/x86/kvm/Kconfig +++ b/arch/x86/kvm/Kconfig @@ -21,6 +21,8 @@ config KVM tristate "Kernel-based Virtual Machine (KVM) support" depends on HIGH_RES_TIMERS depends on X86_LOCAL_APIC + # KVM relies on static_call_cond() being the same as static_call() + depends on HAVE_STATIC_CALL_INLINE select KVM_COMMON select KVM_GENERIC_MMU_NOTIFIER select HAVE_KVM_IRQCHIP KVM disappears from 32-bit kernels. :) So I haven't checked but I suspect this breaks 32-bit? Paolo
On Tue, Jul 16, 2024, Paolo Bonzini wrote: > On 7/13/24 01:56, Sean Christopherson wrote: > > Here's a massage pull request for the static_call() changes, just in case you > > want to go this route instead of applying patches directly after merging > > everything else for 6.11 (it was easy to generate this). If you want to go the > > patches route, I'll post 'em next week. > > > > The following changes since commit c1c8a908a5f4c372f8a8dca0501b56ffc8d260fe: > > > > Merge branch 'vmx' (2024-06-28 22:22:53 +0000) > > > > are available in the Git repository at: > > > > https://github.com/kvm-x86/linux.git tags/kvm-x86-static_calls-6.11 > > > > for you to fetch changes up to b528de209c858f61953023b405a4abbf9a9933da: > > > > KVM: x86/pmu: Add kvm_pmu_call() to simplify static calls of kvm_pmu_ops (2024-06-28 15:23:49 -0700) > > Thanks, indeed there was no straggler static_call() after applying > this. However, there might be a problem: static_call_cond() is equal > to static_call() only if CONFIG_HAVE_STATIC_CALL_INLINE, No, I think you misread the #if-#elif-#else. It's only the !HAVE_STATIC_CALL case that requires use of static_call_cond(). From include/linux/static_call.h: #ifdef CONFIG_HAVE_STATIC_CALL_INLINE #define static_call_cond(name) (void)__static_call(name) #elif defined(CONFIG_HAVE_STATIC_CALL) #define static_call_cond(name) (void)__static_call(name) #else #define static_call_cond(name) (void)__static_call_cond(name) #endif And per Josh, from an old RFC[*] to yank out static_call_cond(): : Static calling a NULL pointer is a NOP, unless you're one of those poor : souls running on an arch (or backported x86 monstrosity) with : CONFIG_HAVE_STATIC_CALL=n, then it's a panic. I double checked that 32-bit KVM works on Intel (which is guaranteed to have a NULL guest_memory_reclaimed()). I also verified that the generated code is identical for both static_call() and static_call_cond(), i.e. the READ_ONCE() of the func at runtime that's present in __static_call_cond() isn't showing up. Dump of assembler code for function kvm_arch_guest_memory_reclaimed: 0xc1042094 <+0>: call 0xc10ce650 <__fentry__> 0xc1042099 <+5>: push %ebp 0xc104209a <+6>: mov %esp,%ebp 0xc104209c <+8>: call 0xc1932d8c <__SCT__kvm_x86_guest_memory_reclaimed> 0xc10420a1 <+13>: pop %ebp 0xc10420a2 <+14>: ret End of assembler dump. Dump of assembler code for function __SCT__kvm_x86_guest_memory_reclaimed: 0xc1932d8c <+0>: ret 0xc1932d8d <+1>: int3 0xc1932d8e <+2>: nop 0xc1932d8f <+3>: nop 0xc1932d90 <+4>: nop 0xc1932d91 <+5>: ud1 %esp,%ecx End of assembler dump. [*] https://lore.kernel.org/all/cover.1678474914.git.jpoimboe@kernel.org
On 7/16/24 17:46, Sean Christopherson wrote: > No, I think you misread the #if-#elif-#else. It's only the !HAVE_STATIC_CALL > case that requires use of static_call_cond(). Oh, of course - the "select HAVE_STATIC_CALL" is right above the "select HAVE_STATIC_CALL_INLINE if HAVE_OBJTOOL" line. I was definitely overthinking it. Paolo