Re: [PATCH] target/i386: Re-introduce few KVM stubs for Clang debug builds

[Kevin Wolf]

Am 11.09.2023 um 13:15 hat Stefan Hajnoczi geschrieben:
> On Mon, 11 Sept 2023 at 06:39, Philippe Mathieu-Daudé <philmd@linaro.org> 
> wrote:
> >
> > Since commits 3adce820cf..ef1cf6890f, When building on
> > a x86 host configured as:
> >
> >   $ ./configure --cc=clang \
> >     --target-list=x86_64-linux-user,x86_64-softmmu \
> >     --enable-debug
> >
> > we get:
> >
> >   [71/71] Linking target qemu-x86_64
> >   FAILED: qemu-x86_64
> >   /usr/bin/ld: libqemu-x86_64-linux-user.fa.p/target_i386_cpu.c.o: in 
> > function `cpu_x86_cpuid':
> >   cpu.c:(.text+0x1374): undefined reference to 
> > `kvm_arch_get_supported_cpuid'
> >   /usr/bin/ld: libqemu-x86_64-linux-user.fa.p/target_i386_cpu.c.o: in 
> > function `x86_cpu_filter_features':
> >   cpu.c:(.text+0x81c2): undefined reference to 
> > `kvm_arch_get_supported_cpuid'
> >   /usr/bin/ld: cpu.c:(.text+0x81da): undefined reference to 
> > `kvm_arch_get_supported_cpuid'
> >   /usr/bin/ld: cpu.c:(.text+0x81f2): undefined reference to 
> > `kvm_arch_get_supported_cpuid'
> >   /usr/bin/ld: cpu.c:(.text+0x820a): undefined reference to 
> > `kvm_arch_get_supported_cpuid'
> >   /usr/bin/ld: 
> > libqemu-x86_64-linux-user.fa.p/target_i386_cpu.c.o:cpu.c:(.text+0x8225): 
> > more undefined references to `kvm_arch_get_supported_cpuid' follow
> >   clang: error: linker command failed with exit code 1 (use -v to see 
> > invocation)
> >   ninja: build stopped: subcommand failed.
> >
> > '--enable-debug' disables optimizations (CFLAGS=-O0).
> >
> > While at this (un)optimization level GCC eliminate the
> > following dead code:
> >
> >   if (0 && foo()) {
> >       ...
> >   }
> >
> > Clang does not. Therefore restore a pair of stubs for
> > unoptimized Clang builds.
> >
> > Reported-by: Kevin Wolf <kwolf@redhat.com>
> > Fixes: 3adce820cf ("target/i386: Remove unused KVM stubs")
> > Fixes: ef1cf6890f ("target/i386: Allow elision of 
> > kvm_hv_vpindex_settable()")
> > Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> > ---
> >  target/i386/kvm/kvm_i386.h | 21 ++++++++++++++++++---
> >  1 file changed, 18 insertions(+), 3 deletions(-)
> >
> > diff --git a/target/i386/kvm/kvm_i386.h b/target/i386/kvm/kvm_i386.h
> > index 55d4e68c34..0b62ac628f 100644
> > --- a/target/i386/kvm/kvm_i386.h
> > +++ b/target/i386/kvm/kvm_i386.h
> > @@ -32,7 +32,6 @@
> >
> >  bool kvm_has_smm(void);
> >  bool kvm_enable_x2apic(void);
> > -bool kvm_hv_vpindex_settable(void);
> >  bool kvm_has_pit_state2(void);
> >
> >  bool kvm_enable_sgx_provisioning(KVMState *s);
> > @@ -41,8 +40,6 @@ bool kvm_hyperv_expand_features(X86CPU *cpu, Error 
> > **errp);
> >  void kvm_arch_reset_vcpu(X86CPU *cs);
> >  void kvm_arch_after_reset_vcpu(X86CPU *cpu);
> >  void kvm_arch_do_init_vcpu(X86CPU *cs);
> > -uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function,
> > -                                      uint32_t index, int reg);
> >  uint64_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t index);
> >
> >  void kvm_set_max_apic_id(uint32_t max_apic_id);
> > @@ -60,6 +57,10 @@ void kvm_put_apicbase(X86CPU *cpu, uint64_t value);
> >
> >  bool kvm_has_x2apic_api(void);
> >  bool kvm_has_waitpkg(void);
> > +bool kvm_hv_vpindex_settable(void);
> > +
> > +uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t function,
> > +                                      uint32_t index, int reg);
> >
> >  uint64_t kvm_swizzle_msi_ext_dest_id(uint64_t address);
> >  void kvm_update_msi_routes_all(void *private, bool global,
> > @@ -76,6 +77,20 @@ typedef struct kvm_msr_handlers {
> >  bool kvm_filter_msr(KVMState *s, uint32_t msr, QEMURDMSRHandler *rdmsr,
> >                      QEMUWRMSRHandler *wrmsr);
> >
> > +#elif defined(__clang__) && !defined(__OPTIMIZE__)
> 
> Another approach is a static library with a .o file containing the
> stubs so the linker only includes it in the executable if the compiler
> emitted the symbols. That way there is no need for defined(__clang__)
> && !defined(__OPTIMIZE__) and it will work with other
> compilers/optimization levels. It's more work to set up though.

Isn't that exactly how it was before the stubs were removed? It would be
a simple revert of that commit.

The approach with static inline functions defined only for a very
specific configuration looks a lot more fragile to me. In fact, I'm
surprised that it works because I think it requires that the header
isn't used in any files that are shared between user space and system
emulation - and naively cpu.c sounded like something that could be
shared. Looks like this patch only works because the linux-user target
uses a separate build of the same CPU emulation source file.

So I think reverting the commit that removed the stubs would be much
more obvious.

Kevin