[PATCH] RISC-V: KVM: Fix compilation without RISCV_ISA_ZICBOM

[Andrew Jones]

On Mon, Oct 10, 2022 at 11:36:24AM -0700, Palmer Dabbelt wrote:
> On Mon, 10 Oct 2022 02:40:29 PDT (-0700), ajones at ventanamicro.com wrote:
> > Fix undefined reference of riscv_cbom_block_size when compiling KVM
> > without RISCV_ISA_ZICBOM. Note, RISCV_ISA_ZICBOM is a sufficient
> > guard as it selects RISCV_DMA_NONCOHERENT, which is needed to compile
> > dma-noncoherent.c (which is the file where riscv_cbom_block_size and
> > its initializer live).
> >
> > Fixes: afd5dde9a186 ("RISC-V: KVM: Provide UAPI for Zicbom block size")
> > Reported-by: kernel test robot <lkp at intel.com>
> > Signed-off-by: Andrew Jones <ajones at ventanamicro.com>
> > ---
> >  arch/riscv/kvm/vcpu.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/arch/riscv/kvm/vcpu.c b/arch/riscv/kvm/vcpu.c
> > index a032c4f0d600..e4453caba728 100644
> > --- a/arch/riscv/kvm/vcpu.c
> > +++ b/arch/riscv/kvm/vcpu.c
> > @@ -265,11 +265,13 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
> >  	case KVM_REG_RISCV_CONFIG_REG(isa):
> >  		reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
> >  		break;
> > +#ifdef CONFIG_RISCV_ISA_ZICBOM
> >  	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
> >  		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOM))
> >  			return -EINVAL;
> >  		reg_val = riscv_cbom_block_size;
> >  		break;
> > +#endif
> >  	default:
> >  		return -EINVAL;
> >  	}
> 
> Thanks, looks like this is tripping up the kernelci builds too: 
> https://linux.kernelci.org/build/id/6343b37c6fd2dcbb5dcab5f3/logs/ .  
> This symbol is actually a bit odd, as we don't actually just use it 
> under Zicbom because of the T-Head stuff.  Looks like that's generically 
> broken, though, so IMO we need something like this
> 
> diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
> index 6cb7d96ad9c7..fba86fa3a628 100644
> --- a/arch/riscv/mm/cacheflush.c
> +++ b/arch/riscv/mm/cacheflush.c
> @@ -5,6 +5,9 @@
> 
>  #include <asm/cacheflush.h>
> 
> +unsigned int riscv_cbom_block_size;
> +EXPORT_SYMBOL(riscv_cbom_block_size);
> +
>  #ifdef CONFIG_SMP
> 
>  #include <asm/sbi.h>
> diff --git a/arch/riscv/mm/dma-noncoherent.c b/arch/riscv/mm/dma-noncoherent.c
> index e3f9bdf47c5f..33d21ace0971 100644
> --- a/arch/riscv/mm/dma-noncoherent.c
> +++ b/arch/riscv/mm/dma-noncoherent.c
> @@ -12,7 +12,6 @@
>  #include <linux/of_device.h>
>  #include <asm/cacheflush.h>
> 
> -unsigned int riscv_cbom_block_size;
>  static bool noncoherent_supported;
> 
>  void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,
> 
> which should be
> 
> Fixes: 8f7e001e0325 ("RISC-V: Clean up the Zicbom block size probing").

Hi Palmer,

Should I post your suggestion as a v2 or were you planning to post it?
I'm happy to do so, but I'd prefer to also move
riscv_init_cbom_blocksize() and drop its #ifdef CONFIG_RISCV_ISA_ZICBOM
guard. The reason I'd prefer we always try to parse the block size is
that we always parse the ISA string, and, when zicbom is present, we set
it in the ISA extension bitmap, even when CONFIG_RISCV_ISA_ZICBOM isn't
enabled. This is a reasonable behavior if CONFIG_RISCV_ISA_ZICBOM is meant
to decide whether or not the kernel uses the feature for itself, as its
Kconfig text implies, rather than to mean we should pretend the feature
isn't present, whether it is or not. So, if we're not trying to completely
avoid the feature by disabling CONFIG_RISCV_ISA_ZICBOM, then KVM can still
inform the guest about it. When KVM does that it also needs the block
size.

Thanks,
drew