Re: [Resend RFC PATCH V4 09/13] x86/Swiotlb/HV: Add Swiotlb bounce buffer remap function for HV IVM

[Tianyu Lan]

Thanks for review.

On 7/20/2021 9:54 PM, Christoph Hellwig wrote:
> Please split the swiotlb changes into a separate patch from the
> consumer.

OK. Will update.

> >   }
> > +
> > +/*
> > + * hv_map_memory - map memory to extra space in the AMD SEV-SNP Isolation VM.
> > + */
> > +unsigned long hv_map_memory(unsigned long addr, unsigned long size)
> > +{
> > +	unsigned long *pfns = kcalloc(size / HV_HYP_PAGE_SIZE,
> > +				      sizeof(unsigned long),
> > +		       GFP_KERNEL);
> > +	unsigned long vaddr;
> > +	int i;
> > +
> > +	if (!pfns)
> > +		return (unsigned long)NULL;
> > +
> > +	for (i = 0; i < size / HV_HYP_PAGE_SIZE; i++)
> > +		pfns[i] = virt_to_hvpfn((void *)addr + i * HV_HYP_PAGE_SIZE) +
> > +			(ms_hyperv.shared_gpa_boundary >> HV_HYP_PAGE_SHIFT);
> > +
> > +	vaddr = (unsigned long)vmap_pfn(pfns, size / HV_HYP_PAGE_SIZE,
> > +					PAGE_KERNEL_IO);
> > +	kfree(pfns);
> > +
> > +	return vaddr;
>
> This seems to miss a 'select VMAP_PFN'. 

VMAP_PFN has been selected in the previous patch "RFC PATCH V4 08/13]
HV/Vmbus: Initialize VMbus ring buffer for Isolation VM"

> But more importantly I don't
> think this actually works.  Various DMA APIs do expect a struct page
> backing, so how is this going to work with say dma_mmap_attrs or
> dma_get_sgtable_attrs?

dma_mmap_attrs() and dma_get_sgtable_attrs() get input virtual address
belonging to backing memory with struct page and returns bounce buffer
dma physical address which is below shared_gpa_boundary(vTOM) and passed
to Hyper-V via vmbus protocol.

The new map virtual address is only to access bounce buffer in swiotlb
code and will not be used other places. It's stored in the mem->vstart.
So the new API set_memory_decrypted_map() in this series is only called
in the swiotlb code. Other platforms may replace set_memory_decrypted()
with set_memory_decrypted_map() as requested.

> > +static unsigned long __map_memory(unsigned long addr, unsigned long size)
> > +{
> > +	if (hv_is_isolation_supported())
> > +		return hv_map_memory(addr, size);
> > +
> > +	return addr;
> > +}
> > +
> > +static void __unmap_memory(unsigned long addr)
> > +{
> > +	if (hv_is_isolation_supported())
> > +		hv_unmap_memory(addr);
> > +}
> > +
> > +unsigned long set_memory_decrypted_map(unsigned long addr, unsigned long size)
> > +{
> > +	if (__set_memory_enc_dec(addr, size / PAGE_SIZE, false))
> > +		return (unsigned long)NULL;
> > +
> > +	return __map_memory(addr, size);
> > +}
> > +
> > +int set_memory_encrypted_unmap(unsigned long addr, unsigned long size)
> > +{
> > +	__unmap_memory(addr);
> > +	return __set_memory_enc_dec(addr, size / PAGE_SIZE, true);
> > +}
>
> Why this obsfucation into all kinds of strange helpers?  Also I think
> we want an ops vectors (or alternative calls) instead of the random
> if checks here.

Yes, agree and will add ops for different platforms to map/unmap memory.

> > + * @vstart:	The virtual start address of the swiotlb memory pool. The swiotlb
> > + *		memory pool may be remapped in the memory encrypted case and store
>
> Normall we'd call this vaddr or cpu_addr.

OK. Will update.

> > -	set_memory_decrypted((unsigned long)vaddr, bytes >> PAGE_SHIFT);
> > -	memset(vaddr, 0, bytes);
> > +	mem->vstart = (void *)set_memory_decrypted_map((unsigned long)vaddr, bytes);
>
> Please always pass kernel virtual addresses as pointers.
>
> And I think these APIs might need better names, e.g.
>
> arch_dma_map_decrypted and arch_dma_unmap_decrypted.
>
> Also these will need fallback versions for non-x86 architectures that
> currently use memory encryption.

Sure. Will update in the next version.