Re: [PATCH 23/32] KVM: s390: pci: handle refresh of PCI translations

[Matthew Rosato]

On 12/8/21 5:30 AM, Niklas Schnelle wrote:
> On Tue, 2021-12-07 at 15:57 -0500, Matthew Rosato wrote:
> > Add a routine that will perform a shadow operation between a guest
> > and host IOAT.  A subsequent patch will invoke this in response to
> > an 04 RPCIT instruction intercept.
> >
> > Signed-off-by: Matthew Rosato <mjrosato@xxxxxxxxxxxxx>
> > ---
> >   arch/s390/include/asm/kvm_pci.h |   1 +
> >   arch/s390/include/asm/pci_dma.h |   1 +
> >   arch/s390/kvm/pci.c             | 191 ++++++++++++++++++++++++++++++++
> >   arch/s390/kvm/pci.h             |   4 +-
> >   4 files changed, 196 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/s390/include/asm/kvm_pci.h b/arch/s390/include/asm/kvm_pci.h
> > index 254275399f21..97e3a369135d 100644
> > --- a/arch/s390/include/asm/kvm_pci.h
> > +++ b/arch/s390/include/asm/kvm_pci.h
> > @@ -30,6 +30,7 @@ struct kvm_zdev_ioat {
> >   struct kvm_zdev {
> >   	struct zpci_dev *zdev;
> >   	struct kvm *kvm;
> > +	u64 rpcit_count;
> >   	struct kvm_zdev_ioat ioat;
> >   	struct zpci_fib fib;
> >   };
> > diff --git a/arch/s390/include/asm/pci_dma.h b/arch/s390/include/asm/pci_dma.h
> > index e1d3c1d3fc8a..0ca15e5db3d9 100644
> > --- a/arch/s390/include/asm/pci_dma.h
> > +++ b/arch/s390/include/asm/pci_dma.h
> > @@ -52,6 +52,7 @@ enum zpci_ioat_dtype {
> >   #define ZPCI_TABLE_ENTRIES		(ZPCI_TABLE_SIZE / ZPCI_TABLE_ENTRY_SIZE)
> >   #define ZPCI_TABLE_PAGES		(ZPCI_TABLE_SIZE >> PAGE_SHIFT)
> >   #define ZPCI_TABLE_ENTRIES_PAGES	(ZPCI_TABLE_ENTRIES * ZPCI_TABLE_PAGES)
> > +#define ZPCI_TABLE_ENTRIES_PER_PAGE	(ZPCI_TABLE_ENTRIES / ZPCI_TABLE_PAGES)
> >   
> >   #define ZPCI_TABLE_BITS			11
> >   #define ZPCI_PT_BITS			8
> > diff --git a/arch/s390/kvm/pci.c b/arch/s390/kvm/pci.c
> > index a1c0c0881332..858c5ecdc8b9 100644
> > --- a/arch/s390/kvm/pci.c
> > +++ b/arch/s390/kvm/pci.c
> > @@ -123,6 +123,195 @@ int kvm_s390_pci_aen_init(u8 nisc)
> >   	return rc;
> >   }
> >   
> > +static int dma_shadow_cpu_trans(struct kvm_vcpu *vcpu, unsigned long *entry,
> > +				unsigned long *gentry)
> > +{
> > +	unsigned long idx;
> > +	struct page *page;
> > +	void *gaddr = NULL;
> > +	kvm_pfn_t pfn;
> > +	gpa_t addr;
> > +	int rc = 0;
> > +
> > +	if (pt_entry_isvalid(*gentry)) {
> > +		/* pin and validate */
> > +		addr = *gentry & ZPCI_PTE_ADDR_MASK;
> > +		idx = srcu_read_lock(&vcpu->kvm->srcu);
> > +		page = gfn_to_page(vcpu->kvm, gpa_to_gfn(addr));
> > +		srcu_read_unlock(&vcpu->kvm->srcu, idx);
> > +		if (is_error_page(page))
> > +			return -EIO;
> > +		gaddr = page_to_virt(page) + (addr & ~PAGE_MASK);
>
> Hmm, this looks like a virtual vs physical address mixup to me that is
> currently not a problem because kernel virtual addresses are equal to
> their physical address. Here page_to_virt(page) gives us a virtual
> address but the entries in the I/O translation table have to be
> physical (aka absolute) addresses.
>
> With my commit "s390/pci: use physical addresses in DMA tables"
> currently in the s390 feature branch this is also reflected in the
> argument types taken by set_pt_pfaa() below so gaddr should have type
> phys_addr_t not void *. That should also remove the need for the cast
> to unsigned long for the duplicate check.

Right...  Like the other comment re: virtual vs physical address I will 
take a look and fix for v2.