Re: [PATCH 1/3] include/hw/pci: Attach BDF to Memory Attributes

[Jason Chien <jason.chien@sifive.com>]

[-- Attachment #1: Type: text/plain, Size: 3387 bytes --]

Adding requester_id here does not break anything, since pci_dma_map()
passes MEMTXATTRS_UNSPECIFIED to dma_memory_map() and requester_id is
unused.

I'll add the below for the comment:
Attach BDF here for use during subsequent IOMMU translation.

Michael S. Tsirkin <mst@redhat.com> 於 2025年4月14日 週一 下午11:28寫道：

> On Sun, Mar 02, 2025 at 05:12:07PM +0800, Jason Chien wrote:
> > This commit adds the BDF to the memory attributes for DMA operations.
> >
> > Signed-off-by: Jason Chien <jason.chien@sifive.com>
> > ---
> >  include/hw/pci/pci_device.h | 10 ++++++++--
> >  1 file changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/hw/pci/pci_device.h b/include/hw/pci/pci_device.h
> > index add208edfa..968f1ba3e9 100644
> > --- a/include/hw/pci/pci_device.h
> > +++ b/include/hw/pci/pci_device.h
> > @@ -244,6 +244,8 @@ static inline MemTxResult pci_dma_rw(PCIDevice *dev,
> dma_addr_t addr,
> >                                       void *buf, dma_addr_t len,
> >                                       DMADirection dir, MemTxAttrs attrs)
> >  {
> > +    attrs.unspecified = 0;
> > +    attrs.requester_id = pci_requester_id(dev);
> >      return dma_memory_rw(pci_get_address_space(dev), addr, buf, len,
> >                           dir, attrs);
> >  }
> > @@ -292,6 +294,8 @@ static inline MemTxResult pci_dma_write(PCIDevice
> *dev, dma_addr_t addr,
> >                                                 uint##_bits##_t *val, \
> >                                                 MemTxAttrs attrs) \
> >      { \
> > +        attrs.unspecified = 0; \
> > +        attrs.requester_id = pci_requester_id(dev); \
> >          return ld##_l##_dma(pci_get_address_space(dev), addr, val,
> attrs); \
> >      } \
> >      static inline MemTxResult st##_s##_pci_dma(PCIDevice *dev, \
> > @@ -299,6 +303,8 @@ static inline MemTxResult pci_dma_write(PCIDevice
> *dev, dma_addr_t addr,
> >                                                 uint##_bits##_t val, \
> >                                                 MemTxAttrs attrs) \
> >      { \
> > +        attrs.unspecified = 0; \
> > +        attrs.requester_id = pci_requester_id(dev); \
> >          return st##_s##_dma(pci_get_address_space(dev), addr, val,
> attrs); \
> >      }
> >
> > @@ -327,8 +333,8 @@ PCI_DMA_DEFINE_LDST(q_be, q_be, 64);
> >  static inline void *pci_dma_map(PCIDevice *dev, dma_addr_t addr,
> >                                  dma_addr_t *plen, DMADirection dir)
> >  {
> > -    return dma_memory_map(pci_get_address_space(dev), addr, plen, dir,
> > -                          MEMTXATTRS_UNSPECIFIED);
> > +    MemTxAttrs attrs = {.requester_id = pci_requester_id(dev)};
> > +    return dma_memory_map(pci_get_address_space(dev), addr, plen, dir,
> attrs);
> >  }
>
>
> Map is the only issue  - bdf can technically change between map and
> unmap.
> The use in hw/net/net_tx_pkt.c is fine as it's under BQL.
> I don't know about the use in megasas though.
> I think it is probably fine as it seems to deal with commands
> and I think any driver would flush these if changing BDF.
> Cc megasas maintainers just to make sure though.
>
> Also, adding a code comment here can't hurt.
>
>
> >  static inline void pci_dma_unmap(PCIDevice *dev, void *buffer,
> dma_addr_t len,
> > --
> > 2.43.2
>
>

[-- Attachment #2: Type: text/html, Size: 4298 bytes --]