Re: [PATCH v4 2/2] PCI: dwc: Add support for 64-bit MSI target address

From: William McVicker
Date: Thu Aug 25 2022 - 14:39:17 EST


On 08/23/2022, Robin Murphy wrote:
> On 2022-08-12 01:03, Will McVicker wrote:
> > Since not all devices require a 32-bit MSI address, add support to the
> > PCIe host driver to allow setting the DMA mask to 64-bits. This allows
> > kernels to disable ZONE_DMA32 and bounce buffering (swiotlb) without
> > risking not being able to get a 32-bit address during DMA allocation.
> > Basically, in the slim chance that there are no 32-bit allocations
> > available, the current PCIe host driver will fail to allocate the
> > msi_msg page due to a DMA address overflow (seen in [1]). With this
> > patch, the PCIe device can advertise 64-bit support via its MSI
> > capabilities to hint to the PCIe host driver to set the DMA mask to
> > 64-bits.
>
> It doesn't matter so much what the host's own capabilities are, though, what
> matters is that if you configure the MSI doorbell to decode some 64-bit
> address because you can, and later an endpoint shows up whose DMA mask
> doesn't reach that address, you've broken MSIs for that endpoint. While the
> host is probing, it cannot possibly know what future endpoint drivers may or
> may not do.
>
> Now, in the case where no ZONE_DMA{32} is configured, there's a fair
> likelihood that addressing-constrained endpoints are probably going to be
> broken in general, so that should not be fatal for the host. So *if* we fail
> to allocate a 32-bit MSI address, *then* we should fall back to a 64-bit one
> (when the host supports it), but to preserve compatibility we still must
> always attempt the 32-bit allocation first (oh, if there isn't a theme
> emerging here...)
>
> TBH this has come up enough times, and has enough subtle complexity, that I
> think it's high time to factor this all out into a helper between the PCI
> core code and the DMA API for "please give me a bus address that's
> guaranteed not to conflict with any valid DMA address".
>
> Thanks,
> Robin.

Hi Robin,

Thanks for the reviews! I'm happy with your suggestion of trying a 32-bit mask
first for the reasons you mentioned. I have tested this out on my Pixel 6 and
db845c and will push the updated patchset as v5.

Regards,
Will

>
> > [1] https://lore.kernel.org/all/Yo0soniFborDl7+C@xxxxxxxxxx/
> >
> > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > Signed-off-by: Will McVicker <willmcvicker@xxxxxxxxxx>
> > Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
> > ---
> > drivers/pci/controller/dwc/pcie-designware-host.c | 14 ++++++++++++--
> > drivers/pci/controller/dwc/pcie-designware.c | 8 ++++++++
> > drivers/pci/controller/dwc/pcie-designware.h | 1 +
> > 3 files changed, 21 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/pci/controller/dwc/pcie-designware-host.c b/drivers/pci/controller/dwc/pcie-designware-host.c
> > index 39f3b37d4033..d7831040791b 100644
> > --- a/drivers/pci/controller/dwc/pcie-designware-host.c
> > +++ b/drivers/pci/controller/dwc/pcie-designware-host.c
> > @@ -330,6 +330,8 @@ static int dw_pcie_msi_host_init(struct dw_pcie_rp *pp)
> > u64 *msi_vaddr;
> > int ret;
> > u32 ctrl, num_ctrls;
> > + bool msi_64bit = false;
> > + u16 msi_capabilities;
> > for (ctrl = 0; ctrl < MAX_MSI_CTRLS; ctrl++)
> > pp->irq_mask[ctrl] = ~0;
> > @@ -367,9 +369,17 @@ static int dw_pcie_msi_host_init(struct dw_pcie_rp *pp)
> > dw_chained_msi_isr, pp);
> > }
> > - ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
> > + msi_capabilities = dw_pcie_msi_capabilities(pci);
> > + if (msi_capabilities & PCI_MSI_FLAGS_ENABLE)
> > + msi_64bit = msi_capabilities & PCI_MSI_FLAGS_64BIT;
> > +
> > + dev_dbg(dev, "Setting MSI DMA mask to %s-bit.\n",
> > + msi_64bit ? "64" : "32");
> > + ret = dma_set_mask_and_coherent(dev, msi_64bit ?
> > + DMA_BIT_MASK(64) : DMA_BIT_MASK(32));
> > if (ret)
> > - dev_warn(dev, "Failed to set DMA mask to 32-bit. Devices with only 32-bit MSI support may not work properly\n");
> > + dev_warn(dev, "Failed to set DMA mask to %s-bit.\n",
> > + msi_64bit ? "64" : "32");
> > msi_vaddr = dmam_alloc_coherent(dev, sizeof(u64), &pp->msi_data,
> > GFP_KERNEL);
> > diff --git a/drivers/pci/controller/dwc/pcie-designware.c b/drivers/pci/controller/dwc/pcie-designware.c
> > index c6725c519a47..650a7f22f9d0 100644
> > --- a/drivers/pci/controller/dwc/pcie-designware.c
> > +++ b/drivers/pci/controller/dwc/pcie-designware.c
> > @@ -82,6 +82,14 @@ u8 dw_pcie_find_capability(struct dw_pcie *pci, u8 cap)
> > }
> > EXPORT_SYMBOL_GPL(dw_pcie_find_capability);
> > +u16 dw_pcie_msi_capabilities(struct dw_pcie *pci)
> > +{
> > + u8 offset;
> > +
> > + offset = dw_pcie_find_capability(pci, PCI_CAP_ID_MSI);
> > + return dw_pcie_readw_dbi(pci, offset + PCI_MSI_FLAGS);
> > +}
> > +
> > static u16 dw_pcie_find_next_ext_capability(struct dw_pcie *pci, u16 start,
> > u8 cap)
> > {
> > diff --git a/drivers/pci/controller/dwc/pcie-designware.h b/drivers/pci/controller/dwc/pcie-designware.h
> > index a871ae7eb59e..45fcdfc8c035 100644
> > --- a/drivers/pci/controller/dwc/pcie-designware.h
> > +++ b/drivers/pci/controller/dwc/pcie-designware.h
> > @@ -332,6 +332,7 @@ void dw_pcie_version_detect(struct dw_pcie *pci);
> > u8 dw_pcie_find_capability(struct dw_pcie *pci, u8 cap);
> > u16 dw_pcie_find_ext_capability(struct dw_pcie *pci, u8 cap);
> > +u16 dw_pcie_msi_capabilities(struct dw_pcie *pci);
> > int dw_pcie_read(void __iomem *addr, int size, u32 *val);
> > int dw_pcie_write(void __iomem *addr, int size, u32 val);