On Sat, Sep 16, 2023 at 01:41:59PM +0300, Andy Shevchenko wrote: > On Thu, Sep 14, 2023 at 03:31:46PM -0500, Bjorn Helgaas wrote: > > On Thu, Sep 14, 2023 at 10:23:24PM +0300, Andy Shevchenko wrote: > > ... > > > > - addr = virt_to_phys(port->base + PCIE_MSI_VECTOR); > > > + addr = virt_to_phys((__force void *)port->base + PCIE_MSI_VECTOR); > > > > Lots of these drivers use either virt_to_phys() or > > platform_get_resource_byname() to get a physical address that they > > then use as the MSI target. > > > > But I don't think that's quite right -- the MSI is a DMA transaction > > on PCI, and in general there's no guarantee that bus addresses are > > identical to CPU physical addresses, so shouldn't we use a dma_addr_t > > obtained from the DMA API? > > > > dw_pcie_msi_host_init() has a complicated version of this that uses > > dmam_alloc_coherent(). > > Fair enough. I leave it to you then to get it fixed properly. > Consider this as just a heads up for the old report. Sounds good, thanks! Bjorn
