Hi Heiko, On 07/01/19 8:05 PM, Heiko Stuebner wrote: > Am Montag, 7. Januar 2019, 07:41:33 CET schrieb Kishon Vijay Abraham I: >> Hi Lorenzo, >> >> The Endpoint controller driver uses features member in 'struct pci_epc' >> to advertise the list of supported features to the endpoint function >> driver. >> >> There are a few shortcomings with this approach. >> *) Certain endpoint controllers support fixed size BAR (e.g. TI's >> AM654 uses Designware configuration with fixed size BAR). The >> size of each BARs cannot be passed to the endpoint function >> driver. >> *) Too many macros for handling EPC features. >> (EPC_FEATURE_NO_LINKUP_NOTIFIER, EPC_FEATURE_BAR_MASK, >> EPC_FEATURE_MSIX_AVAILABLE, EPC_FEATURE_SET_BAR, >> EPC_FEATURE_GET_BAR) >> *) Endpoint controllers are directly modifying struct pci_epc >> members. (I have plans to move struct pci_epc to >> drivers/pci/endpoint so that pci_epc members are referenced >> only by endpoint core). >> >> To overcome the above shortcomings, introduced pci_epc_get_features() >> API, pci_epc_features structure and a ->get_features() callback. >> >> Also added a patch to set BAR flags in pci_epf_alloc_space and >> remove it from pci-epf-test function driver. >> >> Tested on TI's DRA7xx platform. > > While I don't have that much PCI experience and hence cannot judge > this cleanup as a whole, I can at least say, that my Rockchip rk3399 > still does find its PCIE-connected wifi card, so this series on rk3399 > > Tested-by: Heiko Stuebner <heiko@xxxxxxxxx> Thank you for testing this series with PCIe controller configured in RC mode. It would be great if it could be tested with EP mode too. Thanks for the help. Cheers Kishon
