On Fri, 18 Oct 2024 12:24:05 +0200 Marcin Juszkiewicz <marcin.juszkiew...@linaro.org> wrote:
> W dniu 17.10.2024 o 18:58, Michael S. Tsirkin pisze: > >> diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c > >> index 4b2f0805c6..54c0f1ec67 100644 > >> --- a/hw/pci/pcie.c > >> +++ b/hw/pci/pcie.c > >> @@ -86,7 +86,8 @@ pcie_cap_v1_fill(PCIDevice *dev, uint8_t port, uint8_t > >> type, uint8_t version) > >> * Specification, Revision 1.1., or subsequent PCI Express Base > >> * Specification revisions. > >> */ > >> - pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER); > >> + pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER | > >> + PCI_EXP_DEVCAP_EXT_TAG); > >> > >> pci_set_long(exp_cap + PCI_EXP_LNKCAP, > >> (port << PCI_EXP_LNKCAP_PN_SHIFT) | > > > > We can't change capabilities unconditionally. > > It needs at least a machine type compat thing. > > Started looking and wonder how to pass it from MachineClass level down > to pcie.c/pcie_cap_v1_fill() level. see as an example: fa905f65c5549 hw/nvme: add machine compatibility parameter to enable msix exclusive bar > hw/arm/sbsa-ref.c (the machine I know best) has create_pcie() which > allocates PCI_HOST_BRIDGE and then creates 2 pcie devices (default_nic > (e1000e) and bochs-display gfx). > > If I add "pcie_uses_ext_tags = true" to SBSAMachineState then I need to > have it stored in PCIBus structure so pci_create_simple() will know. > This function would copy it into "dev" (PCIDevice) to make it pass to > pcie_cap_v1_fill() function. > > But that's not right way because other PCIe devices are created in other > places. > > Need to dig deeper. >
