On Fri, Sep 22, 2017 at 08:35:15AM -0500, Bjorn Helgaas wrote: > [+cc Thomas (aardvark maintainer), Lorenzo] > > On Fri, Sep 22, 2017 at 01:24:44PM +0200, Marek Behún wrote: > > Hello, > > we are developing a new board based on Marvell Armada 3720. > > While testing the PCI-e slot, I have enountered this problem: > > > > - lspci always correctly prints information about the PCIe card in the > > slot > > - when a ath9k card is in the slot, ath9k fails to initialize with > > ath9k 0000:00:00.0: request_irq failed > > > > The irq number passed to the request_irq function is from the irq > > member of struct pci_dev: > > request_irq(pdev->irq, ath_isr, IRQF_SHARED, "ath9k", sc); > > > > this number (pdev->irq) is 255 > > > > The request_irq function fails in request_threaded_irq on line: > > desc = irq_to_desc(irq); > > (desc is NULL and thus -EINVAL is returned.) > > > > - what is strange is that another card, Intel Centrino Ultimate-N 6300 > > with iwlwifi works. > > > > - another card (ath10k driver) also does not work. > > > > Is this a problem with the pci-aardvark driver for the Aardvark > > controller? > > What kernel version are you using? Has any kernel version worked > correctly? There have been significant IRQ mapping changes in the > last couple releases. > > dev->irq should be assigned in pci_assign_irq(). Can you turn on the > debug there (and maybe add more) and collect the complete dmesg log? It is very likely this will fix it: https://patchwork.ozlabs.org/patch/811432/ Apologies to you and Thomas again - I went through a number of changes and mechanically missed pci-aardvark. Thanks, Lorenzo