On Wed, Sep 19, 2018 at 10:58 AM Nathan Chancellor <natechancellor@xxxxxxxxx> wrote: > > On Wed, Sep 19, 2018 at 10:35:55AM -0700, Nick Desaulniers wrote: > > On Wed, Sep 19, 2018 at 3:29 AM Lorenzo Pieralisi > > <lorenzo.pieralisi@xxxxxxx> wrote: > > > > > > Xiaowei, Binghui, > > > > > > On Tue, Sep 18, 2018 at 10:38:29PM -0700, Nathan Chancellor wrote: > > > > WARNING: vmlinux.o(.text+0x4758cc): Section mismatch in reference from > > > > the function kirin_pcie_probe() to the function > > > > .init.text:kirin_add_pcie_port() > > > > The function kirin_pcie_probe() references > > > > the function __init kirin_add_pcie_port(). > > > > This is often because kirin_pcie_probe lacks a __init > > > > annotation or the annotation of kirin_add_pcie_port is wrong. > > > > > > > > Remove '__init' from kirin_add_pcie_port so there is no mismatch. > > > > > > I think that instead of removing __init section tag we should add > > > it to kirin_pcie_probe(). > > > > A bunch of these functions have a single call site, with a mix of __init: > > > > kirin_pcie_probe() -> > > kirin_add_pcie_port() -> // __init > > kirin_pcie_add_msi() > > > > But it looks like probe functions should be marked init: > > $ cd drivers; grep -r __init | grep probe > > > > seems to turn up quite a lot of *probe() functions that are marked __init. > > https://www.tldp.org/LDP/lkmpg/2.4/html/x281.htm has more info, and > > from reading that, it seems that it's a no-op for a module that's > > marked loadable. > > > > So I agree; __init should be added to kirin_pcie_probe(), but also > > kirin_pcie_add_msi(). That way if the module is statically compiled > > into the kernel image, these functions with a single call site get > > cleaned up after init. > > > > Nathan, would you mind sending a v2? > > > > Hi Nick and Lorenzo, > > I had tested adding __init to both kirin_pcie_probe and > kirin_pcie_add_msi but ran into this warning: > > WARNING: vmlinux.o(.data+0x68b78): Section mismatch in reference from > the variable kirin_pcie_driver to the function I think we want builtin_platform_driver_probe() here. Try removing the probe reference from the declaration of kirin_pcie_driver, and then calling builtin_platform_driver_probe(kirin_pcie_driver, kirin_pcie_probe); Does that fix the section mismatches, while keeping as much in __init as possible? > .init.text:kirin_pcie_probe() > The variable kirin_pcie_driver references > the function __init kirin_pcie_probe() > > Doing a quick 'grep -r __init | grep platform_driver' doesn't show any > 'static struct platform_driver' constructs marked as __init, leading me > to believe this patch is the proper solution. > > I'm happy to hear otherwise, thank you both for the quick responses, > Nathan > > > > > > > Please let me know and ACK accordingly. > > > > > > Lorenzo > > > > > > > Fixes: fc5165db245a ("PCI: kirin: Add HiSilicon Kirin SoC PCIe controller driver") > > > > Reported-by: Nick Desaulniers <ndesaulniers@xxxxxxxxxx> > > > > Signed-off-by: Nathan Chancellor <natechancellor@xxxxxxxxx> > > > > --- > > > > drivers/pci/controller/dwc/pcie-kirin.c | 4 ++-- > > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/drivers/pci/controller/dwc/pcie-kirin.c b/drivers/pci/controller/dwc/pcie-kirin.c > > > > index 5352e0c3be82..9b599296205d 100644 > > > > --- a/drivers/pci/controller/dwc/pcie-kirin.c > > > > +++ b/drivers/pci/controller/dwc/pcie-kirin.c > > > > @@ -467,8 +467,8 @@ static int kirin_pcie_add_msi(struct dw_pcie *pci, > > > > return 0; > > > > } > > > > > > > > -static int __init kirin_add_pcie_port(struct dw_pcie *pci, > > > > - struct platform_device *pdev) > > > > +static int kirin_add_pcie_port(struct dw_pcie *pci, > > > > + struct platform_device *pdev) > > > > { > > > > int ret; > > > > > > > > -- > > > > 2.19.0 > > > > > > > > > > > > -- > > Thanks, > > ~Nick Desaulniers -- Thanks, ~Nick Desaulniers
