On Sat, Jun 15, 2019 at 02:48:24PM +0800, Yuehaibing wrote: > +cc Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx> Can we drop this patch and merge: https://patchwork.ozlabs.org/patch/1131444/ instead ? Thanks, Lorenzo > > On 2019/6/15 14:18, Yuehaibing wrote: > > > > On 2019/6/2 6:59, Michael Kelley wrote: > >> From: YueHaibing <yuehaibing@xxxxxxxxxx> Sent: Friday, May 31, 2019 8:09 AM > >>> > >>> while building without CONFIG_SYSFS, fails as below: > >>> > >>> drivers/pci/controller/pci-hyperv.o: In function 'hv_pci_assign_slots': > >>> pci-hyperv.c:(.text+0x40a): undefined reference to 'pci_create_slot' > >>> drivers/pci/controller/pci-hyperv.o: In function 'pci_devices_present_work': > >>> pci-hyperv.c:(.text+0xc02): undefined reference to 'pci_destroy_slot' > >>> drivers/pci/controller/pci-hyperv.o: In function 'hv_pci_remove': > >>> pci-hyperv.c:(.text+0xe50): undefined reference to 'pci_destroy_slot' > >>> drivers/pci/controller/pci-hyperv.o: In function 'hv_eject_device_work': > >>> pci-hyperv.c:(.text+0x11f9): undefined reference to 'pci_destroy_slot' > >>> > >>> Select SYSFS while PCI_HYPERV is set to fix this. > >>> > >> > >> I'm wondering if is the right way to fix the problem. Conceptually > >> is it possible to setup & operate virtual PCI devices like > >> pci-hyperv.c does, even if sysfs is not present? Or is it right to > >> always required sysfs? > >> > >> The function pci_dev_assign_slot() in slot.c has a null implementation > >> in include/linux/pci.h when CONFIG_SYSFS is not defined, which > >> seems to be trying to solve the same problem for that function. And > >> if CONFIG_HOTPLUG_PCI is defined but CONFIG_SYSFS is not, > >> pci_hp_create_module_link() and pci_hp_remove_module_link() > >> look like they would have the same problem. Maybe there should > >> be degenerate implementations of pci_create_slot() and > >> pci_destroy_slot() for cases when CONFIG_SYSFS is not defined? > >> > >> But I'll admit I don't know the full story behind how PCI slots > >> are represented and used, so maybe I'm off base. I just noticed > >> the inconsistency in how other functions in slot.c are handled. > >> > >> Thoughts? > > > > 268a03a42d33 ("PCI: drivers/pci/slot.c should depend on CONFIG_SYSFS") > > > > make slot.o depends CONFIG_SYSFS > > > > commit 268a03a42d3377d5fb41e6e7cbdec4e0b65cab2e > > Author: Alex Chiang <achiang@xxxxxx> > > Date: Wed Jun 17 19:03:57 2009 -0600 > > > > PCI: drivers/pci/slot.c should depend on CONFIG_SYSFS > > > > There is no way to interact with a physical PCI slot without > > sysfs, so encode the dependency and prevent this build error: > > > > drivers/pci/slot.c: In function 'pci_hp_create_module_link': > > drivers/pci/slot.c:327: error: 'module_kset' undeclared > > > > This patch _should_ make pci-sysfs.o depend on CONFIG_SYSFS too, > > but we cannot (yet) because the PCI core merrily assumes the > > existence of sysfs: > > > > drivers/built-in.o: In function `pci_bus_add_device': > > drivers/pci/bus.c:89: undefined reference to `pci_create_sysfs_dev_files' > > drivers/built-in.o: In function `pci_stop_dev': > > drivers/pci/remove.c:24: undefined reference to `pci_remove_sysfs_dev_files' > > > > So do the minimal bit for now and figure out how to untangle it > > later. > > > > If No CONFIG_SYSFS, slot.o is not build > > > >> > >> Michael > >> > >> > > > > > > . > > >
