On Wed, Jul 3, 2013 at 10:53 PM, Haicheng Li <haicheng.li@xxxxxxxxxxxxxxx> wrote: > On Wed, Jul 03, 2013 at 10:09:32AM -0600, Bjorn Helgaas wrote: >> On Wed, Jul 3, 2013 at 9:16 AM, Haicheng Li <haicheng.li@xxxxxxxxxxxxxxx> wrote: >> > With more and more SOCs having pci device integrated into chip (e.g. Intel >> > Atom series), it's useful to add an interface to cleanly hide pci devices from >> > pci device scanning, which is because: >> > >> > 1. phone or tablet OEMs may choose disabling some pci device in the SOC, >> > such as camera ISP in Intel Atom Z2580 chip, and etc. >> > 2. if such disabled devices are not cleanly removed from pci device tree, >> > then pci-core will still try to operate on the relative device control >> > registers while S3 suspend and resume. >> > 3. so hiding such devices from early begining will not only reduce the kernel >> > boot time, but also optimize the latency of system suspend and resume. >> >> Normally the chip provides a way to disable devices by writing a >> configuration register. Then the device doesn't respond when Linux >> enumerates devices, so nothing special is required in the kernel. > Agreed, this is true. > >> What's different about the Z2580? I'd be surprised if Intel forgot to >> include such a register. > A pci shim faked by firmware was introduced to help easily port Linux onto Z2XXX > SOC chips, which enumerates both real and fake PCI devices inside the SOC (The camera ISP > I mentioned above is a real PCI device in this case) > > A detailed tech talk about this technology by Jacob Pan in elc2010 can be found online > here: http://elinux.org/images/e/ee/Jacob-Pan-x86MID-elc2010.pdf > >> Maybe the firmware just isn't smart enough >> to disable the device? If so, it would be better to fix the firmware >> than to add kludges in the kernel. > On PC or server, end-user/OEM can disable/hide a pci device easily thru BIOS setting or > by hacking BIOS code directly when they find some device is broken or useless. > > However on phone or tablet equipment, there is no BIOS-setting alike UI exposed to end-user/developer > to disable broken device easily (and physically removing the device is not doable on Phone or tablet) > > OTOH, this i/f is really *helpful* for kernel developer to power-on a new platform, debug system > problem, or to do performance tuning of suspend/resume. At least it makes my daily job easier:). > > So if people are strongly against this, I would still suggest accept this i/f as a debug i/f > at least:). > >> > To hide pci devices, just pass such parameters to kernel at boot stage: >> > pci=hide=[<domain>:]<bus>:<slot>.<func>[; ...] This doesn't seem generally useful to me. It might be useful as a development aid, and you can easily maintain it as an out-of-tree patch for that. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
