Re: [RFC][PATCH V4] axi: add AXI bus driver

From: Greg KH
Date: Wed Apr 13 2011 - 17:03:35 EST


On Wed, Apr 13, 2011 at 09:39:54PM +0200, RafaÅ MiÅecki wrote:
> 2011/4/13 Greg KH <greg@xxxxxxxxx>:
> >> diff --git a/drivers/axi/axi_pci_bridge.c b/drivers/axi/axi_pci_bridge.c
> >> new file mode 100644
> >> index 0000000..17e882c
> >> --- /dev/null
> >> +++ b/drivers/axi/axi_pci_bridge.c
> >> @@ -0,0 +1,33 @@
> >> +/*
> >> + * AXI PCI bridge module
> >> + *
> >> + * Licensed under the GNU/GPL. See COPYING for details.
> >> + */
> >> +
> >> +#include "axi_private.h"
> >> +
> >> +#include <linux/axi/axi.h>
> >> +#include <linux/pci.h>
> >> +
> >> +static DEFINE_PCI_DEVICE_TABLE(axi_pci_bridge_tbl) = {
> >> + Â Â { PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, 0x4331) },
> >> + Â Â { PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, 0x4353) },
> >> + Â Â { PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, 0x4727) },
> >> + Â Â { 0, },
> >> +};
> >> +MODULE_DEVICE_TABLE(pci, axi_pci_bridge_tbl);
> >> +
> >> +static struct pci_driver axi_pci_bridge_driver = {
> >> + Â Â .name = "axi-pci-bridge",
> >> + Â Â .id_table = axi_pci_bridge_tbl,
> >> +};
> >> +
> >> +int __init axi_pci_bridge_init(void)
> >> +{
> >> + Â Â return axi_host_pci_register(&axi_pci_bridge_driver);
> >> +}
> >> +
> >> +void __exit axi_pci_bridge_exit(void)
> >> +{
> >> + Â Â axi_host_pci_unregister(&axi_pci_bridge_driver);
> >> +}
> >
> > You register a pci driver that does nothing? ÂThat's not right, you need
> > to then base your axi bus off of that pci device, so it is hooked up
> > correctly in the /sys/devices/ tree. ÂOtherwise you are somewhere up in
> > the virtual location for your axi bus, right?
>
> Please take a look at:
> driver->probe = axi_host_pci_probe;
> driver->remove = axi_host_pci_remove;
> return pci_register_driver(driver);

Odd, why not just set up those functions in that file? Or move all of
this to that file and do it there? This seems like a very small file :)

> >> +bool axi_core_is_enabled(struct axi_device *core)
> >> +{
> >> + Â Â if ((axi_aread32(core, AXI_IOCTL) & (AXI_IOCTL_CLK | AXI_IOCTL_FGC))
> >> + Â Â Â Â != AXI_IOCTL_CLK)
> >> + Â Â Â Â Â Â return false;
> >> + Â Â if (axi_aread32(core, AXI_RESET_CTL) & AXI_RESET_CTL_RESET)
> >> + Â Â Â Â Â Â return false;
> >> + Â Â return true;
> >> +}
> >> +EXPORT_SYMBOL(axi_core_is_enabled);
> >
> > EXPORT_SYMBOL_GPL()?
> >
> > What module uses this? ÂAnd why would it care?
> >
> >> +EXPORT_SYMBOL(axi_core_enable);
> >
> > EXPORT_SYMBOL_GPL()?
> >
> > Same goes for your other exports, just want you to be sure here.
>
> Hm, I'm not sure. Using EXPORT_SYMBOL_GPL will forbid closed source
> drivers from using our bus driver, right? I'm don't have preferences
> on this, if you prefer us to force GPL, I can.

It's totally up to you, it's your code, not mine. Just wanted to remind
you of the option.

> >> +u32 xaxi_chipco_gpio_control(struct axi_drv_cc *cc, u32 mask, u32 value)
> >> +{
> >> + Â Â return axi_cc_write32_masked(cc, AXI_CC_GPIOCTL, mask, value);
> >> +}
> >> +EXPORT_SYMBOL(xaxi_chipco_gpio_control);
> >
> > "xaxi"? ÂShouldn't that be consistant with the other exports and start
> > with "axi"?
>
> Left from old tests/rewrites/splitting. Thanks.
>
>
> >> +static u8 axi_host_pci_read8(struct axi_device *core, u16 offset)
> >> +{
> >> + Â Â if (unlikely(core->bus->mapped_core != core))
> >
> > Are you sure about the use of unlikely in this, and other functions?
> > The compiler almost always does a better job than we do for these types
> > of calls, just let it do it's job.
> >
> >> + Â Â Â Â Â Â axi_host_pci_switch_core(core);
> >> + Â Â return ioread8(core->bus->mmio + offset);
> >
> > I think because of that unlikely, you just slowed down all pci devices,
> > right? ÂThat's not very nice :)
>
> Hm, my logic suggests it is alright, but please consider this once
> more with me ;)
>
> For the most of the time mapped_core (active core) do not change. We
> perform few hundreds of operations on one core in a row. This way
> mapped_core points to passed core for most of the time. Condition
> (mapped_core != core) is unlikely to happen.
>
> Is there anything wrong in my logic?

Drivers almost _never_ need to use likely or unlikely in their code.
The CPU can schedule things better and so can the compiler, so I would
just drop them, _unless_ you can show a benchmark where it matters.

thanks,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/