Re: [PATCH 1/3] driver core: Mark impossible return values of bus_type's match() with unlikely()

From: Zijun Hu
Date: Wed Sep 04 2024 - 11:45:47 EST


On 2024/9/4 21:53, Greg Kroah-Hartman wrote:
> On Wed, Sep 04, 2024 at 08:56:42PM +0800, Zijun Hu wrote:
>> From: Zijun Hu <quic_zijuhu@xxxxxxxxxxx>
>>
>> Bus_type's match() should return bool type compatible integer 0 or 1
>> ideally since its main operations are lookup and comparison normally
>> actually, this rule is followed by ALL bus_types but @amba_bustype within
>> current v6.10 kernel tree, for @amba_bustype, ONLY extra -EPROBE_DEFER
>> may be returned, so mark those impossible or rare return values with
>> unlikely() to help readers understand device and driver binding logic.
>
> unlikely() and likely() should ONLY be used when you can measure the
> performance impact. And any "scan all devices in the bus" function is
> NOT performance critical at all. So this is not the place for that,
> sorry.
>

make sense
thank you for these explanation.

>> Signed-off-by: Zijun Hu <quic_zijuhu@xxxxxxxxxxx>
>> ---
>> drivers/base/dd.c | 16 ++++++++++++----
>> include/linux/device/bus.h | 9 ++++-----
>> 2 files changed, 16 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
>> index 9b745ba54de1..288e19c9854b 100644
>> --- a/drivers/base/dd.c
>> +++ b/drivers/base/dd.c
>> @@ -928,7 +928,11 @@ static int __device_attach_driver(struct device_driver *drv, void *_data)
>> if (ret == 0) {
>> /* no match */
>> return 0;
>> - } else if (ret == -EPROBE_DEFER) {
>> + } else if (unlikely(ret == -EPROBE_DEFER)) {
>> + /*
>> + * Only match() of @amba_bustype may return this error
>> + * in current v6.10 tree, so also give unlikely() here.
>
> version numbers in the kernel source mean nothing, and they age
> horribly. This is not going to work out at all.
>
> let's fix up the one user that is doing this wrong and then we don't
> have to worry about it, right? We have the source for everything, let's
> use it :)
>
yes, you are right.

my original motivation is to see if it is possible change bus_type's
match() return value type to bool type based on below reasons:

(1)
it is not good time for bus_type's match() to operate device such as
I/O since device may be not ready to operate at that time.

(2)
match() is called without holding device lock firstly by driver_attach().

(3)
all bus_type's match() only do lookup and comparison and return bool
type but @amba_bustype which operate device and maybe return extra
-EPROBE_DEFER

> thanks,
>
> greg k-h