On 12/6/20 12:30 PM, Leon Romanovsky wrote:
On Sun, Dec 06, 2020 at 11:33:40AM +0100, Maximilian Luz wrote:
On 12/6/20 10:06 AM, Leon Romanovsky wrote:> On Sun, Dec 06, 2020 at 05:58:32PM +0900, Blaž Hrastnik wrote:
More on that, the whole purpose of proposed interface is to debug and
not intended to be used by any user space code.
The purpose is to provide raw access to the Surface Serial Hub protocol,
just like we provide raw access to USB devices and have hidraw devices.
So this goes a litle beyond just debugging; and eventually the choice
may be made to implement some functionality with userspace drivers,
just like we do for some HID and USB devices.
Still I agree with you that adding new userspace API is something which
needs to be considered carefully. So I will look at this closely when
reviewing this set.
To add to that: this was previously a debugfs interface but was moved to misc after review on the initial RFC:
https://lkml.org/lkml/2020/9/24/96
There is a huge difference between the suggestion and final implementation.
Greg suggested to add new debug module to the drivers/misc that will
open char device explicitly after user loaded that module to debug this
hub. However, the author added full blown char device as a first citizen
that has all not-break-user constrains.
This module still needs to be loaded explicitly. And (I might be wrong
about this) the "not-break-user constraints" hold as soon as I register
a misc device at all, no?
I don't think so, files in drivers/misc/* don't have such strict policy.
Can I get a link to the documentation stating that or someone else
confirming that?
Also I don't think it makes sense to have a platform/surface device in
drivers/misc, after we've explicitly decided to move this code out of
there. IIRC drivers/misc is not a place for misc-devices, but the
directory for devices that don't have any good place elsewhere.
than previously discussed with Greg and b) how the uapi header now
introduces any not-break-user constraints that would not be there
without it.
There is a huge difference between char device for the debug and
exposed UAPI header. The first requires from the user to build and
explicitly run it, while header allows to reliably build on top of
it various applications that we don't control. The not-break-rule
talks about the second.
So it's okay to break stuff that's not explicitly in include/uapi/?
Again, can I get someone to confirm that for me?
As already said, I'm okay with moving the definitions from the header to
the module itself (if there is a consensus on that, CC Greg, Hans),
however both allow you to build user-space tools against the API. Case
in point my python scripts, which don't use the header. Or any other
non-C-based tool. So unless there's a rule that anything without a
header in uapi is fair game, I fail to see your point.
This interface is intended as a stable interface. That's something that
I committed to as soon as I decided to implement this via a misc-device.
Sure, I can move the definitions in the uapi header to the module
itself, but I don't see any benefit in that. If someone really wants to
use this interface, they can just as well copy the definitions from the
module source itself. So why not be upfront about it and make life
easier for everyone?
Because you are actually making life harder for everyone who cares about
UAPIs exposed by the Linux and they definitely different in numbers from
those who needs debug interface for the Microsoft Surface board.
This point again depends on the interface not being stable. Unless, of
course, you want me to remove the interface completely and just maintain
it out of tree...
I'm happy to be told otherwise by the authorities here, but from past
conversations it seems that basically everything providing some sort of
user-space access falls under the "don't break user-space" rule as soon
as somebody uses it.
Regards,
Max
