On Fri, Jun 14, 2019 at 06:13:39PM +0200, Greg Kroah-Hartman wrote:
On Fri, Jun 14, 2019 at 04:34:05PM +0100, Mark Brown wrote:
On Fri, Jun 14, 2019 at 11:47:56AM +0200, Greg Kroah-Hartman wrote:
The majority of this is removing error prints rather than code that actively does something different. If this was like kmalloc() where the API is itself reported errors then this wouldn't be an issue but unless
No, the long-term goal is for debugfs_create_file() to just return void. I have already done enough cleanups in my local tree to do that already for many of the helper functions.
No one should care one bit if a debugfs file succeeds or not, no logic should ever change, nor should it matter. debugfs is for debugging kernel code, not for anything that anyone should ever rely on.
I don't think this would be a helpful change. Nobody should care about debugfs file creation for *production* use but that doesn't mean that people using it for debugging shouldn't care about it. Debugging tools that are fragile can be very misleading for developers and hence intensely frustrating; it's never fun to be building on quicksand. This is particularly true for code like here where the core is providing a bunch of internal data that is hopefully useful to people developing drivers or system integrations, the users have a hopefully more black box view of the core than someone directly working on the core would. If that view of the data ends up being corrupted because some of the files or directories don't get created that is something we should be flagging up to people to try to help them avoid being mislead by what they are seeing. Developers need to be able to trust their debugging tools.
So yes, this patch does remove a bunch of error messages (that as you point out can never be triggered), but the main goal here is to not check the return value of the function at all, which is what this patch does.
Like I said in reply to the other patch the error checking here did the right thing up until January when debugfs was changed to return error pointers instead of NULL. I know that I've found this error checking to be helpful in the past, not having it would be a loss.
If there were some visible error reporting in debugfs I'd not be bothered about this quite so much (though it'd still be less clear) but there isn't so it really feels like a step back.