Re: [PATCH v5 1/3] xhci: hub: export symbol on xhci_hub_control

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2867 bytes --]

On Fri, Oct 28, 2022 at 12:37:15PM +0200, gregkh@linuxfoundation.org wrote:
> On Fri, Oct 28, 2022 at 11:50:47AM +0200, Thierry Reding wrote:
> > On Fri, Oct 28, 2022 at 08:04:54AM +0200, gregkh@linuxfoundation.org wrote:
> > > On Fri, Oct 28, 2022 at 05:36:41AM +0000, Jim Lin wrote:
> > > > On Thu, 2022-10-27 at 15:58 +0200, Greg KH wrote:
> > > > > External email: Use caution opening links or attachments
> > > > > 
> > > > > 
> > > > > On Thu, Oct 27, 2022 at 09:31:25PM +0800, Jim Lin wrote:
> > > > > > Add EXPORT_SYMBOL_GPL on xhci_hub_control() for other driver module
> > > > > > to invoke and avoid linking error.
> > > > > 
> > > > > What other driver module?
> > > > > 
> > > > > There is no user here :(
> > > > > 
> > > > > confused,
> > > > > 
> > > > 
> > > > In arch/arm/configs/multi_v7_defconfig
> > > > It defines
> > > > CONFIG_USB_XHCI_TEGRA=m
> > > > 
> > > > If I don't add EXPORT_SYMBOL_GPL on xhci_hub_control()
> > > > , I will get compile/linking error like
> > > > 
> > > > ERROR: modpost: "xhci_hub_control" [drivers/usb/host/xhci-tegra.ko]
> > > > undefined!
> > > > 
> > > > if patch
> > > > "[PATCH v5,3/3] xhci: tegra: USB2 pad power controls"
> > > > 
> > > > https://patchwork.kernel.org/project/linux-usb/patch/20221027133127.27592-4-jilin@nvidia.com/
> > > > is added in xhci-tegra.c to invoke xhci_hub_control()
> > > > 
> > > > Should I integrate this patch with [PATCH v5,3/3] as one?
> > > 
> > > Yes, do not add something that is not needed for that specific commit,
> > > otherwise it causes reviewers to be confused.
> > 
> > Other subsystem maintainers prefer core changes to be split from driver
> > changes, so this type of split is commonly encountered elsewhere.
> > 
> > Obviously, since this is your turf you get to make the rules. I'm just
> > trying to say that this kind of advice can be confusing for contributors
> > because when they then sent driver and code changes mixed for their next
> > submission, the subsystem maintainer might tell them otherwise.
> 
> Sure, but if you do split it up like this, DOCUMENT WHY THE EXPORT IS
> NEEDED.  That didn't happen here so I had no idea why this was even an
> issue.
> 
> And yes, I am very sensitive to this, we have had LOTS of people trying
> to export xhci symbols in the past few years for no in-kernel users,
> despite us constantly telling them that this is not allowed.  It
> happened again, just yesterday:
> 	https://lore.kernel.org/r/20221027004050.4192111-1-albertccwang@google.com
> 
> And at first glance, I assumed this was much the same as there was no
> description of why this was needed at all.

Agreed. I suppose this could've been spelled out more explicitly in the
cover letter or in patch 1.

Jim, please make sure to describe this dependency explicitly in v6.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]