Re: [PATCH] firmware_loader: re-export fw_fallback_config into firmware_loader's own namespace

[Luis Chamberlain]

On Fri, Apr 24, 2020 at 11:21:19AM +0200, Greg KH wrote:
> On Thu, Apr 23, 2020 at 08:31:40PM +0000, Luis R. Rodriguez wrote:
> > From: Luis Chamberlain <mcgrof@xxxxxxxxxx>
> > 
> > Christoph's recent patch "firmware_loader: remove unused exports", which
> > is not merged upstream yet, removed two exported symbols. One is fine to
> > remove since only built-in code uses it but the other is incorrect.
> > 
> > If CONFIG_FW_LOADER=m so the firmware_loader is modular but
> > CONFIG_FW_LOADER_USER_HELPER=y we fail at mostpost with:
> > 
> > ERROR: modpost: "fw_fallback_config" [drivers/base/firmware_loader/firmware_class.ko] undefined!
> > 
> > This happens because the variable fw_fallback_config is built into the
> > kernel if CONFIG_FW_LOADER_USER_HELPER=y always, so we need to grant
> > access to the firmware loader module by exporting it.
> > 
> > Instead of just exporting it as we used to, take advantage of the new
> > kernel symbol namespacing functionality, and export the symbol only to
> > the firmware loader private namespace. This would prevent misuses from
> > other drivers and makes it clear the goal is to keep this private to
> > the firmware loader alone.
> > 
> > Cc: Christoph Hellwig <hch@xxxxxx>
> > Cc: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> > Cc: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> > Fixes: "firmware_loader: remove unused exports"
> > Reported-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> > Signed-off-by: Luis Chamberlain <mcgrof@xxxxxxxxxx>
> > ---
> >  drivers/base/firmware_loader/fallback.c       | 3 +++
> >  drivers/base/firmware_loader/fallback_table.c | 1 +
> >  2 files changed, 4 insertions(+)
> > 
> > diff --git a/drivers/base/firmware_loader/fallback.c b/drivers/base/firmware_loader/fallback.c
> > index 1e9c96e3ed63..d9ac7296205e 100644
> > --- a/drivers/base/firmware_loader/fallback.c
> > +++ b/drivers/base/firmware_loader/fallback.c
> > @@ -9,6 +9,7 @@
> >  #include <linux/umh.h>
> >  #include <linux/sysctl.h>
> >  #include <linux/vmalloc.h>
> > +#include <linux/module.h>
> >  
> >  #include "fallback.h"
> >  #include "firmware.h"
> > @@ -17,6 +18,8 @@
> >   * firmware fallback mechanism
> >   */
> >  
> > +MODULE_IMPORT_NS(FIRMWARE_LOADER_PRIVATE);
> > +
> >  extern struct firmware_fallback_config fw_fallback_config;
> >  
> >  /* These getters are vetted to use int properly */
> 
> While nice, that does not fix the existing build error that people are
> having, right?

It does.

> > diff --git a/drivers/base/firmware_loader/fallback_table.c b/drivers/base/firmware_loader/fallback_table.c
> > index 0a737349f78f..46a731dede6f 100644
> > --- a/drivers/base/firmware_loader/fallback_table.c
> > +++ b/drivers/base/firmware_loader/fallback_table.c
> > @@ -21,6 +21,7 @@ struct firmware_fallback_config fw_fallback_config = {
> >  	.loading_timeout = 60,
> >  	.old_timeout = 60,
> >  };
> > +EXPORT_SYMBOL_NS_GPL(fw_fallback_config, FIRMWARE_LOADER_PRIVATE);
> 
> 
> How about you send a patch that just reverts the single symbol change
> first, and then a follow-on patch that does this namespace addition.  I
> can queue the first one up now, for 5.7-final, and the second one for
> 5.8-rc1.

Sure.

  Luis