| Message ID | 20111227212253.GA8161@merkur.ravnborg.org |
|---|---|
| State | Accepted |
| Delegated to: | David Miller |
| Headers | show |
From: Sam Ravnborg <sam@ravnborg.org> Date: Tue, 27 Dec 2011 22:22:53 +0100 >>From b82ccce349dee62fe5acd3167a19490bbe9a7e18 Mon Sep 17 00:00:00 2001 > From: Sam Ravnborg <sam@ravnborg.org> > Date: Tue, 27 Dec 2011 22:07:23 +0100 > Subject: [PATCH] apbuart: fix section mismatch warning > > Fix following warnings: > > WARNING: drivers/tty/serial/built-in.o(.text+0x7370): Section mismatch in reference from the function grlib_apbuart_configure() to the variable .init.data:apbuart_match > The function grlib_apbuart_configure() references > the variable __initdata apbuart_match. > This is often because grlib_apbuart_configure lacks a __initdata > annotation or the annotation of apbuart_match is wrong. > > + 3 more warnings like this. > > There is no guarantee that grlib_apbuart_of_driver.of_match_table > is only used at __init time - so drop the __initdata annotation. > > grlib_apbuart_configure() is only used during __init so add __init > to this method too. > > Signed-off-by: Sam Ravnborg <sam@ravnborg.org> > Cc: Daniel Hellstrom <daniel@gaisler.com> > Cc: Greg Kroah-Hartman <gregkh@suse.de> Applied to sparc-next, thanks. Taking a look at the output of "git grep of_device_id | grep __initdata" shows that a lot of drivers take liberties in this area, perhaps assuming that they are built in or something like that. But that isn't even a safe situation, because even built-in drivers can have their of_device_id accessed later after we bootup, for example when a bus is hot-plugged in or even via sysfs dumps of the avaialble drivers. Probably the thing to do is to remove all of these __initdata tags, because as far as I can tell they are _all_ wrong. And perhaps add 'const' where missing as well. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Dec 27, 2011 at 04:35:22PM -0500, David Miller wrote: > From: Sam Ravnborg <sam@ravnborg.org> > Date: Tue, 27 Dec 2011 22:22:53 +0100 > > >>>From b82ccce349dee62fe5acd3167a19490bbe9a7e18 Mon Sep 17 00:00:00 2001 > > From: Sam Ravnborg <sam@ravnborg.org> > > Date: Tue, 27 Dec 2011 22:07:23 +0100 > > Subject: [PATCH] apbuart: fix section mismatch warning > > > > Fix following warnings: > > > > WARNING: drivers/tty/serial/built-in.o(.text+0x7370): Section mismatch in reference from the function grlib_apbuart_configure() to the variable .init.data:apbuart_match > > The function grlib_apbuart_configure() references > > the variable __initdata apbuart_match. > > This is often because grlib_apbuart_configure lacks a __initdata > > annotation or the annotation of apbuart_match is wrong. > > > > + 3 more warnings like this. > > > > There is no guarantee that grlib_apbuart_of_driver.of_match_table > > is only used at __init time - so drop the __initdata annotation. > > > > grlib_apbuart_configure() is only used during __init so add __init > > to this method too. > > > > Signed-off-by: Sam Ravnborg <sam@ravnborg.org> > > Cc: Daniel Hellstrom <daniel@gaisler.com> > > Cc: Greg Kroah-Hartman <gregkh@suse.de> > > Applied to sparc-next, thanks. > > Taking a look at the output of "git grep of_device_id | grep __initdata" > shows that a lot of drivers take liberties in this area, perhaps assuming > that they are built in or something like that. I took a quick look at this. In the places I found the of_device_id() was passed to for example of_platform_bus_probe() which was only called at __init time. And the function did not save the argument. So at least some call sites are OK. But I did not audit nearly all of them - so some bugs are lingering here. Sam -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
From: Sam Ravnborg <sam@ravnborg.org> Date: Tue, 27 Dec 2011 23:20:46 +0100 > In the places I found the of_device_id() was passed to for > example of_platform_bus_probe() which was only called at __init time. > And the function did not save the argument. > > So at least some call sites are OK. Ok, good to know. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/tty/serial/apbuart.c b/drivers/tty/serial/apbuart.c index 77554fd..7162f70 100644 --- a/drivers/tty/serial/apbuart.c +++ b/drivers/tty/serial/apbuart.c @@ -577,7 +577,7 @@ static int __devinit apbuart_probe(struct platform_device *op) return 0; } -static struct of_device_id __initdata apbuart_match[] = { +static struct of_device_id apbuart_match[] = { { .name = "GAISLER_APBUART", }, @@ -597,7 +597,7 @@ static struct platform_driver grlib_apbuart_of_driver = { }; -static int grlib_apbuart_configure(void) +static int __init grlib_apbuart_configure(void) { struct device_node *np; int line = 0;