In arch/arm/kernel/setup.c:setup_ramdisk(), rd_size is set from the
boot tags. The replacement ramdisk driver has rd_size as static
which causes linking to fail when ramdisk is built-in.
Signed-off-by: Byron Bradley <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Russell King <[email protected]>
---
This patch applies against 2.6.24-rc8-mm1 and has not been tested,
it is only know to compile. I'm not sure if there is a better way
to set rd_size or if this is OK.
drivers/block/brd.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/block/brd.c b/drivers/block/brd.c
index 5ef1d26..8536480 100644
--- a/drivers/block/brd.c
+++ b/drivers/block/brd.c
@@ -385,7 +385,7 @@ static struct block_device_operations brd_fops = {
* And now the modules code and kernel interface.
*/
static int rd_nr;
-static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
+int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
module_param(rd_nr, int, 0);
MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
module_param(rd_size, int, 0);
--
1.5.4.rc2.38.gd6da3
On Fri, 18 Jan 2008 02:02:17 +0000 Byron Bradley <[email protected]> wrote:
> In arch/arm/kernel/setup.c:setup_ramdisk(), rd_size is set from the
> boot tags. The replacement ramdisk driver has rd_size as static
> which causes linking to fail when ramdisk is built-in.
>
but...
> diff --git a/drivers/block/brd.c b/drivers/block/brd.c
> index 5ef1d26..8536480 100644
> --- a/drivers/block/brd.c
> +++ b/drivers/block/brd.c
> @@ -385,7 +385,7 @@ static struct block_device_operations brd_fops = {
> * And now the modules code and kernel interface.
> */
> static int rd_nr;
> -static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> +int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> module_param(rd_nr, int, 0);
> MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
> module_param(rd_size, int, 0);
rd_size is a module parameter so it is settable via the
syntax-which-i-can-never-remember. rd.rd_size=1024 or something like that.
If that's all sane, do we have some back-compat reason to continue to
support the special and duplicative rd_size parameter?
(If we never did crap like this:
arch/arm/kernel/setup.c: extern int rd_size, rd_image_start, rd_prompt, rd_doload;
then this sort of problem wouldn't occur so often)
MIPS has the same problem.
On Thu, 2008-01-17 at 18:28 -0800, Andrew Morton wrote:
> On Fri, 18 Jan 2008 02:02:17 +0000 Byron Bradley <[email protected]> wrote:
>
> > In arch/arm/kernel/setup.c:setup_ramdisk(), rd_size is set from the
> > boot tags. The replacement ramdisk driver has rd_size as static
> > which causes linking to fail when ramdisk is built-in.
> >
>
> but...
>
> > diff --git a/drivers/block/brd.c b/drivers/block/brd.c
> > index 5ef1d26..8536480 100644
> > --- a/drivers/block/brd.c
> > +++ b/drivers/block/brd.c
> > @@ -385,7 +385,7 @@ static struct block_device_operations brd_fops = {
> > * And now the modules code and kernel interface.
> > */
> > static int rd_nr;
> > -static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > +int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > module_param(rd_nr, int, 0);
> > MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
> > module_param(rd_size, int, 0);
>
> rd_size is a module parameter so it is settable via the
> syntax-which-i-can-never-remember. rd.rd_size=1024 or something like that.
>
> If that's all sane, do we have some back-compat reason to continue to
> support the special and duplicative rd_size parameter?
Only insofar as we're still supporting ramdisks in the first place.
--
Mathematics is the supreme nostalgia of our time.
On Thu, Jan 17, 2008 at 08:39:23PM -0600, Matt Mackall wrote:
>
> On Thu, 2008-01-17 at 18:28 -0800, Andrew Morton wrote:
> > On Fri, 18 Jan 2008 02:02:17 +0000 Byron Bradley <[email protected]> wrote:
> >
> > > In arch/arm/kernel/setup.c:setup_ramdisk(), rd_size is set from the
> > > boot tags. The replacement ramdisk driver has rd_size as static
> > > which causes linking to fail when ramdisk is built-in.
> > >
> >
> > but...
> >
> > > diff --git a/drivers/block/brd.c b/drivers/block/brd.c
> > > index 5ef1d26..8536480 100644
> > > --- a/drivers/block/brd.c
> > > +++ b/drivers/block/brd.c
> > > @@ -385,7 +385,7 @@ static struct block_device_operations brd_fops = {
> > > * And now the modules code and kernel interface.
> > > */
> > > static int rd_nr;
> > > -static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > > +int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > > module_param(rd_nr, int, 0);
> > > MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
> > > module_param(rd_size, int, 0);
> >
> > rd_size is a module parameter so it is settable via the
> > syntax-which-i-can-never-remember. rd.rd_size=1024 or something like that.
> >
> > If that's all sane, do we have some back-compat reason to continue to
> > support the special and duplicative rd_size parameter?
>
> Only insofar as we're still supporting ramdisks in the first place.
I don't care about initrd or even a backward compatible API myself, I
do have my own reason want this new rd driver in the tree...
Would be nice to get rid of the arch stuff, but it's not too terrible
(at least from the POV of drivers/block/brd.c. So thanks for the patch,
Byron.
On Thu, Jan 17, 2008 at 06:28:51PM -0800, Andrew Morton wrote:
> On Fri, 18 Jan 2008 02:02:17 +0000 Byron Bradley <[email protected]> wrote:
>
> > In arch/arm/kernel/setup.c:setup_ramdisk(), rd_size is set from the
> > boot tags. The replacement ramdisk driver has rd_size as static
> > which causes linking to fail when ramdisk is built-in.
> >
>
> but...
>
> > diff --git a/drivers/block/brd.c b/drivers/block/brd.c
> > index 5ef1d26..8536480 100644
> > --- a/drivers/block/brd.c
> > +++ b/drivers/block/brd.c
> > @@ -385,7 +385,7 @@ static struct block_device_operations brd_fops = {
> > * And now the modules code and kernel interface.
> > */
> > static int rd_nr;
> > -static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > +int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
> > module_param(rd_nr, int, 0);
> > MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
> > module_param(rd_size, int, 0);
>
> rd_size is a module parameter so it is settable via the
> syntax-which-i-can-never-remember. rd.rd_size=1024 or something like that.
>
> If that's all sane, do we have some back-compat reason to continue to
> support the special and duplicative rd_size parameter?
>
>
> (If we never did crap like this:
>
> arch/arm/kernel/setup.c: extern int rd_size, rd_image_start, rd_prompt, rd_doload;
>
> then this sort of problem wouldn't occur so often)
We do "crap like this" because x86 did, and people wanted to set these
parameters at boot time, and in the old days of Linux (eg, 2.0), these
parameters were never available as command line arguments. So the only
way to set them from boot loaders via architecture code is for architecture
code to reference the variable directly.
It looks like things have moved on in x86 land, but because there's an
utter lack of communication about these changes, the "planned obsolescence"
of rd_size hasn't been noticed. So, I recommend that we keep rd_size
visible and take it through the now established feature-removal cycle,
including issuing kernel warnings about it if it gets used.