Hi Jeff,
since drivers/net/tulip/de2104x.c does currently not compile in 2.5.29 due
to a .text.exit error when the driver is compiled into a kernel without
hotplug support I'm wondering whether the patch below would be correct to
make this PCI driver hotpluggable. Is my approach to change __init to
__devinit and __exit to __devexit correct or is there something I've
overseen?
cu
Adrian
--- drivers/net/tulip/de2104x.c.old Sat Jul 27 11:43:47 2002
+++ drivers/net/tulip/de2104x.c Sat Jul 27 11:46:00 2002
@@ -50,7 +50,7 @@
#include <asm/unaligned.h>
/* These identify the driver base version and may not be removed. */
-static char version[] __initdata =
+static char version[] __devinitdata =
KERN_INFO DRV_NAME " PCI Ethernet driver v" DRV_VERSION " (" DRV_RELDATE ")\n";
MODULE_AUTHOR("Jeff Garzik <[email protected]>");
@@ -330,7 +330,7 @@
static unsigned int de_ok_to_advertise (struct de_private *de, u32 new_media);
-static struct pci_device_id de_pci_tbl[] __initdata = {
+static struct pci_device_id de_pci_tbl[] __devinitdata = {
{ PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_DEC_TULIP,
PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
{ PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_DEC_TULIP_PLUS,
@@ -1746,7 +1746,7 @@
return rc;
}
-static void __init de21040_get_mac_address (struct de_private *de)
+static void __devinit de21040_get_mac_address (struct de_private *de)
{
unsigned i;
@@ -1763,7 +1763,7 @@
}
}
-static void __init de21040_get_media_info(struct de_private *de)
+static void __devinit de21040_get_media_info(struct de_private *de)
{
unsigned int i;
@@ -1790,7 +1790,7 @@
}
/* Note: this routine returns extra data bits for size detection. */
-static unsigned __init tulip_read_eeprom(void *regs, int location, int addr_len)
+static unsigned __devinit tulip_read_eeprom(void *regs, int location, int addr_len)
{
int i;
unsigned retval = 0;
@@ -1825,7 +1825,7 @@
return retval;
}
-static void __init de21041_get_srom_info (struct de_private *de)
+static void __devinit de21041_get_srom_info (struct de_private *de)
{
unsigned i, sa_offset = 0, ofs;
u8 ee_data[DE_EEPROM_SIZE + 6] = {};
@@ -1981,7 +1981,7 @@
goto fill_defaults;
}
-static int __init de_init_one (struct pci_dev *pdev,
+static int __devinit de_init_one (struct pci_dev *pdev,
const struct pci_device_id *ent)
{
struct net_device *dev;
@@ -2136,7 +2136,7 @@
return rc;
}
-static void __exit de_remove_one (struct pci_dev *pdev)
+static void __devexit de_remove_one (struct pci_dev *pdev)
{
struct net_device *dev = pci_get_drvdata(pdev);
struct de_private *de = dev->priv;
@@ -2216,14 +2216,14 @@
name: DRV_NAME,
id_table: de_pci_tbl,
probe: de_init_one,
- remove: de_remove_one,
+ remove: __devexit_p(de_remove_one),
#ifdef CONFIG_PM
suspend: de_suspend,
resume: de_resume,
#endif
};
-static int __init de_init (void)
+static int __devinit de_init (void)
{
#ifdef MODULE
printk("%s", version);
@@ -2231,7 +2231,7 @@
return pci_module_init (&de_driver);
}
-static void __exit de_exit (void)
+static void __devexit de_exit (void)
{
pci_unregister_driver (&de_driver);
}
Adrian Bunk wrote:
> Hi Jeff,
>
> since drivers/net/tulip/de2104x.c does currently not compile in 2.5.29 due
> to a .text.exit error when the driver is compiled into a kernel without
> hotplug support I'm wondering whether the patch below would be correct to
> make this PCI driver hotpluggable. Is my approach to change __init to
> __devinit and __exit to __devexit correct or is there something I've
> overseen?
This driver is intentionally not hot-pluggable. I'll convert when
someone actually tells me they are trying to hot-plug such a card.
> -static int __init de_init (void)
> +static int __devinit de_init (void)
> {
> #ifdef MODULE
> printk("%s", version);
> @@ -2231,7 +2231,7 @@
> return pci_module_init (&de_driver);
> }
>
> -static void __exit de_exit (void)
> +static void __devexit de_exit (void)
> {
> pci_unregister_driver (&de_driver);
> }
This is incorrect in any case -- the module init/exit functions are
always __init and __exit.
Jeff
On Sat, 27 Jul 2002, Jeff Garzik wrote:
> > since drivers/net/tulip/de2104x.c does currently not compile in 2.5.29 due
> > to a .text.exit error when the driver is compiled into a kernel without
> > hotplug support I'm wondering whether the patch below would be correct to
> > make this PCI driver hotpluggable. Is my approach to change __init to
> > __devinit and __exit to __devexit correct or is there something I've
> > overseen?
>
>
> This driver is intentionally not hot-pluggable. I'll convert when
> someone actually tells me they are trying to hot-plug such a card.
It's OK for me not to make it hot-pluggable, but in this case you need the
following patch to make it possible to compile the driver statically into
the kernel (currently it fails with a .text.exit error at the final
linking):
--- drivers/net/tulip/de2104x.c.old Sun May 12 13:55:28 2002
+++ drivers/net/tulip/de2104x.c Sun May 12 13:56:09 2002
@@ -2216,7 +2216,9 @@
name: DRV_NAME,
id_table: de_pci_tbl,
probe: de_init_one,
+#ifdef MODULE
remove: de_remove_one,
+#endif
#ifdef CONFIG_PM
suspend: de_suspend,
resume: de_resume,
> > -static int __init de_init (void)
> > +static int __devinit de_init (void)
> > {
> > #ifdef MODULE
> > printk("%s", version);
> > @@ -2231,7 +2231,7 @@
> > return pci_module_init (&de_driver);
> > }
> >
> > -static void __exit de_exit (void)
> > +static void __devexit de_exit (void)
> > {
> > pci_unregister_driver (&de_driver);
> > }
>
>
> This is incorrect in any case -- the module init/exit functions are
> always __init and __exit.
Ah thanks, I had the feeling that something was wrong with my patch but I
didn't know what...
> Jeff
cu
Adrian
--
You only think this is a free country. Like the US the UK spends a lot of
time explaining its a free country because its a police state.
Alan Cox