On Tue, Aug 17, 2004 at 09:38:16AM -0400, James Morris wrote:
Yes, although I'm not clear on what should be done. Returning an error
via an initcall does not do anything, so if these were built statically,
then the kernel would go on running after they failed. This is a general
problem. e.g. IPv6, which is commonly built as a module, will panic if
kmem_cache_create() fails during module load in several places.
The ipv6 behaviour is definnitly bad. OOM situations shouldn't panic
the kernel.
If something is can be built modular it surely isn't important enough to
panic the kernel on bootup if it can't initialize - after all people can
run a kernel without the module loaded just fine.
Em Tue, Aug 17, 2004 at 03:07:39PM +0100, Christoph Hellwig escreveu:
On Tue, Aug 17, 2004 at 09:38:16AM -0400, James Morris wrote:
quoted
Yes, although I'm not clear on what should be done. Returning an error
via an initcall does not do anything, so if these were built statically,
then the kernel would go on running after they failed. This is a general
problem. e.g. IPv6, which is commonly built as a module, will panic if
kmem_cache_create() fails during module load in several places.
The ipv6 behaviour is definnitly bad. OOM situations shouldn't panic
the kernel.
If something is can be built modular it surely isn't important enough to
panic the kernel on bootup if it can't initialize - after all people can
run a kernel without the module loaded just fine.
Agreed, not because it is "not important", but because panicing at module
load, even in very rare cases is unnacceptable IMHO.
- Arnaldo
On Wed, 18 Aug 2004 01:28:00 -0300
Arnaldo Carvalho de Melo [off-list ref] wrote:
Agreed, not because it is "not important", but because panicing at module
load, even in very rare cases is unnacceptable IMHO.
I think even non-modular cases should fail gracefully.
If the ipv4 one fails, no ipv4 networking sorry. :-)