Messages in this thread | | | Date | Fri, 06 Jan 2006 23:46:16 +0100 | From | Eric Dumazet <> | Subject | Re: [patch 0/4] Series to allow a "const" file_operations struct |
| |
Arjan van de Ven a écrit : > On Fri, 2006-01-06 at 22:45 +0100, Arjan van de Ven wrote: >> Hi, >> >> this series allows drivers to have "const" file_operations, by making >> the f_ops field in the inode const. This has another benefit, there have > > ok there was a sentence missing here. The first benefit is that this > moves these hot datastructures to the rodata section, which means they > won't accidentally be doing false cacheline sharing. >
Definitly a good thing I agree.
But your patches miss to really declare all 'struct file_operations' as const, dont they ?
On my x86_64 machine, I managed to reduce by 10% .data section by moving all file_operations, but also 'address_space_operations', 'inode_operations, super_operations, dentry_operations, seq_operations, ... to rodata section.
size vmlinux* text data bss dec hex filename 2476156 522236 244868 3243260 317cfc vmlinux 2588685 571348 246692 3406725 33fb85 vmlinux.old
Eric
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |