Messages in this thread | | | Date | Thu, 26 Feb 2004 15:21:39 -0800 | From | Andrew Morton <> | Subject | Re: [PATCH] __cacheline_aligned always in own section |
| |
Denis Vlasenko <vda@port.imtp.ilyichevsk.odessa.ua> wrote: > > I compile my kernels for 486 but buils system aligns > functions and labels to 16 bytes, with results like this: > > 00000730 <islpci_eth_tx_timeout>: > 730: 55 push %ebp > 731: 89 e5 mov %esp,%ebp > 733: 8b 45 08 mov 0x8(%ebp),%eax > 736: 8b 40 64 mov 0x64(%eax),%eax > 739: 05 14 03 00 00 add $0x314,%eax > 73e: ff 40 14 incl 0x14(%eax) > 741: 5d pop %ebp > 742: c3 ret > 743: 90 nop > 744: 90 nop > 745: 90 nop > 746: 90 nop > 747: 90 nop > 748: 90 nop > 749: 90 nop > 74a: 90 nop > 74b: 90 nop > 74c: 90 nop > 74d: 90 nop > 74e: 90 nop > 74f: 90 nop > > Losing on average 15/2 bytes to alignment, my kernel lose > # echo $((`cat System.map | grep '0 ' | wc -l`*15/2)) > 149632 > bytes only due to function alignment, not counting jump target > alighment. > > Is there any way to prevent this?
Yes, there are ways of turning off a lot of this alignment padding and it makes a significant different in code size. You need to dig around in the gcc documentation for the several -*align* options.
IIRC, not all of the padding could be turned off (the gcc options were not complete) but it's a couple of years since I investigated this.
- 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/
| |