At Mon, 8 Oct 2001 10:19:04 -0700, Andrew Loewenstern wrote: > > that patch didn't work, it needs "unlikely" and including > <linux/compiler.h> breaks other things... just putting in the define > for unlikely (mapped to __builtin_expect) eventually causes the linking > of the kernel to fail (__builtin_expect)... turns out that compiler.h > defines __builtin_expect as "__builtin_expect(x, expected_value) (x)" > so putting both of those in buffer.c made it compile but I'm not sure > it's working properly.
Oh, sorry, it seems that old patch for pe kernel remained there. The fixed patch is attached here. Josh, please take this one. > in any event, I'm getting zero underruns now using 2.4.10 with Andrew > Morton's LL patches and direct IO (I'm primarily doing long captures). > This is on a pretty wimpy 300mhz cyrix mediagx (nsc geode). The real > test is the 100mhz MachZF based sbc which I haven't hooked up yet. Do you mean that you achieved it without patch to buffer.c? ciao, Takashi
buffer.c-ll.dif
Description: Binary data
