Date: Tue, 6 Nov 2007 16:42:36 GMT From: Robert Watson <rwatson@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 128745 for review Message-ID: <200711061642.lA6GgaHf092156@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=128745 Change 128745 by rwatson@rwatson_fledge on 2007/11/06 16:42:16 Update two comments. Affected files ... .. //depot/projects/zcopybpf/src/sys/net/bpf_zerocopy.c#21 edit Differences ... ==== //depot/projects/zcopybpf/src/sys/net/bpf_zerocopy.c#21 (text+ko) ==== @@ -74,7 +74,8 @@ * sf_bufs. Even though the memory is contiguous in user space, it may not * be mapped contiguously in the kernel (i.e., a set of physically * non-contiguous pages in the direct map region) so we must implement - * scatter-gather copying. + * scatter-gather copying. One significant mitigating factor is that on + * systems with a direct memory map, we can avoid TLB misses. * * At the front of the shared memor region is a bpf_zbuf_header, which * contains shared control data to allow user space and the kernel to @@ -402,9 +403,8 @@ } /* - * For now, allow bpfread() to rotate the buffers, but don't perform a copy - * operation or return a value. If we want to copy, we'll need to implement - * scatter-gather copying with a series of uiomove calls here. + * read() is unimplemented for zero-copy BPF buffers, even though in + * principle you could combine the two semantics. */ int bpf_zerocopy_uiomove(struct bpf_d *d, caddr_t buf, u_int len,
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200711061642.lA6GgaHf092156>