Date: Fri, 6 Jun 2008 09:36:20 +0200 From: Marc =?iso-8859-1?q?L=F6rner?= <marc.loerner@hob.de> To: "Bruce M. Simpson" <bms@FreeBSD.org> Cc: freebsd-net@freebsd.org Subject: Re: Probable Bug in tcp.h Message-ID: <200806060936.20113.marc.loerner@hob.de> In-Reply-To: <48480FA5.3020800@FreeBSD.org> References: <200806051712.47048.marc.loerner@hob.de> <48480FA5.3020800@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday 05 June 2008 18:09, Bruce M. Simpson wrote: > Marc Lörner wrote: > > .. > > First of all I have the problam of misalignment of th_off. Because in > > this way always 4 bytes are read and the the bits of th_off are replaced. > > Then the 4 bytes are written back. > > > > But should (th_x and th_off) not only be 1 byte in whole -> only read and > > write 1 byte? > > Which machine architecture are you attempting to compile this code on? > ia64/itanium > On FreeBSD Tier 1 platforms, the access is probably going to come out of > L2 cache anyway, so the fields in question will be read by a burst cycle. > I know! But the problem (on itanium) is that bitfields are always accessed as integers => 4-byte access But th_x/th_off may not always be aligned to 4-bytes => I get an unalignment reference fault If access of th_x/th_off could be changed to 1-byte => 1-byte aligned => my unaligned reference fault would go away > It is worth noting that NetBSD changed the base type of tcphdr's > bitfields to uint8_t, however this shuffles the compiler dependency into > the treatment of the "char" type. Most modern C compilers support > "unsigned char". Does this really change the access to 1-byte? As in "Programming in C" by Kernighan and Ritchie is stated that bitfields must and will always be defined as ints.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200806060936.20113.marc.loerner>