> On May 27, 2016, at 13:34, Conrad Meyer <c...@freebsd.org> wrote: > > On Fri, May 27, 2016 at 1:12 PM, Garrett Cooper <n...@freebsd.org> wrote: >> Author: ngie >> Date: Fri May 27 20:12:32 2016 >> New Revision: 300868 >> URL: https://svnweb.freebsd.org/changeset/base/300868 >> >> Log: >> Remove note about bogus chain-len maximum >> >> There's no current limit on chain-len with Broadwell DE chips; it isn't >> enforced in software, and there doesn't appear to be a hardware limitation >> either on the Intel Xeon D-1527 (Broadwell-DE) chip. > > Hi Ngie, > > The note isn't bogus, it's just not what you think it is—the limit is > in the ioat_test code, not a limit of the hardware. > > Before this commit which documented it (r289733), the limit *was* 4. > However, in the same commit I bumped the limit up to 128 > (IOAT_MAX_BUFS / 2). (I suspect I wrote the documentation first, > before deciding to raise the limit.) > > So the current limit is 128, and should be documented.
Ah… that makes sense. Would it be a better idea to make this limit into a readonly sysctl in ioat_test(4), along with the other limits? If so, I’ll put that out for CR. Thanks, -Ngie
signature.asc
Description: Message signed with OpenPGP using GPGMail
