Date: Fri, 6 Dec 2013 13:52:06 -0700 From: Warner Losh <imp@bsdimp.com> To: John-Mark Gurney <jmg@funkthat.com> Cc: Eitan Adler <lists@eitanadler.com>, "freebsd-arch@freebsd.org" <arch@freebsd.org> Subject: Re: hw.pci.do_power_nodriver=3 Message-ID: <91CBC153-C832-4350-8E19-24783A1CFA63@bsdimp.com> In-Reply-To: <20131206202616.GH55638@funkthat.com> References: <CAF6rxg=cSkTGLFZCeeg6C0V=frQN1iYtpLZBrNAu4WAQCX63Lw@mail.gmail.com> <CAF6rxgmeBAa9LK%2BHWU7NKUkW6RZvk_dNGDpN%2BJ43L=BdUsqHwQ@mail.gmail.com> <E24C9357-B0A8-48D1-8D80-DAC5B09580EF@bsdimp.com> <20131206202616.GH55638@funkthat.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Dec 6, 2013, at 1:26 PM, John-Mark Gurney wrote: > Warner Losh wrote this message on Thu, Dec 05, 2013 at 22:43 -0700: >>=20 >> On Dec 5, 2013, at 9:12 PM, Eitan Adler wrote: >>> Is there any reason we can not set hw.pci.do_power_nodriver=3D3 by = default? >>>=20 >>> My understanding is that there were problems with hardware being >>> powered off and not being powered back on when drivers were loaded. >>> Is this still a concern? If yes, can we flip the switch in HEAD and >>> fix the drivers? >>=20 >> The reason it was for Adaptec RAID controllers. >>=20 >> They had a weird topology: >>=20 >> <-------------------- aac based card = --------------------------> >> pci bus ---- pci bridge ---- pci bus ---+----- some chip with = driver >> = +----- chip without driver >>=20 >> so, when the enumeration code saw that there was no driver attached = to the second chip, it would power it down. Turns out, this chip, while = it didn't have a driver, was critical to the proper functioning of the = RAID card. Scott Long turned off the default power saving because he was = worried there were other parts like this. In addition, in an abundance = of caution, he also created stub drivers for the second chip for each of = the then known aac cards. >>=20 >> Since then, it is unknown if others have followed this design or not, = so it is unknown our exposure if we were to flip this to have a = different default. >=20 > Should we flip this on by default in HEAD to help expose these issues? > It is expected that people running HEAD spend a little time helping us > debug issues, and if they don't want to take the risk, they can change > the default... >=20 > Then maybe after a few years, maybe not 11, but for 12, we can keep it > on by default for a release? Scott and I talked about a possible solution that would be safe. Once I = have that implemented, we can likely switch things over... Warner
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?91CBC153-C832-4350-8E19-24783A1CFA63>