Date: Sun, 11 Feb 2001 19:09:53 +0100 (CET) From: =?ISO-8859-1?Q?G=E9rard_Roudier?= <groudier@club-internet.fr> To: Mark Abene <phiber@radicalmedia.com> Cc: alpha@FreeBSD.ORG Subject: Re: symbios card... Message-ID: <Pine.LNX.4.10.10102111838090.433-100000@linux.local> In-Reply-To: <20010211115723.G1040@mail.radicalmedia.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 11 Feb 2001, Mark Abene wrote: > Thanks! What are the pros/cons of using the newer sym driver versus the > older ncr driver? Which would you recommend with a 53C825? Does it matt= er? For a 53C825 I would simply suggest to use a driver that works for you. :) But I would prefer SYM to be tried first. I have currently a portable beta driver based on SYM (and named SYM-2) and this will be lot of time spare for me to have a single driver to maintain rather that four (2 Under FreeBSD and 2 under Linux) or even more. My testing did show that SYM has less latency than NCR for small IOs using a 53C810 (non-A) and a 53C895. For the 895 I tried both the NCR-generic SCRIPTS and the LOAD/STORE based SCRIPTS. But the difference should not be significant for real work performed with a 825, IMO. For chips that support hardware phase mismatch (896 and later chips), the SYM handles the context from SCRIPTS and ensures that no more than 1 interrupt per IO will be delivered to the CPU and that the SCSI SCRIPTS=20 never stalls waiting for CPU intervention (excepted on error recovery). Also SYM does not scan lists either from CPU or SCRIPTS processor and uses= =20 a completion queue to interface with the C code. This makes it full scalable regarding the number of outstanding IOs. In other words, SYM runs O(1) in normal situations (i.e. excepted for error recovery). SYM is currently configured for 448 outstanding IOs per initiator, but is= =20 in fact not so limited. I just wanted not to allocate more than 1 page at a time for the start and completion queues that use 8 bytes per entry. And allocating the queues contiguously is simple. Btw, the queues could=20 be allocated in several chunks but the additionnal complexity is not=20 yet worthwhile. (Btw, with 8K pages, a simple define change will increase the queue size to 1022 entries, without any other software changes) G=E9rard. > Cheers, > -Mark >=20 > On Sun, Feb 11, 2001 at 04:46:34PM +0100, G=E9rard Roudier wrote: > >=20 > > On Sun, 11 Feb 2001, Mark Abene wrote: > >=20 > > > Would anyone care to comment on whether the Symbios 53C825 (NOT 825A)= is > > > supported by FreeBSD? I saw mention of the 825A in the hardware list= , but > > > not a plain 825. > >=20 > > The NCR53C825 is supported by FreeBSD since at least 6 years. :) > > (Was a NCR product at this time) > >=20 > > Anyway, the hardware list needs to be updated for newer chips (i.e. I > > should have updated it). > >=20 > > At the moment and at least for RELENG4: > >=20 > > - The SYM driver supports all 53C8XX A and not-A chips and also the > > 53C1010-33, 53C1010-66, 53C1010R, 53C1000 and 53C1000R > > Ultra3 controllers. > >=20 > > - The NCR driver supports all 53C8XX A and not-A chips, may-be excepted= =20 > > the 53C875A. > > (Adding the 53C875A chip description will make it support the chip). > >=20 > > For Ultra3 controllers under RELENG3, I would suggest to use an updated > > SYM driver, as the sym-2.1.6 available from > > http://people.freebsd.org/~groudier/ > >=20 > > G=E9rard. >=20 >=20 > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-alpha" in the body of the message To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.4.10.10102111838090.433-100000>