From mboxrd@z Thu Jan 1 00:00:00 1970 Message-Id: <200007301031.LAA32665@hyperion.valhalla.net> Date: Sun, 30 Jul 2000 11:31:33 +0100 Subject: Re: Stealth port problems From: "Iain Sandoe" To: Giuliano Pochini , linuxppc-dev@lists.linuxppc.org Mime-version: 1.0 Content-type: text/plain; charset="US-ASCII" Sender: owner-linuxppc-dev@lists.linuxppc.org List-Id: Sun, Jul 30, 2000, Giuliano Pochini wrote: > Giuliano Pochini wrote: >> >> > >Hi! I have troubles with my stealth port on a G3-blue. When it runs at >> > 115200 >> > >ifconfig >> > >shows a lot of overruns. At 57600 it works fine, but it's slow :-(( > > The problem is definitely X. I made a lot of tests: with dma and non-dma > serial driver, X accelerated, non accelerated, X bare.bone (started with > xinit). The problem is X. When X is running the serial ports looses packets > :-(( > Am I the only one who is having this problem ??? Why does X cause overruns ?? If you can try 2.2.17pre10ben2 or pre13ben2 (linuxcare linux-pmac-benh) then you can apply the IRQ latency test and see if it is that which is blocking you. Scan the dev list for "[ANN} IRQ". It is possible (there was a discussion in the ANN thread about fb IRQ blocks) - X may aggravate this. I have seen 2ms of IRQ blocking on the Lombard (using IDE CDROM) and around 450-750 us of IRQ blocking on my other machines (SCSI). [when heavily loaded]. The RX buffer in an 85230 (which is actually the chip - IIRC - an enhanced 8530) is 8 characters deep - so any more that about 700 us IRQ block will cause a drop out at 115200 continuous... So it looks like music might not be the only pressure to tune the IRQ latencies :-) Note that an x86 kernel guru said on the lad list "a well tuned kernel should not exhibit more than about 60 us worst-case IRQ block" we have a ways to go to get to that. Iain. ** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/