Date: Tue, 13 Feb 1996 17:30:13 +0100 From: se@zpr.uni-koeln.de (Stefan Esser) To: "Justin T. Gibbs" <gibbs@freefall.freebsd.org> Cc: Richard J Kuhns <rjk@sparcmill.grauel.com>, freebsd-stable@freebsd.org Subject: Re: One problem && one question Message-ID: <199602131630.AA09319@Sysiphos> In-Reply-To: "Justin T. Gibbs" <gibbs@freefall.freebsd.org> "Re: One problem && one question" (Feb 13, 8:10)
next in thread | previous in thread | raw e-mail | index | archive | help
On Feb 13, 8:10, "Justin T. Gibbs" wrote:
} Subject: Re: One problem && one question
} >On Feb 13, 9:39, Richard J Kuhns wrote:
} >} Subject: Re: One problem && one question
} >} Justin T. Gibbs writes:
} >} > Patch round two:
} >}
} >} Sorry, still no cigar. I rebooted this machine several times this morning,
} >} trying to copy the boot messages. I guess I'd better set up a minimal
} >} configuration to test this so the bt messages will remain on-screen.
} >}
} >} At any rate, I believe it looked like:
} >}
} >} <Buslogic id info> int a irq 10 pci 0:13
} >} mapreg[10] type 1 addr 0x00006000 size ffff0004
} >} bt0: Invalid base address
} >
} >The region size calculation seems broken!
}
} The driver only cares about the I/O port, which isn't reported above. I've
} seen boot -v output from bt946 cards before, and they at least report a
} sane I/O port address like 0x330 (when listed as a device without a driver)
The BIOS is free to assign any port address, and
0x330 seems to be reasonable to emulate the ISA
version of the card ...
} - never looked at the size, but it should be 4. I wouldn't bother hacking
} up the memory size calculation just for this card since we won't even use
} it. I just sent him an updated patch, that corrects some mistakes I made
The PCI code uses it as a parameter to the pci_map_io()
call, but since this is not about VM, it doesn't have
much of an effect ...
} in the last one, but I did notice something odd in the PCI code:
}
} if (bootverbose) {
} for (reg=PCI_MAP_REG_START; reg<PCI_MAP_REG_END; reg+=4) {
} data = pcibus->pb_read (tag, reg);
} if ((data&~7)==0) continue;
} switch (data&7) {
}
} case 1:
} case 5:
} printf (" map(%x): io(%lx)\n",
} reg, data & ~3);
} break;
}
} I guess that everything must be quad-word aligned so that is why you
} & ~3? Is this in the spec even for I/O ports?
In PCI there are a number of map types, and those are
encoded in those (assumed to be '0') bits. I.e. the type
1 encoding has a '01' in the two low order bits, which
have to be masked out ...
The problem with the wrong size being reported is caused
by the fact, that the bt946c has only a 16 bit port base
register. And for this reason, it returns 0x0000fffd if
a 0xffffffff had been written ...
The 'd' is a 'c' (i.e. -4) plus the mapping type 1 code.
Since the PCI spec defines type 1 as a 32bit I/O port
mapping (IIRC) the 946 really should return 0xfffffffd.
It has no right to ignore the upper 16bits.
But since ISA only supports 16 bits of port address
anyway, it doesn't appear to make much of a difference
to make the code always ignore those upper half for
port mappings ...
Regards, STefan
--
Stefan Esser, Zentrum fuer Paralleles Rechnen Tel: +49 221 4706021
Universitaet zu Koeln, Weyertal 80, 50931 Koeln FAX: +49 221 4705160
==============================================================================
http://www.zpr.uni-koeln.de/~se <se@ZPR.Uni-Koeln.DE>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199602131630.AA09319>