From owner-freebsd-hackers  Mon Mar 10 15:15:58 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id PAA06040
          for hackers-outgoing; Mon, 10 Mar 1997 15:15:58 -0800 (PST)
Received: from main.cyclades.com (root@main.cyclades.com [208.138.19.2])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id PAA06035
          for <hackers@freebsd.org>; Mon, 10 Mar 1997 15:15:56 -0800 (PST)
Received: from main.cyclades.com ([208.138.19.230]) by main.cyclades.com
	 with smtp id m0w4EI5-000SizC
	(Debian Smail-3.2 1996-Jul-4 #2); Mon, 10 Mar 1997 15:15:37 -0800 (PST)
Message-ID: <33249768.7287@cyclades.com>
Date: Mon, 10 Mar 1997 15:21:12 -0800
From: Marcio Saito <marcio@cyclades.com>
Reply-To: marcio@cyclades.com
Organization: Cyclades Corporation
X-Mailer: Mozilla 3.0 (Win95; I)
MIME-Version: 1.0
To: Brian McGovern <bmcgover@cisco.com>
CC: Mark Tinguely <tinguely@plains.nodak.edu>, hackers@freebsd.org
Subject: Re: Question on mapping PCI memory...
References: <199703102125.QAA16038@bmcgover-pc.cisco.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Brian McGovern wrote:
> 
> According to the docs I have, they define the register set as:
> 
> 0x00    Vendor ID (bits 0-15), device id (bits 16-31)
> 0x04    Status register/command register
> 0x08    Rev ID (bits 0-7), Class code (bits 8-31)
> 0x10    Base Address 1 (Memory address for 9060 controller chip)
> 0x14    Base Address 2 (Memory address for 512KB memory window)
> 0x18    Base Address 3 (unused)
> 0x3C    IRQ (0-7)

This is defined for each particular PCI device.

For the PLX9060 (the PCI bridge chip we use), we have:

0x10 - 9060 registers (memory mapped)
0x14 - 9060 registers (I/O mapped)
0x18 - 512kb memory window.

(the PXL chip registers can be accessed either through memory
or I/O addressing).
 
> I then have the lines:
> 
> #define CYZ_PCI_BASE_ADDR1 0x10   /* 9060 Controller memory address */
> #define CYZ_PCI_BASE_ADDR2 0x14   /* Memory Window */
> #define CYZ_PCI_IRQ        0x3C   /* IRQ */
> 
> to make the registers easier to work with. I then have a code chunk that
> does:
> 
> vm_offset_t paddr, vaddr;
> 
> if (pci_map_mem(config_id, CYZ_PCI_BASE_ADDR2, &vaddr, &paddr) == 0)
>   {
>     printf("Error mapping shared memory\n");
>     return;
>   }
> 
> When run, it complains with: pci_map_mem failed: device's memrange
> 0xf880-0xf8ff is incompatible with its bridge's memrange 0x2000000-0xffffffff.

0xf880 is probably an I/O address.

0x18 is the offset associated to the memory window.

Marcio.

> Also, the routine keeps running, so should I think that pci_map_mem returning
> 0 on error (as according to pcivar.h) may be incorrect, or is this just
> considered a warning?
> 
> Also, based on this message, is it possible that these pointers need to
> be shifted left by 16 bits, so it fits in the window?
> 
> In any event, when I try to write to the memory, it has a parity error, and
> panics. I'm assumeing this is because the memory range really is invalid.
> 
> Any suggestions?
>         -Brian

-- 

----------------------------------------------------------------------
Marcio Saito                             e-mail:  marcio@cyclades.com
SW Engineer/Product Manager               phone: (510)770-9727 x214
Cyclades Corporation                        fax: (510)770-0355
----------------------------------------------------------------------