Message-ID: <D107B6A48391D111A7B500805FF5F5FFF81D63@basil.stone.ndsuk.com>
From: "Isaacson, Tom" <TIsaacson AT ndsuk DOT com>
To: "'djgpp AT delorie DOT com'" <djgpp AT delorie DOT com>
Subject: Physical to virtual memory map problems
Date: Thu, 21 Jan 1999 13:51:48 -0000
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2232.9)
Content-Type: text/plain;
	charset="iso-8859-1"
Reply-To: djgpp AT delorie DOT com

I've just inherited some old code from a previous employee, but I can't get
it to work. It's a test program, running under DOS, which initialises a PCI
card and then attempts to perform a series of tests on it. The PCI
initialisation code works fine and the card is configured correctly, but
when I map the memory on the card the writes don't have any effect, and the
reads return -1. I know the card works because we have a Windows NT driver
for it that works fine.

Here is the memory mapping code for the first buffer of 512 bytes:

	// Map physical address for S0
	meminfo.handle = 0x0;
	meminfo.size = local_reg_range;	// 512 bytes
	meminfo.address = s0_base_add;	// base address

	// Allocate descriptor
	local_reg_selector_n = __dpmi_allocate_ldt_descriptors(1);
	if (local_reg_selector_n == -1)
	{
		printf("Allocate local_reg_selector_n FAILED\n");
	}

	// This perform the Physical to Virtual memory map !!
	if (__dpmi_physical_address_mapping(&meminfo) == -1)
	{
		printf("Physical to linear address map of Internal Control
Reg FAILED\n");
	}

	// Set the limit
	if (__dpmi_set_segment_limit(local_reg_selector_n, local_reg_range -
1) == -1)
	{
		printf("Set segment limit of local_reg_selector_n
FAILED\n");
	}

	// Set the base address
	if (__dpmi_set_segment_base_address(local_reg_selector_n,
meminfo.address) == -1)
	{
		printf("Set base address of local_reg_selector_n FAILED\n");
	}

This is repeated for a second buffer of 4MB. The reads are done like this:

	data_count_n = 0;
	for (i = 0; i < 0x200; i += 4)
	{
		temp = _farpeekl(local_reg_selector_n, i);
		data_ula[data_count_n++] = temp;
	}

The writes use _farpeekl. Can anyone see anything wrong with this method?
The only other problem I can think of is the original program was written to
use the motherboard's PCI bus and the computer we are using now has a PCI
extender chip, and DPMI cannot handle memory over the extended PCI bus, but
this seems unlikely.

Tom Isaacson
mailto:TIsaacson AT ndsuk DOT com