linux-mips
[Top] [All Lists]

Re: Au1550 - Problem access Shared memory from PCI card

To: Alexey Shinkin <alexshinkin@hotmail.com>
Subject: Re: Au1550 - Problem access Shared memory from PCI card
From: Pete Popov <ppopov@embeddedalley.com>
Date: Thu, 16 Sep 2004 00:50:22 -0700
Cc: linux-mips@linux-mips.org
In-reply-to: <BAY15-F40hAOylhgnFZ0009083e@hotmail.com>
Organization: Embedded Alley Solutions, Inc
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <BAY15-F40hAOylhgnFZ0009083e@hotmail.com>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mozilla Thunderbird 0.7.3 (X11/20040803)


After I recompiled the driver and the application for MIPS platform I got the following:
1. The driver on the MIPS board (host) finds the PCI card (target)
2. I download an executable code into target and start target application (application starts, debug info is written to target memory). 3. Target starts to communicate with host using shared memory,allocated on host, (target accesses the memory through PCI).

After that I get the following problem:

4. Target writes some values into Host's shared memory and generates interrupt on Host. 5. Host catches the interrupt and in interrupt handler reads the values, written by target.

So the host at this point reads the correct values written by the target?

6. Host writes some replay to the shared memory, generates PCI interrupt on Target.

When you say "host", is this the user application that is writing to the shared memory after remmaping it, or is the host driver itself writing to the shared memory from kernel space?

7. Target gets interrupt but can not see the latest values, written by Host. Reading the shared memory several times doesn' t help.

Looks like target reads values that are cached somewhere . I tried to insert au_sync(), au_sync_delay(), flush_cache_all() on Host side after writting values - nothing helps.


On x86 this approach works fine.

What could be the reason? Could this be hardware problem or can be fixed by fine-tuning

of caching parameters or PCI controller ?

There's no reason why this shouldn't work and it doesn't smell like a hardware problem.

Pete

<Prev in Thread] Current Thread [Next in Thread>