Re: mgag200: Wrong looking code in mga_probe_vram

From: Mathieu Larouche <mathieu.larouche@matrox.com>
To: Michal Srb <msrb@suse.com>, dri-devel@lists.freedesktop.org
Cc: Dave Airlie <airlied@redhat.com>, Matt Russo <Matt.Russo@matrox.com>
Subject: Re: mgag200: Wrong looking code in mga_probe_vram
Date: Wed, 18 Oct 2017 07:49:27 -0400	[thread overview]
Message-ID: <59E73FC7.2010808@matrox.com> (raw)
In-Reply-To: <11691338.pzimoKDfCV@sheogorath.suse.cz>

On 11/10/2017 5:27 AM, Michal Srb wrote:
> Hi,
>
> While reading the mgag200 driver I noticed something that looks wrong.
>
> In mgag200_main.c mga_probe_vram function:
>
>      orig1 = ioread8(mem + offset);
>      orig2 = ioread8(mem + offset + 0x100);
>
>      ...
>
>      iowrite16(orig1, mem + offset);
>      iowrite16(orig2, mem + offset + 0x100);
>
> I do not understand how exactly the probing works, but ioread8 followed by
> iowrite16 does not look correct.
>
> The corresponding function in xf86-video-mga UMS driver is MGACountRam. It
> does the equivalent of read 16, write 16. So in case you can confirm that the
> kernel driver is correct, it would probably mean the user space driver is
> wrong.
>
> Thanks,
> Michal Srb
>

Hi,

I agree with you, there's an issue there. The first two lines should be 
ioread16(). The goal of this routine is to detect the amount of memory 
available since it varies in some products. Basically, it stores the 
initial value, write a value and read it back, then it restores the 
initial value. In this case, only half of the value will be restored.

Mathieu
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel