[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4A81683D.2030408@gefanuc.com>
Date: Tue, 11 Aug 2009 13:46:53 +0100
From: Martyn Welch <martyn.welch@...anuc.com>
To: "Emilio G. Cota" <cota@...ap.org>
CC: Greg K-H <gregkh@...e.de>, linux-kernel@...r.kernel.org,
devel@...uxdriverproject.org,
Sebastien Dugue <sebastien.dugue@...l.net>
Subject: Re: [patch 2/5] Staging: vme: add VME userspace driver
Emilio G. Cota wrote:
> Martyn Welch wrote:
>
>> suggest if you want something more complex that allows the user to just
>> pick a location/size and not worry about windows at all
>>
>
> That's exactly the whole point. I think each bridge should manage
> its resources; putting this on the upper layer would mean the
> layer should have a mechanism of 'discovering' what the bridge
> can/can't do. Anyway this could be revisited later.
> I'm preparing a patch for this.
>
I disagree. The bridge drivers should register their resources with the
core. The core, or a layer above it, can control how those resources are
used. This moves the complexity you want for managing the windows to a
level that will work on all underlying drivers rather than having to be
written explicitly for each one. The mechanism I have provided does this
discovery.
>
>>> - Most accesses are 32-bit accesses. Treating all of them
>>> as 64-bit accesses would decrease performance for most of
>>> them--which happen to be 32-bit.
>>>
>> I'm not - I'm storing them as 64-bit values, which they are, in the
>> structures used in *software*. These are then split *when* a write to
>> the hardware registers is required. Similarly, when the registers are
>> occasionally read they are combined and stored as a 64-bit value. This
>> simplifies all *software* checking and manipulation. By storing these as
>> 2 32-bit values every driver that uses the VME core will need to convert
>> pci addresses, vme addresses and counts to 2 32-bit values. That is
>> madness.
>>
>
> I agree with you on that's painful for doing 64-bit accesses.
> However I'm still not convinced on the performance side (I mean
> software), since most of the time the upper 32bits will be empty.
> Will have a look though.
>
>
Regardless of the contents of the upper 32 bits, they will need to be
checked. For example when ensuring a window offset and size fit within a
given address space. To do this the bound needs to be calculated.
To do this with 2 32-bit values. add the lower window offset and size,
which could lead to an over-flow of the lower 32-bit value, which needs
to be added to the upper 32-bit value, which is the sum of the upper
window offset and size. Then potentially check both the upper and lower
values (thinking of 16-bit address range here) to ensure they are below
the maximum address for the address range.
Compared to an addition of the window offset and size and a comparison
to the maximum address for the address range when using one 64-bit value.
Clearly neither of the above basic outlines check that the combination
has wrapped over 64 bits, but this would affect both equally.
So, which is more complex?
Martyn
> Cheers,
> E.
>
--
Martyn Welch MEng MPhil MIET (Principal Software Engineer) T:+44(0)1327322748
GE Fanuc Intelligent Platforms Ltd, |Registered in England and Wales
Tove Valley Business Park, Towcester, |(3828642) at 100 Barbirolli Square,
Northants, NN12 6PF, UK T:+44(0)1327359444 |Manchester,M2 3AB VAT:GB 927559189
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists