MIME-version: 1.0
Content-transfer-encoding: 7BIT
Content-type: TEXT/PLAIN; charset=US-ASCII
Date: Mon, 20 Jun 2011 18:31:13 -0400 (EDT)
From: Nicolas Pitre <nico@fluxnic.net>
To: Alan Stern <stern@rowland.harvard.edu>
Cc: Alexander Holler <holler@ahsoftware.de>, Arnd Bergmann <arnd@arndb.de>,
        linux-arm-kernel@lists.infradead.org, linux-usb@vger.kernel.org,
        gregkh@suse.de, lkml <linux-kernel@vger.kernel.org>,
        Rabin Vincent <rabin@rab.in>
Subject: Re: [PATCH] USB: ehci: use packed,aligned(4) instead of removing the
 packed attribute
In-reply-to: <Pine.LNX.4.44L0.1106201700500.2113-100000@iolanthe.rowland.org>
Message-id: <alpine.LFD.2.00.1106201824320.2142@xanadu.home>
References: <Pine.LNX.4.44L0.1106201700500.2113-100000@iolanthe.rowland.org>
User-Agent: Alpine 2.00 (LFD 1167 2008-08-23)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2971
Lines: 84

On Mon, 20 Jun 2011, Alan Stern wrote:

> On Mon, 20 Jun 2011, Nicolas Pitre wrote:
> 
> > On Mon, 20 Jun 2011, Alan Stern wrote:
> > 
> > > On Mon, 20 Jun 2011, Alexander Holler wrote:
> > > 
> > > > I see it that way: packed is needed to be sure that at least for struct 
> > > > ehci_regs there are no padding bytes inbetween the members.
> > > 
> > > But is it _really_ needed?
> > > 
> > > > It might 
> > > > work without, but that depends on the compiler (-version, architecture, 
> > > > whatever).
> > > 
> > > Have there _ever_ been _any_ combinations of compiler, version, 
> > > architecture, whatever, that had unwanted padding bytes in this 
> > > structure?
> > 
> > This can be determined by simple code inspection.
> > 
> > If you must have struct members which are not aligned to their natural 
> > size then you need __packed.  Example:
> > 
> > struct foo {
> > 	u8  a;
> > 	u16 b;
> > 	u32 c;
> > 	u64 d;
> > };
> > 
> > Without __packed, there will be padding between a and b, and between c 
> > and d.
> 
> One byte of padding between a and b is enough.  No more is needed, and 
> the compiler would have to be pretty stupid to add anything else.

Obviously, my mistake.  I meant to make c a u16 too but failed to 
correct the example before posting.
> >  If the order of the members in this struct were reversed, then 
> > everything would be naturally aligned and no padding between members 
> > would be inserted.
> > 
> > The size of structures is normally rounded up with padding to the size 
> > of the largest basic element it contains.  Example:
> > 
> > struct foo {
> > 	u64 a;
> > 	u8 b;
> > };
> > 
> > Here sizeof(struct foo) would return 16, even if the actual content 
> > occupies 9 bytes only.  That's because the largest basic element is u64 
> > i.e. 8 bytes.  Normally this trailing padding is not an issue, unless 
> > you have an array of such a struct or if it is a member of another 
> > struct.  If you want to get rid of that padding, you need to use 
> > __packed again (which of course would make all subsequent instances of 
> > that structure in your array completely misaligned too).
> > 
> > Two odd exceptions with the old ABI on ARM:
> > 
> > - The alignment of a 64-bit value is always 4 bytes not 8.
> > 
> > - The size of all structures are always rounded up to a 4-byte boundary, 
> >   irrespective of their content.
> > 
> > If you fall into none of the above issues, then you don't need any 
> > __packed, period.
> 
> We don't fall into any of these cases, and therefore as you say, we
> don't need packed.  Arnd and I have both explained this.  So why do you 
> keep arguing that we do need it?

Please show me where I keep arguing that you need it?


Nicolas
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/