linux-kernel - Re: [PATCH v2 00/10] usb: ulpi: remove "dev" field from struct ulpi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20160815120258.GA11370@kuha.fi.intel.com>
Date:	Mon, 15 Aug 2016 15:02:58 +0300
From:	Heikki Krogerus <heikki.krogerus@...ux.intel.com>
To:	Tal Shorer <tal.shorer@...il.com>
Cc:	gregkh@...uxfoundation.org, linux-usb@...r.kernel.org,
	linux-kernel@...r.kernel.org, balbi@...nel.org
Subject: Re: [PATCH v2 00/10] usb: ulpi: remove "dev" field from struct
 ulpi_ops

Hi,

Please forgive me for taking so long to reply. I just returned from
paternal leave.

On Mon, Aug 01, 2016 at 09:15:48PM +0300, Tal Shorer wrote:
> struct ulpi_ops is defined as follows:
> 
> struct ulpi_ops {
>         struct device *dev;
>         int (*read)(struct ulpi_ops *ops, u8 addr);
>         int (*write)(struct ulpi_ops *ops, u8 addr, u8 val);
> };
> 
> Upon calling ulpi_register_interface(), the struct device argument is
> put inside the struct ulpi_ops argument's dev field. Later, when
> calling the actual read()/write() operations, the struct ulpi_ops is
> passed to them and they use the stored device to access whatever
> private data they need.
> 
> This means that if one wishes to reuse the same oprations for multiple
> interfaces (e.g if we have multiple instances of the same controller),
> any but the last interface registered will not operate properly (and
> the one that does work will be at the mercy of the others to not mess
> it up).
> 
> I understand that barely any driver uses this bus right now, but I
> suppose it's there to be used at some point. We might as well fix the
> design here before we hit this bug.
> 
> This series fixes this by passing the given struct device directly to
> the operation functions via ulpi->dev.parent in ulpi_read() and
> ulpi_write(). It also changes the operations struct to be constant
> since now nobody has a reason to modify it.

If there are multiple instances of the same controller, the controller
driver just needs to provide a separate ops for every one of them.
This isn't really a problem as you describe it. But I'm not against
API improvements even if they don't fix anything. I'll test these
tomorrow.


Thanks,

-- 
heikki