[PATCH 1/2] mtd: atmel_nand: remove unneeded ifdef CONFIG_OF
From: Olof Johansson <hidden>
Date: 2013-09-16 22:28:38
Also in:
linux-devicetree
Guys, On Fri, Sep 13, 2013 at 11:21 AM, Brian Norris [off-list ref] wrote:
+ devicetree at vger.kernel.org On Fri, Sep 13, 2013 at 05:28:13PM +0800, Josh Wu wrote:quoted
On 9/12/2013 7:02 AM, Brian Norris wrote:quoted
On Tue, Sep 03, 2013 at 05:20:27PM +0800, Josh Wu wrote:quoted
Since following commit f3b391425d21e6138e57b2432d91134e0bc2b975...quoted
quoted
quoted
(of_mtd: Add no-op stubs to support CONFIG_OF=n) implements the stub for CONFIG_OF=n. Now we can safely remove all CONFIG_OF in atmel_nand. (Thanks to Ezequiel Garcia's for this protype)I'm not quite so sure about this patch, as I was about the pxa3xx patch. With pxa3xx, the compiler can easily tell that pxa3xx_nand_probe_dt() will return 0 without doing anything in the !CONFIG_OF case (and so will likely remove the dead code), so it's no benefit to have the #ifdef. But in this driver, the atmel_of_init_port() function can't be trivially determined to do nothing (and in fact, it does something in either CONFIG_OF=y or =n case). It's only protected by the 'if (pdev->dev.of_node)' check, which the compiler can't predict.I understand your concern here.quoted
So, I don't know if we should remove the #ifdef at the expense of likely significantly larger code. I won't protest, but I won't merge it yet either. Perhaps others have better ideas, or perhaps you can find a good way to work around this -- e.g., check the of_* helpers for -ENOSYS early in atmel_of_init_port()?
Can we please get less fumbling around on this and just merge a fix, please? You guys have broken the PXA3xx builds for the whole merge window, while there's been a patch sitting in linux-next with _exactly_ this contents since August 30, committed by David. If this is't fixed within the next few days I'll just pick that patch up and include it in our next batch of arm-soc fixes. This is ridiculous. -Olof