From: Marek Vasut Date: Tue, 26 May 2020 14:42:08 +0200 Subject: [RFC PATCH] imx: skip unused compatible strings in drivers In-Reply-To: References: <20200525202429.2146-1-agust@denx.de> <4c07f210-eab5-b456-fb1c-3f1b1eb71e0b@web.de> <20200526135829.7f5637eb@crub> Message-ID: <19ca7297-b7a6-62f5-e88f-e7f981386ffb@denx.de> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de On 5/26/20 2:40 PM, Peng Fan wrote: >> Subject: Re: [RFC PATCH] imx: skip unused compatible strings in drivers >> >> On Tue, 26 May 2020 13:25:00 +0200 >> Marek Vasut marex at denx.de wrote: >> >>> On 5/26/20 1:12 PM, Soeren Moch wrote: >>>> On 25.05.20 22:24, Anatolij Gustschin wrote: >>>>> Converting to DM increases binary size and breaks building some >>>>> boards (i.e. tbs2910, gcc 9.2). The approach to address this issue >>>>> via cutting off unused properties/nodes in device tree via custom >>>>> u-boot.dtsi was not welcome, even if the affected boards do not >>>>> pass the built-in device tree to the kernel. >>>> But there is CONFIG_OF_REMOVE_PROPS="dmas dma-names >> interrupt-parent >>>> interrupts interrupts-extended interrupt-names interrupt-map >>>> interrupt-map-mask" in tbs2910_defconfig. So removing properties is >>>> already done via this config option. Removing nodes is not >>>> implemented (yet? AFAIK). But would be welcome if someone implements >>>> this in a similar way, I think. >>> >>> Why don't we turn this the other way around -- put the { compat string >>> , driver } tuple list into a U-Boot linker list, then filter this >>> linker list against a DT which will be used for the platform (what if >>> there are multiple DTs?), and then scrub all unreferrenced symbols. >> >> This might be a topic again when new bloat causes build breakage again. For >> now I've succeeded to build the tbs2910 board with default CI tool chain >> without this patch (by making DM_VIDEO parts optional). Stripping unused >> compatible entries doesn't save much here (around 800 bytes). So the >> question is if it is worth the effort to implement such scrubbing. > > I vote for this. Whether u-boot or SPL, currently the size is larger and larger. Can you send a patch prototyping this ?