Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from phobos.denx.de (phobos.denx.de [85.214.62.61]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 3EEADC433EF for ; Sat, 19 Feb 2022 20:37:20 +0000 (UTC) Received: from h2850616.stratoserver.net (localhost [IPv6:::1]) by phobos.denx.de (Postfix) with ESMTP id E68A083C43; Sat, 19 Feb 2022 21:37:17 +0100 (CET) Authentication-Results: phobos.denx.de; dmarc=fail (p=none dis=none) header.from=xs4all.nl Authentication-Results: phobos.denx.de; spf=pass smtp.mailfrom=u-boot-bounces@lists.denx.de Received: by phobos.denx.de (Postfix, from userid 109) id 7A67E83C45; Sat, 19 Feb 2022 21:37:15 +0100 (CET) Received: from sibelius.xs4all.nl (sibelius.xs4all.nl [83.163.83.176]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)) (No client certificate requested) by phobos.denx.de (Postfix) with ESMTPS id F1D6D83C41 for ; Sat, 19 Feb 2022 21:37:11 +0100 (CET) Authentication-Results: phobos.denx.de; dmarc=fail (p=none dis=none) header.from=xs4all.nl Authentication-Results: phobos.denx.de; spf=fail smtp.mailfrom=mark.kettenis@xs4all.nl Received: from localhost (bloch.sibelius.xs4all.nl [local]) by bloch.sibelius.xs4all.nl (OpenSMTPD) with ESMTPA id 91bb530c; Sat, 19 Feb 2022 21:37:10 +0100 (CET) Date: Sat, 19 Feb 2022 21:37:10 +0100 (CET) From: Mark Kettenis To: Simon Glass Cc: kettenis@openbsd.org, u-boot@lists.denx.de In-Reply-To: (message from Simon Glass on Fri, 18 Feb 2022 17:21:13 -0700) Subject: Re: [PATCH] drivers: serial: Make sure we really return a serial device References: <20220205231018.3050-1-kettenis@openbsd.org> Message-ID: X-BeenThere: u-boot@lists.denx.de X-Mailman-Version: 2.1.39 Precedence: list List-Id: U-Boot discussion List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: u-boot-bounces@lists.denx.de Sender: "U-Boot" X-Virus-Scanned: clamav-milter 0.103.5 at phobos.denx.de X-Virus-Status: Clean > From: Simon Glass > Date: Fri, 18 Feb 2022 17:21:13 -0700 Hi Simon, > Hi Mark, > > On Mon, 7 Feb 2022 at 14:20, Mark Kettenis wrote: > > > > > From: Simon Glass > > > Date: Mon, 7 Feb 2022 13:22:22 -0700 > > > > > > Hi Mark, > > > > > > On Sat, 5 Feb 2022 at 16:10, Mark Kettenis wrote: > > > > > > > > The stdout-path property in the device tree does not necessarily > > > > point at a serial device. The code that binds the device if it > > > > isn't marked to be bound before relocation does not check whether > > > > the device really is a serial device. So check that its uclass is > > > > UCLASS_SERIAL before probing it. > > > > > > > > Signed-off-by: Mark Kettenis > > > > --- > > > > drivers/serial/serial-uclass.c | 3 ++- > > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > This would be better as an assert() to avoid code bloat. Under what > > > circumstances was this wrong? > > > > I'm passing a debive tree to U-Boot that has stdout-path set to > > "/chosen/framebuffer" to indicate that output should go to the > > framebuffer instead of the serial port. This is a node with a > > "simple-framebuffer" compatible, which means the list_bind_fdt() call > > binds the simple_video (simplefb.c) driver and stores a pointer to its > > struct udevice into devp. I've verified that the uclass is indeed > > UCLASS_VIDEO at that point with my device tree. > > > > Since the function returns 0, the caller of this function thinks we > > returned a valid UCLASS_SERIAL device and when u-boot starts using it > > as such it crashes (without printing anything since there is no > > console output device yet). > > > > So an assert won't work. We should simply return an error an run the > > fallback code below that looks for the serial port with the specified > > index. That way U-Boot continues to work as expected (with output on > > both the serial port and the framebuffer). But once my OS is running, > > it knows to pick the framebuffer console and all's fine. > > > > I haven't looked at the binary growth, but it can't be much. But if > > you think it is a problem, maybe we should make this code that binds > > the console driver optional? > > OK I see. It likely isn't much, but in SPL it is still growth. Worth > checking. before: $ size u-boot text data bss dec he 442903 27700 13448 484051 762d3 after: $ size u-boot text data bss dec hex 442925 27700 13448 484073 762e9 So that's 22 bytes. Thew growth in serial-uclass.o is a bit smaller; only 16 byes. > Can you perhaps create a 'console' alias so that the line above finds > the correct one? Probably. But none of the Linux device trees provide a 'console' alias so this seems to be a U-Boot convention. And it doesn't actually fix the bug; it just works around it. > But I'm a bit unsure of how you get the serial console to work if > serial_check_stdout() fails? Is the serial driver not marked for > pre-reloc use? There is something odd going on. The serial driver is marked for pre-reloc use. When serial_check_stdout() fails the code at the tail end of serial_find_console_or_panic() runs and finds the serial device at index 0 and uses that. The "something odd" that's going on here is that I set "/chosen/stdout-path" to point to the framebuffer instead of a serial port. But nowhere does the device tree specification say that "stdout-path" has to point at a serial device. So U-Boot shouldn't crash when this isn't the case. Cheers, Mark