Re: data abort when run 'dhcp'

From: Tom Rini <trini@konsulko.com>
To: Miquel Raynal <miquel.raynal@bootlin.com>
Cc: qianfan <qianfanguijin@163.com>,
	Heinrich Schuchardt <xypron.glpk@gmx.de>,
	u-boot@lists.denx.de
Subject: Re: data abort when run 'dhcp'
Date: Thu, 20 Jul 2023 14:34:52 -0400	[thread overview]
Message-ID: <20230720183452.GM1670569@bill-the-cat> (raw)
In-Reply-To: <20230720183917.74e1a46c@xps-13>

[-- Attachment #1: Type: text/plain, Size: 1724 bytes --]

On Thu, Jul 20, 2023 at 06:39:17PM +0200, Miquel Raynal wrote:
> Hello,
> 
> qianfanguijin@163.com wrote on Fri, 25 Mar 2022 18:04:46 +0800:
> 
> > It's very strange. And I can't detect it's a bug of usb or dlmalloc.
> > 
> > 1. Starting u-boot and dhcp via am335x's ethernet(cpsw driver), it's ok.
> > 
> > 2. Starting u-boot and dhcp via am335x's usb net, data abort.
> > 
> > 3. start fastboot, and CTRL C right now, dhcp via am335x's usb net, it's ok.
> 
> I am sorry to re-open a thread that is one year old but this is
> still an open bug. The BBB is affected. In particular the BBBW
> because there is no Ethernet connector, which makes the Eth-over-USB
> emulation even more important. All U-Boots since 2021 are affected:
> spurious data aborts, usually at the end of network interactions (tftp,
> ping). I could not bisect it because the boot was deeply broken as
> well on a significant range of commits :-/.
> 
> On my side I narrowed it down to an env update which fails in malloc as
> well. If I comment the env update, it fails a bit later. It really
> looks like a stack corruption which is either related to the Ethernet
> USB gadget or the USB controller driver itself. Network transfers on
> the BBBW using regular Ethernet does not trigger any error.
> 
> I also observe the very strange "fix" mentioned above: starting and
> killing fastboot makes all tftp pass... If anyone has more details to
> share, or perhaps a subsequent thread giving more details, I would
> really like to see this fixed upstream, I suppose I am not the only one
> :-)

What happens if you increase the malloc pool from say 32MB (current
value, 0x2000000) to 64MB (so 0x4000000) ?

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]