Re: [PATCH v1 04/10] bootm: add kexec ELF support

[Peter Mamonov <pmamonov@gmail.com>]

On Thu, May 17, 2018 at 09:01:06AM +0200, Sascha Hauer wrote:
> On Thu, May 17, 2018 at 06:52:10AM +0200, Oleksij Rempel wrote:
> > Am 16.05.2018 um 23:34 schrieb Peter Mamonov:
> > > Hi!
> > > 
> > > On Wed, May 16, 2018 at 06:42:27PM +0200, Oleksij Rempel wrote:
> > >> From: Antony Pavlov <antonynpavlov@gmail.com>
> > >>
> > >> +int kexec_load_bootm_data(struct image_data *data)
> > >> +{
> > >> +	int ret;
> > >> +	struct kexec_info info;
> > >> +	char *cmdline;
> > >> +	const char *t;
> > >> +	size_t tlen;
> > >> +	size_t fsize;
> > >> +	char initrd_cmdline[40];
> > >> +	int padded = 0;
> > >> +
> > >> +	memset(&info, 0, sizeof(info));
> > >> +
> > >> +	initrd_cmdline[0] = 0;
> > >> +
> > >> +	ret = kexec_load_one_file(&info, data->os_file);
> > >> +	if (IS_ERR_VALUE(ret)) {
> > >> +		pr_err("Cannot load %s\n", data->os_file);
> > >> +		return ret;
> > >> +	}
> > > 
> > > There is a potential problem here, which I actually hit some time ago. The 
> > > following code places kernel arguments right after os image. This is perfectly 
> > > fine in case of vmlinuX. However, if one boots a vmlinuZ image, there is no 
> > > easily available knowledge of where the decompressed image will reside. In my 
> > > case vmlinux' BSS overlapped with DTB and kernel cmdline segments and was 
> > > zeroed at linux startup. This was fixed by adding an empty 4k segment at 128M, 
> > > so further segments were allocated beyond 128M, far enough from the kernel 
> > > lair:
> > > 
> > > 	+       /* FIXME: allocate 4k segment @ 0x8000000 (128M), so further
> > > 	+        * segments will be allocated beyond this address.  This prevents
> > > 	+        * kernel parameters from being overwritten by the kernel startup code.
> > > 	+        */
> > > 	+       add_segment(&info, (void *)CKSEG0ADDR(0), 4 << 10, 0x8000000, 4 << 10);
> > > 
> > > However this is an ad-hoc solution and, probably, find_unused_base() may take 
> > > care of such cases.
> > 
> > Yes, correct. This and some other issues would be fixed by porting this
> > part of the code to bootm_load_devicetree() + find_unused_base(). Since
> > my time budget is on the limit, I would prefer to mainline current state
> > of the code ("works for me" TM) and provide platform for testing and
> > cooperation.
> 
> Can we at least detect that the image is a compressed one and bail out
> with an error? It would be a pity when someone else has to figure out
> this bug the hard way now that we already know that it exists.

I suggest to print a warning message, containing relevant addresses.  Something 
like this:

	ACHTUNG: booting vmlinuZ image. Beware of overlaps of kernel arguments 
	and vmlinuX sections. List of kernel arguments:
		cmdline @ XXXXXXXX-XXXXXXXX
		dtb @ XXXXXXXX-XXXXXXXX

However I'm not sure if it is possible to distinguish vmlinuz from vmlinux 
reliably, since both are just ELFs, though vmlinuz has only one section.  
Moreover, I guess this issue applies to any form of compressed OS image, not 
only vmlinuz. So, probably linux should take care about its arguments in 
arch/mips/boot/compressed/head.S.

Regards,
Peter

> 
> Sascha
> 
> -- 
> Pengutronix e.K.                           |                             |
> Industrial Linux Solutions                 | http://www.pengutronix.de/  |
> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox