[U-Boot] [PATCH] spl: fit: Enable GZIP compression also for no kernel partitions

[Tom Rini]

On Thu, Jul 26, 2018 at 04:23:14PM +0000, York Sun wrote:
> On 07/25/2018 11:52 PM, Michal Simek wrote:
> > On 25.7.2018 23:18, York Sun wrote:
> >> On 07/24/2018 10:58 PM, Michal Simek wrote:
> >>> On 24.7.2018 18:26, York Sun wrote:
> >>>> On 07/24/2018 06:07 AM, Michal Simek wrote:
> >>>>> There is no reason to limit gzip usage only for OS_BOOT and kernel image
> >>>>> type.
> >>>>>
> >>>>> Signed-off-by: Michal Simek <michal.simek at xilinx.com>
> >>>>> ---
> >>>>>
> >>>>>  common/spl/spl_fit.c | 5 +----
> >>>>>  1 file changed, 1 insertion(+), 4 deletions(-)
> >>>>>
> >>>>> diff --git a/common/spl/spl_fit.c b/common/spl/spl_fit.c
> >>>>> index 9eabb1c1058b..dbf5ac33a845 100644
> >>>>> --- a/common/spl/spl_fit.c
> >>>>> +++ b/common/spl/spl_fit.c
> >>>>> @@ -257,10 +257,7 @@ static int spl_load_fit_image(struct spl_load_info *info, ulong sector,
> >>>>>  	board_fit_image_post_process(&src, &length);
> >>>>>  #endif
> >>>>>  
> >>>>> -	if (IS_ENABLED(CONFIG_SPL_OS_BOOT)	&&
> >>>>> -	    IS_ENABLED(CONFIG_SPL_GZIP)		&&
> >>>>> -	    image_comp == IH_COMP_GZIP		&&
> >>>>> -	    type == IH_TYPE_KERNEL) {
> >>>>> +	if (IS_ENABLED(CONFIG_SPL_GZIP) && image_comp == IH_COMP_GZIP) {
> >>>>>  		size = length;
> >>>>>  		if (gunzip((void *)load_addr, CONFIG_SYS_BOOTM_LEN,
> >>>>>  			   src, &size)) {
> >>>>>
> >>>>
> >>>> This will uncompress ramdisk unnecessarily.
> >>>
> >>> Can you please share your its fragment? Also is there any other image
> >>> which should be exclude?
> >>
> >> I used it for falcon boot. I guess the executable image should have "entry". In
> >> my setup, only kernel image has "entry". Here is my its file.
> >>
> >> /dts-v1/;
> >>
> >> / {
> >> 	description = "Image file for the LS1046A Linux Kernel";
> >> 	#address-cells = <1>;
> >>
> >> 	images {
> >> 		kernel at 1 {
> >> 			description = "ARM64 Linux kernel";
> >> 			data = /incbin/("./arch/arm64/boot/Image.gz");
> >> 			type = "kernel";
> >> 			arch = "arm64";
> >> 			os = "linux";
> >> 			compression = "gzip";
> >> 			load = <0x80080000>;
> >> 			entry = <0x80080000>;
> >> 		};
> >> 		fdt at 1 {
> >> 			description = "Flattened Device Tree blob";
> >> 			data = /incbin/("./fsl-ls1046ardb.dtb");
> >> 			type = "flat_dt";
> >> 			arch = "arm64";
> >> 			compression = "none";
> >> 			load = <0x90000000>;
> >> 		};
> >> 		ramdisk at 1 {
> >> 			description = "Buildroot initramfs";
> >>                         data = /incbin/("./rootfs.cpio.gz");
> >> 			type = "ramdisk";
> >> 			arch = "arm64";
> >> 			os = "linux";
> >> 			compression = "gzip";
> > 
> > I have tested full u-boot and there is also no uncompression for ramdisk
> > when you put gzip compress there.
> > I have even tried gzip compression for fdt with expectation that u-boot
> > will uncompress it.
> > 
> > Based on doc/uImage.FIT/source_file_format.txt:
> > 165  - compression : Compression used by included data. Supported
> > compressions
> > 166     are "gzip" and "bzip2". If no compression is used compression
> > property
> > 167     should be set to "none".
> > 
> > 
> > Based on me this means that data inside fit are compressed and you are
> > asking u-boot in boot to uncompress it. If you are not asking for that
> > you should put none there.
> > But it doesn't look like this is supported at all for fdt/ramdisk and it
> > is only handled for OS.
> > I see here two cases.
> > 1. I want u-boot to uncompress my data in fit image (whatever it is)
> > before passing control to OS that's why I putting there compression method.
> > 2. I want OS to uncompress data but I want pass this data unchanged from
> > u-boot to OS that's why I am putting compression method at "none"
> > 
> > I am expecting when you put "none" there than you will boot in falcon
> > mode without any issue.
> 
> That will work. I can put "none" for the images I don't want U-Boot to
> uncompress.

Please also update the document to be clear that "none" for "don't touch
my compressed data!" is expected.  The existing language isn't clear but
I agree it makes sense (and follows the long standing practice of "-C
none" on compressed ramdisks for legacy style images).

And per Simon Goldschmidt's suggestion, after we've made the docs
clearer, making it so that U-Boot does decompress things would be good,
but we may also need to make that behavior configurable as I can see
people having put the correct compression in and not wanting it to be
uncompressed as we have examples of both none and gzip for ramdisks
today.  It's also possible (and if someone wants to dig back /
experiment and confirm) that long ago it did auto-decompress the data
and now doesn't, it would be a bugfix and no we don't need to make it
configurable.

> > I have no problem to change this patch and include only kernel and fpga
> > image but it sounds to me that we have gaps in implementation that not
> > all images inside the fit image have the same range of functionalities.
> > 
> > Also I think that "load" entry is that one which matters not "entry".
> 
> Not true here. The "entry" matters if you want to run it, for example
> Linux kernel. It may be different from "load".

load / entry matter for various use cases and platforms, yes.  And since
we're talking about FIT images I feel compelled to bring up 'type =
"kernel_noload"' as the way to avoid U-Boot moving the kernel when it
doesn't have to, as part of speeding up boot time.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20180726/1c6ec7b7/attachment.sig>