Since we're storing the firmware blobs in the QEMU git repository, it
would be nice if the blobs would be a little bit smaller. By using -Os
and -fno-asynchronous-unwind-tables the size of the s390-ccw.img can be
decreased by ca. 4kB, and the size of s390-netboot.img even by ca. 45kB.
The change of speed of the firmware should not really matter compared to
a binary that has been compiled with -O2 since the firmware still starts
very fast.

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 pc-bios/s390-ccw/Makefile | 1 +
 1 file changed, 1 insertion(+)

diff --git a/pc-bios/s390-ccw/Makefile b/pc-bios/s390-ccw/Makefile
index 439e3cc..f22a5ca 100644
--- a/pc-bios/s390-ccw/Makefile
+++ b/pc-bios/s390-ccw/Makefile
@@ -13,6 +13,7 @@ OBJECTS = start.o main.o bootmap.o jump2ipl.o sclp.o menu.o \
 	  virtio.o virtio-scsi.o virtio-blkdev.o libc.o
 
 QEMU_CFLAGS := $(filter -W%, $(QEMU_CFLAGS))
+QEMU_CFLAGS += -Os -fno-asynchronous-unwind-tables
 QEMU_CFLAGS += -ffreestanding -fno-delete-null-pointer-checks -msoft-float
 QEMU_CFLAGS += -march=z900 -fPIE -fno-strict-aliasing
 QEMU_CFLAGS += $(call cc-option, $(QEMU_CFLAGS), -fno-stack-protector)
-- 
1.8.3.1

On 06/14/2018 10:48 AM, Thomas Huth wrote:
> Since we're storing the firmware blobs in the QEMU git repository, it
> would be nice if the blobs would be a little bit smaller. By using -Os
> and -fno-asynchronous-unwind-tables the size of the s390-ccw.img can be
> decreased by ca. 4kB, and the size of s390-netboot.img even by ca. 45kB.
> The change of speed of the firmware should not really matter compared to
> a binary that has been compiled with -O2 since the firmware still starts
> very fast.


> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  pc-bios/s390-ccw/Makefile | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/pc-bios/s390-ccw/Makefile b/pc-bios/s390-ccw/Makefile
> index 439e3cc..f22a5ca 100644
> --- a/pc-bios/s390-ccw/Makefile
> +++ b/pc-bios/s390-ccw/Makefile
> @@ -13,6 +13,7 @@ OBJECTS = start.o main.o bootmap.o jump2ipl.o sclp.o menu.o \
>  	  virtio.o virtio-scsi.o virtio-blkdev.o libc.o
>  
>  QEMU_CFLAGS := $(filter -W%, $(QEMU_CFLAGS))
> +QEMU_CFLAGS += -Os -fno-asynchronous-unwind-tables

Hmm, it looks like that the O2 comes after Os for me (which would override Os).

cc -iquote /home/cborntra/REPOS/qemu/build/. -iquote . -iquote /home/cborntra/REPOS/qemu/tcg -iquote /home/cborntra/REPOS/qemu/tcg/s390 -I/home/cborntra/REPOS/qemu/linux-headers -I/home/cborntra/REPOS/qemu/build/linux-headers -iquote . -iquote /home/cborntra/REPOS/qemu -iquote /home/cborntra/REPOS/qemu/accel/tcg -iquote /home/cborntra/REPOS/qemu/include -Werror -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -Wexpansion-to-defined -Wendif-labels -Wno-shift-negative-value -Wno-missing-include-dirs -Wempty-body -Wnested-externs -Wformat-security -Wformat-y2k -Winit-self -Wignored-qualifiers -Wold-style-declaration -Wold-style-definition -Wtype-limits -Os -fno-asynchronous-unwind-tables -ffreestanding -fno-delete-null-pointer-checks -msoft-float -march=z900 -fPIE -fno-strict-aliasing   -fno-stack-protector -MMD -MP -MT libc.o -MF ./libc.d -O2 -g  -c -o libc.o /home/cborntra/REPOS/qemu/pc-bios/s390-ccw/libc.c


so the effect is only because of the -fno-asynchronous-unwind-tables.



>  QEMU_CFLAGS += -ffreestanding -fno-delete-null-pointer-checks -msoft-float
>  QEMU_CFLAGS += -march=z900 -fPIE -fno-strict-aliasing
>  QEMU_CFLAGS += $(call cc-option, $(QEMU_CFLAGS), -fno-stack-protector)
> 

On 14.06.2018 12:55, Christian Borntraeger wrote:
> 
> On 06/14/2018 10:48 AM, Thomas Huth wrote:
>> Since we're storing the firmware blobs in the QEMU git repository, it
>> would be nice if the blobs would be a little bit smaller. By using -Os
>> and -fno-asynchronous-unwind-tables the size of the s390-ccw.img can be
>> decreased by ca. 4kB, and the size of s390-netboot.img even by ca. 45kB.
>> The change of speed of the firmware should not really matter compared to
>> a binary that has been compiled with -O2 since the firmware still starts
>> very fast.
> 
> 
>>
>> Signed-off-by: Thomas Huth <thuth@redhat.com>
>> ---
>>  pc-bios/s390-ccw/Makefile | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/pc-bios/s390-ccw/Makefile b/pc-bios/s390-ccw/Makefile
>> index 439e3cc..f22a5ca 100644
>> --- a/pc-bios/s390-ccw/Makefile
>> +++ b/pc-bios/s390-ccw/Makefile
>> @@ -13,6 +13,7 @@ OBJECTS = start.o main.o bootmap.o jump2ipl.o sclp.o menu.o \
>>  	  virtio.o virtio-scsi.o virtio-blkdev.o libc.o
>>  
>>  QEMU_CFLAGS := $(filter -W%, $(QEMU_CFLAGS))
>> +QEMU_CFLAGS += -Os -fno-asynchronous-unwind-tables
> 
> Hmm, it looks like that the O2 comes after Os for me (which would override Os).
> 
> cc -iquote /home/cborntra/REPOS/qemu/build/. -iquote . -iquote /home/cborntra/REPOS/qemu/tcg -iquote /home/cborntra/REPOS/qemu/tcg/s390 -I/home/cborntra/REPOS/qemu/linux-headers -I/home/cborntra/REPOS/qemu/build/linux-headers -iquote . -iquote /home/cborntra/REPOS/qemu -iquote /home/cborntra/REPOS/qemu/accel/tcg -iquote /home/cborntra/REPOS/qemu/include -Werror -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -Wexpansion-to-defined -Wendif-labels -Wno-shift-negative-value -Wno-missing-include-dirs -Wempty-body -Wnested-externs -Wformat-security -Wformat-y2k -Winit-self -Wignored-qualifiers -Wold-style-declaration -Wold-style-definition -Wtype-limits -Os -fno-asynchronous-unwind-tables -ffreestanding -fno-delete-null-pointer-checks -msoft-float -march=z900 -fPIE -fno-strict-aliasing   -fno-stack-protector -MMD -MP -MT libc.o -MF ./libc.d -O2 -g  -c -o libc.o /home/cborntra/REPOS/qemu/pc-bios/s390-ccw/libc.c
> 
> so the effect is only because of the -fno-asynchronous-unwind-tables.

Ugh, you're right - at least for the objects for s390-ccw.img. Seems
like the -O2 comes from "CFLAGS" and not from "QEMU_CFLAGS" here, so the
flags should be added to the CFLAGS variable instead.

Actually, I was only looking at the object files from s390-netboot.img,
and the libc files from SLOF get *only* compiled with QEMU_CFLAGS there
(see LIBC_CFLAGS in netboot.mak). So the huge part of the
s390-netboot.img binary got compiled without optimizations so far!
That's why I was seeing such a big improvement there! Wow, I wonder why
we did not notice this before... Anyway, I'll send a v2 to fix these issues.

 Thomas