blackfin: Cleanup linker script using new linker script macros.
PATCH 1/1 blackfin: Cleanup linker script using new linker script macros.
--- arch/blackfin/kernel/vmlinux.lds.S | 54 +++-------------------------------- 1 files changed, 5 insertions(+), 49 deletions(-)
arch/blackfin/kernel/vmlinux.lds.S +5 -49
@@ -96,8 +96,7 @@ SECTIONS { __sdata = .; /* This gets done first, so the glob doesn't suck it in */ - . = ALIGN(32); - *(.data.cacheline_aligned) + CACHELINE_ALIGNED_DATA(32) #if !L1_DATA_A_LENGTH . = ALIGN(32);
@@ -116,12 +115,7 @@ SECTIONS DATA_DATA CONSTRUCTORS - /* make sure the init_task is aligned to the - * kernel thread size so we can locate the kernel - * stack properly and quickly. - */ - . = ALIGN(THREAD_SIZE); - *(.init_task.data) + INIT_TASK_DATA(THREAD_SIZE) __edata = .; }
@@ -134,39 +128,10 @@ SECTIONS . = ALIGN(PAGE_SIZE); ___init_begin = .; - .init.text : - { - . = ALIGN(PAGE_SIZE); - __sinittext = .; - INIT_TEXT - __einittext = .; - } - .init.data : - { - . = ALIGN(16); - INIT_DATA - } - .init.setup : - { - . = ALIGN(16); - ___setup_start = .; - *(.init.setup) - ___setup_end = .; - } - .initcall.init : - { - ___initcall_start = .; - INITCALLS - ___initcall_end = .; - } - .con_initcall.init : - { - ___con_initcall_start = .; - *(.con_initcall.init) - ___con_initcall_end = .; - } + INIT_TEXT_SECTION(PAGE_SIZE) + . = ALIGN(16); + INIT_DATA_SECTION(16) PERCPU(4) - SECURITY_INIT /* we have to discard exit text and such at runtime, not link time, to * handle embedded cross-section references (alt instructions, bug
@@ -181,15 +146,6 @@ SECTIONS EXIT_DATA } - .init.ramfs : - { - . = ALIGN(4); - ___initramfs_start = .; - *(.init.ramfs) - . = ALIGN(4); - ___initramfs_end = .; - } - __l1_lma_start = .; .text_l1 L1_CODE_START : AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))
Signature · 1 line
1.6.3.3
the larger padding in the initramfs is kind of annoying as i cant see any need to pad it to PAGE_SIZE. since the initramfs is released with the rest of the init section, it doesnt need whole pages. a quick test shows that it does waste a few kb in reality. default build for BF533-STAMP for example shows 0x1000 difference.
in terms of correctness, this change misses a reference to the now deleted .init.ramfs:
- .init.ramfs :
- {
- .....
- }-
.text_l1 L1_CODE_START : AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))
so that .text_l1 needs to updated to refer to the new section before it (.exit.data in this case). once i make that change, the resulting link looks the same (minus the initramfs thing mentioned earlier), and it does boot.
-mike
Quoted 15 lines · Mike Frysinger
the larger padding in the initramfs is kind of annoying as i cant see any need to pad it to
PAGE_SIZE. since the initramfs is released with the rest of the init section, it doesnt need whole pages. a quick test shows that it does waste a few kb in reality. default build for BF533-STAMP for example shows0x1000difference.in terms of correctness, this change misses a reference to the now deleted .init.ramfs:
arch/blackfin/kernel/vmlinux.lds.S +0
-5- .init.ramfs : - { - ..... - } - .text_l1 L1_CODE_START : AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))
Quoted 4 lines
so that .text_l1 needs to updated to refer to the new section before it (.exit.data in this case). once i make that change, the resulting link looks the same (minus the initramfs thing mentioned earlier), and it does boot.
OK. I guess we should plan to modify the INIT_DATA_SECTION macro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses of INIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument to INIT_DATA_SECTION.
-Tim Abbott
Quoted 7 lines · Tim Abbott
Quoted 5 lines · Mike Frysinger
the larger padding in the initramfs is kind of annoying as i cant see any need to pad it to
PAGE_SIZE. since the initramfs is released with the rest of the init section, it doesnt need whole pages. a quick test shows that it does waste a few kb in reality. default build for BF533-STAMP for example shows0x1000difference.
Quoted 7 lines
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.
sounds good to me. too bad the function alignment isnt specified in arch linkage.h, otherwise i'd suggest we default to that rather than PAGE_SIZE. as seen in the current Blackfin linker script, we only need the normal .text alignment of 4 bytes.
-mike
Quoted 22 lines · Tim Abbott
Quoted 15 lines · Mike Frysinger
the larger padding in the initramfs is kind of annoying as i cant see any need to pad it to
PAGE_SIZE. since the initramfs is released with the rest of the init section, it doesnt need whole pages. a quick test shows that it does waste a few kb in reality. default build for BF533-STAMP for example shows0x1000difference.in terms of correctness, this change misses a reference to the now deleted .init.ramfs:
arch/blackfin/kernel/vmlinux.lds.S +0
-5- .init.ramfs : - { - ..... - } - .text_l1 L1_CODE_START : AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))Quoted 4 lines
so that .text_l1 needs to updated to refer to the new section before it (.exit.data in this case). once i make that change, the resulting link looks the same (minus the initramfs thing mentioned earlier), and it does boot.
Quoted 7 lines
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.
But this is all discarded during runtime so the added alignment has no cost in the end - no?
Sam
Quoted 30 lines · Sam Ravnborg
Quoted 21 lines · Tim Abbott
Quoted 15 lines · Mike Frysinger
the larger padding in the initramfs is kind of annoying as i cant see any need to pad it to
PAGE_SIZE. since the initramfs is released with the rest of the init section, it doesnt need whole pages. a quick test shows that it does waste a few kb in reality. default build for BF533-STAMP for example shows0x1000difference.in terms of correctness, this change misses a reference to the now deleted .init.ramfs:
- .init.ramfs :
- {
- .....
- }
-
.text_l1L1_CODE_START: AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))Quoted 4 lines
so that .text_l1 needs to updated to refer to the new section before it (.exit.data in this case). once i make that change, the resulting link looks the same (minus the initramfs thing mentioned earlier), and it does boot.
Quoted 7 lines
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.
Quoted 1 line
But this is all discarded during runtime so the added alignment has no cost in the end - no?
once things are booted, there should be no difference. but storage/boot costs increase (you have to store/extract/copy that extra data). you know how miserly we embedded people like to be ;). -mike
Quoted 10 lines · Mike Frysinger
Quoted 9 lines · Sam Ravnborg
Quoted 7 lines · Tim Abbott
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.
Quoted 5 lines
Quoted 1 line
But this is all discarded during runtime so the added alignment has no cost in the end - no?
once things are booted, there should be no difference. but storage/boot costs increase (you have to store/extract/copy that extra data). you know how miserly we embedded people like to be ;).
OK, so how do you want to do this? The options I see are:
(1) we merge this patch now, and add the new alignment argument for -rc2
(2) we add the alignment argument sometime after -rc1 and then merge this for -rc2
-Tim Abbott
Quoted 13 lines · Tim Abbott
Quoted 12 lines · Mike Frysinger
Quoted 10 lines · Sam Ravnborg
Quoted 7 lines · Tim Abbott
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.But this is all discarded during runtime so the added alignment has no cost in the end - no?
Quoted 8 lines
Quoted 3 lines
once things are booted, there should be no difference. but storage/boot costs increase (you have to store/extract/copy that extra data). you know how miserly we embedded people like to be ;).
OK, so how do you want to do this? The options I see are:
(1) we merge this patch now, and add the new alignment argument for -rc2
(2) we add the alignment argument sometime after -rc1 and then merge this for -rc2
doing it in two steps is OK by me and sounds like it'd be easier for you -mike
Quoted 18 lines · Mike Frysinger
Quoted 17 lines · Tim Abbott
Quoted 15 lines · Mike Frysinger
Quoted 10 lines · Sam Ravnborg
Quoted 7 lines · Tim Abbott
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.But this is all discarded during runtime so the added alignment has no cost in the end - no?
once things are booted, there should be no difference. but storage/boot costs increase (you have to store/extract/copy that extra data). you know how miserly we embedded people like to be ;).
Quoted 6 lines
Quoted 4 lines
OK, so how do you want to do this? The options I see are:
(1) we merge this patch now, and add the new alignment argument for -rc2
(2) we add the alignment argument sometime after -rc1 and then merge this for -rc2doing it in two steps is OK by me and sounds like it'd be easier for you
Both options involve two steps -- but as (1) is obviously easier for me, I assume that's what you were referring to. Thanks.
So are you going to send this to Linus? I'd be happy to do so, but I'd need your ack.
-Tim Abbott
Quoted 24 lines · Tim Abbott
Quoted 23 lines · Mike Frysinger
Quoted 21 lines · Tim Abbott
Quoted 10 lines · Mike Frysinger
Quoted 9 lines · Sam Ravnborg
Quoted 7 lines · Tim Abbott
OK. I guess we should plan to modify the
INIT_DATA_SECTIONmacro to add another argument specifying an alignment level for .init.ramfs. It'd be inconvenient to add that right now since there are a lot of patches in linux-next or otherwise in flight that introduce uses ofINIT_DATA_SECTION, and those patches would all be broken by changing this now. Once the dust settles on that for this release, I'll submit a patch adding said argument toINIT_DATA_SECTION.Quoted 5 lines
Quoted 1 line
But this is all discarded during runtime so the added alignment has no cost in the end - no?
once things are booted, there should be no difference. but storage/boot costs increase (you have to store/extract/copy that extra data). you know how miserly we embedded people like to be ;).
OK, so how do you want to do this? The options I see are:
(1) we merge this patch now, and add the new alignment argument for -rc2
(2) we add the alignment argument sometime after -rc1 and then merge this for -rc2
Quoted 7 lines
Quoted 1 line
doing it in two steps is OK by me and sounds like it'd be easier for you
Both options involve two steps -- but as (1) is obviously easier for me, I assume that's what you were referring to. Thanks.
So are you going to send this to Linus? I'd be happy to do so, but I'd need your ack.
if you apply the .text_l1 update (.init.ramfs -> .exit.data), then you can add my s-o-b. otherwise the original patch as posted is broken. -mike
Quoted 6 lines · Mike Frysinger
Quoted 2 lines · Tim Abbott
So are you going to send this to Linus? I'd be happy to do so, but I'd need your ack.
if you apply the .text_l1 update (.init.ramfs -> .exit.data), then you can add my s-o-b. otherwise the original patch as posted is broken.
Right, I modified the patch in my git tree when you first mentioned it like so:
- .text_l1 L1_CODE_START : AT(LOADADDR(.init.ramfs) + SIZEOF(.init.ramfs))
+ .text_l1 L1_CODE_START : AT(LOADADDR(.exit.data) + SIZEOF(.exit.data))
-Tim Abbott