switch m68k to use the generic remapping DMA allocator
Hi Geert and Greg,
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
PATCH 1/2 m68k: use the generic dma coherent remap allocator
This switche to using common code for the DMA allocations, including potential use of the CMA allocator if configure. Also add a comment where the existing behavior seems to be lacking.
--- arch/m68k/Kconfig | 2 ++ arch/m68k/kernel/dma.c | 59 ++++++++---------------------------------- 2 files changed, 13 insertions(+), 48 deletions(-)
arch/m68k/Kconfig +2 -0
@@ -3,10 +3,12 @@ config M68K bool default y select ARCH_32BIT_OFF_T + select ARCH_HAS_DMA_MMAP_PGPROT if MMU && !COLDFIRE select ARCH_HAS_SYNC_DMA_FOR_DEVICE if HAS_DMA select ARCH_MIGHT_HAVE_PC_PARPORT if ISA select ARCH_NO_COHERENT_DMA_MMAP if !MMU select ARCH_NO_PREEMPT if !COLDFIRE + select DMA_DIRECT_REMAP if MMU && !COLDFIRE select HAVE_IDE select HAVE_AOUT if MMU select HAVE_DEBUG_BUGVERBOSE
arch/m68k/kernel/dma.c +11 -48
68 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 2/2 m68k: implement arch_dma_prep_coherent
In reply to Christoph HellwigWhen we remap memory as non-cached to be used as a DMA coherent buffer we should writeback all cache and invalidate the cache lines so that we make sure we have a clean slate. Implement this using the cache_push() helper.
--- arch/m68k/Kconfig | 1 + arch/m68k/kernel/dma.c | 5 +++++ 2 files changed, 6 insertions(+)
arch/m68k/Kconfig +1 -0
@@ -4,6 +4,7 @@ config M68K default y select ARCH_32BIT_OFF_T select ARCH_HAS_DMA_MMAP_PGPROT if MMU && !COLDFIRE + select ARCH_HAS_DMA_PREP_COHERENT select ARCH_HAS_SYNC_DMA_FOR_DEVICE if HAS_DMA select ARCH_MIGHT_HAVE_PC_PARPORT if ISA select ARCH_NO_COHERENT_DMA_MMAP if !MMU
arch/m68k/kernel/dma.c +5 -0
@@ -18,6 +18,11 @@#include <asm/pgalloc.h> #if defined(CONFIG_MMU ) && !defined(CONFIG_COLDFIRE ) +void arch_dma_prep_coherent (struct page *page,size_t size) +{ +cache_push (page_to_phys (page), size); +} +pgprot_t arch_dma_mmap_pgprot (struct device *dev,pgprot_t prot,unsigned long attrs) {
Signature · 1 line
2.20.1
Hi Christoph,
Quoted 5 lines · Christoph Hellwig
Hi Geert and Greg,
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
No impact to ColdFire targets, so I'll have to defer to Geert for his thoughts on the legacy m68k impact.
Regards
Greg
Geert Uytterhoeven
maintainerHi Christoph,
Quoted 3 lines · Christoph Hellwig
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
Thanks. But what does this buy us?
bloat-o-meter says:
add/remove: 75/0 grow/shrink: 11/6 up/down: 4122/-82 (4040)
Gr{oetje,eeting}s,
Geert
Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected] In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Quoted 8 lines · Geert Uytterhoeven
Hi Christoph,
Quoted 3 lines · Christoph Hellwig
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
Thanks. But what does this buy us?
A common dma mapping code base with everyone, including supporting DMA allocations from atomic context, which the documentation and API assume are there, but which don't work on m68k.
Quoted 3 lines
bloat-o-meter says:
add/remove: 75/0 grow/shrink: 11/6 up/down: 4122/-82 (4040)
What do these values stand for? The code should grow a little as we now need to include the the pool allocator for the above API fix.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Geert Uytterhoeven
maintainerHi Christoph,
Quoted 11 lines · Christoph Hellwig
Quoted 6 lines · Geert Uytterhoeven
Quoted 3 lines · Christoph Hellwig
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
Thanks. But what does this buy us?
A common dma mapping code base with everyone, including supporting DMA allocations from atomic context, which the documentation and API assume are there, but which don't work on m68k.
OK, thanks!
Quoted 7 lines
Quoted 3 lines
bloat-o-meter says:
add/remove: 75/0 grow/shrink: 11/6 up/down: 4122/-82 (4040)
What do these values stand for? The code should grow a little as we now need to include the the pool allocator for the above API fix.
Last 3 values are "bytes added/removed (net increase)". So this increases the static kernel size by ca. 4 KiB.
Gr{oetje,eeting}s,
Geert
Signature · 9 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected] In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Quoted 10 lines · Geert Uytterhoeven
Quoted 7 lines
Quoted 3 lines
bloat-o-meter says:
add/remove: 75/0 grow/shrink: 11/6 up/down: 4122/-82 (4040)
What do these values stand for? The code should grow a little as we now need to include the the pool allocator for the above API fix.
Last 3 values are "bytes added/removed (net increase)". So this increases the static kernel size by ca. 4 KiB.
That seems a lot for the little bit of pool code. Did m68k not build lib/genalloc.c by default before?
Also I'd be curious what the first 4 values are.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Geert Uytterhoeven
maintainerHi Christoph,
Quoted 5 lines · Christoph Hellwig
Quoted 4 lines · Geert Uytterhoeven
Quoted 4 lines
Quoted 3 lines
bloat-o-meter says:
add/remove: 75/0 grow/shrink: 11/6 up/down: 4122/-82 (4040)
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
What do these values stand for? The code should grow a little as we now need to include the the pool allocator for the above API fix.
Last 3 values are "bytes added/removed (net increase)". So this increases the static kernel size by ca. 4 KiB.
That seems a lot for the little bit of pool code. Did m68k not
Exactly, hence my original question...
Quoted 1 line
build
lib/genalloc.cby default before?
Indeed, CONFIG_GENERIC_ALLOCATOR wasn't enabled before.
.config 2019-06-25 09:59:23.914874446 +0200 renamed from .config.orig 2019-06-25 09:53:35.098691378 +0200 +5 -0
@@ -2401,6 +2401,7 @@ CONFIG_DECOMPRESS_XZ=y CONFIG_DECOMPRESS_LZO=y CONFIG_DECOMPRESS_LZ4=y +CONFIG_GENERIC_ALLOCATOR=y CONFIG_TEXTSEARCH=y CONFIG_TEXTSEARCH_KMP=m CONFIG_TEXTSEARCH_BM=m
@@ -2409,6 +2410,10 @@ CONFIG_HAS_IOMEM=y CONFIG_HAS_DMA=y CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE=y +CONFIG_ARCH_HAS_DMA_PREP_COHERENT=y +CONFIG_ARCH_HAS_DMA_MMAP_PGPROT=y +CONFIG_DMA_REMAP=y +CONFIG_DMA_DIRECT_REMAP=y # CONFIG_DMA_API_DEBUG is not set CONFIG_SGL_ALLOC=y CONFIG_DQL=y
Quoted 1 line
Also I'd be curious what the first 4 values are.
Reading scripts/bloat-o-meter in the kernel source tree tells me number of added/removed symbols (functions or variables), and number of grown/shrunk symbols.
I run it regularly, to catch silly mistakes (cfr. my favorite one, fixed by commit 23c323af0375a7f6 ("SUNRPC: No, I did not intend to create a 256KiB hashtable")).
Full output before/after for an atari_defconfig kernel (numbers are different, baseline has changed, too):
$ bloat-o-meter vmlinux.orig vmlinux add/remove: 75/0 grow/shrink: 1/2 up/down: 4098/-28 (4070) Function old new delta gen_pool_alloc_algo_owner - 392 +392 dma_atomic_pool_init - 360 +360 gen_pool_best_fit - 248 +248 dma_common_contiguous_remap - 238 +238 gen_pool_destroy - 190 +190 gen_pool_free_owner - 184 +184 devm_gen_pool_create - 174 +174 dma_alloc_from_pool - 152 +152 bitmap_clear_ll - 138 +138 arch_dma_free 12 136 +124 dma_common_free_remap - 110 +110 gen_pool_add_owner - 108 +108 gen_pool_fixed_alloc - 100 +100 dma_common_pages_remap - 92 +92 gen_pool_dma_alloc - 78 +78 arch_dma_prep_coherent - 76 +76 clear_bits_ll - 66 +66 dma_free_from_pool - 62 +62 set_bits_ll - 60 +60 devm_gen_pool_match - 56 +56 addr_in_gen_pool - 56 +56 gen_pool_create - 54 +54 gen_pool_virt_to_phys - 52 +52 gen_pool_first_fit_align - 52 +52 gen_pool_for_each_chunk - 42 +42 gen_pool_get - 40 +40 gen_pool_first_fit_order_align - 36 +36 gen_pool_set_algo - 34 +34 early_coherent_pool - 32 +32 __kstrtab_gen_pool_first_fit_order_align - 31 +31 gen_pool_size - 30 +30 arch_dma_mmap_pgprot - 28 +28 dma_in_atomic_pool - 26 +26 __kstrtab_gen_pool_alloc_algo_owner - 26 +26 __kstrtab_gen_pool_first_fit_align - 25 +25 gen_pool_avail - 24 +24 __kstrtab_gen_pool_for_each_chunk - 24 +24 __kstrtab_gen_pool_virt_to_phys - 22 +22 __kstrtab_gen_pool_fixed_alloc - 21 +21 __kstrtab_devm_gen_pool_create - 21 +21 arch_dma_coherent_to_pfn - 20 +20 __kstrtab_gen_pool_free_owner - 20 +20 __kstrtab_gen_pool_first_fit - 19 +19 __kstrtab_gen_pool_dma_alloc - 19 +19 __kstrtab_gen_pool_add_owner - 19 +19 __kstrtab_gen_pool_set_algo - 18 +18 __kstrtab_gen_pool_best_fit - 18 +18 __kstrtab_gen_pool_destroy - 17 +17 __kstrtab_gen_pool_create - 16 +16 __kstrtab_gen_pool_avail - 15 +15 gen_pool_first_fit - 14 +14 devm_gen_pool_release - 14 +14 __setup_str_early_coherent_pool - 14 +14 __kstrtab_gen_pool_size - 14 +14 __kstrtab_gen_pool_get - 13 +13 __setup_early_coherent_pool - 12 +12 __ksymtab_gen_pool_virt_to_phys - 8 +8 __ksymtab_gen_pool_size - 8 +8 __ksymtab_gen_pool_set_algo - 8 +8 __ksymtab_gen_pool_get - 8 +8 __ksymtab_gen_pool_free_owner - 8 +8 __ksymtab_gen_pool_for_each_chunk - 8 +8 __ksymtab_gen_pool_fixed_alloc - 8 +8 __ksymtab_gen_pool_first_fit_order_align - 8 +8 __ksymtab_gen_pool_first_fit_align - 8 +8 __ksymtab_gen_pool_first_fit - 8 +8 __ksymtab_gen_pool_dma_alloc - 8 +8 __ksymtab_gen_pool_destroy - 8 +8 __ksymtab_gen_pool_create - 8 +8 __ksymtab_gen_pool_best_fit - 8 +8 __ksymtab_gen_pool_avail - 8 +8 __ksymtab_gen_pool_alloc_algo_owner - 8 +8 __ksymtab_gen_pool_add_owner - 8 +8 __ksymtab_devm_gen_pool_create - 8 +8 atomic_pool_size - 4 +4 atomic_pool - 4 +4 arch_dma_alloc 312 310 -2 dma_common_mmap 250 224 -26 Total: Before=3724055, After=3728125, chg +0.11%
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Geert Uytterhoeven
maintainerHi Christoph,
Quoted 3 lines · Christoph Hellwig
can you take a look at the (untested) patches below? They convert m68k to use the generic remapping DMA allocator, which is also used by arm64 and csky.
When building for Sun-3:
kernel/dma/remap.o: In function `arch_dma_alloc':
remap.c:(.text+0x316): undefined reference to __dma_direct_alloc_pages' remap.c:(.text+0x32a): undefined reference to arch_dma_prep_coherent' remap.c:(.text+0x34e): undefined reference to arch_dma_mmap_pgprot' remap.c:(.text+0x378): undefined reference to __dma_direct_free_pages' remap.c:(.text+0x3f4): undefined reference to arch_dma_prep_coherent' remap.c:(.text+0x40a): undefined reference to __dma_direct_alloc_pages' kernel/dma/remap.o: In function `arch_dma_free':
remap.c:(.text+0x446): undefined reference to __dma_direct_free_pages' remap.c:(.text+0x4a8): undefined reference to __dma_direct_free_pages' kernel/dma/remap.o: In function `dma_atomic_pool_init':
remap.c:(.init.text+0x66): undefined reference to `arch_dma_prep_coherent'
Doing
- select DMA_DIRECT_REMAP if MMU && !COLDFIRE
+ select DMA_DIRECT_REMAP if MMU && !COLDFIRE && !SUN3
in arch/m68k/Kconfig fixes the build.
Alternatively, you could use:
- select DMA_DIRECT_REMAP if MMU && !COLDFIRE
+ select DMA_DIRECT_REMAP if HAS_DMA && MMU && !COLDFIRE
Gr{oetje,eeting}s,
Geert
Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected] In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Review
Quoted 11 lines · Geert Uytterhoeven
Doing
(quoted diff) +1
-1- select DMA_DIRECT_REMAP if MMU && !COLDFIRE + select DMA_DIRECT_REMAP if MMU && !COLDFIRE && !SUN3in
arch/m68k/Kconfigfixes the build.Alternatively, you could use:
(quoted diff) +1
-1- select DMA_DIRECT_REMAP if MMU && !COLDFIRE + select DMA_DIRECT_REMAP if HAS_DMA && MMU && !COLDFIRE
The latter might be a little cleaner.
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu