RFC: switch to use the generic remapping dma-allocator
Subsystems (2)
Hi all,
currently parisc has it's own implementation of the dma-coherent allocator that sets up PTEs with the uncached bit. I can't see any reason why using vmap to do the same kind of setup shouldn't work. This small series shows how that could work, but due to a lack of actual parisc hardware it is compile tested only and not only needs a careful review but also actual testing on hardware.
Diffstat:
arch/parisc/Kconfig | 4 arch/parisc/include/asm/pgtable.h | 1 arch/parisc/kernel/pci-dma.c | 446 +------------------------------------- include/linux/dma-map-ops.h | 1 kernel/dma/Kconfig | 3 kernel/dma/direct.c | 4 6 files changed, 24 insertions(+), 435 deletions(-)
PATCH 1/2 dma-mapping: add a arch_can_dma_alloc_coherent hook
This hook allows an architecture to reject coherent DMA allocations entirely if they can't be supported. This will be useful to convert parisc and m68knommu to the generic dma-direct code.
--- include/linux/dma-map-ops.h | 1 + kernel/dma/Kconfig | 3 +++ kernel/dma/direct.c | 4 ++++ 3 files changed, 8 insertions(+)
include/linux/dma-map-ops.h +1 -0
@@ -344,6 +344,7 @@ static inline void arch_dma_mark_clean(phys_addr_t paddr, size_t size) }#endif /* ARCH_HAS_DMA_MARK_CLEAN */ +bool arch_can_dma_alloc_coherent (void );void *arch_dma_set_uncached (void *addr,size_t size);void arch_dma_clear_uncached (void *addr,size_t size);
kernel/dma/Kconfig +3 -0
@@ -20,6 +20,9 @@ config DMA_OPS config DMA_OPS_BYPASS bool +config ARCH_HAS_CAN_DMA_ALLOC_COHERENT + bool + # Lets platform IOMMU driver choose between bypass and IOMMU config ARCH_HAS_DMA_MAP_DIRECT bool
kernel/dma/direct.c +4 -0
@@ -232,6 +232,10 @@ void *dma_direct_alloc(struct device *dev, size_t size,return arch_dma_alloc (dev, size, dma_handle, gfp, attrs); +if (IS_ENABLED (CONFIG_ARCH_HAS_CAN_DMA_ALLOC_COHERENT ) && + !arch_can_dma_alloc_coherent ()) +return NULL ; +/* * If there is a global pool, always allocate from it for * non-coherent devices.
Signature · 1 line
2.30.2
PATCH 2/2 parisc: use the generic remapping dma allocator
In reply to Christoph HellwigUse the dma-direct code that uses vmap to create an uncacheable mapping for coherent allocations instead of the hand-crafted bitmap.
--- arch/parisc/Kconfig | 4 + arch/parisc/include/asm/pgtable.h | 1 - arch/parisc/kernel/pci-dma.c | 446 +----------------------------- 3 files changed, 16 insertions(+), 435 deletions(-)
arch/parisc/Kconfig +4 -0
@@ -8,6 +8,8 @@ config PARISC select HAVE_FUNCTION_GRAPH_TRACER select HAVE_SYSCALL_TRACEPOINTS select ARCH_WANT_FRAME_POINTERS + select ARCH_HAS_CAN_DMA_ALLOC_COHERENT + select ARCH_HAS_DMA_PREP_COHERENT select ARCH_HAS_ELF_RANDOMIZE select ARCH_HAS_STRICT_KERNEL_RWX select ARCH_HAS_STRICT_MODULE_RWX
@@ -19,6 +21,8 @@ config PARISC select ARCH_STACKWALK select ARCH_HAS_DEBUG_VM_PGTABLE select HAVE_RELIABLE_STACKTRACE + select DMA_COHERENT_POOL + select DMA_DIRECT_REMAP select DMA_OPS select RTC_CLASS select RTC_DRV_GENERIC
arch/parisc/include/asm/pgtable.h +0 -1
@@ -263,7 +263,6 @@ extern void __update_cache(pte_t pte);#define PAGE_KERNEL_EXEC __pgprot(_PAGE_KERNEL_EXEC)#define PAGE_KERNEL_RWX __pgprot(_PAGE_KERNEL_RWX)#define PAGE_KERNEL_RO __pgprot(_PAGE_KERNEL_RO) -#define PAGE_KERNEL_UNC __pgprot(_PAGE_KERNEL | _PAGE_NO_CACHE)#define PAGE_GATEWAY __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_GATEWAY| _PAGE_READ)
arch/parisc/kernel/pci-dma.c +12 -434
458 lines not on this page · raw message
Signature · 1 line
2.30.2
Helge Deller
maintainerHi Christoph,
Quoted 6 lines · Christoph Hellwig
currently parisc has it's own implementation of the dma-coherent allocator that sets up PTEs with the uncached bit. I can't see any reason why using vmap to do the same kind of setup shouldn't work. This small series shows how that could work, but due to a lack of actual parisc hardware it is compile tested only and not only needs a careful review but also actual testing on hardware.
Very interesting!
I'll give it a spin on real hardware and will come back with the results...
Thanks!
Helge
Quoted 8 lines
Diffstat:
arch/parisc/Kconfig | 4 arch/parisc/include/asm/pgtable.h | 1 arch/parisc/kernel/pci-dma.c | 446 +------------------------------------- include/linux/dma-map-ops.h | 1 kernel/dma/Kconfig | 3 kernel/dma/direct.c | 4 6 files changed, 24 insertions(+), 435 deletions(-)
Helge Deller
maintainerQuoted 10 lines · Helge Deller
Quoted 6 lines · Christoph Hellwig
currently parisc has it's own implementation of the dma-coherent allocator that sets up PTEs with the uncached bit. I can't see any reason why using vmap to do the same kind of setup shouldn't work. This small series shows how that could work, but due to a lack of actual parisc hardware it is compile tested only and not only needs a careful review but also actual testing on hardware.
Very interesting!
I'll give it a spin on real hardware and will come back with the results...
kernel builds & boots with those patches.
SCSI drives are detected, but SCSI communication fails afterwards:
[ 13.509173] zalon_probe: Zalon version 1, IRQ 68 [ 13.564410] ncr53c720-0: rev 0xf irq 68 [ 13.614207] ncr53c720-0: ID 7, Fast-10, Parity Checking, Differential [ 13.696785] scsi host0: ncr53c8xx-3.4.3g [ 17.575885] scsi 0:0:6:0: Direct-Access SEAGATE ST32171W HP03 PQ: 0 ANSI: 2 [ 17.673111] scsi target0:0:6: Beginning Domain Validation [ 17.742479] scsi target0:0:6: asynchronous [ 17.806301] scsi target0:0:6: wide asynchronous [ 17.871736] scsi target0:0:6: FAST-10 WIDE SCSI 20.0 MB/s ST (100 ns, offset 8) [ 17.965723] scsi target0:0:6: Domain Validation skipping write tests [ 18.041560] scsi target0:0:6: Ending Domain Validation [ 20.448322] 53c700: Version 2.8 By [email protected] [ 20.528403] scsi1: 53c710 rev 2 [ 21.567995] scsi host1: LASI SCSI 53c700 [ 21.936685] scsi 1:0:1:0: CD-ROM PHILIPS PCA80SC V3-0 PQ: 0 ANSI: 2 [ 22.033927] scsi target1:0:1: Beginning Domain Validation [ 22.100102] scsi target1:0:1: asynchronous [ 22.161326] scsi target1:0:1: FAST-10 SCSI 10.0 MB/s ST (100 ns, offset 8) [ 22.250962] scsi target1:0:1: Domain Validation skipping write tests [ 22.326829] scsi target1:0:1: Ending Domain Validation [ 23.972351] st: Version 20160209, fixed bufsize 32768, s/g segs 256 [ 24.052265] sd 0:0:6:0: Power-on or device reset occurred [ 24.117020] sr 1:0:1:0: Power-on or device reset occurred [ 24.185309] sd 0:0:6:0: [sda] 4194685 512-byte logical blocks: (2.15 GB/2.00 GiB) [ 24.279279] sd 0:0:6:0: [sda] Write Protect is off [ 24.336921] sr 1:0:1:0: [sr0] scsi-1 drive [ 24.386003] cdrom: Uniform CD-ROM driver Revision: 3.20 [ 24.451015] sd 0:0:6:0: [sda] Write cache: disabled, read cache: enabled, supports DPO and FUA [ 24.569057] sd 0:0:6:0: Attached scsi generic sg0 type 0 [ 24.634080] sr 1:0:1:0: Attached scsi generic sg1 type 5 ... [ 28.246342] EXT4-fs (sda5): mounting ext3 file system using the ext4 subsystem [ 28.338790] ncr53c720-0:6: ERROR (81:0) (0-0-0) (8/1b) @ (mem 43c:00000000). [ 28.343500] ncr53c720-0: regdump: [ 28.343500] da [ 28.343500] 10 [ 28.343500] c0 [ 28.343500] 1b [ 28.343500] 47 [ 28.343500] 08 [ 28.343500] 06 [ 28.343500] 07 [ 28.343500] 45 [ 28.343500] 08 [ 28.343500] 86 [ 28.343500] af [ 28.343500] 80 [ 28.343500] 00 [ 28.343500] 07 [ 28.343500] 0b [ 28.343500] .
Helge