Re: [RFC PATCH v2 07/11] riscv: cmo: Add dma-noncoherency support

From: twd2 <twd2.me@gmail.com>
To: guoren@kernel.org, anup.patel@wdc.com, palmerdabbelt@google.com,
	arnd@arndb.de, wens@csie.org, maxime@cerno.tech,
	drew@beagleboard.org, liush@allwinnertech.com,
	lazyparser@gmail.com, wefu@redhat.com
Cc: linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
	linux-arch@vger.kernel.org, linux-sunxi@lists.linux.dev,
	Guo Ren <guoren@linux.alibaba.com>,
	Christoph Hellwig <hch@lst.de>
Subject: Re: [RFC PATCH v2 07/11] riscv: cmo: Add dma-noncoherency support
Date: Sun, 17 Oct 2021 17:28:43 +0800	[thread overview]
Message-ID: <e926ec8a-e993-da89-3c8b-0c59272f1751@gmail.com> (raw)
In-Reply-To: <1622970249-50770-11-git-send-email-guoren@kernel.org>

On 2021/6/6 17:04, guoren@kernel.org wrote:
> From: Guo Ren <guoren@linux.alibaba.com>
>
> To support DMA device in a non-coherent interconnect SOC system,
> we need the below facilities:
>  - Changing a virtual memory mapping region attributes from
>    cacheable to noncache + strong order which used in DMA
>    descriptors.
>  - Add noncache + weakorder virtual memory attributes for dma
>    mapping.
>  - Syncing the cache with memory before DMA start and after DMA
>    end with vendor custom CMO instructions.
>
> This patch enables linux kernel generic dma-noncoherency
> infrastructure and introduces new sbi_ecall API for dma_sync.
>
> @@ -27,6 +27,7 @@ enum sbi_ext_id {
> +       SBI_EXT_DMA = 0xAB150401,
>
> Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
> Signed-off-by: Liu Shaohua <liush@allwinnertech.com>
> Cc: Palmer Dabbelt <palmerdabbelt@google.com>
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Anup Patel <anup.patel@wdc.com>
> Cc: Arnd Bergmann <arnd@arndb.de>
> Cc: Drew Fustini <drew@beagleboard.org>
> Cc: Wei Fu <wefu@redhat.com>
> Cc: Wei Wu <lazyparser@gmail.com>
> Cc: Chen-Yu Tsai <wens@csie.org>
> Cc: Maxime Ripard <maxime@cerno.tech>
> ---
>  arch/riscv/Kconfig               |  5 ++++
>  arch/riscv/include/asm/pgtable.h | 26 ++++++++++++++++++++
>  arch/riscv/include/asm/sbi.h     | 15 ++++++++++++
>  arch/riscv/kernel/sbi.c          | 19 ++++++++++++++
>  arch/riscv/mm/Makefile           |  1 +
>  arch/riscv/mm/dma-mapping.c      | 53 ++++++++++++++++++++++++++++++++++++++++
>  6 files changed, 119 insertions(+)
>  create mode 100644 arch/riscv/mm/dma-mapping.c
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index 05c4976..817a9bb 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -20,6 +20,10 @@ config RISCV
>  	select ARCH_HAS_DEBUG_VM_PGTABLE
>  	select ARCH_HAS_DEBUG_VIRTUAL if MMU
>  	select ARCH_HAS_DEBUG_WX
> +	select ARCH_HAS_DMA_PREP_COHERENT
> +	select ARCH_HAS_SYNC_DMA_FOR_CPU
> +	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
> +	select ARCH_HAS_DMA_WRITE_COMBINE
>  	select ARCH_HAS_FORTIFY_SOURCE
>  	select ARCH_HAS_GCOV_PROFILE_ALL
>  	select ARCH_HAS_GIGANTIC_PAGE
> @@ -43,6 +47,7 @@ config RISCV
>  	select CLONE_BACKWARDS
>  	select CLINT_TIMER if !MMU
>  	select COMMON_CLK
> +	select DMA_DIRECT_REMAP
>  	select EDAC_SUPPORT
>  	select GENERIC_ARCH_TOPOLOGY if SMP
>  	select GENERIC_ATOMIC64 if !64BIT
> diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
> index 6ddeb49..e1a82b6 100644
> --- a/arch/riscv/include/asm/pgtable.h
> +++ b/arch/riscv/include/asm/pgtable.h
> @@ -462,6 +462,32 @@ static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
>  	return ptep_test_and_clear_young(vma, address, ptep);
>  }
>  
> +#define pgprot_noncached pgprot_noncached
> +static inline pgprot_t pgprot_noncached(pgprot_t _prot)
> +{
> +	unsigned long prot = pgprot_val(_prot);
> +
> +	prot &= ~_PAGE_DMA_MASK;
> +	prot |= _PAGE_DMA_IO;
> +
> +	return __pgprot(prot);
> +}
> +
> +#define pgprot_writecombine pgprot_writecombine
> +static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
> +{
> +	unsigned long prot = pgprot_val(_prot);
> +
> +	prot &= ~_PAGE_DMA_MASK;
> +	prot |= _PAGE_DMA_WC;
> +
> +	return __pgprot(prot);
> +}
> +
> +#define __HAVE_PHYS_MEM_ACCESS_PROT
> +extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
> +				     unsigned long size, pgprot_t vma_prot);
> +
>  /*
>   * Encode and decode a swap entry
>   *
> diff --git a/arch/riscv/include/asm/sbi.h b/arch/riscv/include/asm/sbi.h
> index 0d42693..133e88a 100644
> --- a/arch/riscv/include/asm/sbi.h
> +++ b/arch/riscv/include/asm/sbi.h
> @@ -27,6 +27,7 @@ enum sbi_ext_id {
>  	SBI_EXT_IPI = 0x735049,
>  	SBI_EXT_RFENCE = 0x52464E43,
>  	SBI_EXT_HSM = 0x48534D,
> +	SBI_EXT_DMA = 0xAB150401,
>  };
>  
>  enum sbi_ext_base_fid {
> @@ -63,6 +64,17 @@ enum sbi_ext_hsm_fid {
>  	SBI_EXT_HSM_HART_STATUS,
>  };
>  
> +enum sbi_ext_dma_fid {
> +	SBI_DMA_SYNC = 0,
> +};
> +
> +enum sbi_dma_sync_data_direction {
> +	SBI_DMA_BIDIRECTIONAL = 0,
> +	SBI_DMA_TO_DEVICE = 1,
> +	SBI_DMA_FROM_DEVICE = 2,
> +	SBI_DMA_NONE = 3,
> +};
> +
>  enum sbi_hsm_hart_status {
>  	SBI_HSM_HART_STATUS_STARTED = 0,
>  	SBI_HSM_HART_STATUS_STOPPED,
> @@ -128,6 +140,9 @@ int sbi_remote_hfence_vvma_asid(const unsigned long *hart_mask,
>  				unsigned long size,
>  				unsigned long asid);
>  int sbi_probe_extension(int ext);
> +void sbi_dma_sync(unsigned long start,
> +		  unsigned long size,
> +		  enum sbi_dma_sync_data_direction dir);
>  
>  /* Check if current SBI specification version is 0.1 or not */
>  static inline int sbi_spec_is_0_1(void)
> diff --git a/arch/riscv/kernel/sbi.c b/arch/riscv/kernel/sbi.c
> index 7402a41..c936019 100644
> --- a/arch/riscv/kernel/sbi.c
> +++ b/arch/riscv/kernel/sbi.c
> @@ -521,6 +521,25 @@ int sbi_probe_extension(int extid)
>  }
>  EXPORT_SYMBOL(sbi_probe_extension);
>  
> +void sbi_dma_sync(unsigned long start,
> +		  unsigned long size,
> +		  enum sbi_dma_sync_data_direction dir)
> +{
> +#if 0
> +	sbi_ecall(SBI_EXT_DMA, SBI_DMA_SYNC, start, size, dir,
> +		  0, 0, 0);
> +#else
> +	/* Just for try, it should be in sbi ecall and will be removed before merged */
> +	register unsigned long i asm("a0") = start & ~(L1_CACHE_BYTES - 1);
> +
> +	for (; i < ALIGN(start + size, L1_CACHE_BYTES); i += L1_CACHE_BYTES)
> +		__asm__ __volatile__(".long 0x02b5000b");
> +

Hi, I'm trying to use this patch for my D1 board.

Though the above code will be removed, I notice that the use of the inline assembly is wrong and `i` (i.e. `a0`) might not be correctly passed to the assembly code when we are using some other compilers.

It should be `__asm__ __volatile__(".long 0x02b5000b" : : "r"(i))`.

Thanks,
Wende

> +	__asm__ __volatile__(".long 0x01b0000b");
> +#endif
> +}
> +EXPORT_SYMBOL(sbi_dma_sync);
> +
>  static long __sbi_base_ecall(int fid)
>  {
>  	struct sbiret ret;
> diff --git a/arch/riscv/mm/Makefile b/arch/riscv/mm/Makefile
> index 7ebaef1..ca0ff90 100644
> --- a/arch/riscv/mm/Makefile
> +++ b/arch/riscv/mm/Makefile
> @@ -13,6 +13,7 @@ obj-y += extable.o
>  obj-$(CONFIG_MMU) += fault.o pageattr.o
>  obj-y += cacheflush.o
>  obj-y += context.o
> +obj-y += dma-mapping.o
>  
>  ifeq ($(CONFIG_MMU),y)
>  obj-$(CONFIG_SMP) += tlbflush.o
> diff --git a/arch/riscv/mm/dma-mapping.c b/arch/riscv/mm/dma-mapping.c
> new file mode 100644
> index 00000000..4afd9dc
> --- /dev/null
> +++ b/arch/riscv/mm/dma-mapping.c
> @@ -0,0 +1,53 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include <linux/dma-map-ops.h>
> +#include <asm/sbi.h>
> +
> +void arch_dma_prep_coherent(struct page *page, size_t size)
> +{
> +	void *ptr = page_address(page);
> +
> +	memset(ptr, 0, size);
> +	sbi_dma_sync(page_to_phys(page), size, SBI_DMA_BIDIRECTIONAL);
> +}
> +
> +void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,
> +		enum dma_data_direction dir)
> +{
> +	switch (dir) {
> +	case DMA_TO_DEVICE:
> +	case DMA_FROM_DEVICE:
> +	case DMA_BIDIRECTIONAL:
> +		sbi_dma_sync(paddr, size, dir);
> +		break;
> +	default:
> +		BUG();
> +	}
> +}
> +
> +void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size,
> +		enum dma_data_direction dir)
> +{
> +	switch (dir) {
> +	case DMA_TO_DEVICE:
> +		return;
> +	case DMA_FROM_DEVICE:
> +	case DMA_BIDIRECTIONAL:
> +		sbi_dma_sync(paddr, size, dir);
> +		break;
> +	default:
> +		BUG();
> +	}
> +}
> +
> +pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
> +			      unsigned long size, pgprot_t vma_prot)
> +{
> +	if (!pfn_valid(pfn))
> +		return pgprot_noncached(vma_prot);
> +	else if (file->f_flags & O_SYNC)
> +		return pgprot_writecombine(vma_prot);
> +
> +	return vma_prot;
> +}
> +EXPORT_SYMBOL(phys_mem_access_prot);

-- 
Best wishes!
Sincerely,
twd2
https://twd2.me