[U-Boot] [RFC PATCH] ARM: asm: types: Introduce DMA_ADDR_T_64BIT

dma_addr_t holds any valid DMA address. If the DMA API only uses 32-bit addresses, dma_addr_t need only be 32 bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32 bits, but drivers do memory-mapped I/O to ioremapped kernel virtual addresses, so they don't care about the size of the actual bus addresses. Also 32 bit ARM systems with LPAE enabled can use 64bit address space, but DMA still use 32bit address like in case of DRA7 and Keystone platforms.
This is inspired from the Linux kernel types implementation[1]
[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/include...
Signed-off-by: Lokesh Vutla lokeshvutla@ti.com --- arch/arm/include/asm/types.h | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/arch/arm/include/asm/types.h b/arch/arm/include/asm/types.h index 388058e..d108915 100644 --- a/arch/arm/include/asm/types.h +++ b/arch/arm/include/asm/types.h @@ -46,16 +46,29 @@ typedef unsigned long long u64; #endif /* CONFIG_ARM64 */
#ifdef CONFIG_PHYS_64BIT -typedef unsigned long long dma_addr_t; typedef unsigned long long phys_addr_t; typedef unsigned long long phys_size_t; #else /* DMA addresses are 32-bits wide */ -typedef u32 dma_addr_t; typedef unsigned long phys_addr_t; typedef unsigned long phys_size_t; #endif
+/* + * A dma_addr_t can hold any valid DMA address, i.e., any address returned + * by the DMA API. + * + * If the DMA API only uses 32-bit addresses, dma_addr_t need only be 32 + * bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32 bits, + * but drivers do memory-mapped I/O to ioremapped kernel virtual addresses, + * so they don't care about the size of the actual bus addresses. + */ +#ifdef CONFIG_DMA_ADDR_T_64BIT +typedef unsigned long long dma_addr_t; +#else +typedef u32 dma_addr_t; +#endif + #endif /* __KERNEL__ */
typedef unsigned long resource_size_t;

Hi Lokesh,
dma_addr_t holds any valid DMA address. If the DMA API only uses 32-bit addresses, dma_addr_t need only be 32 bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32 bits, but drivers do memory-mapped I/O to ioremapped kernel virtual addresses, so they don't care about the size of the actual bus addresses. Also 32 bit ARM systems with LPAE enabled can use 64bit address space, but DMA still use 32bit address like in case of DRA7 and Keystone platforms.
I've already stumbled upon this issue...
This is inspired from the Linux kernel types implementation[1]
[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/include...
Signed-off-by: Lokesh Vutla lokeshvutla@ti.com
arch/arm/include/asm/types.h | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/arch/arm/include/asm/types.h b/arch/arm/include/asm/types.h index 388058e..d108915 100644 --- a/arch/arm/include/asm/types.h +++ b/arch/arm/include/asm/types.h @@ -46,16 +46,29 @@ typedef unsigned long long u64; #endif /* CONFIG_ARM64 */
#ifdef CONFIG_PHYS_64BIT -typedef unsigned long long dma_addr_t; typedef unsigned long long phys_addr_t; typedef unsigned long long phys_size_t; #else /* DMA addresses are 32-bits wide */ -typedef u32 dma_addr_t; typedef unsigned long phys_addr_t; typedef unsigned long phys_size_t; #endif
+/*
- A dma_addr_t can hold any valid DMA address, i.e., any address
returned
- by the DMA API.
- If the DMA API only uses 32-bit addresses, dma_addr_t need only
be 32
- bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32
bits,
- but drivers do memory-mapped I/O to ioremapped kernel virtual
addresses,
- so they don't care about the size of the actual bus addresses.
- */
+#ifdef CONFIG_DMA_ADDR_T_64BIT
Generally this approach is correct, but please pay attention to the CONFIG_PHYS_64BIT.
The actual size of dma_addr_t (64 or 32 bits) is decided by defining or undefining CONFIG_PHYS_64BIT at arch/arm/include/asm/config.h. This is based on the status of CONFIG_ARM64.
To avoid regression we need to take into account status of CONFIG_ARM64 to be sure that CONFIG_DMA_ADDR_T_64BIT is set on ARM64 systems.
+typedef unsigned long long dma_addr_t; +#else +typedef u32 dma_addr_t; +#endif
#endif /* __KERNEL__ */
typedef unsigned long resource_size_t;

On Thursday 24 March 2016 02:11 PM, Lukasz Majewski wrote:
Hi Lokesh,
dma_addr_t holds any valid DMA address. If the DMA API only uses 32-bit addresses, dma_addr_t need only be 32 bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32 bits, but drivers do memory-mapped I/O to ioremapped kernel virtual addresses, so they don't care about the size of the actual bus addresses. Also 32 bit ARM systems with LPAE enabled can use 64bit address space, but DMA still use 32bit address like in case of DRA7 and Keystone platforms.
I've already stumbled upon this issue...
This is inspired from the Linux kernel types implementation[1]
[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/include...
Signed-off-by: Lokesh Vutla lokeshvutla@ti.com
arch/arm/include/asm/types.h | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/arch/arm/include/asm/types.h b/arch/arm/include/asm/types.h index 388058e..d108915 100644 --- a/arch/arm/include/asm/types.h +++ b/arch/arm/include/asm/types.h @@ -46,16 +46,29 @@ typedef unsigned long long u64; #endif /* CONFIG_ARM64 */
#ifdef CONFIG_PHYS_64BIT -typedef unsigned long long dma_addr_t; typedef unsigned long long phys_addr_t; typedef unsigned long long phys_size_t; #else /* DMA addresses are 32-bits wide */ -typedef u32 dma_addr_t; typedef unsigned long phys_addr_t; typedef unsigned long phys_size_t; #endif
+/*
- A dma_addr_t can hold any valid DMA address, i.e., any address
returned
- by the DMA API.
- If the DMA API only uses 32-bit addresses, dma_addr_t need only
be 32
- bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32
bits,
- but drivers do memory-mapped I/O to ioremapped kernel virtual
addresses,
- so they don't care about the size of the actual bus addresses.
- */
+#ifdef CONFIG_DMA_ADDR_T_64BIT
Generally this approach is correct, but please pay attention to the CONFIG_PHYS_64BIT.
The actual size of dma_addr_t (64 or 32 bits) is decided by defining or undefining CONFIG_PHYS_64BIT at arch/arm/include/asm/config.h. This is based on the status of CONFIG_ARM64.
To avoid regression we need to take into account status of CONFIG_ARM64 to be sure that CONFIG_DMA_ADDR_T_64BIT is set on ARM64 systems.
Good point. So we need to select DMA_ADDR_T_64BIT by default for all ARM64 systems. I guess the below diff should be sufficient along with the $subject patch?
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index e5f57ef..550ecde 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -7,6 +7,10 @@ config SYS_ARCH config ARM64 bool
+config DMA_ADDR_T_64BIT + bool + default y if ARM64 + config HAS_VBAR bool
Thanks and regards, Lokesh
+typedef unsigned long long dma_addr_t; +#else +typedef u32 dma_addr_t; +#endif
#endif /* __KERNEL__ */
typedef unsigned long resource_size_t;

Hi Lokesh,
On Thursday 24 March 2016 02:11 PM, Lukasz Majewski wrote:
Hi Lokesh,
dma_addr_t holds any valid DMA address. If the DMA API only uses 32-bit addresses, dma_addr_t need only be 32 bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32 bits, but drivers do memory-mapped I/O to ioremapped kernel virtual addresses, so they don't care about the size of the actual bus addresses. Also 32 bit ARM systems with LPAE enabled can use 64bit address space, but DMA still use 32bit address like in case of DRA7 and Keystone platforms.
I've already stumbled upon this issue...
This is inspired from the Linux kernel types implementation[1]
[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/include...
Signed-off-by: Lokesh Vutla lokeshvutla@ti.com
arch/arm/include/asm/types.h | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/arch/arm/include/asm/types.h b/arch/arm/include/asm/types.h index 388058e..d108915 100644 --- a/arch/arm/include/asm/types.h +++ b/arch/arm/include/asm/types.h @@ -46,16 +46,29 @@ typedef unsigned long long u64; #endif /* CONFIG_ARM64 */
#ifdef CONFIG_PHYS_64BIT -typedef unsigned long long dma_addr_t; typedef unsigned long long phys_addr_t; typedef unsigned long long phys_size_t; #else /* DMA addresses are 32-bits wide */ -typedef u32 dma_addr_t; typedef unsigned long phys_addr_t; typedef unsigned long phys_size_t; #endif
+/*
- A dma_addr_t can hold any valid DMA address, i.e., any address
returned
- by the DMA API.
- If the DMA API only uses 32-bit addresses, dma_addr_t need only
be 32
- bits wide. Bus addresses, e.g., PCI BARs, may be wider than 32
bits,
- but drivers do memory-mapped I/O to ioremapped kernel virtual
addresses,
- so they don't care about the size of the actual bus addresses.
- */
+#ifdef CONFIG_DMA_ADDR_T_64BIT
Generally this approach is correct, but please pay attention to the CONFIG_PHYS_64BIT.
The actual size of dma_addr_t (64 or 32 bits) is decided by defining or undefining CONFIG_PHYS_64BIT at arch/arm/include/asm/config.h. This is based on the status of CONFIG_ARM64.
To avoid regression we need to take into account status of CONFIG_ARM64 to be sure that CONFIG_DMA_ADDR_T_64BIT is set on ARM64 systems.
Good point. So we need to select DMA_ADDR_T_64BIT by default for all ARM64 systems. I guess the below diff should be sufficient along with the $subject patch?
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index e5f57ef..550ecde 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -7,6 +7,10 @@ config SYS_ARCH config ARM64 bool
+config DMA_ADDR_T_64BIT
- bool
- default y if ARM64
config HAS_VBAR bool
Thanks and regards, Lokesh
Acked-by: Lukasz Majewski l.majewski@samsung.com
+typedef unsigned long long dma_addr_t; +#else +typedef u32 dma_addr_t; +#endif
#endif /* __KERNEL__ */
typedef unsigned long resource_size_t;
participants (3)
-
Lokesh Vutla
-
Lokesh Vutla
-
Lukasz Majewski