User Tools

Site Tools




ddi_dma_attr - DMA attributes structure


#include <sys/ddidmareq.h>


Solaris DDI specific (Solaris DDI)


A ddi_dma_attr_t structure describes device- and DMA engine-specific attributes necessary to allocate DMA resources for a device. The driver might have to extend the attributes with bus-specific information, depending on the bus to which the device is connected.


uint_t dma_attr_version; /* version number */
uint64_t dma_attr_addr_lo; /* low DMA address range */
uint64_t dma_attr_addr_hi; /* high DMA address range */
uint64_t dma_attr_count_max; /* DMA counter register */
uint64_t dma_attr_align; /* DMA address alignment */
uint_t dma_attr_burstsizes; /* DMA burstsizes */
uint32_t dma_attr_minxfer; /* min effective DMA size */
uint64_t dma_attr_maxxfer; /* max DMA xfer size */
uint64_t dma_attr_seg; /* segment boundary */
int dma_attr_sgllen; /* s/g list length */
uint32_t dma_attr_granular; /* granularity of device */
uint_t dma_attr_flags; /* DMA transfer flags */

The dma_attr_version stores the version number of this DMA attribute structure. It should be set to DMA_ATTR_V0.

The dma_attr_addr_lo and dma_attr_addr_hi fields specify the address range the device’s DMA engine can access. The dma_attr_addr_lo field describes the inclusive lower 64-bit boundary. The dma_attr_addr_hi describes the inclusive upper 64-bit boundary. The system ensures that allocated DMA resources are within the range specified. See ddi_dma_cookie(9S).

The dma_attr_count_max describes an inclusive upper bound for the device’s DMA counter register. For example, 0xFFFFFF would describe a DMA engine with a 24-bit counter register. DMA resource allocation functions have to break up a DMA object into multiple DMA cookies if the size of the object exceeds the size of the DMA counter register.

The dma_attr_align specifies alignment requirements for allocated DMA resources. This field can be used to force more restrictive alignment than imposed by dma_attr_burstsizes or dma_attr_minxfer, such as alignment at a page boundary. Most drivers set this field to 1, indicating byte alignment.

The dma_attr_align only specifies alignment requirements for allocated DMA resources. The buffer passed to ddi_dma_addr_bind_handle(9F) or ddi_dma_buf_bind_handle(9F) must have an equally restrictive alignment (see ddi_dma_mem_alloc(9F)).

The dma_attr_burstsizes field describes the possible burst sizes the DMA engine of a device can accept. The format of the data sizes is binary, encoded in terms of powers of two. When DMA resources are allocated, the system can modify the burstsizes value to reflect the system limits. The driver must use the allowable burstsizes to program the DMA engine. See ddi_dma_burstsizes(9F).

The dma_attr_minxfer field describes the minimum effective DMA access size in units of bytes. DMA resources can be modified, depending on the presence and use of I/O caches and write buffers between the DMA engine and the memory object. This field is used to determine alignment and padding requirements for ddi_dma_mem_alloc(9F).

The dma_attr_maxxfer field describes the maximum effective DMA access size in units of bytes.

The dma_attr_seg field specifies segment boundary restrictions for allocated DMA resources. The system allocates DMA resources for the device so that the object does not span the segment boundary specified by dma_attr_seg. For example, a value of 0xFFFF means a DMA cookie must not cross a 64-Kbyte boundary. DMA resource allocation functions might have to break up a DMA object into multiple DMA cookies to enforce segment boundary restrictions. In this case, the transfer must be performed using scatter-gather I/O or multiple DMA windows.

The dma_attr_sgllen field describes the length of the DMA scatter/gather list of a device. Possible values are as follows:

< 0

Device DMA engine is not constrained by the size, for example, withDMA chaining.

= 0


= 1

Device DMA engine does not support scatter/gather such as third party DMA.

> 1

Device DMA engine uses scatter/gather. The dma_attr_sgllen value is the maximum number of entries in the list.

The dma_attr_granular field gives the granularity in bytes of the DMA transfer ability of the device. This value can be used to specify the sector size of a mass storage device. When a bind operation requires a partial mapping, this field is used to ensure that the sum of the sizes of the cookies in every DMA window, except the last one, is a whole multiple of granularity. However, if the device does not have a scatter-gather capability, it is impossible for the DDI to ensure the granularity. In this case, the value of the dma_attr_granular field should be 1.

The dma_attr_flags field can be set to a combination of:


Some platforms, such as SPARC systems, support what is called Direct Virtual Memory Access (DVMA). On these platforms, the device is provided with a virtual address by the system in order to perform the transfer. In this case, the underlying platform provides an IOMMU, which translates accesses to these virtual addresses into the proper physical addresses. Some of these platforms also support DMA. DDI_DMA_FORCE_PHYSICAL indicates that the system should return physical rather than virtual I/O addresses if the system supports both. If the system does not support physical DMA, the return value from ddi_dma_alloc_handle(9F) is DDI_DMA_BADATTR. In this case, the driver has to clear DDI_DMA_FORCE_PHYSICAL and retry the operation.


Using this value indicates that the driver is hardened: able to cope with the incorrect results of DMA operations that might result from an I/O fault. The value also indicates that the driver will use ddi_fm_dma_err_get(9F) to check DMA handles for faults on a regular basis.

If a DMA error is detected during a DMA access to an area mapped by such a handle, the system should not panic if possible, but should instead mark the DMA handle as having faulted.

This value is advisory: it tells the system that the driver can continue in the face of I/O faults. It does not guarantee that the system will not panic, as that depends on the nature of the fault and the capabilities of the system. It is quite legitimate for an implementation to ignore this flag and panic anyway.


This optional flag can be set if the DMA transactions associated with this handle are not required to observe strong DMA write ordering among themselves, nor with DMA write transactions of other handles.

The flag allows the host bridge to transfer data to and from memory more efficiently and might result in better DMA performance on some platforms.

Drivers for devices with hardware support, such as marking the bus transactions relaxed ordered, should not use this flag. Such drivers should use the hardware capability instead.


Example 1 Initializing the ddi_dma_attr_t Structure

Assume a device has the following DMA characteristics:

oFull 32-bit range addressable
o24-bit DMA counter register
oByte alignment
o4- and 8-byte burst sizes support
oMinimum effective transfer size of 1 bytes
o64 Mbyte minus 1 (26-bit) maximum transfer size limit
oMaximum segment size of 32 Kbyte
o17 scatter/gather list elements
o512-byte device transfer size granularity

The corresponding ddi_dma_attr_t structure is initialized as follows:

static ddi_dma_attr_t dma_attrs = {
DMA_ATTR_V0 /* version number */
(uint64_t)0x0, /* low address */
(uint64_t)0xffffffff, /* high address */
(uint64_t)0xffffff, /* DMA counter max */
(uint64_t)0x1 /* alignment */
0x0c, /* burst sizes */
0x1, /* minimum transfer size */
(uint64_t)0x3ffffff, /* maximum transfer size */
(uint64_t)0x7fff, /* maximum segment size */
17, /* scatter/gather list lgth */
512 /* granularity */
0 /* DMA flags */


See attributes(7) for descriptions of the following attributes:


attributes(7), ddi_dma_addr_bind_handle(9F), ddi_dma_alloc_handle(9F), ddi_dma_buf_bind_handle(9F), ddi_dma_burstsizes(9F), ddi_dma_mem_alloc(9F), ddi_dma_nextcookie(9F), ddi_fm_dma_err_get(9F), ddi_dma_cookie(9S)

Writing Device Drivers in Oracle Solaris 11.4

solaris/ddi_dma_attr.9s.txt · Last modified: 2023/07/19 08:57 by A User Not Logged in