VEO VESHM API (low-level)

VESHM API functions for VEO. More...

Functions
int	veo_shared_mem_open (void *vemva, size_t size, int syncnum, long long mode_flag)
	Register a VESHM area. More...
int	veo_shared_mem_close (void *vemva, size_t size, int syncnum, long long mode_flag)
	Unregister a VESHM area. More...
void *	veo_shared_mem_attach (veo_proc_handle *h, pid_t pid, void *veshm_vemva, size_t size, int syncnum, long long mode_flag)
	Attach a VESHM area. More...
int	veo_shared_mem_detach (veo_proc_handle *h, void *veshm_addr, long long mode_flag)
	Detach a VESHM area. More...

Detailed Description

VESHM API functions for VEO.

This API is a low-level API and is intended to be called by upper layer software. This API is not intended to be called by a user program. To use VEO VESHM API functions, include "ve_offload.h" header and "veo_veshm.h" header.

Function Documentation

int veo_shared_mem_open	(	void *	vemva,
		size_t	size,
		int	syncnum,
		long long	mode_flag
	)

Register a VESHM area.

Parameters

[in]	vemva	Virtual address of VESHM area (HMEM addr)
[in]	size	Size in byte
[in]	syncnum	Pair number of PCISYAR/PCISYMR Physical register number: 0-3 (Supported 0 only)
[in]	mode_flag	Mode flag which is ORed value of the following macros VE_REGISTER_PCI: Set up a memory as VESHM and register the memory with PCIATB. The values of vemva and size need to be aligned on the PCIATB page size (the PCIATB page size of almost all of the models of VEs is 64 MB). A caller process specifies this flag to allow VE processes on remote VEs and the local VE to access the specified memory. VE_REGISTER_NONE: Set up a memory as VESHM and NOT register the memory with PCIATB. The values of vemva and size need to be aligned on the page size of VE memory(default:64MB). A caller process specifies this flag to allow VE processes on the local VE to access the specified memory. VE_PCISYNC: Enable synchronization. VE_REGISTER_PCI must be specified. VE_SHM_RO: Set "Read Only" permission.

Return values

0	On success
-1	On failure (set errno) EINVAL Invalid value. (E.g., it is negative value, or too big) EINVAL Invalid value. (Different paze size from PCIATB) ENOMEM Creating a VESHM failed. EACCESS Permission denied. ECANCELED Operation canceled. ENOTSUP Library load or symbol retrieval failed. EBUSY Starting VESHM creation on VE or wating for result of it failed.

int veo_shared_mem_close	(	void *	vemva,
		size_t	size,
		int	syncnum,
		long long	mode_flag
	)

Unregister a VESHM area.

Parameters

[in]	vemva	Virtual address of VESHM area (HMEM addr)
[in]	size	Size in byte
[in]	syncnum	Pair number of PCISYAR/PCISYMR Physical register number: 0-3 (Supported 0 only)
[in]	mode_flag	Mode flag which is the same value as the argument of veo_shared_mem_open().

Return values

0	On success
-1	On failure (set errno) EINVAL Invalid value. ECANCELED Operation canceled. ENOTSUP Library load or symbol retrieval failed. EBUSY Starting VESHM close on VE or wating for result of it failed.

void* veo_shared_mem_attach	(	veo_proc_handle *	h,
		pid_t	pid,
		void *	veshm_vemva,
		size_t	size,
		int	syncnum,
		long long	mode_flag
	)

Attach a VESHM area.

This function basically does the same as ve_shared_mem_attach(). The difference point is that if VE_REGISTER_VEMVA is specified for the mode flag, HMEM address will be returned as attached address.

Parameters

[in]	h	VEO process handle
[in]	pid	Pid of an owner process that is gotten by veo_get_pid_from_hmem()
[in]	veshm_vemva	Virtual address of VESHM area (HMEM addr)
[in]	size	Size in byte
[in]	syncnum	Pair number of PCISYAR/PCISYMR Physical register number: 0-3 (Supported 0 only)
[in]	mode_flag	Mode flag which is ORed value of one of the following macros and the same value as the argument of veo_shared_mem_open() VE_REGISTER_VEHVA: VESHM on a remote VE will be attached to VEHVA (using DMAATB). A caller process specifies this flag to access the memory registered by a VE process running on a remote VE. The caller process can transfer data using veo_dma_post() or veo_dma_post_wait() with returned address. VE_REGISTER_VEMVA: VESHM on the local VE will be attached to VEMVA (using ATB). A caller process specifies this flag to access the memory registered by a VE process running on the local VE. The caller process can transfer data using veo_hmemcpy() with returned address. VE_MEM_LOCAL: An own memory will be attached to VEHVA (using DMAATB).

Note

A VESHM area is recognized by a combination of (vemva, size, syncnum, mode_flag) which are arguments of this function.

Return values

Attached-address	On success
0xffffffffffffffff	On failure (set errno) EINVAL Invalid value. EFAULT Bad address. ESRCH No such process. ENOENT No such memory. ENOMEM Cannot attach VESHM. EACCESS Cannot attach VESHM. EACCESS Permission denied. ECANCELED Operation canceled. EAGAIN Owner process is swapped-out. ENOTSUP Library load or symbol retrieval failed. EBUSY Starting VESHM attachment on VE or wating for result of it failed.

int veo_shared_mem_detach	(	veo_proc_handle *	h,
		void *	veshm_addr,
		long long	mode_flag
	)

Detach a VESHM area.

This function basically does the same as ve_shared_mem_detach(). The difference point is that if VE_REGISTER_VEMVA is specified for the mode flag, HMEM address is must specified as attached address.

Parameters

[in]	h	VEO process handle
[in]	veshm_addr	Virtual address of an attached VESHM area (HMEM addr in case of VE_REGISTERM_VEMVA)
[in]	mode_flag	Mode flag which is one of the following macros VE_REGISTER_VEHVA: VESHM on a remote VE will be attached to VEHVA (using DMAATB). VE_REGISTER_VEMVA: VESHM on the local VE will be attached to VEMVA (using ATB). VE_MEM_LOCAL: An own memory will be attached to VEHVA (using DMAATB).

Return values

0	On success
-1	On failure (set errno) EINVAL Invalid argument. ECANCELED No such memory. ENOTSUP Library load or symbol retrieval failed. EBUSY Starting VESHM detachment on VE or wating for result of it failed.