Changes between Version 24 and Version 25 of kernel_syscalls

Timestamp:

Dec 29, 2014, 11:20:49 PM (9 years ago)

Author:

alain

Comment:

--

kernel_syscalls

@@ -55 +55 @@
 === 1) int '''_sys_nic_alloc'''( unsigned int is_rx ) ===
 This function allocates a private NIC_RX or NIC_TX channel, the associated chbuf, and a private CMA channel to the calling task. It register these channel indexes in the task context.
- * '''is_isr''' : boolean (RX transfer if non zero)
+ * '''is_rx''' : boolean (RX transfer if non zero)
 Returns the NIC channel index if success. Return -1 if no NIC channel available, or if no CMA channel available.
 
 === 2) int '''_sys_nic_start'''( unsigned int is_rx ) ===
-This function starts the NIC channel allocated to the calling task, and starts the CMA transfer between the internal NIC chbuf and the kernel chbuf allocated to the calling task.
- * '''is_isr''' : boolean (RX transfer if non zero)
+This function initializes the kernel chbuf allocated to the calling task. It allocates the distributed 4 Kbytes containers (2 containers per cluster) from the distributed kernel heap, and initializes the kernel chbuf  describing the containers status. This chbuf descriptor is allocated in the seg_kernel_data segment in cluster [0,0].
+It starts the NIC channel allocated to the calling task, and starts the CMA transfer between the NIC chbuf and the distributed kernel chbuf. The CMA component is configured in OUT_OF_ORDER transfer mode (non blocking when a DST buffer is full.
+ * '''is_rx''' : boolean (RX transfer if non zero)
 Returns 0 if success. Returns -1 if no NIC channel or no CMA channel allocated to the calling task.
 
 === 3) int '''_sys_nic_move'''(  unsigned int nic_channel,  unsigned int is_rx,  void* buffer ) ===
-This function moves one 4kbytes container between the kernel chbuf defined by the nic_channel argument, and an user buffer. 
- * '''nic_channel''' : selected nic_chbuf channel.
- * '''is_isr''' : boolean (RX transfer if non zero)
+This function moves one 4 Kbytes container from the kernel chbuf defined by the ''nic_channel'' argument, to an user buffer defined by the ''buffer'' argument. This blocking function enforces locality and parallelism: it scans the two containers  located in the same cluster as the calling task, and returns only when a full container has been found. 
+ * '''nic_channel''' : selected nic channel.
+ * '''is_rx''' : boolean (RX transfer if non zero)
  * '''buffer''' is the user buffer virtual base address.
-It supports concurrent kernel access to the chbuf, because It takes the lock protecting exclusive access to the chbuf pointer (index field). 
-It is blocking if the kernel chbuf is empty (for an RX transfer) or full (for a TX transfer). The bloking situation
-is bounded by a timeout defined by the GIET_NIC_TIMEOUT parameter defined in the ''giet_config.h'' file.
 Returns 0 if success, returns -1 if the user buffer address is illegal, or if the NIC channel is too large, or if the timeout is elapsed.
 
 === 4) int '''_sys_nic_stop'''( unsigned int is_rx ) ===
 This function stops the NIC and the CMA channels allocated to the calling task.
- * '''is_isr''' : boolean (RX transfer if non zero)
+ * '''is_rx''' : boolean (RX transfer if non zero)
 Returns 0 if success. Returns -1 if no NIC channel or no CMA channel allocated to the calling task.