Changes between Version 77 and Version 78 of library_stdio

Timestamp:

Apr 13, 2015, 9:29:15 AM (9 years ago)

Author:

alain

Comment:

--

library_stdio

@@ -47 +47 @@
 The GIET_VM allows user applications to use hardware accelerators (called coprocessors). These coprocessors can be distributed in the architecture, but there is at most one coprocessor per cluster. To be supported by the GIET_VM, a coprocessor must use the '''vci_mwmr_dma''' component, that is a generic multi-channels DMA controller.
 
-On the coprocessor side, it provides TO_COPROC or FROM_COPROC ports, that implement FIFO interfaces. Coprocessor can request to transfer one or several bursts of 32 bits words on a TO_COPROC or FROM_COPROC port, without address. The burst size (generally a cache line), the number of TO_COPROC and FROM_COPROC ports, and the number of bursts for a given port are hardware parameters, depending on the coprocessor type. Each port define a private communication channel between the coprocessor and a user memory buffer. The total number of channels cannot be larger than 16. A channel is identified by an index, and the TO_COPROC channels have the smallest indexes.
-
-The '''vci_mwmr_dma''' component implements two transfer modes that can be configured by the user application through dedicate system calls.
-The user application must use the ''giet_coproc_alloc()'' system call to get a private hardware coprocessor. Then it must use the ''giet_coproc_channel_init()'' system call to configure the TO_COPROC/FROM_COPROC channels.
-The coprocessor and the communication channels are activated by the ''giet_coproc_run()'' system call. 
+The '''vci_mwmr_dma''' component provide to coprocessor a variable number of TO_COPROC or FROM_COPROC ports, that implement FIFO interfaces. Coprocessor can request to transfer one or several bursts of 32 bits words, without address. The burst size (generally a cache line), the number of TO_COPROC and FROM_COPROC ports, and the number of bursts for a given port are hardware parameters, depending on the coprocessor type. Each port define a private communication channel between the coprocessor and a user memory buffer. The total number of channels cannot be larger than 16. A channel is identified by an index, and the TO_COPROC channels have the smallest indexes.
+
+The '''vci_mwmr_dma''' component implements two transfer modes that can be defined by the user application through dedicate system calls.
+The user application must use the ''giet_coproc_alloc()'' system call to get a private hardware coprocessor. Then it must use the ''giet_coproc_channel_init()'' system call to configure the TO_COPROC/FROM_COPROC channels. Finally, the coprocessor and the communication channels are activated by the ''giet_coproc_run()'' system call. 
  * In '''MODE_MWMR''', each channel FSM  implements the 7 steps MWMR protocol, and transfer an "infinite" data stream, between one coprocessor port and a MWMR software FIFO in memory. The ''giet_coproc_run()'' system call is non blocking, as the synchronisation is done through the MWMR FIFOs (no transfer completion event). The MWR IRQ is only asserted if a VCI error is reported in memory access. 
- * In '''MODE_DMA_IRQ''' or '''MODE_DMA_NO_IRQ''', each channel FSM transfers a single buffer between one coprocessor port and the memory, and keep blocked when the transfer is completed. In both DMA modes, the user application must use the ''giet_coproc_completed()'' system call to detect global completion and report the possible addressing errors.
-   - In '''MODE_DMA_IRQ''', the calling task is descheduled, after coprocessor activation, in the ''giet_coproc_run()'' system call. It is rescheduled by the MWR IRQ signaling the global completion.  
-   - In '''MODE_DMA_NO_IRQ''', the ''giet_coproc_run()'' system call returns after coprocessor activation, and the ''giet_coproc_completed()'' system call uses a polling to detect  completion and report errors.
+ * In '''MODE_DMA_IRQ''' or '''MODE_DMA_NO_IRQ''', each channel FSM transfers a single buffer between one coprocessor port and the memory, and keep blocked when the transfer is completed. 
+   - In '''MODE_DMA_IRQ''', the calling task is descheduled, after coprocessor activation, in the ''giet_coproc_run()'' system call. It is rescheduled by the MWR IRQ signaling the global completion. The _mwr_isr() scan all channels status registers to report possible addressing errors, and reset the communication channels.
+   - In '''MODE_DMA_NO_IRQ''', the ''giet_coproc_run()'' system call returns after coprocessor activation, and the user application must use the blocking ''giet_coproc_completed()'' system call that directly scan the channels registers to detect  completion, report errors, and reset the channels.
 
 '''WARNING''': For a given coprocessor, all channels must be running in the same mode.