I'm probably the wrong guy to answer this question as I'm a noob into how sw os's work, but regarding linux memory access from the fpga:
A (hw) function in fpga (also with dma channels), can address any linux memory location (even sw restricted ones). If needed it is also possible to setup say like a 64KB dual port shared memory block inside the fpga fabric and have both fpga and linux access to that. On Thursday, 26 September 2019 15:48:35 UTC+2, [email protected] wrote: > > Sep 26, 2019, 01:29 by [email protected] <javascript:>: > > > Well current state is that PR (Partial Reconfiguration) is brand new to > the OS (Open Source) world, > > as > IntelfPGA (former Altera) "just" have promised it for their 19.1 > release (no lite version out yet), < > https://github.com/machinekit/mksocfpga/issues/100> > > on the contrary Xilinx have sneaked it out very quietly with their > Vivado 2019.1 release this summer < > https://github.com/machinekit/mksocfpga/issues/100> > > > > So while the idea has had time to settle in this old thread, the > possibility of implementation here in Machinekit is brand new.... :-) > > Michael B > > > Well, > and how it is with the memory? And with the bus connection between hard > ARM processor and FPGA fabric? Because now we have the HAL memory block > locked into RAM with HAL library enabled allocating and memory (alignment) > management from Linux side. But I presume that for FPGA-side components, > that would not be good enough and this memory block will have to be > directly in FPGA fabric so the components can use this space as a > "register", right? Will then be possible to atimically access this memory > (or variables there stored) both from Linux running on an ARM core and > component in FPGA fabric? (I mean as a direct memory access, zero-copy, not > some memory synchronization.) > > Cern. > > > > > On Wednesday, 25 September 2019 20:49:04 UTC+2, [email protected] wrote: > > > >> I am late to the party, I know, sorry, but this idea is very > interesting to me. As I know that perspectives and opinions change, I would > like to inquire about the current state. If all in this thread is still > valid or if it was redacted in some way? > >> > >> I need to wrap my head around this concept, but fundamentally speaking, > I see no reason why it should not be possible and even how it is that much > different from the current state. Because, currently the operation on HAL > is discrete and sequential. But only up to the point. As I see it, the > basic structure of HAL is the input and output of each block (component). > What is happening inside the component is a black box and of no particular > interest to the user or a system. That "happening" is enabled by so called > threads or tasks (on the Linux OS side), but actually from theoretical > point of view are also of no importance. > >> > >> Given the dawn of multicore, we can have multiple threads running > independent on each other on different isolated CPU/cores all reaching the > same memory. There is still the limit that threads on one instance has to > be run in increments of the first one, but I am not sure if that is real > limit or just something nobody changed from LinuxCNC days. (Because really, > it is nonsense.) > >> > >> If you can somehow pass-through the memory (I/O) from FPGA-side HAL to > Linux-side HAL, I think you are pretty much done and you have HAL in FPGA. > (HostMot2 FPGA firmware is also a HAL type, but you have the ugly > read/write functions. I call it the LinuxCNC way of thinking about it.) > >> > >> Because then it will be the same old, same old. > >> > >> And that opens up some very interesting possibilities. > >> > >> BTW, I have only very rough understanding about FPGA development. But > that SystemC looks extremely promising. > >> > >> Cern. > >> > > > > > > > > -- > > website: > http://www.machinekit.io <http://www.machinekit.io>> blog: > > http://blog.machinekit.io <http://blog.machinekit.io>> github: > > https://github.com/machinekit <https://github.com/machinekit> > > --- > > You received this message because you are subscribed to the Google > Groups "Machinekit" group. > > To unsubscribe from this group and stop receiving emails from it, send > an email to > [email protected] <javascript:> <mailto: > [email protected] <javascript:>>> . > > To view this discussion on the web visit > > https://groups.google.com/d/msgid/machinekit/a9420e6d-4f39-46e2-97c1-d4f7af69c89e%40googlegroups.com > > < > https://groups.google.com/d/msgid/machinekit/a9420e6d-4f39-46e2-97c1-d4f7af69c89e%40googlegroups.com?utm_medium=email&utm_source=footer>> > > . > > > > -- website: http://www.machinekit.io blog: http://blog.machinekit.io github: https://github.com/machinekit --- You received this message because you are subscribed to the Google Groups "Machinekit" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/machinekit/b430e32c-13cc-47d2-ab88-3da6aa31a293%40googlegroups.com.
