Zynq UltraScale MPSoC, DMA driver
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
TIA Jorge
On 07/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
alternatively, would it be acceptable - upstream - If I just init ECC as done by FSBL (ie, without a proper DMA driver, just accessing the controller)?
TIA Jorge
Hi,
On 6/7/21 8:41 PM, Jorge Ramirez-Ortiz, Foundries wrote:
On 07/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
alternatively, would it be acceptable - upstream - If I just init ECC as done by FSBL (ie, without a proper DMA driver, just accessing the controller)?
There are 2 things here. The first thing that would be the best to do it proper framework which is likely dma where dma_memset should be likely added. The second that this code should land in SPL which can't be big that's why this feature should be very small.
Based on above I think that would be good to have long term both of them in the tree. It means feel free to send short version and let's look at it and dma one should be hopefully pretty much using the same code via dma framework.
Thanks, Michal
On 08/06/21, Michal Simek wrote:
Hi,
On 6/7/21 8:41 PM, Jorge Ramirez-Ortiz, Foundries wrote:
On 07/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
alternatively, would it be acceptable - upstream - If I just init ECC as done by FSBL (ie, without a proper DMA driver, just accessing the controller)?
There are 2 things here. The first thing that would be the best to do it proper framework which is likely dma where dma_memset should be likely added. The second that this code should land in SPL which can't be big that's why this feature should be very small.
the way I am doing it ATM is just a verbatim of XFsbl_EccInit (grabs a channel and inits all memory to DEADBEEF). About 100 LOC total.
I belive the FSBL copyright notice allows for that (below excerpt).
[..] Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, [..]
but yes this just works as expected.
a bit trickier is where to init from - for the particular product I am working on I can hardcode the addresses and sizes and init right after board_init_f in spl.c...havent had the time to think about something portable yet - if you have a suggestion I could send both patches (one for the spl_ecc_init(addr, len) function and another one calling it)
thanks for your inputs
Based on above I think that would be good to have long term both of them in the tree. It means feel free to send short version and let's look at it and dma one should be hopefully pretty much using the same code via dma framework.
Thanks, Michal
On 08/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
On 08/06/21, Michal Simek wrote:
Hi,
On 6/7/21 8:41 PM, Jorge Ramirez-Ortiz, Foundries wrote:
On 07/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
alternatively, would it be acceptable - upstream - If I just init ECC as done by FSBL (ie, without a proper DMA driver, just accessing the controller)?
There are 2 things here. The first thing that would be the best to do it proper framework which is likely dma where dma_memset should be likely added. The second that this code should land in SPL which can't be big that's why this feature should be very small.
the way I am doing it ATM is just a verbatim of XFsbl_EccInit (grabs a channel and inits all memory to DEADBEEF). About 100 LOC total.
1:1 when ported to SPL (except the coding standards) https://github.com/Xilinx/embeddedsw/blob/master/lib/sw_apps/zynqmp_fsbl/src...
I belive the FSBL copyright notice allows for that (below excerpt).
[..] Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, [..]
but yes this just works as expected.
a bit trickier is where to init from - for the particular product I am working on I can hardcode the addresses and sizes and init right after board_init_f in spl.c...havent had the time to think about something portable yet - if you have a suggestion I could send both patches (one for the spl_ecc_init(addr, len) function and another one calling it)
thanks for your inputs
Based on above I think that would be good to have long term both of them in the tree. It means feel free to send short version and let's look at it and dma one should be hopefully pretty much using the same code via dma framework.
Thanks, Michal
On 6/8/21 2:30 PM, Jorge Ramirez-Ortiz, Foundries wrote:
On 08/06/21, Michal Simek wrote:
Hi,
On 6/7/21 8:41 PM, Jorge Ramirez-Ortiz, Foundries wrote:
On 07/06/21, Jorge Ramirez-Ortiz, Foundries wrote:
hi Michal
um, when we exchanged emails about enabling ECC support for MPSoC, I left with the understanding that there already was a DMA driver available in u-boot that I could use to initialize the memory.
do you have something in the works or will I have to write such a driver? compatible would be 'xlnx,zynqmp-dma-1.0'
alternatively, would it be acceptable - upstream - If I just init ECC as done by FSBL (ie, without a proper DMA driver, just accessing the controller)?
There are 2 things here. The first thing that would be the best to do it proper framework which is likely dma where dma_memset should be likely added. The second that this code should land in SPL which can't be big that's why this feature should be very small.
the way I am doing it ATM is just a verbatim of XFsbl_EccInit (grabs a channel and inits all memory to DEADBEEF). About 100 LOC total.
I belive the FSBL copyright notice allows for that (below excerpt).
[..] Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, [..]
but yes this just works as expected.
You should talk to lawyer about it and I can't give you any recommendation about it.
a bit trickier is where to init from - for the particular product I am working on I can hardcode the addresses and sizes and init right after board_init_f in spl.c...havent had the time to think about something portable yet - if you have a suggestion I could send both patches (one for the spl_ecc_init(addr, len) function and another one calling it)
SPL dtb doesn't have memory node which is a surprise. But I expect ECC is working on the memory which is written first. I am OK with adding memory base/size for ECC to Kconfig till someone has time to take a look at memory controller registers to find out which one has size in it that can be read and use.
Thanks, Michal
participants (2)
-
Jorge Ramirez-Ortiz, Foundries -
Michal Simek