You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@nuttx.apache.org by "danielappiagyei-bc (via GitHub)" <gi...@apache.org> on 2023/10/15 01:12:03 UTC
[I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
danielappiagyei-bc opened a new issue, #10912:
URL: https://github.com/apache/nuttx/issues/10912
## Bug description
When using the LPUART serial driver with DMA for RX, I am seeing several dropped, missing, or otherwise incorrect bytes in the rxfifo (`g_lpuartXfifo`) the DMA places bytes in.
## Partial solution
The major problem looks to be the `imxrt_dma_nextrx()` function [here](https://github.com/apache/nuttx/blob/master/arch/arm/src/imxrt/imxrt_serial.c#L1373 ) which can return one index past the end of the buffer when `dmaresidual` is 0 (all bytes in the dma transfer have been sent). The return value of this function is used without verifying that it's a valid index in the buffer. Correcting this bug causes missed bytes to occur less often (from several times a minute to only once or twice every few minutes) but does not fix the problem completely.
The only permanent fix I found is placing the dma's [rxfifo](https://github.com/apache/nuttx/blob/master/arch/arm/src/imxrt/imxrt_serial.c#L711) inside the IMXRT's tightly coupled memory (TCM) region via `__attribute__((section("<tcm region>")))`, which is non-cacheable and super fast compared to SRAM. No incorrect bytes occur when I do this. This leads me to think there might be a caching issue.
Aside from that, I tried disabling write through cache (`CONFIG_ARMV7M_DCACHE_WRITETHROUGH=n`) since [the chip's errata](https://www.nxp.com/docs/en/errata/IMXRT1064CE_A.pdf) mentions "_Cortex-M7: Write-Through stores and loads may return incorrect data_" but that did not solve the issue.
## Relevant Info
I setup the IMXRT to receive the same ~36 byte message over UART from another processor. The first few bytes of the message are a header containing sync bytes, a length, and a CRC32 precalculated by the other processor. I log anytime there's a mismatch in expected data. When disabling DMA and using interrupts for each character received, there are no incorrect bytes. When using DMA, there are. I've confirmed bytes aren't being dropped by framing, parity, or noise errors. I've also confirmed that the upper layer of the serial driver isn't causing this.
Other info:
- sending the message every 10ms
- 480600 baud rate
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "xiaoxiang781216 (via GitHub)" <gi...@apache.org>.
xiaoxiang781216 commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763284840
@davids5 is more familiar chip driver than me, do you have any suggestion? @davids5 .
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "davids5 (via GitHub)" <gi...@apache.org>.
davids5 commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1777579092
@danielappiagyei-bc This resolve the issue https://github.com/apache/nuttx/pull/11020
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "danielappiagyei-bc (via GitHub)" <gi...@apache.org>.
danielappiagyei-bc commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763724200
> @danielappiagyei-bc
>
> Great problem write up! This will help very much. Thank you for it! Hopefully others can learn to put more detail in their issues and pull requests.
>
> I will work on this is the LPI2C issue this coming week.
Absolutely, and thank you David! We were on nuttx 10.x and were in need of a dma driver to speed up our uart comms. I spent a week reading the processor datasheet and exploring the code in `imxrt_edma.c` in preparation of writing my own driver. Luckily before implementing I stumbled upon your serial dma addition in nuttx 12 and we were able to upgrade and take advantage of all your hard work :-)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "danielappiagyei-bc (via GitHub)" <gi...@apache.org>.
danielappiagyei-bc commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763726864
@davids5 do you have any idea why I would need to put the rxfifo's into non-cacheable TCM for things to work properly? Additionally, have you seen this behavior before? Also also, should I make a commit that puts the rxfifos into TCM since that is what worked for me or should we snoop around to see if there's another explanation for what's going on
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "pkarashchenko (via GitHub)" <gi...@apache.org>.
pkarashchenko commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763314448
What is the size of the RX FIFO buffer in your case? Is that also a multiple of dcache line size?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "davids5 (via GitHub)" <gi...@apache.org>.
davids5 commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1764311675
@danielappiagyei-bc
It has been found in our testing on the 1170 as well. We also found that placing it in no cacheable region fixed it as well. I will be trying to figure out the root cause this week.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "davids5 (via GitHub)" <gi...@apache.org>.
davids5 commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1777212451
@danielappiagyei-bc - I am looking into this. The DMA count when using the channel is circular mode counts from RXDMA_BUFFER_SIZE 32-31....1-32 in HW and is never 0 So that RXDMA_BUFFER_SIZE - dmaresidual will be 0 - 31.
Unless that is a race on the read. I have added an ASSERT and continue to test this.....
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "pkarashchenko (via GitHub)" <gi...@apache.org>.
pkarashchenko commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1764416504
> I have a [PR](https://github.com/apache/nuttx/pull/10911) up which fixes the overrun in `imxrt_dma_nextrx()`. The other half of the fix which I haven't included in the PR is to place the `g_lpuart{1-8}rxfifo` into TCM. It fixes the issue but what I don't understand is why our call to `up_invalidate_dcache()` [here](https://github.com/apache/nuttx/blob/master/arch/arm/src/imxrt/imxrt_serial.c#L2724) isn't sufficient.
>
> @davids5 @xiaoxiang781216 Do either of u have any thoughts on what's going on?
Maybe you can move
```
up_invalidate_dcache((uintptr_t)priv->rxfifo,
(uintptr_t)priv->rxfifo + RXDMA_BUFFER_SIZE);
```
from `imxrt_dma_rxcallback` to `imxrt_dma_setup` before `imxrt_dmach_xfrsetup(priv->rxdma , &config);` call? Not sure if that will work, but worth of trying
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "danielappiagyei-bc (via GitHub)" <gi...@apache.org>.
danielappiagyei-bc commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763226964
I have a [PR](https://github.com/apache/nuttx/pull/10911) up which fixes the overrun in `imxrt_dma_nextrx()`. The other half of the fix which I haven't included in the PR is to place the `g_lpuart{1-8}rxfifo` into TCM. It fixes the issue but what I don't understand is why our call to `up_invalidate_dcache()` [here](https://github.com/apache/nuttx/blob/master/arch/arm/src/imxrt/imxrt_serial.c#L2724) isn't sufficient.
@davids5 @xiaoxiang781216 Do either of u have any thoughts on what's going on?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "danielappiagyei-bc (via GitHub)" <gi...@apache.org>.
danielappiagyei-bc commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763721605
> What is the size of the RX FIFO buffer in your case? Is that also a multiple of dcache line size?
Indeed! I've used 128 and 256 with no success
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "davids5 (via GitHub)" <gi...@apache.org>.
davids5 commented on issue #10912:
URL: https://github.com/apache/nuttx/issues/10912#issuecomment-1763327149
@danielappiagyei-bc
Great problem write up! This will help very much. Thank you for it!
Hopefully others can learn to put more detail in their issues and pull requests.
I will work on this is the LPI2C issue this coming week.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] imxrt: serial dma rx occasionally reports incorrect byte [nuttx]
Posted by "xiaoxiang781216 (via GitHub)" <gi...@apache.org>.
xiaoxiang781216 closed issue #10912: imxrt: serial dma rx occasionally reports incorrect byte
URL: https://github.com/apache/nuttx/issues/10912
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org