ESP32+W5500 corrupt TCP Packets

[ale-trevizoli]

Hi Paul!
I'm using your library for long time and I love it, but now I migrate some codes to ESP32 with W5500 (berfore I using MEGA2560+W5100), but on W5500 some random times I'm getting corrupt tcp data packets, so in WEBSERVER sample, if we make a lot of refresh pages in Browser we get corrupt characters on page. I tried to use ESP32+W5100 and ESP32+W5500 with OLD Ethernet2 library and it's not happen.
Could you help me to fix it?
You can simulate only taking ESP32+W5500. I'm using pins GPIO 5(CS),18(CLK),19(MISO) and 23(MOSI).
W5500 Module:
https://ae01.alicdn.com/kf/H6e6dbc78b77c493fb11edd8e8fa3f4aec/USR-ES1-w5500-chip-novo-spi-para-lan-ethernet-conversor-tcp-ip-mod.jpg

[PaulStoffregen]

Could you help me to fix it?

I do not use ESP, so I can not help if the problem is only on ESP32.

[PaulStoffregen]

I see you've already posted this same question twice on Arduino's repository.
arduino-libraries/Ethernet#120
arduino-libraries/Ethernet#124

[ale-trevizoli]

Sorry Paul, I tried to put this information o arduino-libraries too, to get some quick answer. I'm finishing one long time project and now I have this bug.
I really don't know if this is a ESP32 problem or if is an Library problem, what I know for searching on internet is that have some Registers are wrong on library like explained on this issue link: arduino-libraries/Ethernet#84 .
But I cannot say if is the root cause of this problem (is out of my capacity). I hope someone can say it.
About ESP32, based on my experience, since Ethernet2 library don't have this issue, using exactly same code, only changing ethernet.h to ethernet2.h, have a good chance to be something on library, what you think?
Have two problems to use Ethernet2 library, this is old and not maintained, and have another bugs on it.
But I'm not have enough knowledge to point where is this error on Ethernet library, only think I can say is Ethernet2 library not show this same error on same code and same hardware.
Other think that I could check is with W5100 not show this error too.
But now I have 100 units of W5500 ready to use on this project, so I'm trying to work to fix it.
Sorry my poor english, I'm from Brazil.

[PaulStoffregen]

Do you have a Teensy board? Can you reproduce the problems with a Teensy?

I will not spend time testing ESP32. But if the problem can be reproduced with Teensy, I will investigate.

[workpage2]

Please write an example code for w5500 (ethernet.h) + webserver.h
Everything works with wifi, but it doesn’t work with ethernet.

[Alexandr-zz2not1]

+1 get same problem. In ethernet2 no bugs, in ethernet some packet's have mistakes. I thought it was problem of electric design, but ethernet2 work fine. Any webserver example get unreadable parts of text or html or payload data, some times it'just happen :-/. Esp32+w5500 ethernet.h it seems like there are some problems in transport. P.s. I use that config to download update from another server to esp32 - waiting for problems. May be problem is only in outgoing from esp32 traffic.

[PaulStoffregen]

I am not maintaining Ethernet for non-Teensy boards. If this problem is only on ESP32, I do not intend to investigate.

Maybe someone else will take the time to test on ESP32 and find the cause of problems?

[Alexandr-zz2not1]

Ok. Thanks for unswer. Just looking for information. About library - it's cool tool. Try to find out something. Sorry for disturb. -- Отправлено из Mail.ru для Android понедельник, 12 октября 2020г., 16:21 +07:00 от Paul Stoffregen notifications@github.com :

…

I am not maintaining Ethernet for non-Teensy boards. If this problem is only on ESP32, I do not intend to investigate. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub , or unsubscribe .

[Alexandr-zz2not1]

so, I find some.. ...\utility\w5100.cpp-> uint16_t W5100Class::read(uint16_t addr, uint8_t *buf, uint16_t len): ... [code] SPI.transfer(cmd, 3); memset(buf, 0, len); SPI.transfer(buf, len); resetSS(); [/code] ... looks like esp32 is 32bit and doing something wrong when we ask memset() or something else. Ethernet2.h use [code] SPI.transfer(cmd, 3); for (uint16_t i=0; i<len; i++){ buf[i] = SPI.transfer(0); } resetSS(); [/code] and that works fine. no problem now. -_- пн, 12 окт. 2020 г. в 17:03, Александр Куклин <dizz34all@gmail.com>:

…

Ok. Thanks for unswer. Just looking for information. About library - it's cool tool. Try to find out something. Sorry for disturb. -- Отправлено из Mail.ru для Android понедельник, 12 октября 2020г., 16:21 +07:00 от Paul Stoffregen ***@***.***: I am not maintaining Ethernet for non-Teensy boards. If this problem is only on ESP32, I do not intend to investigate. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#39 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ARK2N575AYWZSVE2CJ2J2TDSKLDCNANCNFSM4KHD42YA> .

[PaulStoffregen]

Sounds like a bug in ESP's SPI library. Maybe report it there?

[Alexandr-zz2not1]

May be.. But spi working fine in my application. In that sketch, and no problem. I think if they have bug, they will find it. So for now, problem is only here. It is not big deal change 1 function. If spi lib change - may be brakes somthing else :-). Since SPI is in ESP32 core, i think it wold be easier find problem in much large problems. In my experience errors was rare. And i have not that much skill, to find problem in SPI.)) -- Отправлено из Mail.ru для Android воскресенье, 18 октября 2020г., 15:02 +07:00 от Paul Stoffregen notifications@github.com :

…

Sounds like a bug in ESP's SPI library. Maybe report it there? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub , or unsubscribe .

[PaulStoffregen]

If 3 calls to SPI.transfer(0) works, but SPI.transfer(buf, 3) fails, that is absolutely a bug in the SPI library.

[Alexandr-zz2not1]

I don't mind) may be there are something more else. Anyway. I just findound the bad place, simple fix it. If someone will find another route: add some libs or fix typos, add definition.. it would be cool) -- Отправлено из Mail.ru для Android воскресенье, 18 октября 2020г., 15:45 +07:00 от Paul Stoffregen notifications@github.com :

…

If 3 calls to SPI.transfer(0) works, but SPI.transfer(buf, 3) fails, that is absolutely a bug in the SPI library. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub , or unsubscribe .

[Alexandr-zz2not1]

Could it be, that memset doing before spi can transmit data, and some of buf became corrupted? Now i add yield(); before memset. And problem don't appear. -- Отправлено из Mail.ru для Android воскресенье, 18 октября 2020г., 15:45 +07:00 от Paul Stoffregen notifications@github.com :

…

If 3 calls to SPI.transfer(0) works, but SPI.transfer(buf, 3) fails, that is absolutely a bug in the SPI library. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub , or unsubscribe .