Update MRAA timer functions

[TMRh20]

• Found MRAA timer functions not working properly
• Copy timing functions from SPIDEV

outdated suggestion

[2bndy5]

Verified working on my RPi4.

Its surprising that this is the solution to this. I noticed it last year, but wasn't sure what the problem actually was.

Should we also revert the timing code for SPIDEV driver? I'm curious if the char-dev API would still need the delayMicrosecond(1) in ce().

[TMRh20]

Should we also revert the timing code for SPIDEV driver?

Not sure what you mean. This is a straight copy of the SPIDEV timing code.

[2bndy5]

Oops. I thought SPIDEV was using only chrono lib.

[2bndy5]

I just tried only chrono approach, and the scanner example doesn't need the the delayMicroSeonds(1) in ce().

patch that I tested with char-dev-irq branch

diff --git a/RF24.cpp b/RF24.cpp
index 13db347..fd41e17 100644
--- a/RF24.cpp
+++ b/RF24.cpp
@@ -108,7 +108,7 @@ void RF24::ce(bool level)
 #endif
         digitalWrite(ce_pin, level);
 #ifdef RF24_LINUX
-        delayMicroseconds(1); // SPIDEV driver's GPIO needs a delay here (for some reason)
+        // delayMicroseconds(1); // SPIDEV driver's GPIO needs a delay here (for some reason)
 #endif
 #ifndef RF24_LINUX
     }
diff --git a/utility/SPIDEV/compatibility.cpp b/utility/SPIDEV/compatibility.cpp
index 6b96f65..8593fbd 100644
--- a/utility/SPIDEV/compatibility.cpp
+++ b/utility/SPIDEV/compatibility.cpp
@@ -1,10 +1,8 @@
 #include "compatibility.h"
-
-long long mtime, seconds, useconds;
-//static struct timeval start, end;
-//struct timespec start, end;
-#include <time.h>
 #include <chrono>
+//static struct timeval start, end;
+//static long mtime, seconds, useconds;
+
 /**********************************************************************/
 /**
  * This function is added in order to simulate arduino delay() function
@@ -12,35 +10,33 @@ long long mtime, seconds, useconds;
  */
 void __msleep(int milisec)
 {
-    struct timespec req; // = {0};
-    req.tv_sec = (time_t)milisec / 1000;
-    req.tv_nsec = (milisec % 1000) * 1000000L;
-    //nanosleep(&req, (struct timespec *)NULL);
-    clock_nanosleep(CLOCK_REALTIME, 0, &req, NULL);
+    struct timespec req = {0};
+    req.tv_sec = 0;
+    req.tv_nsec = milisec * 1000000L;
+    nanosleep(&req, (struct timespec*)NULL);
+    //usleep(milisec*1000);
 }
 
-void __usleep(int microsec)
+void __usleep(int milisec)
 {
-    struct timespec req; // = {0};
-    req.tv_sec = (time_t)microsec / 1000000;
-    req.tv_nsec = (microsec / 1000000) * 1000;
-    //nanosleep(&req, (struct timespec *)NULL);
-    clock_nanosleep(CLOCK_REALTIME, 0, &req, NULL);
+    struct timespec req = {0};
+    req.tv_sec = 0;
+    req.tv_nsec = milisec * 1000L;
+    nanosleep(&req, (struct timespec*)NULL);
+    //usleep(milisec);
 }
 
+auto start = std::chrono::steady_clock::now();
+
 /**
  * This function is added in order to simulate arduino millis() function
  */
-
-bool timerStarted = false;
-
 void __start_timer()
 {
+    //gettimeofday(&start, NULL);
 }
 
-auto start = std::chrono::steady_clock::now();
-
-uint32_t __millis()
+long __millis()
 {
     auto end = std::chrono::steady_clock::now();
 
diff --git a/utility/SPIDEV/compatibility.h b/utility/SPIDEV/compatibility.h
index 7015323..ffc6953 100644
--- a/utility/SPIDEV/compatibility.h
+++ b/utility/SPIDEV/compatibility.h
@@ -3,17 +3,15 @@
  * @author purinda
  *
  * Created on 24 June 2012, 3:08 PM
- * patch for safer monotonic clock & millis() correction for 64bit LDV 2018
  */
 
-#ifndef RF24_UTILITY_SPIDEV_COMPATIBLITY_H_
-#define RF24_UTILITY_SPIDEV_COMPATIBLITY_H_
+#ifndef RF24_UTILITY_MRAA_COMPATIBLITY_H_
+#define RF24_UTILITY_MRAA_COMPATIBLITY_H_
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#include <stdint.h> // for uintXX_t types
 #include <stddef.h>
 #include <time.h>
 #include <sys/time.h>
@@ -24,10 +22,10 @@ void __usleep(int milisec);
 
 void __start_timer();
 
-uint32_t __millis();
+long __millis();
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif // RF24_UTILITY_SPIDEV_COMPATIBLITY_H_
+#endif // RF24_UTILITY_MRAA_COMPATIBLITY_H_

[TMRh20]

Did you test it fully, because that code didn't work properly before? That's why it was changed, the delays don't work properly.

[2bndy5]

I just ran the scanner example, so no I didn't fully vet it. I'm currently looking into the difference between clock_nanosleep() vs nanosleep(). Changing clock_nanosleep() to use CLOCK_MONOTONIC is not the same as using just nanosleep() for some reason.

[2bndy5]

from hosted manpage:

Calling clock_nanosleep() with the value TIMER_ABSTIME not set in the flags argument and with a clock_id of CLOCK_REALTIME is equivalent to calling nanosleep() with the same rqtp and rmtp arguments.

[2bndy5]

Holy crap! It was a math error. This works without extra delay in ce():

@@ -23,7 +23,7 @@ void __usleep(int microsec)
 {
     struct timespec req; // = {0};
     req.tv_sec = (time_t)microsec / 1000000;
-    req.tv_nsec = (microsec / 1000000) * 1000;
+    req.tv_nsec = (microsec % 1000000) * 1000;
     //nanosleep(&req, (struct timespec *)NULL);
     clock_nanosleep(CLOCK_REALTIME, 0, &req, NULL);
 }

microsec / 1000000 results in a decimal value when microsec < 1000000, but since microsec and 1000000 are not floats, this becomes 0. And obviously 0 * 1000 isn't a what we want.

But I might need to sleep on it and check my math when I'm more refreshed.