debugging cache coherency

src/drivers/elphel/histograms.c

@@ -124,7 +124,7 @@
 /**
  * \def MDF21(x) optional debug output 
  */
-#define ELPHEL_INFO 1
+// #define ELPHEL_INFO 1
 #if ELPHEL_DEBUG
 // only when commands are issued
   #define MDF21(x) { if (GLOBALPARS(G_DEBUG) & (1 <<21)) {printk("%s:%d:%s ",__FILE__,__LINE__,__FUNCTION__);x ;} }
@@ -460,7 +460,8 @@ int set_histograms  (int sensor_port,           ///< sensor port number (0..3)
 ///    __cpuc_flush_dcache_area(pElphel_buf->histograms_vaddr, pElphel_buf->histograms_size);
 ///    outer_inv_range(fpga_hist_phys, fpga_hist_phys +  pElphel_buf->histograms_size * PAGE_SIZE); // END - EXCLUSIVE!, CHANGE ORDER (AFTER ____cpuc_flush_dcache_area
 
-
+/// TODO: Cleanup, something is still wrong with caching. test with
+/// http://192.168.0.45/pnghist.cgi?sensor_port=0&sqrt=1&scale=5&average=5&height=128&fillz=1&linterpz=0&draw=2&colors=41&debug=9
     for (i=0; i<4; i++) if (needed & ( 1 << i )) {
         u32 phys_addr= fpga_hist_phys + PAGE_SIZE*(hist_frame + PARS_FRAMES * (sensor_chn + MAX_SENSORS *sensor_port)) + i*256*sizeof(u32); // start of selected color
         u32 * dma_data = &fpga_hist_data[sensor_port][sensor_chn][hist_frame][i][0];

src/drivers/elphel/imu_log393.c

@@ -57,7 +57,8 @@
 #include "cci2c.h"
 #include <uapi/elphel/x393_devices.h>
 
-
+//#define DEBUG_CACHE
+//#define DEBUG_CACHE_VALUE 0xa5
 #if 1
 #define D(x) x
 //#define D0(x) x
@@ -75,6 +76,7 @@
 //#define D1(x) x
 
 
+
 #define IS_103695_REV_A 1
 
 #define MULT_SAXI_CHN 0 ///< using channel 0 of a 4-channel DMA engine
@@ -87,7 +89,7 @@
 #define LOGGER_USE_IRQ 1
 
 //#define LOGGER_MIN_BYTES_AHEAD 128 // actual 64, but got some zeros, let's wait extra 64-byte record
-#define LOGGER_MIN_BYTES_AHEAD 256 // actual 64, but got some zeros, let's wait extra 64-byte record // with 128 was still getting 32x0?
+#define LOGGER_MIN_BYTES_AHEAD 64 // 256 // actual 64, but got some zeros, let's wait extra 64-byte record // with 128 was still getting 32x0?
 #ifdef NC353
 #define EXT_DMA_1_START \
         do { reg_bif_dma_rw_ch1_start c = {.run=1};\
@@ -416,6 +418,12 @@ static void set_logger_params(int which){ // 1 - program IOPINS, 2 - reset first
             dev_info(g_dev_ptr,"Stopping DMA\n");
             dev_dbg(g_dev_ptr,"Stopping DMA\n");
             logger_dma_stop();
+#ifdef DEBUG_CACHE
+            outer_inv_range(logger_phys, logger_phys + logger_size);
+            __cpuc_flush_dcache_area(logger_buffer, logger_size);
+            memset(logger_buffer, DEBUG_CACHE_VALUE, logger_size);
+#endif
+
         }
 
 #ifdef NC353
@@ -773,15 +781,17 @@ static int imu_open(struct inode *inode, struct file *filp) {
             set_logger_params(WHICH_INIT |
                     WHICH_RESET |
                     WHICH_RESET_SPI |
+					/*
                     WHICH_DIVISOR |
                     WHICH_RS232DIV |
                     WHICH_NMEA |
                     WHICH_CONFIG |
                     WHICH_REGISTERS |
                     WHICH_MESSAGE |
+                    */
                     WHICH_PERIOD |
                     WHICH_EN_DMA |
-					WHICH_CAMSYNC |
+//					WHICH_CAMSYNC |
                     WHICH_EN_LOGGER );
             numBytesRead=0;
         } else {
@@ -821,6 +831,14 @@ static int imu_release(struct inode *inode, struct file *filp) {
     int p = MINOR(inode->i_rdev);
     switch ( p ) {
     case DEV393_MINOR(DEV393_LOGGER) :
+#ifdef DEBUG_CACHE
+        dev_dbg(g_dev_ptr,"got LSEEK_IMU_STOP\n");
+    	set_logger_params(WHICH_RESET | WHICH_RESET_SPI);
+    	numBytesRead=0;
+		dev_info(g_dev_ptr,"Closing IMU device in debug mode. Stopping DMA\n");
+    	dev_dbg(g_dev_ptr,"Closing IMU device in debug mode. Stopping DMA\n");
+    	break;
+#endif
     case DEV393_MINOR(DEV393_LOGGER_CTRL):
         dev_info(g_dev_ptr,"Closing IMU device, numBytesWritten=0x%llx,  numBytesRead=0x%llx (only global pointer, does not include files opened in read mode)\n", numBytesWritten, numBytesRead);
         dev_dbg(g_dev_ptr,"Closing IMU device, numBytesWritten=0x%llx,  numBytesRead=0x%llx (only global pointer, does not include files opened in read mode)\n", numBytesWritten, numBytesRead);
@@ -1046,6 +1064,9 @@ static ssize_t imu_read(struct file * file, char * buf, size_t count, loff_t *of
         if (pe > logger_size) pe= logger_size;
 #endif
         /// copy all (or first part)
+        /// invalidate CPU L1 and L2 caches (in this order)
+        outer_inv_range(logger_phys + byteIndexRead, logger_phys + pe);
+        __cpuc_flush_dcache_area(&charDMABuf[byteIndexRead], (pe-byteIndexRead)); // L2_INVAL_SIZE);
         err=copy_to_user(buf, &charDMABuf[byteIndexRead], (pe-byteIndexRead));
         if (err) {
             dev_err(g_dev_ptr,"1. tried to copy 0x%x bytes from offset 0x%llx, result=0x%x\n", (pe-byteIndexRead), *off,err);
@@ -1058,7 +1079,9 @@ static ssize_t imu_read(struct file * file, char * buf, size_t count, loff_t *of
         dev_dbg(g_dev_ptr,"leftToRead=0x%08x thisNumBytesRead=0x%08x\n", leftToRead, (int) thisNumBytesRead);
         ///Do we need to copy from the beginning of the buffer?
         if (leftToRead>0) {
-            //          err=copy_to_user(buf, &charDMABuf[0], leftToRead);
+            /// invalidate CPU L1 and L2 caches (in this order)
+            outer_inv_range(logger_phys, logger_phys + leftToRead);
+            __cpuc_flush_dcache_area(&charDMABuf[0], leftToRead); // L2_INVAL_SIZE);
             err=copy_to_user(&buf[pe-byteIndexRead], &charDMABuf[0], leftToRead);
             byteIndexRead=0;
         }
@@ -1097,7 +1120,7 @@ static irqreturn_t logger_irq_handler(int irq,      ///< [in] interrupt number
                                       void *dev_id) ///< [in] pointer to driver's private data structure
                                                     ///< @return \e IRQ_HANDLED if interrupt was processed and \e IRQ_NONE otherwise
 {
-    x393_mult_saxi_al_t mult_saxi_dwp = x393_mult_saxi_pointers(MULT_SAXI_CHN);
+    x393_mult_saxi_al_t volatile mult_saxi_dwp = x393_mult_saxi_pointers(MULT_SAXI_CHN);
     if (mult_saxi_dwp.addr32 < logger_offs32){
         numBytesWrittenBase += logger_size;
     }

src/drivers/elphel/sensor_common.c

@@ -752,10 +752,11 @@ void tasklet_compressor_function(unsigned long arg)
         outer_inv_range(phys_addr_start, phys_addr_end);
         __cpuc_flush_dcache_area(virt_addr_start, L2_INVAL_SIZE);
     } else {
+    	// from pointer to the end of buffer
         phys_addr_end = phys_addr_start + (ccam_dma_size - DW2BYTE(jptr->fpga_cntr_prev) - 1);
         outer_inv_range(phys_addr_start, phys_addr_end);
         __cpuc_flush_dcache_area(virt_addr_start, ccam_dma_size - DW2BYTE(jptr->fpga_cntr_prev));
-
+        // wrapped-over part in the beginning of the buffer
         phys_addr_start = circbuf_priv_ptr[jptr->chn_num].phys_addr;
         phys_addr_end = phys_addr_start + (sz - ccam_dma_size - 1);
         virt_addr_start = circbuf_priv_ptr[jptr->chn_num].buf_ptr;