local transport mmap write does not complete

sr-cdf commented 6 days ago

If i write all ones or all zeros to the mixer lo phase increment shared bram and read back after a time.sleep(0), i get back what i set.

But if i write random bytes (which better represents the actual formatted lo phase increment values) i do not always get back what i write.

[edit] Increasing the sleep time does not help.

The following loop helps but surprisingly, it can take many repetitions before the data is read back correctly.

write the bytes
sleep
read the bytes
if bytes read == bytes written break
else go to 1

Some example code:

(py3venv) casper@localhost:~$ sudo ~/py3venv/bin/ipython

import souk_mkid_readout
import time
import numpy as np

r = souk_mkid_readout.SoukMkidReadout('localhost',configfile='/home/casper/src/souk-firmware/software/control_sw/config/souk-single-pipeline-krm-2G.yaml',local=False)
r.program()
r.initialize()
r_fast = souk_mkid_readout.SoukMkidReadout('localhost',configfile='/home/casper/src/souk-firmware/software/control_sw/config/souk-single-pipeline-krm-2G.yaml',local=True)

i=0
ramname = f'{r.mixer.prefix}tx_lo{i}_phase_inc'  # 'p0_mix_tx_lo0_phase_inc': Sbram:p0_mix_tx_lo0_phase_inc,
phase_addrs_tx = []
phase_addrs_tx += [r_fast.mixer.host.transport._get_device_address(ramname)]

nbytes = r.mixer._n_serial_chans * 4
maxwrite=512
nwrite = (nbytes // maxwrite)

b0=bytes([0]*nbytes) # all 0x00
b1=bytes([255]*nbytes) # all 0xFF
br=bytes(list(np.floor(np.random.uniform(0,256,nbytes)).astype(int)))

#must write in 512 byte chunks?
for i in range(len(phase_addrs_tx)):
    for j in range(nwrite):
        print(i,j)
        #write all 0x00
        raw0 = b0[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = raw0
        time.sleep(0)
        ret0 = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]

        #write all 0xff
        raw1 = b1[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = raw1
        time.sleep(0)
        ret1 = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]

        #write random bytes
        rawr = br[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = rawr
        time.sleep(0)
        retr = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]

        print(ret0==raw0, ret1==raw1, retr==rawr)

The result is not always the same but generally there are some Falses in the third column:

0 0
True True False
0 1
True True True
0 2
True True True
0 3
True True True
0 4
True True False
0 5
True True True
0 6
True True True
0 7
True True True
0 8
True True True
0 9
True True True
0 10
True True True
0 11
True True True
0 12
True True True
0 13
True True True
0 14
True True False
0 15
True True True

sr-cdf commented 6 days ago

If i write a ramp of 8bit ints, instead of random bytes, i see this:


bramp=bytes((nbytes//256)*list(np.arange(256).astype(int)))
for i in range(len(phase_addrs_tx)):
    for j in range(nwrite):
        print(i,j)
        #write all 0x00
        raw0 = b0[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = raw0
        time.sleep(0)
        ret0 = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]

        #write all 0xff
        raw1 = b1[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = raw1
        time.sleep(0)
        ret1 = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]

        #write random bytes
        rawramp = bramp[j*maxwrite:(j+1)*maxwrite]
        r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite] = rawramp
        time.sleep(0)
        retramp = r_fast.mixer.host.transport.axil_mm[phase_addrs_tx[i]+j*maxwrite : phase_addrs_tx[i] +(j+1)*maxwrite]
        if rawramp!=retramp:
            break

        print(ret0==raw0, ret1==raw1, retramp==rawramp)

raw_vals=np.frombuffer(rawramp,dtype=np.uint8)
ret_vals=np.frombuffer(retramp,dtype=np.uint8)

print(raw_vals)
print(ret_vals)
print(ret_vals!=raw_vals)

[  0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17
  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35
  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53
  54  55  56  57  58  59  60  61  62  63  64  65  66  67  68  69  70  71
  72  73  74  75  76  77  78  79  80  81  82  83  84  85  86  87  88  89
  90  91  92  93  94  95  96  97  98  99 100 101 102 103 104 105 106 107
 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215
 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
 252 253 254 255   0   1   2   3   4   5   6   7   8   9  10  11  12  13
  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49
  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65  66  67
  68  69  70  71  72  73  74  75  76  77  78  79  80  81  82  83  84  85
  86  87  88  89  90  91  92  93  94  95  96  97  98  99 100 101 102 103
 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193
 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
 248 249 250 251 252 253 254 255]

[  0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17
  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35
  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53
  54  55  56  57  58  59  60  61  62  63  64  65  66  67  68  69  70  71
  72  73  74  75  76  77  78  79  80  81  82  83  84  85  86  87  88  89
  90  91  92  93  94  95  96  97  98  99 100 101 102 103 104 105 106 107
 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
 180 181 182 183 184 185 186 187 188 189 190 191  48  49  50  51  52  53
  54  55  56  57  58  59  60  61  62  63 208 209 210 211 212 213 214 215
 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
 252 253 254 255   0   1   2   3   4   5   6   7   8   9  10  11  12  13
  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49
  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65  66  67
  68  69  70  71  72  73  74  75  76  77  78  79  80  81  82  83  84  85
  86  87  88  89  90  91  92  93  94  95  96  97  98  99 100 101 102 103
 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193
 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
 248 249 250 251 252 253 254 255]

[False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
  True  True  True  True  True  True  True  True  True  True  True  True
  True  True  True  True False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False False False False False
 False False False False False False False False]

Any thoughts?

sr-cdf commented 6 days ago

Seems to be a python problem as the following c-code (which pretty much does the same thing) does not crash the cpu and always reads back the data that was written:

test_mmap.c 

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <string.h>

#define MAP_SIZE 0x100000000 // 4GB
#define ADDR 0xa00aa000
#define NBYTES 8192
#define BLOCK_SIZE 4096

int main() {
  int fd;
  unsigned char *axil_mm;
  unsigned char *data;
  int size, written, n_bytes, i;
  int mismatch = 0; // Flag to track  mismatches
  // Open /dev/mem
  fd = open("/dev/mem", O_RDWR | O_SYNC);
  if (fd == -1) {
    perror("open");
    exit(1);
  }

  // Create the memory mapping
  axil_mm = mmap(NULL, MAP_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
  if (axil_mm == MAP_FAILED) {
    perror("mmap");
    exit(1);
  }

  // Allocate memory for data
  size = NBYTES; 
  data = malloc(size);
  if (data == NULL) {
    perror("malloc");
    exit(1);
  }

  // Generate the repeating pattern of 0-255
  for (i = 0; i < size; i++) {
    data[i] = i % 256;
  }

  // Write data to memory
  size = NBYTES;
  written = 0;
  for (i = 0; i < size / BLOCK_SIZE + 1; i++) {
    n_bytes = size - written < BLOCK_SIZE ? size - written : BLOCK_SIZE;
    if (n_bytes > 0) {
      memcpy(axil_mm + ADDR + written, data + written, n_bytes);
    }
    written += n_bytes;
  }

  // Read data from memory and compare
  printf("Data read from memory:\n");
  for (i = 0; i < size; i++) {
    printf("%03d ", axil_mm[ADDR + i]);
    if ((i + 1) % 16 == 0) {
      printf("\n");
    }
    if (axil_mm[ADDR + i] != data[i]) { // Check for mismatch
      mismatch = 1;
    }
  }
  printf("\n");

  // Report mismatches
  if (mismatch) {
    printf("Error: Mismatch detected between written and read data!\n");
  } else {
    printf("Success: All written bytes match read bytes.\n");
  }

  printf("\n");

  // Clean up
  free(data);
  munmap(axil_mm, MAP_SIZE);
  close(fd);

  return 0;
}

jack-h commented 18 hours ago

Does the C code also solve https://github.com/realtimeradio/souk-firmware/issues/64 ?

It also occurs to me that reading back the value and seeing it not update doesn't necessarily imply that the value didn't make it to the FPGA. Since you use the same options to open the file and map its contents I don't have any immediate suggestions. Do you think it is worth abandoning the LocalMemTransport interface at this point (I assume the C will also be much faster for things like tone tracking)? Some time ago I added https://github.com/realtimeradio/souk-firmware/blob/main/software/control_sw/test_scripts/fpg2header.py to scrape the fpg file for memory addresses so you could access them by name.

jack-h commented 18 hours ago

Possible related (or at least a potentially useful comparison): https://github.com/realtimeradio/casperfpga/blob/c65b856531f7572ac8e43494499272e2948e498a/src/transport_localpcie.py#L55

realtimeradio / souk-firmware

local transport mmap write does not complete #65