I'm porting linux 2.6 kernel to a ML 410 board using Xilinx's kernel
tree. Particularly, I would like to use Xilinx's icap driver (as in
driver/char/xilinx_hwicap.
I do not have problems in using it to write a partial bitstream to
ICAP; but I have a problem when performing read back.
I'm trying to read from a device from a user application using:
int* frame = (int*) malloc( 83 * sizeof(int));
FILE *ICAP;
ICAP = fopen("/dev/icap0", "r+b"
;
fwrite(word, sizeof(*word), 18, ICAP);
fread(frame, sizeof(int), 83, ICAP);
The device is opened OK. And the 18 words written to icap is the same
as those in hwicap v1_01_a driver's xhwicap_device_read_frame.c:
1. Dummy 2. Sync 3. Noop 4. Noop
5. CMD 6. RCRC 7. Noop 8. Noop
9. CMD 10. RCFG 11. Noop 12. Noop
13. Noop 14. W FAR 15. frame addr 16. FRDO 83 words
17. Noop 18. Noop
The problem is, in icap read function,
static int buffer_icap_device_read(struct hwicap_drvdata *drvdata,u32
offset, u32 count)
"count" is not the same value as I requested with fread.
Instead of 0x53 (words), it gets 0x1000 (bytes). This value seems to
causes a device read failure (it is not associated with frame
boundary, and it is not the same value as previous command written to
icap requests)- the icap status register after read is 0, not the
normal value 0x7ffc; and data copied to user buffer is not the desired
frame data.
Actually, I am able to work around with this problem by forcing count
to 83 in hwicap_read. This is OK since I mostly only needs to read one
frame. By doing so, icap status register is the correct value of
0x7ffc after read. And in user space, *frame get the desired frame
data.
How can I fix the driver function so that hwicap_read gets the same
"count" as fread passes to it?
