netusb: wait for the transfer to complete before retrying. #6369
Conversation
In the same way as cdc_acm, set a semaphore in the transfer complete interrupt and wait for it in the main loop before retrying. This fixes a bug with >= 64 byte frames when logging is off. If logging is on, then the time to print logging is enough for the transmit to complete. If logging is off, then the 10 retries complete quickly and try_write() gives up. Signed-off-by: Michael Hope <[email protected]>
Codecov Report
@@ Coverage Diff @@
## master #6369 +/- ##
=======================================
Coverage 53.08% 53.08%
=======================================
Files 430 430
Lines 41040 41040
Branches 7908 7908
=======================================
Hits 21785 21785
Misses 16032 16032
Partials 3223 3223Continue to review full report at Codecov.
|
|
Friendly ping... |
| @@ -188,6 +196,8 @@ int try_write(u8_t ep, u8_t *data, u16_t len) | |||
| */ | |||
| if (tries--) { | |||
| SYS_LOG_WRN("Error: EAGAIN. Another try"); | |||
| /* Wait for the transfer to complete */ | |||
| k_sem_take(&netusb.bulk_in_sem, K_MSEC(100)); | |||
| continue; | |||
There was a problem hiding this comment.
Just to check, could the issue be solved by k_yield() (or even k_sleep() despite it does not look nice)? BTW try_write() would be replaced by usb_transfer(), there is a patch for ECM in PR #5983 .
There was a problem hiding this comment.
Ack, thanks.
re: k_yield() it might reduce the problem but it's hard to say. USB is fast so you don't need to spend much time for the transfer to have completed - for example, if logging is on then the SYS_LOG_WRN() is enough.
re: transfers, I'll look into it.
|
This was obsoleted by ccf0b46 |
In the same way as cdc_acm, set a semaphore in the transfer complete
interrupt and wait for it in the main loop before retrying.
This fixes a bug with >= 64 byte frames when logging is off. If
logging is on, then the time to print logging is enough for the
transmit to complete. If logging is off, then the 10 retries complete
quickly and try_write() gives up.
Signed-off-by: Michael Hope [email protected]