uvm_reg_bit_bash_seq always write value 1

[akshay.raoy]

uvm_reg_bit_bash_seq always writes value 1, both the times and always keep the exp value to default value of the register.

in this sequence bash_kth_bit task i tried just adding rg.set(val) after write is done. it started working fine.

i am using uvm-1.0p1 version of UVM package.

Does anyone face this issue in uvm-1.1 also? or is it fixed in new version?

[janick]

What is the type of bit it is trying to bash?

Where there prior errors when bashing this bit or previous bits?

Is the DUT "idle"?

[akshay.raoy]

Its a read/write bit, this is the first bit of bit bash.

yes DUT is idle, and i am able to see writing value 1 to each bit twice in waveform.

[janick]

Looks like the mirrored value is not updated in time for the next cycle. I bet you are using explicit monitoring and the write() or read() returns before the monitoring path has completed the mirror update.

[akshay.raoy]

i am just using uvm_reg_bit_bash_seq which is with UVM package. monitoring is not done in my defined sequence.

this is how i am using it.

uvm_reg_bit_bash_seq pie8_ral_seq = uvm_reg_bit_bash_seq::type_id::create("pie8_ral_seq",this)

pie8_ral_seq.model = tbEnv.regmodel;

pie8_ral_seq.start(null);

[janick]

The pre-defined sequence depends on the mirror being properly updated in a timely manner between consecutive register operations. I suspect that it is that mirror update that is too slow.

Make a copy of the pre-defined sequence and add a delay before a register read or write and see if that fixes it. if it does, your monitoring path is slower than your stimulus path.

[ryz]

My guessing is that you have not called set_auto_predict(1) on the register map and then you need to have an explicit monitor to update the field value.