sometimes we should use "disable fork" instead of "disable block_name"

时间:2023-03-08 22:21:50
sometimes we should use "disable fork" instead of "disable block_name"

A disable named block statement stops the execution of all blocks with that same name in all threads, which probably causes unexpected behavior.

Specific to this test, the “disable PM_LINK_WAIT_USB2_L1” statement will stops all active threads of PM_LINK_WAIT_USB2_L1 blocks, even if these PM_LINK_WAIT_USB2_L1 blocks are in different sequences.

Below I list the process of this issue.

1. In sequence “fch_usb_multi_port_trans_seqs”, sequence “fch_usb_lpm_trans_seqs” was called twice for each Port.

//Start transfers for devices

for(int dn=0;dn<number_of_device;dn++) begin

automatic int j=dn;

fork

begin

for(int k=0;k<2;k++)   begin

usb_lpm_trans_seqs[j].otg_xhc_device_transfer= otg_xhc_device_transfer[j];

usb_lpm_trans_seqs[j].start(p_sequencer, this);

end

end

join_none

end

2. In sequence “fch_usb_lpm_trans_seqs”, we need to read register PORTSC_20 to check whether Port is in L1 LPM state.

//Enter low power

reg_port_num = otg_xhc_device_transfer.device_cfg.port_number;

if( pm_controller.pm_link_state == fch_otg_pm_controller::U1_L1 ||

pm_controller.pm_link_state == fch_otg_pm_controller::U2_L1 ) begin

fork : PM_LINK_WAIT_USB2_L1

begin

do begin

#5us;

expected_pm_link_state = 4'd2;

uvm_ext_fd_reg_access(`uvm_ext_fileline,RD,"dwc_usb3_reg","PORTSC_20",reg_data, reg_port_num);

end while(reg_data[8:5] != 4'd2);

end

begin

if(otg_xhc_device_transfer.device_cfg.connected_bus_speed == svt_usb_types::LS) begin

#3500us;

end else if(otg_xhc_device_transfer.device_cfg.connected_bus_speed == svt_usb_types::FS) begin

#2500us;

end else begin

#500us;

end

`uvm_fatal(get_type_name(), $psprintf("PM: Link, waiting for USB2.0 L1 failed for port %0d", otg_xhc_device_transfer.device_cfg.port_number));

end

join_any

            disable PM_LINK_WAIT_USB2_L1;

`uvm_info(get_type_name(), $psprintf("PM: Link, Port %0d is L1", otg_xhc_device_transfer.device_cfg.port_number), UVM_LOW);

end

3. below waveform snapshot,

At #T1, usb_lpm_trans_seqs_0 read register PORTSC_20, and the RDATA is 32’h603

At #T2, usb_lpm_trans_seqs_1 read register PORTSC_20, and the RDATA is 32’h643.

That is, Port0 is in L1 state, and Port1 is not.

sometimes we should use "disable fork" instead of "disable block_name"

4. However, The sequence mistakenly checked the Port0 and Port1 were both in L1 state, which was reported in below log.

sometimes we should use "disable fork" instead of "disable block_name"