cvw
cvw copied to clipboard
openhwgroup
Remove flush from EBU critical path
The EBU receives flush signals (necessary to solve a logic bug). These are now on the critical path, but not in the version of the critical path reported in Chapter 6.
For example, buscachefsm has Flush affecting HTRANS. See the code and timing report below. In the timing report, a branch misprediction raises FlushD, which affects HTRANS. busfsm is similar.
Rose Thompson feels the right fix is to issue the transaction anyway in the Execute stage, but to send a signal in the memory stage for the controller to cancel the transaction via the AHB protocol. If this isn't feasible, it might be possible to kill HWSTRB in the memory stage so that transaction has no effect.
After the fix, rerun Linux boot in lock step, and regenerate Chapter 6 synthesis results.
assign HTRANS = (CurrState == ADR_PHASE & HREADY & ((|BusRW) | (|CacheBusRW) | BusCMOZero) & ~Flush) |
(CurrState == ATOMIC_READ_DATA_PHASE) |
(CacheAccess & FinalBeatCount & |CacheBusRW & HREADY & ~Flush) ? AHB_NONSEQ : // if we have a pipelined request
(CacheAccess & |BeatCount) ? (BURST_EN ? AHB_SEQ : AHB_NONSEQ) : AHB_IDLE;
PCSrcE (net) 0.005546 0.000000 2.174526 r
ifu/PCSrcE (ifu__598949) 0.000000 2.174526 r
ifu/PCSrcE (net) 0.005546 0.000000 2.174526 r
ifu/U5/Y (sky130_osu_sc_12T_ms__buf_1) 0.076231 0.110984 2.285510 r
ifu/BPWrongE (net) 2 0.007587 0.000000 2.285510 r
ifu/BPWrongE (ifu__598949) 0.000000 2.285510 r
BPWrongE (net) 0.007587 0.000000 2.285510 r
hzu/BPWrongE (hazard__598949) 0.000000 2.285510 r
hzu/BPWrongE (net) 0.007587 0.000000 2.285510 r
hzu/U4/Y (sky130_osu_sc_12T_ms__inv_2) 0.036381 0.048787 2.334297 f
hzu/n8 (net) 3 0.009074 0.000000 2.334297 f
hzu/U3/Y (sky130_osu_sc_12T_ms__inv_2) 0.052211 0.057438 2.391736 r
hzu/FlushD (net) 4 0.010859 0.000000 2.391736 r
hzu/FlushD (hazard__598949) 0.000000 2.391736 r
FlushD (net) 0.010859 0.000000 2.391736 r
ifu/FlushD (ifu__598949) 0.000000 2.391736 r
ifu/FlushD (net) 0.010859 0.000000 2.391736 r
ifu/bus.passthrough.ahbinterface/Flush (ahbinterface_32_0) 0.000000 2.391736 r
ifu/bus.passthrough.ahbinterface/Flush (net) 0.010859 0.000000 2.391736 r
ifu/bus.passthrough.ahbinterface/busfsm/Flush (busfsm_1) 0.000000 2.391736 r
ifu/bus.passthrough.ahbinterface/busfsm/Flush (net) 0.010859 0.000000 2.391736 r
ifu/bus.passthrough.ahbinterface/busfsm/U29/Y (sky130_osu_sc_12T_ms__nor2_1)
0.040983 0.046972 2.438707 f
ifu/bus.passthrough.ahbinterface/busfsm/HTRANS[1] (net)
2 0.004385 0.000000 2.438707 f
ifu/bus.passthrough.ahbinterface/busfsm/HTRANS[1] (busfsm_1) 0.000000 2.438707 f
ifu/bus.passthrough.ahbinterface/HTRANS[1] (net) 0.004385 0.000000 2.438707 f
ifu/bus.passthrough.ahbinterface/HTRANS[1] (ahbinterface_32_0) 0.000000 2.438707 f
ifu/IFUHTRANS[1] (net) 0.004385 0.000000 2.438707 f
ifu/IFUHTRANS[1] (ifu__598949) 0.000000 2.438707 f
IFUHTRANS[1] (net) 0.004385 0.000000 2.438707 f
ebu.ebu/IFUHTRANS[1] (ebu__598949) 0.000000 2.438707 f
ebu.ebu/IFUHTRANS[1] (net) 0.004385 0.000000 2.438707 f
ebu.ebu/IFUInput/HTRANSIn[1] (controllerinput_PA_BITS34) 0.000000 2.438707 f
ebu.ebu/IFUInput/HTRANSIn[1] (net) 0.004385 0.000000 2.438707 f
ebu.ebu/IFUInput/genblk1.RestorMux/d0[35] (mux2_WIDTH43) 0.000000 2.438707 f
ebu.ebu/IFUInput/genblk1.RestorMux/d0[35] (net) 0.004385 0.000000 2.438707 f
ebu.ebu/IFUInput/genblk1.RestorMux/U4/Y (sky130_osu_sc_12T_ms__oai21_l)
0.140213 0.116030 2.554738 r
ebu.ebu/IFUInput/genblk1.RestorMux/n12 (net)
1 0.004894 0.000000 2.554738 r
ebu.ebu/IFUInput/genblk1.RestorMux/U3/Y (sky130_osu_sc_12T_ms__inv_2)
0.055106 0.071135 2.625873 f
ebu.ebu/IFUInput/genblk1.RestorMux/y[35] (net)
5 0.012306 0.000000 2.625873 f
ebu.ebu/IFUInput/genblk1.RestorMux/y[35] (mux2_WIDTH43) 0.000000 2.625873 f
ebu.ebu/IFUInput/N0 (net) 0.012306 0.000000 2.625873 f
ebu.ebu/IFUInput/Request (controllerinput_PA_BITS34) 0.000000 2.625873 f
ebu.ebu/IFUReq (net) 0.012306 0.000000 2.625873 f
ebu.ebu/ebufsmarb/IFUReq (ebufsmarb) 0.000000 2.625873 f
ebu.ebu/ebufsmarb/IFUReq (net) 0.012306 0.000000 2.625873 f
ebu.ebu/ebufsmarb/U16/Y (sky130_osu_sc_12T_ms__nand2_1) 0.057364 0.067131 2.693004 r
ebu.ebu/ebufsmarb/n6 (net) 1 0.004894 0.000000 2.693004 r
ebu.ebu/ebufsmarb/U17/Y (sky130_osu_sc_12T_ms__inv_2) 0.039137 0.050475 2.743479 f
ebu.ebu/ebufsmarb/IFUSave (net) 4 0.011732 0.000000 2.743479 f
ebu.ebu/ebufsmarb/IFUSave (ebufsmarb) 0.000000 2.743479 f
ebu.ebu/IFUSave (net) 0.011732 0.000000 2.743479 f
ebu.ebu/IFUInput/Save (controllerinput_PA_BITS34) 0.000000 2.743479 f
ebu.ebu/IFUInput/Save (net) 0.011732 0.000000 2.743479 f
ebu.ebu/IFUInput/genblk1.SaveReg/en (flopenr_WIDTH43) 0.000000 2.743479 f
ebu.ebu/IFUInput/genblk1.SaveReg/en (net) 0.011732 0.000000 2.743479 f
ebu.ebu/IFUInput/genblk1.SaveReg/U52/Y (sky130_osu_sc_12T_ms__inv_2)
0.029795 0.042582 2.786060 r
ebu.ebu/IFUInput/genblk1.SaveReg/n32 (net)
2 0.005157 0.000000 2.786060 r
ebu.ebu/IFUInput/genblk1.SaveReg/U17/Y (sky130_osu_sc_12T_ms__nand2_1)
0.088478 0.080056 2.866116 f
ebu.ebu/IFUInput/genblk1.SaveReg/n20 (net)
1 0.008764 0.000000 2.866116 f
ebu.ebu/IFUInput/genblk1.SaveReg/U5/Y (sky130_osu_sc_12T_ms__inv_4)
0.087029 0.102538 2.968654 r
ebu.ebu/IFUInput/genblk1.SaveReg/n98 (net)
18 0.035834 0.000000 2.968654 r
ebu.ebu/IFUInput/genblk1.SaveReg/U61/Y (sky130_osu_sc_12T_ms__nand2_l)
0.070212 0.062284 3.030937 f
ebu.ebu/IFUInput/genblk1.SaveReg/n38 (net)
1 0.002488 0.000000 3.030937 f
ebu.ebu/IFUInput/genblk1.SaveReg/U31/Y (sky130_osu_sc_12T_ms__nand2_1)
0.048481 0.059047 3.089985 r
ebu.ebu/IFUInput/genblk1.SaveReg/n46 (net)
1 0.002852 0.000000 3.089985 r
ebu.ebu/IFUInput/genblk1.SaveReg/q_reg[8]/D (sky130_osu_sc_12T_ms__dff_1)
0.048481 0.000000 3.089985 r
data arrival time 3.089985
clock clk (rise edge) 2.525253 2.525253
clock network delay (ideal) 0.000000 2.525253
ebu.ebu/IFUInput/genblk1.SaveReg/q_reg[8]/CK (sky130_osu_sc_12T_ms__dff_1)
0.000000 2.525253 r
library setup time -0.135478 2.389775
data required time 2.389775
----------------------------------------------------------------------------------------------
data required time 2.389775
data arrival time -3.089985
----------------------------------------------------------------------------------------------
slack (VIOLATED) -0.700210
Resolved in a feature branch ahbopt. Improved Arty A7 clock speed from 20 Mhz to 25 MHz. However, we are saving this for after tag 1.0 as this change won't match the book.
