| Message ID | 1452137792-24062-2-git-send-email-ruscur@russell.cc |
|---|---|
| State | Superseded |
| Headers | show |
Hey, On Thu, 7 Jan 2016 14:36:29 Russell Currey wrote: <snip> > + > + /* Reset error registers. TODO: are there more we should clear here? */ > + npu_ioda_sel(npu_dev->npu, NPU_IODA_TBL_PESTB, 0, true); > + for (val = 0; val < npu_dev->npu->total_devices; val++) The PESTB is indexed by PE number right? So there could be more devices than PESTB entries in the system if some devices are sharing PE#s (which the latest kernel patches do) so we should be looping up to NPU_NUM_OF_PES rather than npu->total_devices. - Alistair > + out_be64(npu_dev->npu->at_regs + NPU_IODA_DATA0, 0); > + > return PROCEDURE_COMPLETE; > } > DEFINE_PROCEDURE(reset_npu_dl); >
diff --git a/hw/npu-hw-procedures.c b/hw/npu-hw-procedures.c index 24f3b2c..c2a2be0 100644 --- a/hw/npu-hw-procedures.c +++ b/hw/npu-hw-procedures.c @@ -154,6 +154,12 @@ static uint32_t reset_npu_dl(struct npu_dev *npu_dev) xscom_write(npu_dev->npu->chip_id, npu_dev->xscom + NX_TL_CMD_D_CR, PPC_BIT(0)); xscom_write(npu_dev->npu->chip_id, npu_dev->xscom + NX_TL_RSP_CR, PPC_BIT(15)); xscom_write(npu_dev->npu->chip_id, npu_dev->xscom + NX_TL_RSP_D_CR, PPC_BIT(15)); + + /* Reset error registers. TODO: are there more we should clear here? */ + npu_ioda_sel(npu_dev->npu, NPU_IODA_TBL_PESTB, 0, true); + for (val = 0; val < npu_dev->npu->total_devices; val++) + out_be64(npu_dev->npu->at_regs + NPU_IODA_DATA0, 0); + return PROCEDURE_COMPLETE; } DEFINE_PROCEDURE(reset_npu_dl);
Once the NPU has been reset, it should no longer have any fixable errors. Clear any relevant error registers at the end of the reset procedure. There may be other error registers that should be cleared, if found they will be added later. Signed-off-by: Russell Currey <ruscur@russell.cc> --- hw/npu-hw-procedures.c | 6 ++++++ 1 file changed, 6 insertions(+)