| Message ID | 1463731546-15232-1-git-send-email-Shengzhou.Liu@nxp.com |
|---|---|
| State | Superseded |
| Headers | show |
On 05/20/2016 01:15 AM, Shengzhou Liu wrote: > Default address parity is disabled for DDR4 UDIMM for considing > performance, it needs to enable parity for A-009803 workaround. > > Signed-off-by: Shengzhou Liu <Shengzhou.Liu@nxp.com> > --- > drivers/ddr/fsl/options.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/ddr/fsl/options.c b/drivers/ddr/fsl/options.c > index d0075ff..dcef0bb 100644 > --- a/drivers/ddr/fsl/options.c > +++ b/drivers/ddr/fsl/options.c > @@ -1016,6 +1016,9 @@ unsigned int populate_memctl_options(const common_timing_params_t *common_dimm, > } > } > > +#ifdef CONFIG_SYS_FSL_ERRATUM_A009803 > + popts->ap_en = 1; > +#endif > /* > * BSTTOPRE precharge interval > * > Shengzhou, It looks backward. You should detect if the condition is right before applying the workaround, not by forcing the condition. York
> -----Original Message----- > From: York Sun [mailto:york.sun@nxp.com] > Sent: Friday, May 20, 2016 11:29 PM > To: Shengzhou Liu <shengzhou.liu@nxp.com>; u-boot@lists.denx.de > Subject: Re: [PATCH] driver/ddr/fsl: Force enabling parity for A-009803 > > > > Shengzhou, > > It looks backward. You should detect if the condition is right before > applying the workaround, not by forcing the condition. > > York If ERRATUM_A009803 is defined, ap_en should always be enabled whatever user configures parity=on in hwconfig. As ERRATUM_A009803 is enabled only in arch-fsl-layerscape/config.h for LS2 SoC, I don't think we need to check anything other. What condition do you mean? -Shengzhou
On 05/23/2016 02:18 AM, Shengzhou Liu wrote: >> -----Original Message----- >> From: York Sun [mailto:york.sun@nxp.com] >> Sent: Friday, May 20, 2016 11:29 PM >> To: Shengzhou Liu <shengzhou.liu@nxp.com>; u-boot@lists.denx.de >> Subject: Re: [PATCH] driver/ddr/fsl: Force enabling parity for A-009803 >>> >> >> Shengzhou, >> >> It looks backward. You should detect if the condition is right before >> applying the workaround, not by forcing the condition. >> >> York > > If ERRATUM_A009803 is defined, ap_en should always be enabled whatever user configures parity=on in hwconfig. > As ERRATUM_A009803 is enabled only in arch-fsl-layerscape/config.h for LS2 SoC, I don't think we need to check anything other. > What condition do you mean? > Shengzhou, My point is you should force ap=1. Do you mean if ERRATUM_A009803 is enabled, users are forced to use address parity? That doesn't sound right. We have been running UDIMM without address parity for a long time. York
> -----Original Message----- > From: York Sun [mailto:york.sun@nxp.com] > Sent: Monday, May 23, 2016 11:33 PM > To: Shengzhou Liu <shengzhou.liu@nxp.com>; u-boot@lists.denx.de > Subject: Re: [PATCH] driver/ddr/fsl: Force enabling parity for A-009803 > Shengzhou, > > My point is you should force ap=1. Do you mean if ERRATUM_A009803 is > enabled, users are forced to use address parity? That doesn't sound right. > We have been running UDIMM without address parity for a long time. > > York > York, My understanding is that ERRATUM_A009803 may still happen whatever ap_en is enabled or disabled. To apply the workaround of A009803, it requires ap_en is enabled. Is your understanding that if we disable ap_en, ERRATUM_A009803 will never happen? The CE document doesn't explain clearly this. In last mail, did you mean we should force ap_en = 0 in case of A-009803? Shengzhou
On 05/23/2016 10:15 PM, Shengzhou Liu wrote: > >> -----Original Message----- >> From: York Sun [mailto:york.sun@nxp.com] >> Sent: Monday, May 23, 2016 11:33 PM >> To: Shengzhou Liu <shengzhou.liu@nxp.com>; u-boot@lists.denx.de >> Subject: Re: [PATCH] driver/ddr/fsl: Force enabling parity for A-009803 >> Shengzhou, >> >> My point is you should force ap=1. Do you mean if ERRATUM_A009803 is >> enabled, users are forced to use address parity? That doesn't sound right. >> We have been running UDIMM without address parity for a long time. >> >> York >> > York, > My understanding is that ERRATUM_A009803 may still happen whatever ap_en is enabled or disabled. > To apply the workaround of A009803, it requires ap_en is enabled. Is your understanding that if we > disable ap_en, ERRATUM_A009803 will never happen? The CE document doesn't explain clearly this. > In last mail, did you mean we should force ap_en = 0 in case of A-009803? > Sorry I had a typo. I meant you should NOT force ap=1. Let me explain. The erratum tells you _if_ address parity is used, for either UDIMM or RDIMM, you need to implement the workaround, as step 1, 2, 3, ... We understand users don't have a choice for RDIMM, the address parity is always enabled. But for UDIMM, users can choose not to enable it. Your _this_ patch forces the address parity to be true, regardless of user's choice. I think this is wrong. The erratum always applies to affected SoCs, but the address parity is not always enabled. That's what I meant for "condition". York
diff --git a/drivers/ddr/fsl/options.c b/drivers/ddr/fsl/options.c index d0075ff..dcef0bb 100644 --- a/drivers/ddr/fsl/options.c +++ b/drivers/ddr/fsl/options.c @@ -1016,6 +1016,9 @@ unsigned int populate_memctl_options(const common_timing_params_t *common_dimm, } } +#ifdef CONFIG_SYS_FSL_ERRATUM_A009803 + popts->ap_en = 1; +#endif /* * BSTTOPRE precharge interval *
Default address parity is disabled for DDR4 UDIMM for considing performance, it needs to enable parity for A-009803 workaround. Signed-off-by: Shengzhou Liu <Shengzhou.Liu@nxp.com> --- drivers/ddr/fsl/options.c | 3 +++ 1 file changed, 3 insertions(+)