| Submitter | Rudolf Marek |
|---|---|
| Date | 2011-02-02 21:20:28 |
| Message ID | <4D49CA9C.8000402@assembler.cz> |
| Download | mbox | patch |
| Permalink | /patch/2605/ |
| State | Superseded |
| Headers | show |
Comments
src/northbridge/amd/amdk8/coherent_ht_car.c was renamed to someting else since then (I wasn't around then), and the patch can no longer be applied. Can you please check this? Alex On 02/02/2011 11:20 PM, Rudolf Marek wrote: > While we are at it. I'm attaching some very old patch (r2978 ;) > which contains some errata fixes too, if you have some spare time please > try to check if it is correct. > > Signed-off-by: Rudolf Marek <r.marek@assembler.cz> > > > Thanks, > Rudolf >
Am 02.02.2011 22:20, schrieb Rudolf Marek: > While we are at it. I'm attaching some very old patch (r2978 ;) > which contains some errata fixes too, if you have some spare time > please try to check if it is correct. > > Signed-off-by: Rudolf Marek <r.marek@assembler.cz> > > > Thanks, > Rudolf > Hi, are you sure ? -#if K8_REV_F_SUPPORT == 1 - /* Erratum 131... */ + /* Erratum 131 is fixed together with erratum 169 */ msr = rdmsr(NB_CFG_MSR); - msr.lo |= 1<< 20; + msr.lo |= 1<< 32;<======================= wrmsr(NB_CFG_MSR, msr); Josef
On 02/03/2011 10:17 AM, Josef Kellermann wrote: >> e are at it. I'm attaching some very old patch (r2978 ;) >> which contains some errata fixes too, if you have some spare time >> please try to check if it is correct. >> >> Signed-off-by: Rudolf Marek <r.marek@assembler.cz> >> >> >> Thanks, >> Rudolf >> > Hi, > are you sure ? You're right: AMD Publication #33610, page34 "[Regarding erratum #131] Systems implementing the workaround for erratum #169 should not apply this workaround." So it seems AMD recommends either or, but not both. Page 59: "[Regarding erratum #169] The workaround for this erratum supersedes the workaround for erratum #131. When implementing this workaround, the workaround for erratum #131 should not be applied." So we better implement only 169, and explain in the comments why 131 is left out. Alex
Am 03.02.2011 10:14, schrieb Alex G.: > On 02/03/2011 10:17 AM, Josef Kellermann wrote: >>> e are at it. I'm attaching some very old patch (r2978 ;) >>> which contains some errata fixes too, if you have some spare time >>> please try to check if it is correct. >>> >>> Signed-off-by: Rudolf Marek<r.marek@assembler.cz> >>> >>> >>> Thanks, >>> Rudolf >>> >> Hi, >> are you sure ? > You're right: > > AMD Publication #33610, page34 > > "[Regarding erratum #131] > Systems implementing the workaround for erratum #169 should not apply > this workaround." > > So it seems AMD recommends either or, but not both. > > Page 59: > "[Regarding erratum #169] > The workaround for this erratum supersedes the workaround for erratum > #131. When implementing > this workaround, the workaround for erratum #131 should not be applied." > > So we better implement only 169, and explain in the comments why 131 is > left out. > > Alex > > Hi, sorry for the misunderstanding. Setting bit 32 in msr should be -> 'msr.hi |= (1<< 0)', no? Josef
> Hi, > sorry for the misunderstanding. > Setting bit 32 in msr should be -> 'msr.hi |= (1<< 0)', no? LOL! Nice catch. This is what happens when you're still up at 6AM, obsessive drawing lines on a Google Earth printscreen in order to finish a project due in a few hours. :P It's funny to see those logic mistakes made with so much ease. As for #169, we should set 'msr.hi |= (1<< 0)' :P, "set F0x68[22:21](DsNpReqLmt0) to 01b," and drop #131 altogether. I'm assuming the second part means 0:18.0, register 0x6b-0x68[22:21], or Rx6a[6-5]. reg = pci_read_config8(0:18.0, 0x6a); reg |= 0x20; pci_write_config8(0:18.0, 0x6a, reg); Alex
Patch
Index: src/cpu/amd/model_fxx/model_fxx_init.c =================================================================== --- src/cpu/amd/model_fxx/model_fxx_init.c (revision 2978) +++ src/cpu/amd/model_fxx/model_fxx_init.c (working copy) @@ -368,6 +368,7 @@ static inline void k8_errata(void) { msr_t msr; + u32 cpuid; #if K8_REV_F_SUPPORT == 0 if (is_cpu_pre_c0()) { /* Erratum 63... */ @@ -439,7 +440,7 @@ #endif #if K8_REV_F_SUPPORT == 0 - if (!is_cpu_pre_e0()) + if (!is_cpu_pre_d0()) #endif { /* Erratum 110 ... */ @@ -453,13 +454,19 @@ msr.lo |= 1 << 6; wrmsr(HWCR_MSR, msr); -#if K8_REV_F_SUPPORT == 1 - /* Erratum 131... */ + /* Erratum 131 is fixed together with erratum 169 */ msr = rdmsr(NB_CFG_MSR); - msr.lo |= 1 << 20; + msr.lo |= 1 << 32; wrmsr(NB_CFG_MSR, msr); -#endif + cpuid = cpuid_eax(0x1); + + /* Fixme erratum 113 JH-E1 DH-E3 */ + if ((cpuid == 0x20f10) || (cpuid == 0x20FF0) || (cpuid == 0x20FC0)) { + msr = rdmsr_amd(BU_CFG_MSR); + msr.hi |= (1 << (48 - 32)); + wrmsr_amd(BU_CFG_MSR, msr); + } } extern void model_fxx_update_microcode(unsigned cpu_deviceid); =================================================================== --- src/northbridge/amd/amdk8/coherent_ht.c (revision 2978) +++ src/northbridge/amd/amdk8/coherent_ht.c (working copy) @@ -1704,6 +1704,12 @@ needs_reset = 1; /* Needed? */ } } + /* Errata 169 */ + dev = NODE_HT(node); + cmd = pci_read_config32(dev, 0x68); + cmd &= ~(3 << 21); + cmd |= (1 << 21); + pci_write_config32(dev, 0x68, cmd); #endif } return needs_reset; Index: src/northbridge/amd/amdk8/amdk8.h =================================================================== --- src/northbridge/amd/amdk8/amdk8.h (revision 2978) +++ src/northbridge/amd/amdk8/amdk8.h (working copy) @@ -135,6 +135,8 @@ #define DCL_QFC_EN (1<<2) #define DCL_DisDqsHys (1<<3) #define DCL_Burst2Opt (1<<5) +/* Check errata 115 for JH-E1 */ +/* Check errata 116 */ #define DCL_DramInit (1<<8) #define DCL_DualDIMMen (1<<9) #define DCL_DramEnable (1<<10) Index: src/northbridge/amd/amdk8/coherent_ht_car.c =================================================================== --- src/northbridge/amd/amdk8/coherent_ht_car.c (revision 2978) +++ src/northbridge/amd/amdk8/coherent_ht_car.c (working copy) @@ -1703,6 +1703,12 @@ needs_reset = 1; /* Needed? */ } } + /* Errata 169 */ + dev = NODE_HT(node); + cmd = pci_read_config32(dev, 0x68); + cmd &= ~(3 << 21); + cmd |= (1 << 21); + pci_write_config32(dev, 0x68, cmd); #endif } return needs_reset;
While we are at it. I'm attaching some very old patch (r2978 ;) which contains some errata fixes too, if you have some spare time please try to check if it is correct. Signed-off-by: Rudolf Marek <r.marek@assembler.cz> Thanks, Rudolf