amd64_edac: Carve out ECC-related hw settings

This is in preparation for the init path reorganization where we want
only to

1) test whether a particular node supports ECC
2) can it be enabled

and only then do the necessary allocation/initialization. For that,
we need to decouple the ECC settings of the node from the instance's
descriptor.

The should be no functional change introduced by this patch.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
This commit is contained in:
Borislav Petkov 2010-10-14 16:01:30 +02:00
parent f1db274e1b
commit ae7bb7c679
2 changed files with 49 additions and 24 deletions

View file

@ -18,6 +18,7 @@ static struct msr __percpu *msrs;
/* Per-node driver instances */ /* Per-node driver instances */
static struct mem_ctl_info **mcis; static struct mem_ctl_info **mcis;
static struct amd64_pvt **pvts; static struct amd64_pvt **pvts;
static struct ecc_settings **ecc_stngs;
/* /*
* Address to DRAM bank mapping: see F2x80 for K8 and F2x[1,0]80 for Fam10 and * Address to DRAM bank mapping: see F2x80 for K8 and F2x[1,0]80 for Fam10 and
@ -2293,7 +2294,7 @@ static bool amd64_nb_mce_bank_enabled_on_node(int nid)
return ret; return ret;
} }
static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on) static int amd64_toggle_ecc_err_reporting(struct ecc_settings *s, u8 nid, bool on)
{ {
cpumask_var_t cmask; cpumask_var_t cmask;
int cpu; int cpu;
@ -2303,7 +2304,7 @@ static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on)
return false; return false;
} }
get_cpus_on_this_dct_cpumask(cmask, pvt->mc_node_id); get_cpus_on_this_dct_cpumask(cmask, nid);
rdmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs); rdmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs);
@ -2313,14 +2314,14 @@ static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on)
if (on) { if (on) {
if (reg->l & K8_MSR_MCGCTL_NBE) if (reg->l & K8_MSR_MCGCTL_NBE)
pvt->flags.nb_mce_enable = 1; s->flags.nb_mce_enable = 1;
reg->l |= K8_MSR_MCGCTL_NBE; reg->l |= K8_MSR_MCGCTL_NBE;
} else { } else {
/* /*
* Turn off NB MCE reporting only when it was off before * Turn off NB MCE reporting only when it was off before
*/ */
if (!pvt->flags.nb_mce_enable) if (!s->flags.nb_mce_enable)
reg->l &= ~K8_MSR_MCGCTL_NBE; reg->l &= ~K8_MSR_MCGCTL_NBE;
} }
} }
@ -2334,18 +2335,20 @@ static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on)
static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
{ {
struct amd64_pvt *pvt = mci->pvt_info; struct amd64_pvt *pvt = mci->pvt_info;
u8 nid = pvt->mc_node_id;
struct ecc_settings *s = ecc_stngs[nid];
u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
amd64_read_pci_cfg(pvt->F3, K8_NBCTL, &value); amd64_read_pci_cfg(pvt->F3, K8_NBCTL, &value);
/* turn on UECCn and CECCEn bits */ /* turn on UECCEn and CECCEn bits */
pvt->old_nbctl = value & mask; s->old_nbctl = value & mask;
pvt->nbctl_mcgctl_saved = 1; s->nbctl_valid = true;
value |= mask; value |= mask;
pci_write_config_dword(pvt->F3, K8_NBCTL, value); pci_write_config_dword(pvt->F3, K8_NBCTL, value);
if (amd64_toggle_ecc_err_reporting(pvt, ON)) if (amd64_toggle_ecc_err_reporting(s, nid, ON))
amd64_warn("Error enabling ECC reporting over MCGCTL!\n"); amd64_warn("Error enabling ECC reporting over MCGCTL!\n");
amd64_read_pci_cfg(pvt->F3, K8_NBCFG, &value); amd64_read_pci_cfg(pvt->F3, K8_NBCFG, &value);
@ -2357,7 +2360,7 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
if (!(value & K8_NBCFG_ECC_ENABLE)) { if (!(value & K8_NBCFG_ECC_ENABLE)) {
amd64_warn("DRAM ECC disabled on this node, enabling...\n"); amd64_warn("DRAM ECC disabled on this node, enabling...\n");
pvt->flags.nb_ecc_prev = 0; s->flags.nb_ecc_prev = 0;
/* Attempt to turn on DRAM ECC Enable */ /* Attempt to turn on DRAM ECC Enable */
value |= K8_NBCFG_ECC_ENABLE; value |= K8_NBCFG_ECC_ENABLE;
@ -2372,7 +2375,7 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
amd64_info("Hardware accepted DRAM ECC Enable\n"); amd64_info("Hardware accepted DRAM ECC Enable\n");
} }
} else { } else {
pvt->flags.nb_ecc_prev = 1; s->flags.nb_ecc_prev = 1;
} }
debugf0("NBCFG(2)= 0x%x CHIPKILL= %s ECC_ENABLE= %s\n", value, debugf0("NBCFG(2)= 0x%x CHIPKILL= %s ECC_ENABLE= %s\n", value,
@ -2384,26 +2387,28 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt) static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt)
{ {
u8 nid = pvt->mc_node_id;
struct ecc_settings *s = ecc_stngs[nid];
u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
if (!pvt->nbctl_mcgctl_saved) if (!s->nbctl_valid)
return; return;
amd64_read_pci_cfg(pvt->F3, K8_NBCTL, &value); amd64_read_pci_cfg(pvt->F3, K8_NBCTL, &value);
value &= ~mask; value &= ~mask;
value |= pvt->old_nbctl; value |= s->old_nbctl;
pci_write_config_dword(pvt->F3, K8_NBCTL, value); pci_write_config_dword(pvt->F3, K8_NBCTL, value);
/* restore previous BIOS DRAM ECC "off" setting which we force-enabled */ /* restore previous BIOS DRAM ECC "off" setting we force-enabled */
if (!pvt->flags.nb_ecc_prev) { if (!s->flags.nb_ecc_prev) {
amd64_read_pci_cfg(pvt->F3, K8_NBCFG, &value); amd64_read_pci_cfg(pvt->F3, K8_NBCFG, &value);
value &= ~K8_NBCFG_ECC_ENABLE; value &= ~K8_NBCFG_ECC_ENABLE;
pci_write_config_dword(pvt->F3, K8_NBCFG, value); pci_write_config_dword(pvt->F3, K8_NBCFG, value);
} }
/* restore the NB Enable MCGCTL bit */ /* restore the NB Enable MCGCTL bit */
if (amd64_toggle_ecc_err_reporting(pvt, OFF)) if (amd64_toggle_ecc_err_reporting(s, nid, OFF))
amd64_warn("Error restoring NB MCGCTL settings!\n"); amd64_warn("Error restoring NB MCGCTL settings!\n");
} }
@ -2654,6 +2659,8 @@ static int __devinit amd64_init_one_instance(struct pci_dev *pdev,
const struct pci_device_id *mc_type) const struct pci_device_id *mc_type)
{ {
int ret = 0; int ret = 0;
u8 nid = get_node_id(pdev);
struct ecc_settings *s;
ret = pci_enable_device(pdev); ret = pci_enable_device(pdev);
if (ret < 0) { if (ret < 0) {
@ -2661,9 +2668,16 @@ static int __devinit amd64_init_one_instance(struct pci_dev *pdev,
return -EIO; return -EIO;
} }
ret = -ENOMEM;
s = kzalloc(sizeof(struct ecc_settings), GFP_KERNEL);
if (!s)
return ret;
ecc_stngs[nid] = s;
ret = amd64_probe_one_instance(pdev); ret = amd64_probe_one_instance(pdev);
if (ret < 0) if (ret < 0)
amd64_err("Error probing instance: %d\n", get_node_id(pdev)); amd64_err("Error probing instance: %d\n", nid);
return ret; return ret;
} }
@ -2688,6 +2702,9 @@ static void __devexit amd64_remove_one_instance(struct pci_dev *pdev)
amd_report_gart_errors(false); amd_report_gart_errors(false);
amd_unregister_ecc_decoder(amd64_decode_bus_error); amd_unregister_ecc_decoder(amd64_decode_bus_error);
kfree(ecc_stngs[pvt->mc_node_id]);
ecc_stngs[pvt->mc_node_id] = NULL;
/* Free the EDAC CORE resources */ /* Free the EDAC CORE resources */
mci->pvt_info = NULL; mci->pvt_info = NULL;
mcis[pvt->mc_node_id] = NULL; mcis[pvt->mc_node_id] = NULL;
@ -2767,9 +2784,10 @@ static int __init amd64_edac_init(void)
goto err_ret; goto err_ret;
err = -ENOMEM; err = -ENOMEM;
pvts = kzalloc(amd_nb_num() * sizeof(pvts[0]), GFP_KERNEL); pvts = kzalloc(amd_nb_num() * sizeof(pvts[0]), GFP_KERNEL);
mcis = kzalloc(amd_nb_num() * sizeof(mcis[0]), GFP_KERNEL); mcis = kzalloc(amd_nb_num() * sizeof(mcis[0]), GFP_KERNEL);
if (!(pvts && mcis)) ecc_stngs = kzalloc(amd_nb_num() * sizeof(ecc_stngs[0]), GFP_KERNEL);
if (!(pvts && mcis && ecc_stngs))
goto err_ret; goto err_ret;
msrs = msrs_alloc(); msrs = msrs_alloc();
@ -2820,6 +2838,9 @@ static void __exit amd64_edac_exit(void)
pci_unregister_driver(&amd64_pci_driver); pci_unregister_driver(&amd64_pci_driver);
kfree(ecc_stngs);
ecc_stngs = NULL;
kfree(mcis); kfree(mcis);
mcis = NULL; mcis = NULL;

View file

@ -461,17 +461,21 @@ struct amd64_pvt {
/* place to store error injection parameters prior to issue */ /* place to store error injection parameters prior to issue */
struct error_injection injection; struct error_injection injection;
/* Save old hw registers' values before we modified them */
u32 nbctl_mcgctl_saved; /* When true, following 2 are valid */
u32 old_nbctl;
/* DCT per-family scrubrate setting */ /* DCT per-family scrubrate setting */
u32 min_scrubrate; u32 min_scrubrate;
/* family name this instance is running on */ /* family name this instance is running on */
const char *ctl_name; const char *ctl_name;
/* misc settings */ };
/*
* per-node ECC settings descriptor
*/
struct ecc_settings {
u32 old_nbctl;
bool nbctl_valid;
struct flags { struct flags {
unsigned long nb_mce_enable:1; unsigned long nb_mce_enable:1;
unsigned long nb_ecc_prev:1; unsigned long nb_ecc_prev:1;