Messages in this thread | | | Date | Fri, 29 May 2015 18:00:40 +0200 | From | Borislav Petkov <> | Subject | Re: [PATCH 6/6] edac, mce_amd_inj: Inject errors on NBC for bank 4 errors |
| |
On Wed, May 27, 2015 at 02:03:38PM -0500, Aravind Gopalakrishnan wrote: > For bank 4 errors, MCE logging and reporting is done only on > node base cores. Refer D18F3x44[NbMcaToMstCpuEn] field in > Fam10h and later BKDGs. > > This patch ensures that we inject the error on the node base core > for bank 4 errors. Otherwise, triggering #MC or apic interrupts on > a non node base core would not have any effect on the system. > (i.e), we would not see any relevant output on kernel logs for > the error we just injected. > > Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@amd.com> > --- > drivers/edac/mce_amd_inj.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 76 insertions(+) > > diff --git a/drivers/edac/mce_amd_inj.c b/drivers/edac/mce_amd_inj.c > index ca5b29f..bd44ba4 100644 > --- a/drivers/edac/mce_amd_inj.c > +++ b/drivers/edac/mce_amd_inj.c > @@ -17,9 +17,12 @@ > #include <linux/cpu.h> > #include <linux/string.h> > #include <linux/uaccess.h> > +#include <linux/pci.h> > #include <asm/mce.h> > +#include <asm/amd_nb.h> > > #include "mce_amd.h" > +#include "amd64_edac.h" > > /* > * Collect all the MCi_XXX settings > @@ -205,6 +208,65 @@ static void trigger_thr_int(void *info) > asm volatile("int $249"); > } > > +static u32 amd_get_num_nodes(void) > +{ > + u32 nodes = 1; > + > + if (cpu_has_topoext) { > + u32 ecx; > + > + ecx = cpuid_ecx(0x8000001e); > + nodes = ((ecx >> 8) & 7) + 1; > + } else if (static_cpu_has(X86_FEATURE_NODEID_MSR)) { > + u64 value; > + > + rdmsrl(MSR_FAM10H_NODE_ID, value); > + nodes = ((value >> 3) & 7) + 1; > + }
So we already do that (and more) in amd_get_topology(). I'm thinking you should take this function out of the CONFIG_X86_HT ifdeffery (also in its caller amd_detect_cmp()) and you should save "nodes" in a local static variable called nodes_per_processor and a small accessor called amd_get_nodes_cnt() should return it. Similar to amd_get_nb_id().
Don't forget to add a comment explaning what that nodes_per_processor means.
And then amd_mce_inj.c will simply use it instead of duplicating that information here.
Please do that in 2 pre-patches.
> + > + return nodes; > +} > + > +static u32 amd_get_nbc_for_node(int node_id) > +{ > + struct cpuinfo_x86 *c = &boot_cpu_data; > + u32 nodes, cores_per_node; > + > + nodes = amd_get_num_nodes(); > + cores_per_node = c->x86_max_cores / nodes; > + > + return cores_per_node * node_id; > +} > + > +static void toggle_nb_mca_mst_cpu(u16 nid) > +{ > + struct pci_dev *F3 = node_to_amd_nb(nid)->misc; > + u32 val; > + int err; > + > + if (!F3) > + return; > + > + err = pci_read_config_dword(F3, NBCFG, &val); > + if (err) { > + pr_err("%s: Error reading F%dx%03x.\n", __func__, > + PCI_FUNC(F3->devfn), > + NBCFG); > + return; > + } > + > + if (!(val & BIT(27))) { > + pr_err("%s: BIOS not setting D18F3x44[NbMcaToMstCpuEn]." > + "Doing that here\n", __func__);
WARNING: quoted string split across lines #99: FILE: drivers/edac/mce_amd_inj.c:260: + pr_err("%s: BIOS not setting D18F3x44[NbMcaToMstCpuEn]." + "Doing that here\n", __func__);
Do integrate checkpatch.pl into your workflow. It is sometimes right.
> + val |= BIT(27); > + err = pci_write_config_dword(F3, NBCFG, val); > + if (err) > + pr_err("%s: Error writing F%dx%03x.\n", __func__, > + PCI_FUNC(F3->devfn), > + NBCFG); > + } > +}
-- Regards/Gruss, Boris.
ECO tip #101: Trim your mails when you reply. --
| |