Skip to content

Commit 584e097

Browse files
Kyle Meyeraegl
authored andcommitted
EDAC/{i10nm,skx,skx_common}: Support UV systems
The 3-bit source IDs in PCI configuration space registers, used to map devices to sockets, are limited to 8 unique IDs, and each ID is local to a UPI/QPI domain. Source IDs cannot be used to map devices to sockets on UV systems because they can exceed 8 sockets and have multiple UPI/QPI domains with identical, repeating source IDs. Use NUMA information to get package IDs instead of source IDs on UV systems, and use package/source IDs to name IMC information structures. Signed-off-by: Kyle Meyer <[email protected]> Signed-off-by: Tony Luck <[email protected]> Tested-by: Qiuxu Zhuo <[email protected]> Reviewed-by: Qiuxu Zhuo <[email protected]> Link: https://lore.kernel.org/all/[email protected]/
1 parent 2e55bb9 commit 584e097

File tree

4 files changed

+40
-30
lines changed

4 files changed

+40
-30
lines changed

drivers/edac/i10nm_base.c

Lines changed: 3 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1011,7 +1011,7 @@ static struct notifier_block i10nm_mce_dec = {
10111011

10121012
static int __init i10nm_init(void)
10131013
{
1014-
u8 mc = 0, src_id = 0, node_id = 0;
1014+
u8 mc = 0, src_id = 0;
10151015
const struct x86_cpu_id *id;
10161016
struct res_config *cfg;
10171017
const char *owner;
@@ -1071,19 +1071,14 @@ static int __init i10nm_init(void)
10711071
if (rc < 0)
10721072
goto fail;
10731073

1074-
rc = skx_get_node_id(d, &node_id);
1075-
if (rc < 0)
1076-
goto fail;
1077-
1078-
edac_dbg(2, "src_id = %d node_id = %d\n", src_id, node_id);
1074+
edac_dbg(2, "src_id = %d\n", src_id);
10791075
for (i = 0; i < imc_num; i++) {
10801076
if (!d->imc[i].mdev)
10811077
continue;
10821078

10831079
d->imc[i].mc = mc++;
10841080
d->imc[i].lmc = i;
1085-
d->imc[i].src_id = src_id;
1086-
d->imc[i].node_id = node_id;
1081+
d->imc[i].src_id = src_id;
10871082
if (d->imc[i].hbm_mc) {
10881083
d->imc[i].chan_mmio_sz = cfg->hbm_chan_mmio_sz;
10891084
d->imc[i].num_channels = cfg->hbm_chan_num;

drivers/edac/skx_base.c

Lines changed: 3 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -600,7 +600,7 @@ static int __init skx_init(void)
600600
const struct munit *m;
601601
const char *owner;
602602
int rc = 0, i, off[3] = {0xd0, 0xd4, 0xd8};
603-
u8 mc = 0, src_id, node_id;
603+
u8 mc = 0, src_id;
604604
struct skx_dev *d;
605605

606606
edac_dbg(2, "\n");
@@ -650,15 +650,12 @@ static int __init skx_init(void)
650650
rc = skx_get_src_id(d, 0xf0, &src_id);
651651
if (rc < 0)
652652
goto fail;
653-
rc = skx_get_node_id(d, &node_id);
654-
if (rc < 0)
655-
goto fail;
656-
edac_dbg(2, "src_id=%d node_id=%d\n", src_id, node_id);
653+
654+
edac_dbg(2, "src_id = %d\n", src_id);
657655
for (i = 0; i < SKX_NUM_IMC; i++) {
658656
d->imc[i].mc = mc++;
659657
d->imc[i].lmc = i;
660658
d->imc[i].src_id = src_id;
661-
d->imc[i].node_id = node_id;
662659
rc = skx_register_mci(&d->imc[i], d->imc[i].chan[0].cdev,
663660
"Skylake Socket", EDAC_MOD_STR,
664661
skx_get_dimm_config, cfg);

drivers/edac/skx_common.c

Lines changed: 33 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -19,6 +19,7 @@
1919
#include <linux/adxl.h>
2020
#include <acpi/nfit.h>
2121
#include <asm/mce.h>
22+
#include <asm/uv/uv.h>
2223
#include "edac_module.h"
2324
#include "skx_common.h"
2425

@@ -221,33 +222,51 @@ void skx_set_decode(skx_decode_f decode, skx_show_retry_log_f show_retry_log)
221222
}
222223
EXPORT_SYMBOL_GPL(skx_set_decode);
223224

224-
int skx_get_src_id(struct skx_dev *d, int off, u8 *id)
225+
static int skx_get_pkg_id(struct skx_dev *d, u8 *id)
225226
{
226-
u32 reg;
227+
int node;
228+
int cpu;
227229

228-
if (pci_read_config_dword(d->util_all, off, &reg)) {
229-
skx_printk(KERN_ERR, "Failed to read src id\n");
230-
return -ENODEV;
230+
node = pcibus_to_node(d->util_all->bus);
231+
if (numa_valid_node(node)) {
232+
for_each_cpu(cpu, cpumask_of_pcibus(d->util_all->bus)) {
233+
struct cpuinfo_x86 *c = &cpu_data(cpu);
234+
235+
if (c->initialized && cpu_to_node(cpu) == node) {
236+
*id = c->topo.pkg_id;
237+
return 0;
238+
}
239+
}
231240
}
232241

233-
*id = GET_BITFIELD(reg, 12, 14);
234-
return 0;
242+
skx_printk(KERN_ERR, "Failed to get package ID from NUMA information\n");
243+
return -ENODEV;
235244
}
236-
EXPORT_SYMBOL_GPL(skx_get_src_id);
237245

238-
int skx_get_node_id(struct skx_dev *d, u8 *id)
246+
int skx_get_src_id(struct skx_dev *d, int off, u8 *id)
239247
{
240248
u32 reg;
241249

242-
if (pci_read_config_dword(d->util_all, 0xf4, &reg)) {
243-
skx_printk(KERN_ERR, "Failed to read node id\n");
250+
/*
251+
* The 3-bit source IDs in PCI configuration space registers are limited
252+
* to 8 unique IDs, and each ID is local to a UPI/QPI domain.
253+
*
254+
* Source IDs cannot be used to map devices to sockets on UV systems
255+
* because they can exceed 8 sockets and have multiple UPI/QPI domains
256+
* with identical, repeating source IDs.
257+
*/
258+
if (is_uv_system())
259+
return skx_get_pkg_id(d, id);
260+
261+
if (pci_read_config_dword(d->util_all, off, &reg)) {
262+
skx_printk(KERN_ERR, "Failed to read src id\n");
244263
return -ENODEV;
245264
}
246265

247-
*id = GET_BITFIELD(reg, 0, 2);
266+
*id = GET_BITFIELD(reg, 12, 14);
248267
return 0;
249268
}
250-
EXPORT_SYMBOL_GPL(skx_get_node_id);
269+
EXPORT_SYMBOL_GPL(skx_get_src_id);
251270

252271
static int get_width(u32 mtr)
253272
{
@@ -507,7 +526,7 @@ int skx_register_mci(struct skx_imc *imc, struct pci_dev *pdev,
507526
pvt->imc = imc;
508527

509528
mci->ctl_name = kasprintf(GFP_KERNEL, "%s#%d IMC#%d", ctl_name,
510-
imc->node_id, imc->lmc);
529+
imc->src_id, imc->lmc);
511530
if (!mci->ctl_name) {
512531
rc = -ENOMEM;
513532
goto fail0;

drivers/edac/skx_common.h

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -103,7 +103,7 @@ struct skx_dev {
103103
bool hbm_mc;
104104
u8 mc; /* system wide mc# */
105105
u8 lmc; /* socket relative mc# */
106-
u8 src_id, node_id;
106+
u8 src_id;
107107
struct skx_channel {
108108
struct pci_dev *cdev;
109109
struct pci_dev *edev;
@@ -244,7 +244,6 @@ void skx_set_mem_cfg(bool mem_cfg_2lm);
244244
void skx_set_res_cfg(struct res_config *cfg);
245245

246246
int skx_get_src_id(struct skx_dev *d, int off, u8 *id);
247-
int skx_get_node_id(struct skx_dev *d, u8 *id);
248247

249248
int skx_get_all_bus_mappings(struct res_config *cfg, struct list_head **list);
250249

0 commit comments

Comments
 (0)