Hi all,
This set adds two pieces of debug functionality.
1) Saving the system physical address of a recorded error.
2) Printing record entries through a debugfs file.
I'd like to include Murali, Naveen, and Sathya as co-developers, since
this is based on their previous work from here:
https://lore.kernel.org/r/[email protected]
Thanks,
Yazen
Yazen Ghannam (3):
RAS: Export ras_debugfs_dir
RAS/AMD/FMPM: Save SPA values
RAS/AMD/FMPM: Add debugfs interface to print record entries
drivers/ras/amd/fmpm.c | 183 +++++++++++++++++++++++++++++++++++++++++
drivers/ras/debugfs.c | 1 +
2 files changed, 184 insertions(+)
base-commit: 3513ecaa685c6627a943b1f610421754734301fa
--
2.34.1
..so that RAS modules can use it.
Signed-off-by: Yazen Ghannam <[email protected]>
---
drivers/ras/debugfs.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/ras/debugfs.c b/drivers/ras/debugfs.c
index ffb973c328e3..fd3d4595c122 100644
--- a/drivers/ras/debugfs.c
+++ b/drivers/ras/debugfs.c
@@ -4,6 +4,7 @@
#include "debugfs.h"
struct dentry *ras_debugfs_dir;
+EXPORT_SYMBOL_GPL(ras_debugfs_dir);
static atomic_t trace_count = ATOMIC_INIT(0);
--
2.34.1
The system physical address (SPA) of an error is not a stable value. It
will change depending on the location of the memory: parts can be
swapped. And it will change depending on memory topology: NUMA nodes
and/or interleaving can be adjusted.
Therefore, the SPA value is not part of the "FRU Memory Poison" record
format. And it will not be saved to persistent storage.
However, the SPA values can be helpful during debug and for system
admins during run time.
Save the SPA values in a separate structure. Update this when records
are restored and when new errors are saved.
Signed-off-by: Yazen Ghannam <[email protected]>
---
drivers/ras/amd/fmpm.c | 54 ++++++++++++++++++++++++++++++++++++++++++
1 file changed, 54 insertions(+)
diff --git a/drivers/ras/amd/fmpm.c b/drivers/ras/amd/fmpm.c
index 80dd112b720a..04961d36dd9f 100644
--- a/drivers/ras/amd/fmpm.c
+++ b/drivers/ras/amd/fmpm.c
@@ -111,6 +111,9 @@ struct fru_rec {
*/
static struct fru_rec **fru_records;
+#define INVALID_SPA ~0ULL
+static u64 *sys_addrs;
+
#define CPER_CREATOR_FMP \
GUID_INIT(0xcd5c2993, 0xf4b2, 0x41b2, 0xb5, 0xd4, 0xf9, 0xc3, \
0xa0, 0x33, 0x08, 0x75)
@@ -140,6 +143,9 @@ static unsigned int max_nr_fru;
/* Total length of record including headers and list of descriptor entries. */
static size_t max_rec_len;
+/* Total number of entries for the entire system. */
+static unsigned int sys_nr_entries;
+
/*
* Protect the local records cache in fru_records and prevent concurrent
* writes to storage. This is only needed after init once notifier block
@@ -269,6 +275,40 @@ static bool rec_has_fpd(struct fru_rec *rec, struct cper_fru_poison_desc *fpd)
return false;
}
+static void save_spa(struct fru_rec *rec, unsigned int entry,
+ u64 addr, u64 id, unsigned int cpu)
+{
+ unsigned int i, fru_idx, sys_entry;
+ unsigned long sys_addr;
+ struct atl_err a_err;
+
+ memset(&a_err, 0, sizeof(struct atl_err));
+
+ a_err.addr = addr;
+ a_err.ipid = id;
+ a_err.cpu = cpu;
+
+ sys_addr = amd_convert_umc_mca_addr_to_sys_addr(&a_err);
+ if (IS_ERR_VALUE(sys_addr)) {
+ pr_debug("Failed to get system address\n");
+ return;
+ }
+
+ for (i = 0; i < sys_nr_entries; i += max_nr_entries) {
+ fru_idx = i / max_nr_entries;
+ if (fru_records[fru_idx] != rec)
+ continue;
+
+ sys_entry = i + entry;
+ if (sys_entry < sys_nr_entries) {
+ sys_addrs[sys_entry] = sys_addr;
+ pr_debug("fru_idx: %u, entry: %u, sys_entry: %u, sys_addr: 0x%016llx\n",
+ fru_idx, entry, sys_entry, sys_addrs[sys_entry]);
+ break;
+ }
+ }
+}
+
static void update_fru_record(struct fru_rec *rec, struct mce *m)
{
struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
@@ -301,6 +341,7 @@ static void update_fru_record(struct fru_rec *rec, struct mce *m)
entry = fmp->nr_entries;
save_fpd:
+ save_spa(rec, entry, m->addr, m->ipid, m->extcpu);
fpd_dest = &rec->entries[entry];
memcpy(fpd_dest, &fpd, sizeof(struct cper_fru_poison_desc));
@@ -385,6 +426,7 @@ static void retire_mem_fmp(struct fru_rec *rec)
continue;
retire_dram_row(fpd->addr, fpd->hw_id, err_cpu);
+ save_spa(rec, i, fpd->addr, fpd->hw_id, err_cpu);
}
}
@@ -696,6 +738,8 @@ static int get_system_info(void)
if (!max_nr_entries)
max_nr_entries = FMPM_DEFAULT_MAX_NR_ENTRIES;
+ sys_nr_entries = max_nr_fru * max_nr_entries;
+
max_rec_len = sizeof(struct fru_rec);
max_rec_len += sizeof(struct cper_fru_poison_desc) * max_nr_entries;
@@ -714,6 +758,7 @@ static void free_records(void)
kfree(rec);
kfree(fru_records);
+ kfree(sys_addrs);
}
static int allocate_records(void)
@@ -734,6 +779,15 @@ static int allocate_records(void)
}
}
+ sys_addrs = kcalloc(sys_nr_entries, sizeof(u64), GFP_KERNEL);
+ if (!sys_addrs) {
+ ret = -ENOMEM;
+ goto out_free;
+ }
+
+ for (i = 0; i < sys_nr_entries; i++)
+ sys_addrs[i] = INVALID_SPA;
+
return ret;
out_free:
--
2.34.1
On Mon, Feb 26, 2024 at 09:29:39AM -0600, Yazen Ghannam wrote:
> ...so that RAS modules can use it.
>
> Signed-off-by: Yazen Ghannam <[email protected]>
> ---
> drivers/ras/debugfs.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/ras/debugfs.c b/drivers/ras/debugfs.c
> index ffb973c328e3..fd3d4595c122 100644
> --- a/drivers/ras/debugfs.c
> +++ b/drivers/ras/debugfs.c
> @@ -4,6 +4,7 @@
> #include "debugfs.h"
>
> struct dentry *ras_debugfs_dir;
> +EXPORT_SYMBOL_GPL(ras_debugfs_dir);
I am never a fan of exporting things so let's add a helper this first,
in a separate pre-patch.
Also, I'd like to have this:
- fmpm_dfs_dir = debugfs_create_dir(KBUILD_MODNAME, ras_debugfs_dir);
+ fmpm_dfs_dir = debugfs_create_dir("fmpm", dfs);
so that we don't accidentally change the module name and fsck up scripts
in the process.
Thx.
---
diff --git a/drivers/ras/amd/fmpm.c b/drivers/ras/amd/fmpm.c
index e0811a71e6e7..b8e3e4c81f7e 100644
--- a/drivers/ras/amd/fmpm.c
+++ b/drivers/ras/amd/fmpm.c
@@ -913,10 +913,13 @@ static const struct file_operations fmpm_fops = {
static void setup_debugfs(void)
{
- if (!ras_debugfs_dir)
+ struct dentry *dfs;
+
+ dfs = ras_get_debugfs_root();
+ if (!dfs)
return;
- fmpm_dfs_dir = debugfs_create_dir(KBUILD_MODNAME, ras_debugfs_dir);
+ fmpm_dfs_dir = debugfs_create_dir("fmpm", dfs);
if (!fmpm_dfs_dir)
return;
diff --git a/drivers/ras/cec.c b/drivers/ras/cec.c
index 321af498ee11..e440b15fbabc 100644
--- a/drivers/ras/cec.c
+++ b/drivers/ras/cec.c
@@ -480,9 +480,15 @@ DEFINE_SHOW_ATTRIBUTE(array);
static int __init create_debugfs_nodes(void)
{
- struct dentry *d, *pfn, *decay, *count, *array;
+ struct dentry *d, *pfn, *decay, *count, *array, *dfs;
- d = debugfs_create_dir("cec", ras_debugfs_dir);
+ dfs = ras_get_debugfs_root();
+ if (!dfs) {
+ pr_warn("Error getting RAS debugfs root!\n");
+ return -1;
+ }
+
+ d = debugfs_create_dir("cec", dfs);
if (!d) {
pr_warn("Error creating cec debugfs node!\n");
return -1;
diff --git a/drivers/ras/debugfs.c b/drivers/ras/debugfs.c
index fd3d4595c122..42afd3de68b2 100644
--- a/drivers/ras/debugfs.c
+++ b/drivers/ras/debugfs.c
@@ -3,11 +3,16 @@
#include <linux/ras.h>
#include "debugfs.h"
-struct dentry *ras_debugfs_dir;
-EXPORT_SYMBOL_GPL(ras_debugfs_dir);
+static struct dentry *ras_debugfs_dir;
static atomic_t trace_count = ATOMIC_INIT(0);
+struct dentry *ras_get_debugfs_root(void)
+{
+ return ras_debugfs_dir;
+}
+EXPORT_SYMBOL_GPL(ras_get_debugfs_root);
+
int ras_userspace_consumers(void)
{
return atomic_read(&trace_count);
diff --git a/drivers/ras/debugfs.h b/drivers/ras/debugfs.h
index c07443b462ad..4749ccdeeba1 100644
--- a/drivers/ras/debugfs.h
+++ b/drivers/ras/debugfs.h
@@ -4,6 +4,6 @@
#include <linux/debugfs.h>
-extern struct dentry *ras_debugfs_dir;
+struct dentry *ras_get_debugfs_root(void);
#endif /* __RAS_DEBUGFS_H__ */
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Mon, Feb 26, 2024 at 09:29:40AM -0600, Yazen Ghannam wrote:
> @@ -111,6 +111,9 @@ struct fru_rec {
> */
> static struct fru_rec **fru_records;
>
> +#define INVALID_SPA ~0ULL
> +static u64 *sys_addrs;
Let's do:
/* system physical addresses array */
static u64 *spa_addrs;
so that it is self-documenting.
> +
> #define CPER_CREATOR_FMP \
> GUID_INIT(0xcd5c2993, 0xf4b2, 0x41b2, 0xb5, 0xd4, 0xf9, 0xc3, \
> 0xa0, 0x33, 0x08, 0x75)
> @@ -140,6 +143,9 @@ static unsigned int max_nr_fru;
> /* Total length of record including headers and list of descriptor entries. */
> static size_t max_rec_len;
>
> +/* Total number of entries for the entire system. */
> +static unsigned int sys_nr_entries;
sys_ things are always related to syscalls etc. Since it is a static
var, just call it:
/* Number of SPA entries */
nr_entries
or so.
I was gonna say "nr_err_records" but we're calling them entries so...
> /*
> * Protect the local records cache in fru_records and prevent concurrent
> * writes to storage. This is only needed after init once notifier block
> @@ -269,6 +275,40 @@ static bool rec_has_fpd(struct fru_rec *rec, struct cper_fru_poison_desc *fpd)
> return false;
> }
>
> +static void save_spa(struct fru_rec *rec, unsigned int entry,
> + u64 addr, u64 id, unsigned int cpu)
> +{
> + unsigned int i, fru_idx, sys_entry;
> + unsigned long sys_addr;
> + struct atl_err a_err;
> +
> + memset(&a_err, 0, sizeof(struct atl_err));
> +
> + a_err.addr = addr;
> + a_err.ipid = id;
> + a_err.cpu = cpu;
> +
> + sys_addr = amd_convert_umc_mca_addr_to_sys_addr(&a_err);
> + if (IS_ERR_VALUE(sys_addr)) {
> + pr_debug("Failed to get system address\n");
> + return;
> + }
> +
> + for (i = 0; i < sys_nr_entries; i += max_nr_entries) {
> + fru_idx = i / max_nr_entries;
> + if (fru_records[fru_idx] != rec)
> + continue;
> +
> + sys_entry = i + entry;
> + if (sys_entry < sys_nr_entries) {
> + sys_addrs[sys_entry] = sys_addr;
> + pr_debug("fru_idx: %u, entry: %u, sys_entry: %u, sys_addr: 0x%016llx\n",
> + fru_idx, entry, sys_entry, sys_addrs[sys_entry]);
> + break;
> + }
else {
WARN_ON_ONCE("... indexing wrong bla... ");
}
to catch crap.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
It is helpful to see the saved record entries during run time in
human-readable format. This is useful for testing during module
development. And it can be used by system admins to quickly and easily
see the state of the system.
Provide a sequential file in debugfs to print fields of interest from
the FRU records and their entries.
Don't fail to load the module if the debugfs interface is not available.
This is a convenience feature which does not affect other module
functionality.
The new interface reads the record entries and should hold the mutex.
Expand the mutex code comment to clarify when it should be held.
Signed-off-by: Yazen Ghannam <[email protected]>
---
drivers/ras/amd/fmpm.c | 129 +++++++++++++++++++++++++++++++++++++++++
1 file changed, 129 insertions(+)
diff --git a/drivers/ras/amd/fmpm.c b/drivers/ras/amd/fmpm.c
index 04961d36dd9f..967af84b7fe9 100644
--- a/drivers/ras/amd/fmpm.c
+++ b/drivers/ras/amd/fmpm.c
@@ -54,6 +54,8 @@
#include <asm/cpu_device_id.h>
#include <asm/mce.h>
+#include "../debugfs.h"
+
#define INVALID_CPU UINT_MAX
/* Validation Bits */
@@ -114,6 +116,9 @@ static struct fru_rec **fru_records;
#define INVALID_SPA ~0ULL
static u64 *sys_addrs;
+static struct dentry *fmpm_dfs_dir;
+static struct dentry *fmpm_dfs_entries;
+
#define CPER_CREATOR_FMP \
GUID_INIT(0xcd5c2993, 0xf4b2, 0x41b2, 0xb5, 0xd4, 0xf9, 0xc3, \
0xa0, 0x33, 0x08, 0x75)
@@ -150,6 +155,11 @@ static unsigned int sys_nr_entries;
* Protect the local records cache in fru_records and prevent concurrent
* writes to storage. This is only needed after init once notifier block
* registration is done.
+ *
+ * The majority of a record is fixed at module init and will not change
+ * during run time. The entries within a record will be updated as new
+ * errors are reported. The mutex should be held whenever the entries are
+ * accessed during run time.
*/
static DEFINE_MUTEX(fmpm_update_mutex);
@@ -799,6 +809,122 @@ static int allocate_records(void)
return ret;
}
+static void *fmpm_start(struct seq_file *f, loff_t *pos)
+{
+ if (*pos >= (sys_nr_entries + 1))
+ return NULL;
+ return pos;
+}
+
+static void *fmpm_next(struct seq_file *f, void *data, loff_t *pos)
+{
+ if (++(*pos) >= (sys_nr_entries + 1))
+ return NULL;
+ return pos;
+}
+
+static void fmpm_stop(struct seq_file *f, void *data)
+{
+}
+
+#define SHORT_WIDTH 8
+#define U64_WIDTH 18
+#define TIMESTAMP_WIDTH 19
+#define LONG_WIDTH 24
+#define U64_PAD (LONG_WIDTH - U64_WIDTH)
+#define TS_PAD (LONG_WIDTH - TIMESTAMP_WIDTH)
+static int fmpm_show(struct seq_file *f, void *data)
+{
+ unsigned int fru_idx, entry, sys_entry, line;
+ struct cper_fru_poison_desc *fpd;
+ struct fru_rec *rec;
+
+ line = *(loff_t *)data;
+ if (line == 0) {
+ seq_printf(f, "%-*s", SHORT_WIDTH, "fru_idx");
+ seq_printf(f, "%-*s", LONG_WIDTH, "fru_id");
+ seq_printf(f, "%-*s", SHORT_WIDTH, "entry");
+ seq_printf(f, "%-*s", LONG_WIDTH, "timestamp");
+ seq_printf(f, "%-*s", LONG_WIDTH, "hw_id");
+ seq_printf(f, "%-*s", LONG_WIDTH, "addr");
+ seq_printf(f, "%-*s", LONG_WIDTH, "spa");
+ goto out_newline;
+ }
+
+ sys_entry = line - 1;
+ fru_idx = sys_entry / max_nr_entries;
+ entry = sys_entry % max_nr_entries;
+
+ rec = fru_records[fru_idx];
+ if (!rec)
+ goto out;
+
+ seq_printf(f, "%-*u", SHORT_WIDTH, fru_idx);
+ seq_printf(f, "0x%016llx%-*s", rec->fmp.fru_id, U64_PAD, "");
+ seq_printf(f, "%-*u", SHORT_WIDTH, entry);
+
+ mutex_lock(&fmpm_update_mutex);
+
+ if (entry >= rec->fmp.nr_entries) {
+ seq_printf(f, "%-*s", LONG_WIDTH, "*");
+ seq_printf(f, "%-*s", LONG_WIDTH, "*");
+ seq_printf(f, "%-*s", LONG_WIDTH, "*");
+ seq_printf(f, "%-*s", LONG_WIDTH, "*");
+ goto out_unlock;
+ }
+
+ fpd = &rec->entries[entry];
+
+ seq_printf(f, "%ptT%-*s", &fpd->timestamp, TS_PAD, "");
+ seq_printf(f, "0x%016llx%-*s", fpd->hw_id, U64_PAD, "");
+ seq_printf(f, "0x%016llx%-*s", fpd->addr, U64_PAD, "");
+
+ if (sys_addrs[sys_entry] == INVALID_SPA)
+ seq_printf(f, "%-*s", LONG_WIDTH, "*");
+ else
+ seq_printf(f, "0x%016llx%-*s", sys_addrs[sys_entry], U64_PAD, "");
+
+out_unlock:
+ mutex_unlock(&fmpm_update_mutex);
+out_newline:
+ seq_putc(f, '\n');
+out:
+ return 0;
+}
+
+static const struct seq_operations fmpm_seq_ops = {
+ .start = fmpm_start,
+ .next = fmpm_next,
+ .stop = fmpm_stop,
+ .show = fmpm_show,
+};
+
+static int fmpm_open(struct inode *inode, struct file *file)
+{
+ return seq_open(file, &fmpm_seq_ops);
+}
+
+static const struct file_operations fmpm_fops = {
+ .open = fmpm_open,
+ .release = seq_release,
+ .read = seq_read,
+ .llseek = seq_lseek,
+};
+
+static void setup_debugfs(void)
+{
+ if (!ras_debugfs_dir)
+ return;
+
+ fmpm_dfs_dir = debugfs_create_dir(KBUILD_MODNAME, ras_debugfs_dir);
+ if (!fmpm_dfs_dir)
+ return;
+
+ fmpm_dfs_entries = debugfs_create_file("entries", 0400, fmpm_dfs_dir, NULL, &fmpm_fops);
+ if (!fmpm_dfs_entries)
+ debugfs_remove(fmpm_dfs_dir);
+}
+
static const struct x86_cpu_id fmpm_cpuids[] = {
X86_MATCH_VENDOR_FAM(AMD, 0x19, NULL),
{ }
@@ -840,6 +966,8 @@ static int __init fru_mem_poison_init(void)
if (ret)
goto out_free;
+ setup_debugfs();
+
retire_mem_records();
mce_register_decode_chain(&fru_mem_poison_nb);
@@ -856,6 +984,7 @@ static int __init fru_mem_poison_init(void)
static void __exit fru_mem_poison_exit(void)
{
mce_unregister_decode_chain(&fru_mem_poison_nb);
+ debugfs_remove(fmpm_dfs_dir);
free_records();
}
--
2.34.1