SLFO-pool/powerpc-utils
SHA256
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Sync from SUSE:SLFO:Main powerpc-utils revision a7538321d8f190c196916ed56ef7f51d

Browse Source
This commit is contained in:
Adrian Schröter 2024-11-28 17:45:33 +01:00
parent d16dfc0a8c
commit f64e800c72
16 changed files with 121 additions and 1347 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
0001-drmgr-Return-from-get_node_by_name-if-matched-DRC-in.patch
View File

@ -1,35 +0,0 @@
From 1dc1ecf7dce7825d352b045c98aa51711b58aaca Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Fri, 21 Jun 2024 15:39:42 -0700
Subject: [PATCH] drmgr: Return from get_node_by_name() if matched DRC index
Upstream: merged, expected in v1.3.13
Git-commit: 1dc1ecf7dce7825d352b045c98aa51711b58aaca
get_node_by_name() should return dr_node if the DRC name or DRC
index is matched. But the current code returns only if the DRC
name is matched. This patch fixes this issue and returns dr_node
if the index is matched.
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
src/drmgr/common_pci.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/drmgr/common_pci.c b/src/drmgr/common_pci.c
index c6dcfdf..2e0e5fb 100644
--- a/src/drmgr/common_pci.c
+++ b/src/drmgr/common_pci.c
@@ -969,7 +969,7 @@ get_node_by_name(const char *drc_name, uint32_t node_type)
/* See if the drc index was specified */
drc_index = strtoul(drc_name, NULL, 0);
if (node->drc_index == drc_index)
- continue;
+ break;
for (child = node->children; child; child = child->next) {
if (strcmp(drc_name, child->drc_name) == 0)
--
2.45.2

99
0001-drmgr-pci-Enable-in-kernel-functionality-to-update-d.patch
View File

@ -1,99 +0,0 @@
From a6a7d375d38823a08d358b50866a95f6f326907e Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Tue, 13 Aug 2024 14:40:23 -0700
Subject: [PATCH 1/4] drmgr/pci: Enable in-kernel functionality to update
device tree
drmgr updates the device tree by writing to /proc/ppc64/ofdt. Also
invokes configure_connector RTAS call to retrieve new device nodes
for IO ADD. But this functionality need /dev/mem access which is
restricted under system lockdown.
The kernel updates provided a sysfs file (/sys/kernel/dlpar) that
will allow drmgr command invoke the following interfaces to update
the device tree.
dt add index <DRC index> ---> To add new device nodes to the device
tree which is used for IO ADD.
dt remove index <DRC index> ---> To remove device nodes for IO
REMOVE
This patch checks the kernel interface for the availability of
device tree update feature and adds do_dt_kernel_dlpar() to invoke
the above kernel interfaces.
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
src/drmgr/common.c | 6 ++++++
src/drmgr/common_pci.c | 29 +++++++++++++++++++++++++++++
src/drmgr/dr.h | 1 +
3 files changed, 36 insertions(+)
diff --git a/src/drmgr/common.c b/src/drmgr/common.c
index bfec0b9bb966..70f4dfda92a5 100644
--- a/src/drmgr/common.c
+++ b/src/drmgr/common.c
@@ -1504,6 +1504,12 @@ int kernel_dlpar_exists(void)
if (strstr(buf, "cpu"))
return 1;
break;
+ case DRC_TYPE_PCI:
+ case DRC_TYPE_PHB:
+ case DRC_TYPE_SLOT:
+ if (strstr(buf, "dt"))
+ return 1;
+ break;
default:
return 0;
}
diff --git a/src/drmgr/common_pci.c b/src/drmgr/common_pci.c
index 1ef4ec9653b0..374129cf33e0 100644
--- a/src/drmgr/common_pci.c
+++ b/src/drmgr/common_pci.c
@@ -1619,3 +1619,32 @@ int disable_hp_children(char *drc_name)
}
return 0;
}
+
+/*
+ * kernel interface to update device tree nodes.
+ * dlpar dt [add|remove] index <#drc index>
+ */
+int do_dt_kernel_dlpar(uint32_t index, int action)
+{
+ char cmdbuf[256];
+ int offset;
+
+ offset = sprintf(cmdbuf, "%s ", "dt");
+
+ switch (action) {
+ case ADD:
+ offset += sprintf(cmdbuf + offset, "add ");
+ break;
+ case REMOVE:
+ offset += sprintf(cmdbuf + offset, "remove ");
+ break;
+ default:
+ /* Should not happen */
+ say(ERROR, "Invalid action type specified\n");
+ return -EINVAL;
+ }
+
+ offset += sprintf(cmdbuf + offset, "index 0x%x", index);
+
+ return do_kernel_dlpar(cmdbuf, offset);
+}
diff --git a/src/drmgr/dr.h b/src/drmgr/dr.h
index 60c31c44b7e3..72ede55547a3 100644
--- a/src/drmgr/dr.h
+++ b/src/drmgr/dr.h
@@ -188,4 +188,5 @@ static inline int do_kernel_dlpar(const char *cmd, int len)
{
return do_kernel_dlpar_common(cmd, len, 0);
}
+int do_dt_kernel_dlpar(uint32_t, int);
#endif
--
2.46.0

72
0002-drmgr-SLOT-Add-kernel-interface-support-for-device-t.patch
View File

@ -1,72 +0,0 @@
From e2388681df4aec97ab9a3f883baf37f51a710a00 Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Tue, 13 Aug 2024 14:40:26 -0700
Subject: [PATCH 2/4] drmgr/SLOT: Add kernel interface support for device tree
update
Use the following kernel interfaces for SLOT device type to update
the device tree if this feature is enabled in the kernel.
dt add index <DRC index> --> for IO add
dt remove index <DRC index> --> for IO remove
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
src/drmgr/drslot_chrp_slot.c | 25 +++++++++++++++++--------
1 file changed, 17 insertions(+), 8 deletions(-)
diff --git a/src/drmgr/drslot_chrp_slot.c b/src/drmgr/drslot_chrp_slot.c
index 0966c256aa01..180b10857add 100644
--- a/src/drmgr/drslot_chrp_slot.c
+++ b/src/drmgr/drslot_chrp_slot.c
@@ -71,7 +71,10 @@ release_slot(struct dr_node *slot)
if (rc)
return rc;
- rc = remove_device_tree_nodes(slot->ofdt_path);
+ if (kernel_dlpar_exists())
+ rc = do_dt_kernel_dlpar(slot->drc_index, REMOVE);
+ else
+ rc = remove_device_tree_nodes(slot->ofdt_path);
if (rc) {
acquire_drc(slot->drc_index);
return rc;
@@ -160,7 +163,6 @@ static int
acquire_slot(char *drc_name, struct dr_node **slot)
{
struct dr_connector drc;
- struct of_node *of_nodes;
char path[DR_PATH_MAX];
int rc;
@@ -180,14 +182,21 @@ acquire_slot(char *drc_name, struct dr_node **slot)
if (rc)
return rc;
- of_nodes = configure_connector(drc.index);
- if (of_nodes == NULL) {
- release_drc(drc.index, PCI_DLPAR_DEV);
- return -1;
+ if (kernel_dlpar_exists()) {
+ rc = do_dt_kernel_dlpar(drc.index, ADD);
+ } else {
+ struct of_node *of_nodes;
+
+ of_nodes = configure_connector(drc.index);
+ if (of_nodes == NULL) {
+ release_drc(drc.index, PCI_DLPAR_DEV);
+ return -1;
+ }
+
+ rc = add_device_tree_nodes(path, of_nodes);
+ free_of_node(of_nodes);
}
- rc = add_device_tree_nodes(path, of_nodes);
- free_of_node(of_nodes);
if (rc) {
say(ERROR, "add_device_tree_nodes failed at %s\n", path);
release_drc(drc.index, PCI_DLPAR_DEV);
--
2.46.0

236
0002-lparstat-Fix-Idle-and-busy-PURR-SPURR.patch
View File

@ -1,236 +0,0 @@
From 2af8c0b9a285e8a6104560d0f482819e56060443 Mon Sep 17 00:00:00 2001
From: Saket Kumar Bhaskar <skb99@linux.ibm.com>
Date: Thu, 18 Jul 2024 00:39:24 +0530
Subject: [PATCH] lparstat: Fix Idle and busy PURR/SPURR
Upstream: merged, expected in v1.3.13
Git-commit: 2af8c0b9a285e8a6104560d0f482819e56060443
lparstat -E gives %busy and %idle for actual(PURR based) and normalized
(SPURR based).Idle and busy PURR/SPURR values are not adding upto 100%
in case of dedicated-donate and shared partitions, with the present
formula. Because of this, users might get a false impression of resource
utilisation. This is expected because a core can be in either
idle or busy state out of total of 100(core's shared resource can
either be consumed or be left idle). When lpar is in dedicated-donate
or shared,the purr values are not being counted when the CPU is ceded.
The idle_purr is calculated by taking snapshots of purr values at
every idle entry and idle exit. So, when a CPU is ceded, the calculation
for idle_purr will be wrong as purr is not being counted.
Before Change:
|-----------------------------------------------------------------|
| Dedicated-donate (8 cores) : |
|----------------------|---------------------|--------------------|
| | Actual | Normalized |
| Stress-ng threads |---------------------|--------------------|
| | %busy | %idle | %busy | %idle |
|----------------------|----------|----------|---------|----------|
| 0 threads | 0.02 | 0.05 | 0.02 | 0.05 |
|----------------------|----------|----------|---------|----------|
| 8 threads | 32.64 | 17.37 | 35.25 | 18.77 |
|----------------------|----------|----------|---------|----------|
| 16 threads | 58.61 | 16.42 | 63.29 | 17.74 |
|----------------------|----------|----------|---------|----------|
| 24 threads | 78.14 | 21.86 | 84.39 | 23.61 |
|----------------------|----------|----------|---------|----------|
| 32 threads | 83.60 | 16.40 | 90.30 | 17.71 |
|----------------------|----------|----------|---------|----------|
| 40 threads | 91.90 | 6.94 | 98.31 | 7.46 |
|----------------------|----------|----------|---------|----------|
| 48 threads | 96.08 | 3.92 | 102.79 | 4.21 |
|----------------------|----------|----------|---------|----------|
| 56 threads | 98.42 | 1.57 | 105.31 | 1.69 |
|----------------------|----------|----------|---------|----------|
| 64 threads | 100.00 | 0.00 | 106.00 | 0.00 |
|----------------------|----------|----------|---------|----------|
|-----------------------------------------------------------------|
| Shared Capped (8 VP / 4 EC) : |
|----------------------|---------------------|--------------------|
| | Actual | Normalized |
| Stress-ng threads |---------------------|--------------------|
| | %busy | %idle | %busy | %idle |
|----------------------|----------|----------|---------|----------|
| 0 threads | 0.04 | 0.18 | 0.03 | 0.19 |
|----------------------|----------|----------|---------|----------|
| 8 threads | 35.90 | 14.09 | 38.77 | 15.21 |
|----------------------|----------|----------|---------|----------|
| 16 threads | 35.25 | 14.84 | 38.08 | 16.02 |
|----------------------|----------|----------|---------|----------|
| 24 threads | 40.13 | 9.73 | 42.93 | 10.43 |
|----------------------|----------|----------|---------|----------|
| 32 threads | 44.13 | 5.73 | 47.22 | 6.14 |
|----------------------|----------|----------|---------|----------|
| 40 threads | 46.47 | 3.42 | 50.18 | 3.69 |
|----------------------|----------|----------|---------|----------|
| 48 threads | 48.03 | 1.83 | 51.39 | 1.96 |
|----------------------|----------|----------|---------|----------|
| 56 threads | 49.04 | 0.86 | 52.47 | 0.93 |
|----------------------|----------|----------|---------|----------|
| 64 threads | 49.87 | 0.00 | 53.36 | 0.00 |
|----------------------|----------|----------|---------|----------|
This commit, rather than considering delta_idle_purr for calculation of
idle ticks, takes (delta_tb - delta_purr + delta_idle_purr) as total
ticks for which the CPUs were idle. Here, since delta_idle_purr will
also contain some idle ticks, thats why it is added to the formula.
Since, the output was correct for dedicated capped mode, changes has
been made only for shared and dedicated-donate mode.
Further, no changes has been made for calculation of %busy.
Similar changes has been done for SPURR.
After Change:
|-----------------------------------------------------------------|
| Dedicated-donate (8 cores) : |
|----------------------|---------------------|--------------------|
| | Actual | Normalized |
| Stress-ng threads |---------------------|--------------------|
| | %busy | %idle | %busy | %idle |
|----------------------|----------|----------|---------|----------|
| 0 threads | 0.02 | 99.98 | 0.02 | 100.04 |
|----------------------|----------|----------|---------|----------|
| 8 threads | 35.97 | 64.03 | 38.84 | 61.51 |
|----------------------|----------|----------|---------|----------|
| 16 threads | 58.60 | 41.40 | 63.28 | 37.08 |
|----------------------|----------|----------|---------|----------|
| 24 threads | 78.14 | 21.86 | 84.39 | 23.61 |
|----------------------|----------|----------|---------|----------|
| 32 threads | 83.60 | 16.41 | 90.29 | 17.71 |
|----------------------|----------|----------|---------|----------|
| 40 threads | 92.96 | 7.04 | 100.39 | 7.61 |
|----------------------|----------|----------|---------|----------|
| 48 threads | 96.08 | 3.92 | 103.77 | 4.24 |
|----------------------|----------|----------|---------|----------|
| 56 threads | 98.42 | 1.58 | 105.31 | 1.68 |
|----------------------|----------|----------|---------|----------|
| 64 threads | 100.00 | 0.00 | 107.00 | 0.00 |
|----------------------|----------|----------|---------|----------|
|-----------------------------------------------------------------|
| Shared Capped (8 VP / 4 EC) : |
|----------------------|---------------------|--------------------|
| | Actual | Normalized |
| Stress-ng threads |---------------------|--------------------|
| | %busy | %idle | %busy | %idle |
|----------------------|----------|----------|---------|----------|
| 0 threads | 0.03 | 99.97 | 0.19 | 99.44 |
|----------------------|----------|----------|---------|----------|
| 8 threads | 35.91 | 64.09 | 38.78 | 61.58 |
|----------------------|----------|----------|---------|----------|
| 16 threads | 36.83 | 63.17 | 39.78 | 60.55 |
|----------------------|----------|----------|---------|----------|
| 24 threads | 40.16 | 59.84 | 43.37 | 56.95 |
|----------------------|----------|----------|---------|----------|
| 32 threads | 44.47 | 55.53 | 48.02 | 52.38 |
|----------------------|----------|----------|---------|----------|
| 40 threads | 46.55 | 53.45 | 50.27 | 50.04 |
|----------------------|----------|----------|---------|----------|
| 48 threads | 48.13 | 51.87 | 52.48 | 47.82 |
|----------------------|----------|----------|---------|----------|
| 56 threads | 49.01 | 50.99 | 52.93 | 47.41 |
|----------------------|----------|----------|---------|----------|
| 64 threads | 49.90 | 50.10 | 53.40 | 46.19 |
|----------------------|----------|----------|---------|----------|
Before Change:
%idle = delta_idle_purr / delta_tb * 100
After Change:
%idle = (delta_tb - delta_purr + delta_idle_purr) / delta_tb * 100
Signed-off-by: Saket Kumar Bhaskar <skb99@linux.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
src/lparstat.c | 49 ++++++++++++++++++++++++++++++++++++++++++-------
1 file changed, 42 insertions(+), 7 deletions(-)
diff --git a/src/lparstat.c b/src/lparstat.c
index d2fdb3f..9d9ba1b 100644
--- a/src/lparstat.c
+++ b/src/lparstat.c
@@ -515,11 +515,17 @@ void get_cpu_idle_purr(struct sysentry *unused_se, char *buf)
{
double delta_tb, delta_purr, delta_idle_purr;
double physc, idle;
+ char *descr;
+ char mode[32];
delta_tb = get_scaled_tb();
delta_purr = get_delta_value("purr");
delta_idle_purr = get_delta_value("idle_purr");
+ get_sysdata("shared_processor_mode", &descr, mode);
+ if (!strcmp(mode, "Dedicated"))
+ get_sysdata("DedDonMode", &descr, mode);
+
/*
* Given that these values are read from different
* sources (purr from lparcfg and idle_purr from sysfs),
@@ -528,10 +534,23 @@ void get_cpu_idle_purr(struct sysentry *unused_se, char *buf)
*/
if (delta_idle_purr > delta_purr)
delta_idle_purr = delta_purr;
-
- physc = (delta_purr - delta_idle_purr) / delta_tb;
- idle = (delta_purr / delta_tb) - physc;
- idle *= 100.00;
+ /*
+ * Round down delta_purr to delta_tb if delta_tb - delta_purr
+ * error is under -1%.
+ */
+ if (((delta_tb - delta_purr + delta_idle_purr) / delta_tb * 100) > -1 && ((delta_tb - delta_purr + delta_idle_purr) / delta_tb * 100) < 0)
+ delta_purr = delta_tb;
+
+ if (!strcmp(mode, "Capped")) {
+ /* For dedicated - capped mode */
+ physc = (delta_purr - delta_idle_purr) / delta_tb;
+ idle = (delta_purr / delta_tb) - physc;
+ idle *= 100.00;
+ } else {
+ /* For shared and dedicated - donate mode */
+ idle = (delta_tb - delta_purr + delta_idle_purr) / delta_tb;
+ idle *= 100.00;
+ }
sprintf(buf, "%.2f", idle);
}
@@ -559,14 +578,30 @@ void get_cpu_idle_spurr(struct sysentry *unused_se, char *buf)
double delta_tb, delta_spurr, delta_idle_spurr;
double physc, idle;
double rfreq;
+ char *descr;
+ char mode[32];
delta_tb = get_scaled_tb();
delta_spurr = get_delta_value("spurr");
delta_idle_spurr = get_delta_value("idle_spurr");
- physc = (delta_spurr - delta_idle_spurr) / delta_tb;
- idle = (delta_spurr / delta_tb) - physc;
- idle *= 100.00;
+ get_sysdata("shared_processor_mode", &descr, mode);
+ if (!strcmp(mode, "Dedicated"))
+ get_sysdata("DedDonMode", &descr, mode);
+
+ if (delta_spurr > delta_tb)
+ delta_spurr = delta_tb;
+
+ if (!strcmp(mode, "Capped")) {
+ /* For dedicated - capped mode */
+ physc = (delta_spurr - delta_idle_spurr) / delta_tb;
+ idle = (delta_spurr / delta_tb) - physc;
+ idle *= 100.00;
+ } else {
+ /* For shared and dedicated - donate mode */
+ idle = (delta_tb - delta_spurr + delta_idle_spurr) / delta_tb;
+ idle *= 100.00;
+ }
rfreq = round_off_freq();
idle += ((idle * rfreq) / 100);
--
2.45.2

79
0003-drmgr-pci-Add-kernel-interface-support-for-device-tr.patch
View File

@ -1,79 +0,0 @@
From 7038756642711fa53143906b3b3f29900eb4a4ea Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Tue, 13 Aug 2024 14:40:25 -0700
Subject: [PATCH 3/4] drmgr/pci: Add kernel interface support for device tree
update
Use the following kernel interfaces for PCI device type to update
the device tree if this feature is enabled in the kernel.
dt add index <DRC index> --> for IO add
dt remove index <DRC index> --> for IO remove
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
src/drmgr/drslot_chrp_pci.c | 32 ++++++++++++++++++++++----------
1 file changed, 22 insertions(+), 10 deletions(-)
diff --git a/src/drmgr/drslot_chrp_pci.c b/src/drmgr/drslot_chrp_pci.c
index 3b48de30be66..56f8bb397e67 100644
--- a/src/drmgr/drslot_chrp_pci.c
+++ b/src/drmgr/drslot_chrp_pci.c
@@ -366,7 +366,6 @@ static int add_work(struct dr_node *node)
int pow_state; /* Tells us if power was turned on when */
int iso_state; /* Tells us isolation state after */
int rc;
- struct of_node *new_nodes;/* nodes returned from configure_connector */
/* if we're continuing, set LED_ON and see if a card is really there. */
if (process_led(node, LED_ON))
@@ -425,16 +424,26 @@ static int add_work(struct dr_node *node)
* the return status requires a message, print it out
* and exit, otherwise, add the nodes to the OF tree.
*/
- new_nodes = configure_connector(node->drc_index);
- if (new_nodes == NULL) {
- rtas_set_indicator(ISOLATION_STATE, node->drc_index, ISOLATE);
- set_power(node->drc_power, POWER_OFF);
- return -1;
+ if (kernel_dlpar_exists()) {
+ rc = do_dt_kernel_dlpar(node->drc_index, ADD);
+ } else {
+ struct of_node *new_nodes; /* nodes returned from */
+ /* configure_connector */
+
+ new_nodes = configure_connector(node->drc_index);
+ if (new_nodes == NULL) {
+ rtas_set_indicator(ISOLATION_STATE, node->drc_index,
+ ISOLATE);
+ set_power(node->drc_power, POWER_OFF);
+ return -1;
+ }
+
+ say(DEBUG, "Adding %s to %s\n", new_nodes->name,
+ node->ofdt_path);
+ rc = add_device_tree_nodes(node->ofdt_path, new_nodes);
+ free_of_node(new_nodes);
}
- say(DEBUG, "Adding %s to %s\n", new_nodes->name, node->ofdt_path);
- rc = add_device_tree_nodes(node->ofdt_path, new_nodes);
- free_of_node(new_nodes);
if (rc) {
say(DEBUG, "add_device_tree_nodes failed at %s\n",
node->ofdt_path);
@@ -692,7 +701,10 @@ static struct dr_node *remove_work(struct dr_node *all_nodes)
* the device tree.
*/
for (child = node->children; child; child = child->next) {
- rc = remove_device_tree_nodes(child->ofdt_path);
+ if (kernel_dlpar_exists())
+ rc = do_dt_kernel_dlpar(child->drc_index, REMOVE);
+ else
+ rc = remove_device_tree_nodes(child->ofdt_path);
if (rc) {
say(ERROR, "%s", sw_error);
rtas_set_indicator(ISOLATION_STATE, node->drc_index,
--
2.46.0

87
0003-lparstat-app-Use-pic-value-at-boot-for-accurate-boot.patch
View File

@ -1,87 +0,0 @@
From 81c51b59aacaa66d90d571ee19a2deeda1d45271 Mon Sep 17 00:00:00 2001
From: Shrikanth Hegde <sshegde@linux.ibm.com>
Date: Tue, 14 May 2024 20:46:43 +0530
Subject: [PATCH] lparstat: app: Use pic value at boot for accurate boot time
reporting
Upstream: merged, expected in v1.3.13
Git-commit: 81c51b59aacaa66d90d571ee19a2deeda1d45271
When there are no options specified for lparstat, it is expected to
give reports since LPAR(Logical Partition) boot. APP(Available Physical
Processors) is an indicator for available cores in an Shared Processor
LPAR(SPLPAR). APP is derived using pool_idle_time which is obtained
using H_PIC call.
The interval based reports show correct APP value while since boot
report shows very high APP values. This happens because in that case APP
is obtained by dividing pool idle time by LPAR uptime. Since pool idle
time is reported by the PowerVM hypervisor since its boot, it need not
align with LPAR boot.
To fix that use the boot pool idle time added newly in the lparcfg as
below.
APP = (pool idle time - boot pool idle time) / (uptime * timebase)
*This depends on "powerpc/pseries: Add pool idle time at LPAR boot" be
merged into kernel*
Results: (Observe APP values)
========================================================================
lparstat
System Configuration
type=Shared mode=Uncapped smt=8 lcpu=12 mem=15573440 kB cpus=37 ent=12.00
reboot
stress-ng --cpu=$(nproc) -t 600
sleep 600
So in this case app is expected to close to 37-6=31.
====== 6.9-rc1 and lparstat 1.3.10 =============
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
47.48 0.01 0.00 52.51 0.00 0.00 47.49 69099.72 541547 21
=== With this patch and this patch to do the above equation ===
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
47.48 0.01 0.00 52.51 5.73 47.75 47.49 31.21 541753 21
Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
src/lparstat.c | 3 ++-
src/lparstat.h | 2 ++
2 files changed, 4 insertions(+), 1 deletion(-)
diff --git a/src/lparstat.c b/src/lparstat.c
index 9d9ba1b..30889b2 100644
--- a/src/lparstat.c
+++ b/src/lparstat.c
@@ -460,7 +460,8 @@ void get_cpu_app(struct sysentry *unused_se, char *buf)
se = get_sysentry("pool_idle_time");
new_app = strtoll(se->value, NULL, 0);
if (se->old_value[0] == '\0') {
- old_app = 0;
+ se = get_sysentry("boot_pool_idle_time");
+ old_app = strtoll(se->value, NULL, 0);
} else {
old_app = strtoll(se->old_value, NULL, 0);
}
diff --git a/src/lparstat.h b/src/lparstat.h
index b7c88e9..77203e1 100644
--- a/src/lparstat.h
+++ b/src/lparstat.h
@@ -124,6 +124,8 @@ struct sysentry system_data[] = {
.get = &get_percent_entry},
{.name = "pool_idle_time",
.descr = "Shared Processor Pool Idle Time"},
+ {.name = "boot_pool_idle_time",
+ .descr = "Shared Processor Pool Idle Time"},
{.name = "pool_num_procs",
.descr = "Shared Processor Pool Processors"},
{.name = "unallocated_capacity_weight",
--
2.45.2

142
0004-drmgr-phb-Add-kernel-interface-support-for-device-tr.patch
View File

@ -1,142 +0,0 @@
From b8ab373035984dd8aba29f1a0c03dcef6546f004 Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Tue, 13 Aug 2024 14:40:24 -0700
Subject: [PATCH 4/4] drmgr/phb: Add kernel interface support for device tree
update
Use the following kernel interfaces for PHB device type to update
the device tree if this feature is enabled in the kernel.
dt add index <DRC index> --> for IO add
dt remove index <DRC index> --> for IO remove
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
src/drmgr/common_pci.c | 29 +++++++++++++++++++++-------
src/drmgr/drslot_chrp_phb.c | 38 ++++++++++++++++++++-----------------
2 files changed, 43 insertions(+), 24 deletions(-)
diff --git a/src/drmgr/common_pci.c b/src/drmgr/common_pci.c
index 374129cf33e0..540864574569 100644
--- a/src/drmgr/common_pci.c
+++ b/src/drmgr/common_pci.c
@@ -1390,7 +1390,6 @@ print_node_list(struct dr_node *first_node)
static int
acquire_hp_resource(struct dr_connector *drc, char *of_path)
{
- struct of_node *new_nodes;
int state;
int rc;
@@ -1429,12 +1428,21 @@ acquire_hp_resource(struct dr_connector *drc, char *of_path)
}
if (state == PRESENT) {
- new_nodes = configure_connector(drc->index);
- if (new_nodes == NULL)
- return -1;
+ /*
+ * Use kernel DLPAR interface if it is enabled
+ */
+ if (kernel_dlpar_exists()) {
+ rc = do_dt_kernel_dlpar(drc->index, ADD);
+ } else {
+ struct of_node *new_nodes;
+
+ new_nodes = configure_connector(drc->index);
+ if (new_nodes == NULL)
+ return -1;
- rc = add_device_tree_nodes(of_path, new_nodes);
- free_of_node(new_nodes);
+ rc = add_device_tree_nodes(of_path, new_nodes);
+ free_of_node(new_nodes);
+ }
if (rc) {
say(ERROR, "add nodes failed for 0x%x\n", drc->index);
return rc;
@@ -1490,7 +1498,14 @@ release_hp_resource(struct dr_node *node)
{
int rc;
- rc = remove_device_tree_nodes(node->ofdt_path);
+ /*
+ * Use kernel DLPAR interface if it is enabled
+ */
+ if (kernel_dlpar_exists())
+ rc = do_dt_kernel_dlpar(node->drc_index, REMOVE);
+ else
+ rc = remove_device_tree_nodes(node->ofdt_path);
+
if (rc) {
say(ERROR, "failed to remove kernel nodes for index 0x%x\n",
node->drc_index);
diff --git a/src/drmgr/drslot_chrp_phb.c b/src/drmgr/drslot_chrp_phb.c
index f59baa4f9e27..ffa17d8f6b7d 100644
--- a/src/drmgr/drslot_chrp_phb.c
+++ b/src/drmgr/drslot_chrp_phb.c
@@ -108,17 +108,16 @@ release_phb(struct dr_node *phb)
{
int rc;
- rc = remove_device_tree_nodes(phb->ofdt_path);
- if (rc)
- return rc;
-
- if (phb->phb_ic_ofdt_path[0] != '\0') {
- rc = remove_device_tree_nodes(phb->phb_ic_ofdt_path);
- if (rc)
- return rc;
+ if (kernel_dlpar_exists())
+ rc = do_dt_kernel_dlpar(phb->drc_index, REMOVE);
+ else {
+ rc = remove_device_tree_nodes(phb->ofdt_path);
+ if (!rc && (phb->phb_ic_ofdt_path[0] != '\0'))
+ rc = remove_device_tree_nodes(phb->phb_ic_ofdt_path);
}
- rc = release_drc(phb->drc_index, PHB_DEV);
+ if (!rc)
+ rc = release_drc(phb->drc_index, PHB_DEV);
return rc;
}
@@ -371,7 +370,6 @@ phb_remove_error:
static int acquire_phb(char *drc_name, struct dr_node **phb)
{
struct dr_connector drc;
- struct of_node *of_nodes;
char path[DR_PATH_MAX];
int rc;
@@ -386,14 +384,20 @@ static int acquire_phb(char *drc_name, struct dr_node **phb)
if (rc)
return rc;
- of_nodes = configure_connector(drc.index);
- if (of_nodes == NULL) {
- release_drc(drc.index, PHB_DEV);
- return -1;
- }
+ if (kernel_dlpar_exists()) {
+ rc = do_dt_kernel_dlpar(drc.index, ADD);
+ } else {
+ struct of_node *of_nodes;
- rc = add_device_tree_nodes(path, of_nodes);
- free_of_node(of_nodes);
+ of_nodes = configure_connector(drc.index);
+ if (of_nodes == NULL) {
+ release_drc(drc.index, PHB_DEV);
+ return -1;
+ }
+
+ rc = add_device_tree_nodes(path, of_nodes);
+ free_of_node(of_nodes);
+ }
if (rc) {
say(ERROR, "add_device_tree_nodes failed at %s\n", path);
release_drc(drc.index, PHB_DEV);
--
2.46.0

238
0004-lparstat-Use-CLOCK_BOOTTIME-for-get_time-interface-a.patch
View File

@ -1,238 +0,0 @@
From 9572f8c2022fed9783f0f606cbe778ffe0d93fef Mon Sep 17 00:00:00 2001
From: Shrikanth Hegde <sshegde@linux.ibm.com>
Date: Tue, 14 May 2024 20:46:44 +0530
Subject: [PATCH] lparstat: Use CLOCK_BOOTTIME for get_time interface and
Deprecate get_sys_upttime
Upstream: merged, expected in v1.3.13
Git-commit: 9572f8c2022fed9783f0f606cbe778ffe0d93fef
"time" is used in lparstat.c to find the time elapsed either since boot
or between two intervals. But it is using gettimeofday which returns the
time elapsed since Epoch. This works for intervals calculations but it
doesn't work for since boot reports.
Instead use the CLOCK_BOOTTIME interface to get the elapsed time. This
fixes physc, utilization based on purr being wrong since boot.
Remove "uptime" interface since there are no users of it. One can get
the system uptime by calling "time" itself.
=============================== ::Test:: ==========================
reboot
stress-ng --cpu=$(nproc) -t 600
sleep 600
Results::
==================== Shared LPAR ==================================
System Configuration
type=Shared mode=Uncapped smt=8 lcpu=12 mem=15573440 kB cpus=37 ent=12.00
lparstat -E <-- Observe utilization values
====== 6.9-rc1 and lparstat 1.3.10 =============
---Actual--- -Normalized-
%busy %idle Frequency %busy %idle
------ ------ ------------- ------ ------
0.00 0.00 3.87GHz[106%] 0.00 0.00
==== With this patch and patch 2/3 =============
---Actual--- -Normalized-
%busy %idle Frequency %busy %idle
------ ------ ------------- ------ ------
38.72 0.11 3.87GHz[106%] 41.04 0.12
lparstat <-- Observe physc values
====== 6.9-rc1 and lparstat 1.3.10 ===================================
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
47.48 0.01 0.00 52.51 0.00 0.00 47.49 69099.72 541547 21
=== With this patch and this patch ================================ ===
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
47.48 0.01 0.00 52.51 5.73 47.75 47.49 31.21 541753 21
==================== Dedicated LPAR ==================================
System Configuration
type=Dedicated mode=Capped smt=8 lcpu=12 mem=15573248 kB cpus=0 ent=12.00
::lparstat -E:: <-- Observe utilization values.
======= 6.9-rc1 and lparstat 1.3.10 =============
---Actual--- -Normalized-
%busy %idle Frequency %busy %idle
------ ------ ------------- ------ ------
0.00 0.00 3.87GHz[106%] 0.00 0.00
=== With this patch and powerpc-utils patch to do the above equation ===
---Actual--- -Normalized-
%busy %idle Frequency %busy %idle
------ ------ ------------- ------ ------
48.87 51.51 3.87GHz[106%] 51.81 54.60
::lparstat:: <-- Observe physc values.
======= 6.9-rc1 and lparstat 1.3.10 =============
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
48.38 0.01 0.00 51.61 0.03 0.25 48.39 0.00 344661 8
=== With this patch and powerpc-utils patch to do the above equation ===
%user %sys %wait %idle physc %entc lbusy app vcsw phint
----- ----- ----- ----- ----- ----- ----- ----- ----- -----
48.38 0.01 0.00 51.61 12.05 100.42 48.39 0.00 344877 8
=============================================================================
Interval based lparstat values are same. With this patch the physc and
busy purr/idle purr values show correctly for since boot reports.
Note: this patch doesn't fix the idle purr being incorrect. That is
currently being investigated.
Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
src/lparstat.c | 57 +++++++++-----------------------------------------
src/lparstat.h | 6 ------
2 files changed, 10 insertions(+), 53 deletions(-)
diff --git a/src/lparstat.c b/src/lparstat.c
index 30889b2..3e9169d 100644
--- a/src/lparstat.c
+++ b/src/lparstat.c
@@ -36,6 +36,7 @@
#include "lparstat.h"
#include "pseries_platform.h"
#include "cpu_info_helpers.h"
+#include <time.h>
#define LPARCFG_FILE "/proc/ppc64/lparcfg"
#define SE_NOT_FOUND "???"
@@ -255,14 +256,17 @@ long long get_delta_value(char *se_name)
void get_time()
{
- struct timeval t;
struct sysentry *se;
+ struct timespec ts;
+ int err;
- gettimeofday(&t, 0);
+ err = clock_gettime(CLOCK_BOOTTIME, &ts);
+ if (err)
+ return;
se = get_sysentry("time");
sprintf(se->value, "%lld",
- (long long)t.tv_sec * 1000000LL + (long long)t.tv_usec);
+ (long long)ts.tv_sec);
}
int get_time_base()
@@ -304,7 +308,6 @@ double get_scaled_tb(void)
online_cores = atoi(se->value);
elapsed = get_delta_value("time");
- elapsed = elapsed / 1000000.0;
se = get_sysentry("timebase");
timebase = atoi(se->value);
@@ -312,31 +315,6 @@ double get_scaled_tb(void)
return (timebase * elapsed) * online_cores;
}
-void get_sys_uptime(struct sysentry *unused_se, char *uptime)
-{
- FILE *f;
- char buf[80];
-
- f = fopen("/proc/uptime", "r");
- if (!f) {
- fprintf(stderr, "Could not open /proc/uptime\n");
- sprintf(uptime, SE_NOT_VALID);
- return;
- }
-
- if ((fgets(buf, 80, f)) != NULL) {
- char *value;
-
- value = strchr(buf, ' ');
- *value = '\0';
- sprintf(uptime, "%s", buf);
- } else {
- sprintf(uptime, SE_NOT_VALID);
- }
-
- fclose(f);
-}
-
int get_nominal_frequency(void)
{
FILE *f;
@@ -403,13 +381,12 @@ void get_cpu_physc(struct sysentry *unused_se, char *buf)
delta_purr = get_delta_value("purr");
se = get_sysentry("tbr");
- if (se->value[0] != '\0') {
+ if (se->old_value[0] != '\0') {
delta_tb = get_delta_value("tbr");
physc = delta_purr / delta_tb;
} else {
elapsed = get_delta_value("time");
- elapsed = elapsed / 1000000.0;
se = get_sysentry("timebase");
timebase = atoi(se->value);
@@ -436,23 +413,9 @@ void get_cpu_app(struct sysentry *unused_se, char *buf)
{
struct sysentry *se;
float timebase, app, elapsed_time;
- long long new_app, old_app, delta_time;
- char *descr, uptime[32];
+ long long new_app, old_app;
- se = get_sysentry("time");
- if (se->old_value[0] == '\0') {
- /* Single report since boot */
- get_sysdata("uptime", &descr, uptime);
-
- if (!strcmp(uptime, SE_NOT_VALID)) {
- sprintf(buf, "-");
- return;
- }
- elapsed_time = atof(uptime);
- } else {
- delta_time = get_delta_value("time");
- elapsed_time = delta_time / 1000000.0;
- }
+ elapsed_time = get_delta_value("time");
se = get_sysentry("timebase");
timebase = atof(se->value);
diff --git a/src/lparstat.h b/src/lparstat.h
index 77203e1..86e45e4 100644
--- a/src/lparstat.h
+++ b/src/lparstat.h
@@ -60,7 +60,6 @@ extern void get_cpu_stat(struct sysentry *, char *);
extern void get_cpu_physc(struct sysentry *, char *);
extern void get_per_entc(struct sysentry *, char *);
extern void get_cpu_app(struct sysentry *, char *);
-extern void get_sys_uptime(struct sysentry *, char *);
extern void get_cpu_util_purr(struct sysentry *unused_se, char *buf);
extern void get_cpu_idle_purr(struct sysentry *unused_se, char *buf);
extern void get_cpu_util_spurr(struct sysentry *unused_se, char *buf);
@@ -272,11 +271,6 @@ struct sysentry system_data[] = {
{.name = "phint",
.descr = "Phantom Interrupts"},
- /* /proc/uptime */
- {.name = "uptime",
- .descr = "System Uptime",
- .get = &get_sys_uptime},
-
/* /sys/devices/system/cpu/cpu<n>/ */
/* Sum of per CPU SPURR registers */
{.name = "spurr",
--
2.45.2

63
0005-ofpathname-skip-devices-with-no-devspec-when-coorela.patch
View File

@ -1,63 +0,0 @@
From a56028561ad9c01b5d3a767e50455c561e09191c Mon Sep 17 00:00:00 2001
From: Tyrel Datwyler <tyreld@linux.ibm.com>
Date: Fri, 26 Jul 2024 16:17:47 -0700
Subject: [PATCH] ofpathname: skip devices with no devspec when coorelating FC
OF paths
Upstream: merged, expected in v1.3.13
Git-commit: a56028561ad9c01b5d3a767e50455c561e09191c
When using ofpathname to find the logical device associated to a Fibre
Channel or Virtual Fibre Channel OF path an ERR_NO_SYSFS_DEVINFO error
can be returned if a device with no devspec attribute is in the list of
devices to cross reference.
If a device with no devspec is found simply skip it an continue the
for-loop device search.
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
scripts/ofpathname | 16 ++++++++++++----
1 file changed, 12 insertions(+), 4 deletions(-)
diff --git a/scripts/ofpathname b/scripts/ofpathname
index 833d03f..711ab62 100755
--- a/scripts/ofpathname
+++ b/scripts/ofpathname
@@ -1649,10 +1649,14 @@ of2l_vfc()
link=$PWD
local device_dir=${PWD##/*/}
- goto_dir $PWD "devspec"
+ goto_dir $PWD "devspec" 0
+ if [[ $? -eq 1 ]]; then
+ continue;
+ fi
OF_PATH=`$CAT $PWD/devspec`
+
if [[ -z $OF_PATH ]]; then
- err $ERR_NO_LOGDEV
+ err $ERR_NO_LOGDEV
fi
# Skip if this is not the correct FC port
@@ -1727,10 +1731,14 @@ of2l_fc()
cd $link
local device_dir=${PWD##/*/}
- goto_dir $PWD "devspec"
+ goto_dir $PWD "devspec" 0
+ if [[ $? -eq 1 ]]; then
+ continue;
+ fi
OF_PATH=`$CAT devspec`
+
if [[ -z $OF_PATH ]]; then
- err $ERR_NO_LOGDEV
+ err $ERR_NO_LOGDEV;
fi
# Skip if this is not the correct FC port
--
2.45.2

46
0006-drmgr-Free-nodes-returned-from-configure_connector.patch
View File

@ -1,46 +0,0 @@
From 5db2df531f9c242b13ef6520814c99685144c6d4 Mon Sep 17 00:00:00 2001
From: Haren Myneni <haren@linux.ibm.com>
Date: Sat, 29 Jun 2024 14:14:10 -0700
Subject: [PATCH] drmgr: Free nodes returned from configure_connector
Upstream: merged, expected in v1.3.13
Git-commit: 5db2df531f9c242b13ef6520814c99685144c6d4
of_nodes returned from configure_connector should be freed after
updating the device tree and is missing in acquire_hp_resource()
and add_work(). This patch calls free_of_node() in these functions.
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
---
src/drmgr/common_pci.c | 1 +
src/drmgr/drslot_chrp_pci.c | 1 +
2 files changed, 2 insertions(+)
diff --git a/src/drmgr/common_pci.c b/src/drmgr/common_pci.c
index 2411641..759589a 100644
--- a/src/drmgr/common_pci.c
+++ b/src/drmgr/common_pci.c
@@ -1434,6 +1434,7 @@ acquire_hp_resource(struct dr_connector *drc, char *of_path)
return -1;
rc = add_device_tree_nodes(of_path, new_nodes);
+ free_of_node(new_nodes);
if (rc) {
say(ERROR, "add nodes failed for 0x%x\n", drc->index);
return rc;
diff --git a/src/drmgr/drslot_chrp_pci.c b/src/drmgr/drslot_chrp_pci.c
index ac078db..ec3c77c 100644
--- a/src/drmgr/drslot_chrp_pci.c
+++ b/src/drmgr/drslot_chrp_pci.c
@@ -454,6 +454,7 @@ static int add_work(struct dr_node *node, bool partner_device)
say(DEBUG, "Adding %s to %s\n", new_nodes->name, node->ofdt_path);
rc = add_device_tree_nodes(node->ofdt_path, new_nodes);
+ free_of_node(new_nodes);
if (rc) {
say(DEBUG, "add_device_tree_nodes failed at %s\n",
node->ofdt_path);
--
2.45.2

BIN
powerpc-utils-1.3.12.tar.gz (Stored with Git LFS)
View File

Binary file not shown.

BIN
powerpc-utils-1.3.13.tar.gz (Stored with Git LFS) Normal file
View File

Binary file not shown.

19
powerpc-utils.changes
View File

@ -1,3 +1,22 @@
-------------------------------------------------------------------
Thu Nov 21 08:10:24 UTC 2024 - Michal Suchanek <msuchanek@suse.de>
- Update to version 1.3.13 (jsc#PED-9917)
* multipath - drmgr support (jsc#PED-9914)
- Remove upstreamed patches
* ppc64_cpu-Clean-up-sysfs-smt-control-error-handling.patch
* ppc64_cpu-Support-partial-SMT-level-through-SYS-FS-s.patch
* 0005-ofpathname-skip-devices-with-no-devspec-when-coorela.patch
* 0002-lparstat-Fix-Idle-and-busy-PURR-SPURR.patch
* 0003-lparstat-app-Use-pic-value-at-boot-for-accurate-boot.patch
* 0004-lparstat-Use-CLOCK_BOOTTIME-for-get_time-interface-a.patch
* 0001-drmgr-Return-from-get_node_by_name-if-matched-DRC-in.patch
* 0006-drmgr-Free-nodes-returned-from-configure_connector.patch
* 0001-drmgr-pci-Enable-in-kernel-functionality-to-update-d.patch
* 0002-drmgr-SLOT-Add-kernel-interface-support-for-device-t.patch
* 0004-drmgr-phb-Add-kernel-interface-support-for-device-tr.patch
* 0003-drmgr-pci-Add-kernel-interface-support-for-device-tr.patch
-------------------------------------------------------------------
Fri Sep 20 06:15:46 UTC 2024 - Michal Suchanek <msuchanek@suse.de>

14
powerpc-utils.spec
View File

@ -17,7 +17,7 @@
Name: powerpc-utils
Version: 1.3.12
Version: 1.3.13
Release: 0
Summary: Utilities for PowerPC Hardware
License: GPL-2.0-or-later
@ -29,18 +29,6 @@ Patch1: powerpc-utils-lsprop.patch
Patch2: ofpathname_powernv.patch
Patch3: fix_kexec_service_name_for_suse.patch
Patch4: libvirt-service-dep.patch
Patch5: 0001-drmgr-Return-from-get_node_by_name-if-matched-DRC-in.patch
Patch6: 0006-drmgr-Free-nodes-returned-from-configure_connector.patch
Patch7: 0005-ofpathname-skip-devices-with-no-devspec-when-coorela.patch
Patch8: ppc64_cpu-Support-partial-SMT-level-through-SYS-FS-s.patch
Patch9: ppc64_cpu-Clean-up-sysfs-smt-control-error-handling.patch
Patch10: 0002-lparstat-Fix-Idle-and-busy-PURR-SPURR.patch
Patch11: 0003-lparstat-app-Use-pic-value-at-boot-for-accurate-boot.patch
Patch12: 0004-lparstat-Use-CLOCK_BOOTTIME-for-get_time-interface-a.patch
Patch13: 0001-drmgr-pci-Enable-in-kernel-functionality-to-update-d.patch
Patch14: 0002-drmgr-SLOT-Add-kernel-interface-support-for-device-t.patch
Patch15: 0003-drmgr-pci-Add-kernel-interface-support-for-device-tr.patch
Patch16: 0004-drmgr-phb-Add-kernel-interface-support-for-device-tr.patch
BuildRequires: autoconf
BuildRequires: automake
BuildRequires: libnuma-devel

65
ppc64_cpu-Clean-up-sysfs-smt-control-error-handling.patch
View File

@ -1,65 +0,0 @@
From 8d613e0e81561ce0b1d6ea834b07c73f5f9251a1 Mon Sep 17 00:00:00 2001
From: Michal Suchanek <msuchanek@suse.de>
Date: Fri, 9 Feb 2024 13:12:33 +0100
Subject: [PATCH] ppc64_cpu: Clean up sysfs smt/control error handling
When the kernel does not support the sysfs intercface do not report an
arror, fall back to the old method silently.
Suggested-by: Nathan Lynch<nathanl@linux.ibm.com>
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
---
v3: retry is needed on ENODEV to support powernv
---
src/ppc64_cpu.c | 24 +++++++++++++++++++-----
1 file changed, 19 insertions(+), 5 deletions(-)
diff --git a/src/ppc64_cpu.c b/src/ppc64_cpu.c
index c318928..688152b 100644
--- a/src/ppc64_cpu.c
+++ b/src/ppc64_cpu.c
@@ -364,14 +364,28 @@ static int is_dscr_capable(void)
/*
* Depends on kernel's CONFIG_HOTPLUG_CPU
+ * Return -1 for fatal error, -2 to retry.
*/
static int set_smt_control(int smt_state)
{
if (set_attribute(SYS_SMT_CONTROL, "%d", smt_state)) {
- /* Silently ignore kernel not supporting this feature */
- if (errno != ENODEV)
- perror(SYS_SMT_CONTROL);
- return -1;
+ switch (errno) {
+ case ENOENT:
+ /*
+ * The kernel does not have the interface.
+ * Try the old method.
+ */
+ return -2;
+ case ENODEV:
+ /*
+ * Setting SMT state not supported by this interface.
+ * eg. powernv
+ */
+ return -2;
+ default:
+ perror(SYS_SMT_CONTROL);
+ return -1;
+ }
}
return 0;
}
@@ -405,7 +419,7 @@ static int do_smt(char *state, bool numeric)
}
/* Try using smt/control if failing, fall back to the legacy way */
- if (set_smt_control(smt_state))
+ if ((rc = set_smt_control(smt_state)) == -2)
rc = set_smt_state(smt_state);
}
--
2.44.0

71
ppc64_cpu-Support-partial-SMT-level-through-SYS-FS-s.patch
View File

@ -1,71 +0,0 @@
From 46c524be975a108d2b8d1cadb95003b9c2670c8e Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.ibm.com>
Date: Thu, 29 Jun 2023 16:41:37 +0200
Subject: [PATCH] ppc64_cpu: Support partial SMT level through SYS FS
smt/control files
The next kernel release will support partial SMT level [1] though the SYS
FS file "devices/system/cpu/smt/control". This allows the SMT level to be
recorded in the kernel. With the current SMT level stored in the kernel,
when a new CPU is added, only the necessary threads are brought online.
The legacy way to active threads through the SYS FS files
'devices/system/cpu/cpu<n>/online', is still used in the case the new SYS
FS API is not available. This allows compatibility with the previous kernel
versions.
[1] https://lore.kernel.org/linuxppc-dev/20230705145143.40545-1-ldufour@linux.ibm.com/
Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com>
---
src/ppc64_cpu.c | 20 +++++++++++++++++++-
1 file changed, 19 insertions(+), 1 deletion(-)
diff --git a/src/ppc64_cpu.c b/src/ppc64_cpu.c
index 5fdf86a..c33a293 100644
--- a/src/ppc64_cpu.c
+++ b/src/ppc64_cpu.c
@@ -56,6 +56,8 @@
#define DIAGNOSTICS_RUN_MODE 42
#define CPU_OFFLINE -1
+#define SYS_SMT_CONTROL "/sys/devices/system/cpu/smt/control"
+
#ifdef HAVE_LINUX_PERF_EVENT_H
struct cpu_freq {
int offline;
@@ -360,6 +362,20 @@ static int is_dscr_capable(void)
return 0;
}
+/*
+ * Depends on kernel's CONFIG_HOTPLUG_CPU
+ */
+static int set_smt_control(int smt_state)
+{
+ if (set_attribute(SYS_SMT_CONTROL, "%d", smt_state)) {
+ /* Silently ignore kernel not supporting this feature */
+ if (errno != ENODEV)
+ perror(SYS_SMT_CONTROL);
+ return -1;
+ }
+ return 0;
+}
+
static int do_smt(char *state, bool numeric)
{
int rc = 0;
@@ -388,7 +404,9 @@ static int do_smt(char *state, bool numeric)
return -1;
}
- rc = set_smt_state(smt_state);
+ /* Try using smt/control if failing, fall back to the legacy way */
+ if (set_smt_control(smt_state))
+ rc = set_smt_state(smt_state);
}
return rc;
--
2.41.0