powerpc/pseries: Simplify cpu readd to use drc_index
PATCH 1/3 powerpc/pseries: Simplify cpu readd to use drc_index
The current dlpar_cpu_readd() takes in a cpu_id and uses that to look up the cpus device_node so that we can get at the ibm,my-drc-index property. The only user of cpu readd is an OF notifier call back. This call back already has a reference to the device_node and therefore can retrieve the drc_index from the device_node.
This patch simplifies dlpar_cpu_readd() to take a drc_index directly and does away with an uneccsary device_node lookup.
--- arch/powerpc/include/asm/topology.h | 2 +- arch/powerpc/mm/numa.c | 6 +++--- arch/powerpc/platforms/pseries/hotplug-cpu.c | 10 +--------- 3 files changed, 5 insertions(+), 13 deletions(-)
arch/powerpc/include/asm/topology.h +1 -1
@@ -133,7 +133,7 @@ static inline void shared_proc_topology_init(void) {}#define topology_core_cpumask (cpu) (per_cpu(cpu_core_map, cpu))#define topology_core_id (cpu) (cpu_to_core_id(cpu)) -int dlpar_cpu_readd (int cpu); +int dlpar_cpu_readd (u32 drc_index);#endif #endif
arch/powerpc/mm/numa.c +3 -3
@@ -1479,9 +1479,9 @@ static int dt_update_callback(struct notifier_block *nb,case OF_RECONFIG_UPDATE_PROPERTY :if (of_node_is_type (update->dn,"cpu" ) && !of_prop_cmp (update->prop->name,"ibm,associativity" )) { -u32 core_id; -of_property_read_u32 (update->dn,"reg" , &core_id); - rc =dlpar_cpu_readd (core_id); +u32 drc_index; +of_property_read_u32 (update->dn,"ibm,my-drc-index" , &drc_index); + rc =dlpar_cpu_readd (drc_index); rc =NOTIFY_OK ; }break ;
arch/powerpc/platforms/pseries/hotplug-cpu.c +1 -9
@@ -802,18 +802,10 @@ static int dlpar_cpu_add_by_count(u32 cpus_to_add)return rc; } -int dlpar_cpu_readd (int cpu) +int dlpar_cpu_readd (u32 drc_index) { -struct device_node *dn; -struct device *dev; -u32 drc_index;int rc; - dev =get_cpu_device (cpu); - dn = dev->of_node; - - rc =of_property_read_u32 (dn,"ibm,my-drc-index" , &drc_index); - rc =dlpar_cpu_remove_by_index (drc_index);if (!rc) rc =dlpar_cpu_add (drc_index);
Signature · 1 line
2.18.1
PATCH 2/3 powerpc/pseries: Disable PRRN memory device tree trigger
Memory affintiy updates as currently implemented have proved unstable.
This patch comments out the PRRN hook for the time being while we investigate how to either stablize the current implementation or find a better approach.
--- arch/powerpc/platforms/pseries/mobility.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/powerpc/platforms/pseries/mobility.c +5 -2
@@ -242,13 +242,15 @@ static int add_dt_node(__be32 parent_phandle, __be32 drc_index)static void prrn_update_node (__be32 phandle) { +/* PRRN Memory Updates have proved unstable. Disable for the time being. +* struct pseries_hp_errorlog hp_elog; struct device_node *dn; -/* +* * If a node is found from a the given phandle, the phandle does not * represent the drc index of an LMB and we can ignore. -*/ +* dn = of_find_node_by_phandle(be32_to_cpu(phandle)); if (dn) { of_node_put(dn);
@@ -261,6 +263,7 @@ static void prrn_update_node(__be32 phandle)hp_elog._drc_u.drc_index = phandle; handle_dlpar_errorlog(&hp_elog); +*/ }int pseries_devicetree_update (s32 scope)
Signature · 1 line
2.18.1
PATCH 3/3 powerpc/pseries: Don't update cpu topology after PRRN event
In reply to Tyrel DatwylerWhen we receive a PRRN event through the event-scan interface we call pseries_devicetree_update() to update the affinty properties in our device tree via RTAS. Following this our implemenation attempts to both frob the existing kernel cpu numa affinities of the live system with the new device tree properties while also performing a full cpu hotplug readd of the affected cpus in resposne to the a OF property notifier triggerd by the device tree update.
This patch does away with the topology update call to frob the associativity since the DLPAR readd will put the cpu in the proper numa node when its added back.
--- arch/powerpc/kernel/rtasd.c | 1 - 1 file changed, 1 deletion(-)
arch/powerpc/kernel/rtasd.c +0 -1
@@ -285,7 +285,6 @@ static void handle_prrn_event(s32 scope)* the RTAS event. */ pseries_devicetree_update (-scope); -numa_update_cpu_topology (false); }static void handle_rtas_event (const struct rtas_error_log *log)
Signature · 1 line
2.18.1
Quoted 5 lines · Tyrel Datwyler
The current
dlpar_cpu_readd()takes in acpu_idand uses that to look up the cpusdevice_nodeso that we can get at the ibm,my-drc-index property. The only user of cpu readd is an OF notifier call back. This call back already has a reference to thedevice_nodeand therefore can retrieve thedrc_indexfrom thedevice_node.
dlpar_cpu_readd is a hack to try to change the CPU-node relationship at runtime without destabilizing the system. It doesn't accomplish that and it should just be removed (and I'm working on that).
Quoted 11 lines · Nathan Lynch
Quoted 5 lines · Tyrel Datwyler
The current
dlpar_cpu_readd()takes in acpu_idand uses that to look up the cpusdevice_nodeso that we can get at the ibm,my-drc-index property. The only user of cpu readd is an OF notifier call back. This call back already has a reference to thedevice_nodeand therefore can retrieve thedrc_indexfrom thedevice_node.
dlpar_cpu_readdis a hack to try to change the CPU-node relationship at runtime without destabilizing the system. It doesn't accomplish that and it should just be removed (and I'm working on that).
I will politely disagree. We've done exactly this from userspace for years. My experience still suggests that memory affinity is the problem area, and that the work to push this all into the kernel originally was poorly tested.
-Tyrel
Quoted 13 lines · Tyrel Datwyler
Quoted 11 lines · Nathan Lynch
Quoted 5 lines · Tyrel Datwyler
The current
dlpar_cpu_readd()takes in acpu_idand uses that to look up the cpusdevice_nodeso that we can get at the ibm,my-drc-index property. The only user of cpu readd is an OF notifier call back. This call back already has a reference to thedevice_nodeand therefore can retrieve thedrc_indexfrom thedevice_node.
dlpar_cpu_readdis a hack to try to change the CPU-node relationship at runtime without destabilizing the system. It doesn't accomplish that and it should just be removed (and I'm working on that).
Quoted 4 lines
I will politely disagree. We've done exactly this from userspace for years. My experience still suggests that memory affinity is the problem area, and that the work to push this all into the kernel originally was poorly tested.
Kernel implementation details aside, how do you change the cpu-node relationship at runtime without breaking NUMA-aware applications? Is this not a fundamental issue to address before adding code like this?
Quoted 15 lines · Nathan Lynch
Quoted 13 lines · Tyrel Datwyler
Quoted 11 lines · Nathan Lynch
Quoted 5 lines · Tyrel Datwyler
The current
dlpar_cpu_readd()takes in acpu_idand uses that to look up the cpusdevice_nodeso that we can get at the ibm,my-drc-index property. The only user of cpu readd is an OF notifier call back. This call back already has a reference to thedevice_nodeand therefore can retrieve thedrc_indexfrom thedevice_node.
dlpar_cpu_readdis a hack to try to change the CPU-node relationship at runtime without destabilizing the system. It doesn't accomplish that and it should just be removed (and I'm working on that).
Quoted 9 lines
Quoted 4 lines
I will politely disagree. We've done exactly this from userspace for years. My experience still suggests that memory affinity is the problem area, and that the work to push this all into the kernel originally was poorly tested.
Kernel implementation details aside, how do you change the cpu-node relationship at runtime without breaking NUMA-aware applications? Is this not a fundamental issue to address before adding code like this?
If that is the concern then hotplug in general already breaks them. Take for example the removal of a faulty processor and then adding a new processor back. It is quite possible that the new processor is in a different NUMA node. Keep in mind that in this scenario the new processor and threads gets the same logical cpu ids as the faulty processor we just removed.
Now we have to ask the question who is right and who is wrong. In this case the kernel data structures reflect the correct NUMA topology. However, did the NUMA aware application or libnuma make an assumption that specific sets of logical cpu ids are always in the same NUMA node?
-Tyrel
Quoted 6 lines · Tyrel Datwyler
Quoted 4 lines · Nathan Lynch
Kernel implementation details aside, how do you change the cpu-node relationship at runtime without breaking NUMA-aware applications? Is this not a fundamental issue to address before adding code like this?
Quoted 6 lines
If that is the concern then hotplug in general already breaks them. Take for example the removal of a faulty processor and then adding a new processor back. It is quite possible that the new processor is in a different NUMA node. Keep in mind that in this scenario the new processor and threads gets the same logical cpu ids as the faulty processor we just removed.
Yes, the problem is re-use of a logical CPU id with a node id that differs from the one it was initially assigned, and there are several ways to get into that situation on this platform. We probably need to be more careful in how we allocate a spot in the CPU maps for a newly-added processor. I believe the algorithm is simple first-fit right now, and it doesn't take into account prior NUMA relationships.
Quoted 5 lines
Now we have to ask the question who is right and who is wrong. In this case the kernel data structures reflect the correct NUMA topology. However, did the NUMA aware application or libnuma make an assumption that specific sets of logical cpu ids are always in the same NUMA node?
Yes, and that assumption is widespread because people tend to develop on an architecture where this kind of stuff doesn't happen (at least not yet).
And I don't really agree that the current behavior reflects what is actually going on. When Linux running in a PowerVM LPAR receives a notification to change the NUMA properties of a processor at runtime, it's because the platform has changed the physical characteristics of the partition. I.e. you're now using a different physical processor, with different relationships to the other resources in the system. Even if it didn't destabilize the kernel (by changing the result of cpu_to_node() when various subsystems assume it will be static), continuing to use the logical CPU ids on the new processor obscures what has actually happened. And we have developers that have told us that this behavior - changing the logical cpu<->node relationship at runtime
- is something their existing NUMA-aware applications cannot handle.