linux/arch/powerpc/mm/numa.c

1809 lines
44 KiB
C
Raw Normal View History

/*
* pSeries NUMA support
*
* Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/threads.h>
#include <linux/bootmem.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <linux/mmzone.h>
#include <linux/export.h>
#include <linux/nodemask.h>
#include <linux/cpu.h>
#include <linux/notifier.h>
#include <linux/memblock.h>
#include <linux/of.h>
powerpc/mm: Fix numa reserve bootmem page selection Fix the powerpc NUMA reserve bootmem page selection logic. commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb (powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes) changed the logic for how the powerpc LMB reserved regions were converted to bootmen reserved regions. As the folowing discussion reports, the new logic was not correct. mark_reserved_regions_for_nid() goes through each LMB on the system that specifies a reserved area. It searches for active regions that intersect with that LMB and are on the specified node. It attempts to bootmem-reserve only the area where the active region and the reserved LMB intersect. We can not reserve things on other nodes as they may not have bootmem structures allocated, yet. We base the size of the bootmem reservation on two possible things. Normally, we just make the reservation start and stop exactly at the start and end of the LMB. However, the LMB reservations are not aware of NUMA nodes and on occasion a single LMB may cross into several adjacent active regions. Those may even be on different NUMA nodes and will require separate calls to the bootmem reserve functions. So, the bootmem reservation must be trimmed to fit inside the current active region. That's all fine and dandy, but we trim the reservation in a page-aligned fashion. That's bad because we start the reservation at a non-page-aligned address: physbase. The reservation may only span 2 bytes, but that those bytes may span two pfns and cause a reserve_size of 2*PAGE_SIZE. Take the case where you reserve 0x2 bytes at 0x0fff and where the active region ends at 0x1000. You'll jump into that if() statment, but node_ar.end_pfn=0x1 and start_pfn=0x0. You'll end up with a reserve_size=0x1000, and then call reserve_bootmem_node(node, physbase=0xfff, size=0x1000); 0x1000 may not be on the same node as 0xfff. Oops. In almost all the vm code, end_<anything> is not inclusive. If you have an end_pfn of 0x1234, page 0x1234 is not included in the range. Using PFN_UP instead of the (>> >> PAGE_SHIFT) will make this consistent with the other VM code. We also need to do math for the reserved size with physbase instead of start_pfn. node_ar.end_pfn << PAGE_SHIFT is *precisely* the end of the node. However, (start_pfn << PAGE_SHIFT) is *NOT* precisely the beginning of the reserved area. That is, of course, physbase. If we don't use physbase here, the reserve_size can be made too large. From: Dave Hansen <dave@linux.vnet.ibm.com> Tested-by: Geoff Levand <geoffrey.levand@am.sony.com> Tested on PS3. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2009-02-12 13:36:04 +01:00
#include <linux/pfn.h>
#include <linux/cpuset.h>
#include <linux/node.h>
#include <linux/stop_machine.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/uaccess.h>
Merge branch 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup Pull cgroup updates from Tejun Heo: - Fixes and a lot of cleanups. Locking cleanup is finally complete. cgroup_mutex is no longer exposed to individual controlelrs which used to cause nasty deadlock issues. Li fixed and cleaned up quite a bit including long standing ones like racy cgroup_path(). - device cgroup now supports proper hierarchy thanks to Aristeu. - perf_event cgroup now supports proper hierarchy. - A new mount option "__DEVEL__sane_behavior" is added. As indicated by the name, this option is to be used for development only at this point and generates a warning message when used. Unfortunately, cgroup interface currently has too many brekages and inconsistencies to implement a consistent and unified hierarchy on top. The new flag is used to collect the behavior changes which are necessary to implement consistent unified hierarchy. It's likely that this flag won't be used verbatim when it becomes ready but will be enabled implicitly along with unified hierarchy. The option currently disables some of broken behaviors in cgroup core and also .use_hierarchy switch in memcg (will be routed through -mm), which can be used to make very unusual hierarchy where nesting is partially honored. It will also be used to implement hierarchy support for blk-throttle which would be impossible otherwise without introducing a full separate set of control knobs. This is essentially versioning of interface which isn't very nice but at this point I can't see any other options which would allow keeping the interface the same while moving towards hierarchy behavior which is at least somewhat sane. The planned unified hierarchy is likely to require some level of adaptation from userland anyway, so I think it'd be best to take the chance and update the interface such that it's supportable in the long term. Maintaining the existing interface does complicate cgroup core but shouldn't put too much strain on individual controllers and I think it'd be manageable for the foreseeable future. Maybe we'll be able to drop it in a decade. Fix up conflicts (including a semantic one adding a new #include to ppc that was uncovered by header the file changes) as per Tejun. * 'for-3.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (45 commits) cpuset: fix compile warning when CONFIG_SMP=n cpuset: fix cpu hotplug vs rebuild_sched_domains() race cpuset: use rebuild_sched_domains() in cpuset_hotplug_workfn() cgroup: restore the call to eventfd->poll() cgroup: fix use-after-free when umounting cgroupfs cgroup: fix broken file xattrs devcg: remove parent_cgroup. memcg: force use_hierarchy if sane_behavior cgroup: remove cgrp->top_cgroup cgroup: introduce sane_behavior mount option move cgroupfs_root to include/linux/cgroup.h cgroup: convert cgroupfs_root flag bits to masks and add CGRP_ prefix cgroup: make cgroup_path() not print double slashes Revert "cgroup: remove bind() method from cgroup_subsys." perf: make perf_event cgroup hierarchical cgroup: implement cgroup_is_descendant() cgroup: make sure parent won't be destroyed before its children cgroup: remove bind() method from cgroup_subsys. devcg: remove broken_hierarchy tag cgroup: remove cgroup_lock_is_held() ...
2013-04-30 04:14:20 +02:00
#include <linux/slab.h>
#include <asm/cputhreads.h>
#include <asm/sparsemem.h>
#include <asm/prom.h>
#include <asm/smp.h>
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
#include <asm/cputhreads.h>
#include <asm/topology.h>
#include <asm/firmware.h>
#include <asm/paca.h>
#include <asm/hvcall.h>
#include <asm/setup.h>
#include <asm/vdso.h>
static int numa_enabled = 1;
static char *cmdline __initdata;
static int numa_debug;
#define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
int numa_cpu_lookup_table[NR_CPUS];
cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
struct pglist_data *node_data[MAX_NUMNODES];
EXPORT_SYMBOL(numa_cpu_lookup_table);
EXPORT_SYMBOL(node_to_cpumask_map);
EXPORT_SYMBOL(node_data);
static int min_common_depth;
static int n_mem_addr_cells, n_mem_size_cells;
static int form1_affinity;
#define MAX_DISTANCE_REF_POINTS 4
static int distance_ref_points_depth;
static const __be32 *distance_ref_points;
static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS];
/*
* Allocate node_to_cpumask_map based on number of available nodes
* Requires node_possible_map to be valid.
*
* Note: cpumask_of_node() is not valid until after this is done.
*/
static void __init setup_node_to_cpumask_map(void)
{
unsigned int node;
/* setup nr_node_ids if not done yet */
if (nr_node_ids == MAX_NUMNODES)
setup_nr_node_ids();
/* allocate the map */
for (node = 0; node < nr_node_ids; node++)
alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]);
/* cpumask_of_node() will now work */
dbg("Node to cpumask map for %d nodes\n", nr_node_ids);
}
static int __init fake_numa_create_new_node(unsigned long end_pfn,
unsigned int *nid)
{
unsigned long long mem;
char *p = cmdline;
static unsigned int fake_nid;
static unsigned long long curr_boundary;
/*
* Modify node id, iff we started creating NUMA nodes
* We want to continue from where we left of the last time
*/
if (fake_nid)
*nid = fake_nid;
/*
* In case there are no more arguments to parse, the
* node_id should be the same as the last fake node id
* (we've handled this above).
*/
if (!p)
return 0;
mem = memparse(p, &p);
if (!mem)
return 0;
if (mem < curr_boundary)
return 0;
curr_boundary = mem;
if ((end_pfn << PAGE_SHIFT) > mem) {
/*
* Skip commas and spaces
*/
while (*p == ',' || *p == ' ' || *p == '\t')
p++;
cmdline = p;
fake_nid++;
*nid = fake_nid;
dbg("created new fake_node with id %d\n", fake_nid);
return 1;
}
return 0;
}
powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes If there are multiple reserved memory blocks via lmb_reserve() that are contiguous addresses and on different NUMA nodes we are losing track of which address ranges to reserve in bootmem on which node. I discovered this when I recently got to try 16GB huge pages on a system with more then 2 nodes. When scanning the device tree in early boot we call lmb_reserve() with the addresses of the 16G pages that we find so that the memory doesn't get used for something else. For example the addresses for the pages could be 4000000000, 4400000000, 4800000000, 4C00000000, etc - 8 pages, one on each of eight nodes. In the lmb after all the pages have been reserved it will look something like the following: lmb_dump_all: memory.cnt = 0x2 memory.size = 0x3e80000000 memory.region[0x0].base = 0x0 .size = 0x1e80000000 memory.region[0x1].base = 0x4000000000 .size = 0x2000000000 reserved.cnt = 0x5 reserved.size = 0x3e80000000 reserved.region[0x0].base = 0x0 .size = 0x7b5000 reserved.region[0x1].base = 0x2a00000 .size = 0x78c000 reserved.region[0x2].base = 0x328c000 .size = 0x43000 reserved.region[0x3].base = 0xf4e8000 .size = 0xb18000 reserved.region[0x4].base = 0x4000000000 .size = 0x2000000000 The reserved.region[0x4] contains the 16G pages. In arch/powerpc/mm/num.c: do_init_bootmem() we loop through each of the node numbers looking for the reserved regions that belong to the particular node. It is not able to identify region 0x4 as being a part of each of the 8 nodes. It is assuming that a reserved region is only on a single node. This patch takes out the reserved region loop from inside the loop that goes over each node. It looks up the active region containing the start of the reserved region. If it extends past that active region then it adjusts the size and gets the next active region containing it. Signed-off-by: Jon Tollefson <kniht@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2008-10-09 12:18:40 +02:00
/*
* get_node_active_region - Return active region containing pfn
* Active range returned is empty if none found.
* @pfn: The page to return the region for
* @node_ar: Returned set to the active region containing @pfn
powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes If there are multiple reserved memory blocks via lmb_reserve() that are contiguous addresses and on different NUMA nodes we are losing track of which address ranges to reserve in bootmem on which node. I discovered this when I recently got to try 16GB huge pages on a system with more then 2 nodes. When scanning the device tree in early boot we call lmb_reserve() with the addresses of the 16G pages that we find so that the memory doesn't get used for something else. For example the addresses for the pages could be 4000000000, 4400000000, 4800000000, 4C00000000, etc - 8 pages, one on each of eight nodes. In the lmb after all the pages have been reserved it will look something like the following: lmb_dump_all: memory.cnt = 0x2 memory.size = 0x3e80000000 memory.region[0x0].base = 0x0 .size = 0x1e80000000 memory.region[0x1].base = 0x4000000000 .size = 0x2000000000 reserved.cnt = 0x5 reserved.size = 0x3e80000000 reserved.region[0x0].base = 0x0 .size = 0x7b5000 reserved.region[0x1].base = 0x2a00000 .size = 0x78c000 reserved.region[0x2].base = 0x328c000 .size = 0x43000 reserved.region[0x3].base = 0xf4e8000 .size = 0xb18000 reserved.region[0x4].base = 0x4000000000 .size = 0x2000000000 The reserved.region[0x4] contains the 16G pages. In arch/powerpc/mm/num.c: do_init_bootmem() we loop through each of the node numbers looking for the reserved regions that belong to the particular node. It is not able to identify region 0x4 as being a part of each of the 8 nodes. It is assuming that a reserved region is only on a single node. This patch takes out the reserved region loop from inside the loop that goes over each node. It looks up the active region containing the start of the reserved region. If it extends past that active region then it adjusts the size and gets the next active region containing it. Signed-off-by: Jon Tollefson <kniht@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2008-10-09 12:18:40 +02:00
*/
static void __init get_node_active_region(unsigned long pfn,
struct node_active_region *node_ar)
powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes If there are multiple reserved memory blocks via lmb_reserve() that are contiguous addresses and on different NUMA nodes we are losing track of which address ranges to reserve in bootmem on which node. I discovered this when I recently got to try 16GB huge pages on a system with more then 2 nodes. When scanning the device tree in early boot we call lmb_reserve() with the addresses of the 16G pages that we find so that the memory doesn't get used for something else. For example the addresses for the pages could be 4000000000, 4400000000, 4800000000, 4C00000000, etc - 8 pages, one on each of eight nodes. In the lmb after all the pages have been reserved it will look something like the following: lmb_dump_all: memory.cnt = 0x2 memory.size = 0x3e80000000 memory.region[0x0].base = 0x0 .size = 0x1e80000000 memory.region[0x1].base = 0x4000000000 .size = 0x2000000000 reserved.cnt = 0x5 reserved.size = 0x3e80000000 reserved.region[0x0].base = 0x0 .size = 0x7b5000 reserved.region[0x1].base = 0x2a00000 .size = 0x78c000 reserved.region[0x2].base = 0x328c000 .size = 0x43000 reserved.region[0x3].base = 0xf4e8000 .size = 0xb18000 reserved.region[0x4].base = 0x4000000000 .size = 0x2000000000 The reserved.region[0x4] contains the 16G pages. In arch/powerpc/mm/num.c: do_init_bootmem() we loop through each of the node numbers looking for the reserved regions that belong to the particular node. It is not able to identify region 0x4 as being a part of each of the 8 nodes. It is assuming that a reserved region is only on a single node. This patch takes out the reserved region loop from inside the loop that goes over each node. It looks up the active region containing the start of the reserved region. If it extends past that active region then it adjusts the size and gets the next active region containing it. Signed-off-by: Jon Tollefson <kniht@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2008-10-09 12:18:40 +02:00
{
unsigned long start_pfn, end_pfn;
int i, nid;
for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
if (pfn >= start_pfn && pfn < end_pfn) {
node_ar->nid = nid;
node_ar->start_pfn = start_pfn;
node_ar->end_pfn = end_pfn;
break;
}
}
powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes If there are multiple reserved memory blocks via lmb_reserve() that are contiguous addresses and on different NUMA nodes we are losing track of which address ranges to reserve in bootmem on which node. I discovered this when I recently got to try 16GB huge pages on a system with more then 2 nodes. When scanning the device tree in early boot we call lmb_reserve() with the addresses of the 16G pages that we find so that the memory doesn't get used for something else. For example the addresses for the pages could be 4000000000, 4400000000, 4800000000, 4C00000000, etc - 8 pages, one on each of eight nodes. In the lmb after all the pages have been reserved it will look something like the following: lmb_dump_all: memory.cnt = 0x2 memory.size = 0x3e80000000 memory.region[0x0].base = 0x0 .size = 0x1e80000000 memory.region[0x1].base = 0x4000000000 .size = 0x2000000000 reserved.cnt = 0x5 reserved.size = 0x3e80000000 reserved.region[0x0].base = 0x0 .size = 0x7b5000 reserved.region[0x1].base = 0x2a00000 .size = 0x78c000 reserved.region[0x2].base = 0x328c000 .size = 0x43000 reserved.region[0x3].base = 0xf4e8000 .size = 0xb18000 reserved.region[0x4].base = 0x4000000000 .size = 0x2000000000 The reserved.region[0x4] contains the 16G pages. In arch/powerpc/mm/num.c: do_init_bootmem() we loop through each of the node numbers looking for the reserved regions that belong to the particular node. It is not able to identify region 0x4 as being a part of each of the 8 nodes. It is assuming that a reserved region is only on a single node. This patch takes out the reserved region loop from inside the loop that goes over each node. It looks up the active region containing the start of the reserved region. If it extends past that active region then it adjusts the size and gets the next active region containing it. Signed-off-by: Jon Tollefson <kniht@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2008-10-09 12:18:40 +02:00
}
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
static void reset_numa_cpu_lookup_table(void)
{
unsigned int cpu;
for_each_possible_cpu(cpu)
numa_cpu_lookup_table[cpu] = -1;
}
static void update_numa_cpu_lookup_table(unsigned int cpu, int node)
{
numa_cpu_lookup_table[cpu] = node;
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
}
static void map_cpu_to_node(int cpu, int node)
{
update_numa_cpu_lookup_table(cpu, node);
dbg("adding cpu %d to node %d\n", cpu, node);
if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node])))
cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
}
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PPC_SPLPAR)
static void unmap_cpu_from_node(unsigned long cpu)
{
int node = numa_cpu_lookup_table[cpu];
dbg("removing cpu %lu from node %d\n", cpu, node);
if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) {
cpumask_clear_cpu(cpu, node_to_cpumask_map[node]);
} else {
printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
cpu, node);
}
}
#endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */
/* must hold reference to node during call */
static const __be32 *of_get_associativity(struct device_node *dev)
{
return of_get_property(dev, "ibm,associativity", NULL);
}
/*
* Returns the property linux,drconf-usable-memory if
* it exists (the property exists only in kexec/kdump kernels,
* added by kexec-tools)
*/
static const __be32 *of_get_usable_memory(struct device_node *memory)
{
const __be32 *prop;
u32 len;
prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
if (!prop || len < sizeof(unsigned int))
return NULL;
return prop;
}
int __node_distance(int a, int b)
{
int i;
int distance = LOCAL_DISTANCE;
if (!form1_affinity)
return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE);
for (i = 0; i < distance_ref_points_depth; i++) {
if (distance_lookup_table[a][i] == distance_lookup_table[b][i])
break;
/* Double the distance for each NUMA level */
distance *= 2;
}
return distance;
}
static void initialize_distance_lookup_table(int nid,
const __be32 *associativity)
{
int i;
if (!form1_affinity)
return;
for (i = 0; i < distance_ref_points_depth; i++) {
const __be32 *entry;
entry = &associativity[be32_to_cpu(distance_ref_points[i])];
distance_lookup_table[nid][i] = of_read_number(entry, 1);
}
}
/* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
* info is found.
*/
static int associativity_to_nid(const __be32 *associativity)
{
int nid = -1;
if (min_common_depth == -1)
goto out;
if (of_read_number(associativity, 1) >= min_common_depth)
nid = of_read_number(&associativity[min_common_depth], 1);
/* POWER4 LPAR uses 0xffff as invalid node */
if (nid == 0xffff || nid >= MAX_NUMNODES)
nid = -1;
if (nid > 0 &&
of_read_number(associativity, 1) >= distance_ref_points_depth)
initialize_distance_lookup_table(nid, associativity);
out:
return nid;
}
/* Returns the nid associated with the given device tree node,
* or -1 if not found.
*/
static int of_node_to_nid_single(struct device_node *device)
{
int nid = -1;
const __be32 *tmp;
tmp = of_get_associativity(device);
if (tmp)
nid = associativity_to_nid(tmp);
return nid;
}
/* Walk the device tree upwards, looking for an associativity id */
int of_node_to_nid(struct device_node *device)
{
struct device_node *tmp;
int nid = -1;
of_node_get(device);
while (device) {
nid = of_node_to_nid_single(device);
if (nid != -1)
break;
tmp = device;
device = of_get_parent(tmp);
of_node_put(tmp);
}
of_node_put(device);
return nid;
}
EXPORT_SYMBOL_GPL(of_node_to_nid);
static int __init find_min_common_depth(void)
{
int depth;
struct device_node *root;
if (firmware_has_feature(FW_FEATURE_OPAL))
root = of_find_node_by_path("/ibm,opal");
else
root = of_find_node_by_path("/rtas");
if (!root)
root = of_find_node_by_path("/");
/*
* This property is a set of 32-bit integers, each representing
* an index into the ibm,associativity nodes.
*
* With form 0 affinity the first integer is for an SMP configuration
* (should be all 0's) and the second is for a normal NUMA
* configuration. We have only one level of NUMA.
*
* With form 1 affinity the first integer is the most significant
* NUMA boundary and the following are progressively less significant
* boundaries. There can be more than one level of NUMA.
*/
distance_ref_points = of_get_property(root,
"ibm,associativity-reference-points",
&distance_ref_points_depth);
if (!distance_ref_points) {
dbg("NUMA: ibm,associativity-reference-points not found.\n");
goto err;
}
distance_ref_points_depth /= sizeof(int);
if (firmware_has_feature(FW_FEATURE_OPAL) ||
firmware_has_feature(FW_FEATURE_TYPE1_AFFINITY)) {
dbg("Using form 1 affinity\n");
form1_affinity = 1;
}
if (form1_affinity) {
depth = of_read_number(distance_ref_points, 1);
} else {
if (distance_ref_points_depth < 2) {
printk(KERN_WARNING "NUMA: "
"short ibm,associativity-reference-points\n");
goto err;
}
depth = of_read_number(&distance_ref_points[1], 1);
}
/*
* Warn and cap if the hardware supports more than
* MAX_DISTANCE_REF_POINTS domains.
*/
if (distance_ref_points_depth > MAX_DISTANCE_REF_POINTS) {
printk(KERN_WARNING "NUMA: distance array capped at "
"%d entries\n", MAX_DISTANCE_REF_POINTS);
distance_ref_points_depth = MAX_DISTANCE_REF_POINTS;
}
of_node_put(root);
return depth;
err:
of_node_put(root);
return -1;
}
static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
{
struct device_node *memory = NULL;
memory = of_find_node_by_type(memory, "memory");
if (!memory)
panic("numa.c: No memory nodes found!");
*n_addr_cells = of_n_addr_cells(memory);
*n_size_cells = of_n_size_cells(memory);
of_node_put(memory);
}
static unsigned long read_n_cells(int n, const __be32 **buf)
{
unsigned long result = 0;
while (n--) {
result = (result << 32) | of_read_number(*buf, 1);
(*buf)++;
}
return result;
}
/*
* Read the next memblock list entry from the ibm,dynamic-memory property
* and return the information in the provided of_drconf_cell structure.
*/
static void read_drconf_cell(struct of_drconf_cell *drmem, const __be32 **cellp)
{
const __be32 *cp;
drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
cp = *cellp;
drmem->drc_index = of_read_number(cp, 1);
drmem->reserved = of_read_number(&cp[1], 1);
drmem->aa_index = of_read_number(&cp[2], 1);
drmem->flags = of_read_number(&cp[3], 1);
*cellp = cp + 4;
}
/*
* Retrieve and validate the ibm,dynamic-memory property of the device tree.
*
* The layout of the ibm,dynamic-memory property is a number N of memblock
* list entries followed by N memblock list entries. Each memblock list entry
* contains information as laid out in the of_drconf_cell struct above.
*/
static int of_get_drconf_memory(struct device_node *memory, const __be32 **dm)
{
const __be32 *prop;
u32 len, entries;
prop = of_get_property(memory, "ibm,dynamic-memory", &len);
if (!prop || len < sizeof(unsigned int))
return 0;
entries = of_read_number(prop++, 1);
/* Now that we know the number of entries, revalidate the size
* of the property read in to ensure we have everything
*/
if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
return 0;
*dm = prop;
return entries;
}
/*
* Retrieve and validate the ibm,lmb-size property for drconf memory
* from the device tree.
*/
static u64 of_get_lmb_size(struct device_node *memory)
{
const __be32 *prop;
u32 len;
prop = of_get_property(memory, "ibm,lmb-size", &len);
if (!prop || len < sizeof(unsigned int))
return 0;
return read_n_cells(n_mem_size_cells, &prop);
}
struct assoc_arrays {
u32 n_arrays;
u32 array_sz;
const __be32 *arrays;
};
/*
* Retrieve and validate the list of associativity arrays for drconf
* memory from the ibm,associativity-lookup-arrays property of the
* device tree..
*
* The layout of the ibm,associativity-lookup-arrays property is a number N
* indicating the number of associativity arrays, followed by a number M
* indicating the size of each associativity array, followed by a list
* of N associativity arrays.
*/
static int of_get_assoc_arrays(struct device_node *memory,
struct assoc_arrays *aa)
{
const __be32 *prop;
u32 len;
prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
if (!prop || len < 2 * sizeof(unsigned int))
return -1;
aa->n_arrays = of_read_number(prop++, 1);
aa->array_sz = of_read_number(prop++, 1);
/* Now that we know the number of arrays and size of each array,
* revalidate the size of the property read in.
*/
if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
return -1;
aa->arrays = prop;
return 0;
}
/*
* This is like of_node_to_nid_single() for memory represented in the
* ibm,dynamic-reconfiguration-memory node.
*/
static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
struct assoc_arrays *aa)
{
int default_nid = 0;
int nid = default_nid;
int index;
if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
!(drmem->flags & DRCONF_MEM_AI_INVALID) &&
drmem->aa_index < aa->n_arrays) {
index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
nid = of_read_number(&aa->arrays[index], 1);
if (nid == 0xffff || nid >= MAX_NUMNODES)
nid = default_nid;
}
return nid;
}
/*
* Figure out to which domain a cpu belongs and stick it there.
* Return the id of the domain used.
*/
static int numa_setup_cpu(unsigned long lcpu)
{
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
int nid;
struct device_node *cpu;
/*
* If a valid cpu-to-node mapping is already available, use it
* directly instead of querying the firmware, since it represents
* the most recent mapping notified to us by the platform (eg: VPHN).
*/
if ((nid = numa_cpu_lookup_table[lcpu]) >= 0) {
map_cpu_to_node(lcpu, nid);
return nid;
}
cpu = of_get_cpu_node(lcpu, NULL);
if (!cpu) {
WARN_ON(1);
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
nid = 0;
goto out;
}
nid = of_node_to_nid_single(cpu);
if (nid < 0 || !node_online(nid))
nid = first_online_node;
out:
map_cpu_to_node(lcpu, nid);
of_node_put(cpu);
return nid;
}
static void verify_cpu_node_mapping(int cpu, int node)
{
int base, sibling, i;
/* Verify that all the threads in the core belong to the same node */
base = cpu_first_thread_sibling(cpu);
for (i = 0; i < threads_per_core; i++) {
sibling = base + i;
if (sibling == cpu || cpu_is_offline(sibling))
continue;
if (cpu_to_node(sibling) != node) {
WARN(1, "CPU thread siblings %d and %d don't belong"
" to the same node!\n", cpu, sibling);
break;
}
}
}
static int cpu_numa_callback(struct notifier_block *nfb, unsigned long action,
void *hcpu)
{
unsigned long lcpu = (unsigned long)hcpu;
int ret = NOTIFY_DONE, nid;
switch (action) {
case CPU_UP_PREPARE:
case CPU_UP_PREPARE_FROZEN:
nid = numa_setup_cpu(lcpu);
verify_cpu_node_mapping((int)lcpu, nid);
ret = NOTIFY_OK;
break;
#ifdef CONFIG_HOTPLUG_CPU
case CPU_DEAD:
case CPU_DEAD_FROZEN:
case CPU_UP_CANCELED:
case CPU_UP_CANCELED_FROZEN:
unmap_cpu_from_node(lcpu);
break;
ret = NOTIFY_OK;
#endif
}
return ret;
}
/*
* Check and possibly modify a memory region to enforce the memory limit.
*
* Returns the size the region should have to enforce the memory limit.
* This will either be the original value of size, a truncated value,
* or zero. If the returned value of size is 0 the region should be
* discarded as it lies wholly above the memory limit.
*/
static unsigned long __init numa_enforce_memory_limit(unsigned long start,
unsigned long size)
{
/*
* We use memblock_end_of_DRAM() in here instead of memory_limit because
* we've already adjusted it for the limit and it takes care of
* having memory holes below the limit. Also, in the case of
* iommu_is_off, memory_limit is not set but is implicitly enforced.
*/
if (start + size <= memblock_end_of_DRAM())
return size;
if (start >= memblock_end_of_DRAM())
return 0;
return memblock_end_of_DRAM() - start;
}
/*
* Reads the counter for a given entry in
* linux,drconf-usable-memory property
*/
static inline int __init read_usm_ranges(const __be32 **usm)
{
/*
* For each lmb in ibm,dynamic-memory a corresponding
* entry in linux,drconf-usable-memory property contains
* a counter followed by that many (base, size) duple.
* read the counter from linux,drconf-usable-memory
*/
return read_n_cells(n_mem_size_cells, usm);
}
/*
* Extract NUMA information from the ibm,dynamic-reconfiguration-memory
* node. This assumes n_mem_{addr,size}_cells have been set.
*/
static void __init parse_drconf_memory(struct device_node *memory)
{
const __be32 *uninitialized_var(dm), *usm;
unsigned int n, rc, ranges, is_kexec_kdump = 0;
unsigned long lmb_size, base, size, sz;
int nid;
struct assoc_arrays aa = { .arrays = NULL };
n = of_get_drconf_memory(memory, &dm);
if (!n)
return;
lmb_size = of_get_lmb_size(memory);
if (!lmb_size)
return;
rc = of_get_assoc_arrays(memory, &aa);
if (rc)
return;
/* check if this is a kexec/kdump kernel */
usm = of_get_usable_memory(memory);
if (usm != NULL)
is_kexec_kdump = 1;
for (; n != 0; --n) {
struct of_drconf_cell drmem;
read_drconf_cell(&drmem, &dm);
/* skip this block if the reserved bit is set in flags (0x80)
or if the block is not assigned to this partition (0x8) */
if ((drmem.flags & DRCONF_MEM_RESERVED)
|| !(drmem.flags & DRCONF_MEM_ASSIGNED))
continue;
base = drmem.base_addr;
size = lmb_size;
ranges = 1;
if (is_kexec_kdump) {
ranges = read_usm_ranges(&usm);
if (!ranges) /* there are no (base, size) duple */
continue;
}
do {
if (is_kexec_kdump) {
base = read_n_cells(n_mem_addr_cells, &usm);
size = read_n_cells(n_mem_size_cells, &usm);
}
nid = of_drconf_to_nid_single(&drmem, &aa);
fake_numa_create_new_node(
((base + size) >> PAGE_SHIFT),
&nid);
node_set_online(nid);
sz = numa_enforce_memory_limit(base, size);
if (sz)
memblock_set_node(base, sz,
&memblock.memory, nid);
} while (--ranges);
}
}
static int __init parse_numa_properties(void)
{
struct device_node *memory;
int default_nid = 0;
unsigned long i;
if (numa_enabled == 0) {
printk(KERN_WARNING "NUMA disabled by user\n");
return -1;
}
min_common_depth = find_min_common_depth();
if (min_common_depth < 0)
return min_common_depth;
dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
/*
* Even though we connect cpus to numa domains later in SMP
* init, we need to know the node ids now. This is because
* each node to be onlined must have NODE_DATA etc backing it.
*/
for_each_present_cpu(i) {
struct device_node *cpu;
int nid;
cpu = of_get_cpu_node(i, NULL);
BUG_ON(!cpu);
nid = of_node_to_nid_single(cpu);
of_node_put(cpu);
/*
* Don't fall back to default_nid yet -- we will plug
* cpus into nodes once the memory scan has discovered
* the topology.
*/
if (nid < 0)
continue;
node_set_online(nid);
}
get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
for_each_node_by_type(memory, "memory") {
unsigned long start;
unsigned long size;
int nid;
int ranges;
const __be32 *memcell_buf;
unsigned int len;
memcell_buf = of_get_property(memory,
"linux,usable-memory", &len);
if (!memcell_buf || len <= 0)
memcell_buf = of_get_property(memory, "reg", &len);
if (!memcell_buf || len <= 0)
continue;
/* ranges in cell */
ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
new_range:
/* these are order-sensitive, and modify the buffer pointer */
start = read_n_cells(n_mem_addr_cells, &memcell_buf);
size = read_n_cells(n_mem_size_cells, &memcell_buf);
/*
* Assumption: either all memory nodes or none will
* have associativity properties. If none, then
* everything goes to default_nid.
*/
nid = of_node_to_nid_single(memory);
if (nid < 0)
nid = default_nid;
fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
node_set_online(nid);
if (!(size = numa_enforce_memory_limit(start, size))) {
if (--ranges)
goto new_range;
else
continue;
}
memblock_set_node(start, size, &memblock.memory, nid);
if (--ranges)
goto new_range;
}
/*
* Now do the same thing for each MEMBLOCK listed in the
* ibm,dynamic-memory property in the
* ibm,dynamic-reconfiguration-memory node.
*/
memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
if (memory)
parse_drconf_memory(memory);
return 0;
}
static void __init setup_nonnuma(void)
{
unsigned long top_of_ram = memblock_end_of_DRAM();
unsigned long total_ram = memblock_phys_mem_size();
unsigned long start_pfn, end_pfn;
unsigned int nid = 0;
struct memblock_region *reg;
printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
top_of_ram, total_ram);
printk(KERN_DEBUG "Memory hole size: %ldMB\n",
(top_of_ram - total_ram) >> 20);
for_each_memblock(memory, reg) {
start_pfn = memblock_region_memory_base_pfn(reg);
end_pfn = memblock_region_memory_end_pfn(reg);
fake_numa_create_new_node(end_pfn, &nid);
memblock_set_node(PFN_PHYS(start_pfn),
PFN_PHYS(end_pfn - start_pfn),
&memblock.memory, nid);
node_set_online(nid);
}
}
void __init dump_numa_cpu_topology(void)
{
unsigned int node;
unsigned int cpu, count;
if (min_common_depth == -1 || !numa_enabled)
return;
for_each_online_node(node) {
printk(KERN_DEBUG "Node %d CPUs:", node);
count = 0;
/*
* If we used a CPU iterator here we would miss printing
* the holes in the cpumap.
*/
for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
if (cpumask_test_cpu(cpu,
node_to_cpumask_map[node])) {
if (count == 0)
printk(" %u", cpu);
++count;
} else {
if (count > 1)
printk("-%u", cpu - 1);
count = 0;
}
}
if (count > 1)
printk("-%u", nr_cpu_ids - 1);
printk("\n");
}
}
static void __init dump_numa_memory_topology(void)
{
unsigned int node;
unsigned int count;
if (min_common_depth == -1 || !numa_enabled)
return;
for_each_online_node(node) {
unsigned long i;
printk(KERN_DEBUG "Node %d Memory:", node);
count = 0;
for (i = 0; i < memblock_end_of_DRAM();
i += (1 << SECTION_SIZE_BITS)) {
if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
if (count == 0)
printk(" 0x%lx", i);
++count;
} else {
if (count > 0)
printk("-0x%lx", i);
count = 0;
}
}
if (count > 0)
printk("-0x%lx", i);
printk("\n");
}
}
/*
* Allocate some memory, satisfying the memblock or bootmem allocator where
* required. nid is the preferred node and end is the physical address of
* the highest address in the node.
*
* Returns the virtual address of the memory.
*/
static void __init *careful_zallocation(int nid, unsigned long size,
unsigned long align,
unsigned long end_pfn)
{
void *ret;
int new_nid;
unsigned long ret_paddr;
ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT);
/* retry over all memory */
if (!ret_paddr)
ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM());
if (!ret_paddr)
panic("numa.c: cannot allocate %lu bytes for node %d",
size, nid);
ret = __va(ret_paddr);
/*
* We initialize the nodes in numeric order: 0, 1, 2...
* and hand over control from the MEMBLOCK allocator to the
* bootmem allocator. If this function is called for
* node 5, then we know that all nodes <5 are using the
* bootmem allocator instead of the MEMBLOCK allocator.
*
* So, check the nid from which this allocation came
* and double check to see if we need to use bootmem
* instead of the MEMBLOCK. We don't free the MEMBLOCK memory
* since it would be useless.
*/
new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
if (new_nid < nid) {
ret = __alloc_bootmem_node(NODE_DATA(new_nid),
size, align, 0);
dbg("alloc_bootmem %p %lx\n", ret, size);
}
memset(ret, 0, size);
return ret;
}
static struct notifier_block ppc64_numa_nb = {
.notifier_call = cpu_numa_callback,
.priority = 1 /* Must run before sched domains notifier. */
};
static void __init mark_reserved_regions_for_nid(int nid)
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
{
struct pglist_data *node = NODE_DATA(nid);
struct memblock_region *reg;
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
for_each_memblock(reserved, reg) {
unsigned long physbase = reg->base;
unsigned long size = reg->size;
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
unsigned long start_pfn = physbase >> PAGE_SHIFT;
powerpc/mm: Fix numa reserve bootmem page selection Fix the powerpc NUMA reserve bootmem page selection logic. commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb (powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes) changed the logic for how the powerpc LMB reserved regions were converted to bootmen reserved regions. As the folowing discussion reports, the new logic was not correct. mark_reserved_regions_for_nid() goes through each LMB on the system that specifies a reserved area. It searches for active regions that intersect with that LMB and are on the specified node. It attempts to bootmem-reserve only the area where the active region and the reserved LMB intersect. We can not reserve things on other nodes as they may not have bootmem structures allocated, yet. We base the size of the bootmem reservation on two possible things. Normally, we just make the reservation start and stop exactly at the start and end of the LMB. However, the LMB reservations are not aware of NUMA nodes and on occasion a single LMB may cross into several adjacent active regions. Those may even be on different NUMA nodes and will require separate calls to the bootmem reserve functions. So, the bootmem reservation must be trimmed to fit inside the current active region. That's all fine and dandy, but we trim the reservation in a page-aligned fashion. That's bad because we start the reservation at a non-page-aligned address: physbase. The reservation may only span 2 bytes, but that those bytes may span two pfns and cause a reserve_size of 2*PAGE_SIZE. Take the case where you reserve 0x2 bytes at 0x0fff and where the active region ends at 0x1000. You'll jump into that if() statment, but node_ar.end_pfn=0x1 and start_pfn=0x0. You'll end up with a reserve_size=0x1000, and then call reserve_bootmem_node(node, physbase=0xfff, size=0x1000); 0x1000 may not be on the same node as 0xfff. Oops. In almost all the vm code, end_<anything> is not inclusive. If you have an end_pfn of 0x1234, page 0x1234 is not included in the range. Using PFN_UP instead of the (>> >> PAGE_SHIFT) will make this consistent with the other VM code. We also need to do math for the reserved size with physbase instead of start_pfn. node_ar.end_pfn << PAGE_SHIFT is *precisely* the end of the node. However, (start_pfn << PAGE_SHIFT) is *NOT* precisely the beginning of the reserved area. That is, of course, physbase. If we don't use physbase here, the reserve_size can be made too large. From: Dave Hansen <dave@linux.vnet.ibm.com> Tested-by: Geoff Levand <geoffrey.levand@am.sony.com> Tested on PS3. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2009-02-12 13:36:04 +01:00
unsigned long end_pfn = PFN_UP(physbase + size);
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
struct node_active_region node_ar;
unsigned long node_end_pfn = pgdat_end_pfn(node);
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
/*
* Check to make sure that this memblock.reserved area is
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
* within the bounds of the node that we care about.
* Checking the nid of the start and end points is not
* sufficient because the reserved area could span the
* entire node.
*/
if (end_pfn <= node->node_start_pfn ||
start_pfn >= node_end_pfn)
continue;
get_node_active_region(start_pfn, &node_ar);
while (start_pfn < end_pfn &&
node_ar.start_pfn < node_ar.end_pfn) {
unsigned long reserve_size = size;
/*
* if reserved region extends past active region
* then trim size to active region
*/
if (end_pfn > node_ar.end_pfn)
reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
powerpc/mm: Fix numa reserve bootmem page selection Fix the powerpc NUMA reserve bootmem page selection logic. commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb (powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes) changed the logic for how the powerpc LMB reserved regions were converted to bootmen reserved regions. As the folowing discussion reports, the new logic was not correct. mark_reserved_regions_for_nid() goes through each LMB on the system that specifies a reserved area. It searches for active regions that intersect with that LMB and are on the specified node. It attempts to bootmem-reserve only the area where the active region and the reserved LMB intersect. We can not reserve things on other nodes as they may not have bootmem structures allocated, yet. We base the size of the bootmem reservation on two possible things. Normally, we just make the reservation start and stop exactly at the start and end of the LMB. However, the LMB reservations are not aware of NUMA nodes and on occasion a single LMB may cross into several adjacent active regions. Those may even be on different NUMA nodes and will require separate calls to the bootmem reserve functions. So, the bootmem reservation must be trimmed to fit inside the current active region. That's all fine and dandy, but we trim the reservation in a page-aligned fashion. That's bad because we start the reservation at a non-page-aligned address: physbase. The reservation may only span 2 bytes, but that those bytes may span two pfns and cause a reserve_size of 2*PAGE_SIZE. Take the case where you reserve 0x2 bytes at 0x0fff and where the active region ends at 0x1000. You'll jump into that if() statment, but node_ar.end_pfn=0x1 and start_pfn=0x0. You'll end up with a reserve_size=0x1000, and then call reserve_bootmem_node(node, physbase=0xfff, size=0x1000); 0x1000 may not be on the same node as 0xfff. Oops. In almost all the vm code, end_<anything> is not inclusive. If you have an end_pfn of 0x1234, page 0x1234 is not included in the range. Using PFN_UP instead of the (>> >> PAGE_SHIFT) will make this consistent with the other VM code. We also need to do math for the reserved size with physbase instead of start_pfn. node_ar.end_pfn << PAGE_SHIFT is *precisely* the end of the node. However, (start_pfn << PAGE_SHIFT) is *NOT* precisely the beginning of the reserved area. That is, of course, physbase. If we don't use physbase here, the reserve_size can be made too large. From: Dave Hansen <dave@linux.vnet.ibm.com> Tested-by: Geoff Levand <geoffrey.levand@am.sony.com> Tested on PS3. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2009-02-12 13:36:04 +01:00
- physbase;
/*
* Only worry about *this* node, others may not
* yet have valid NODE_DATA().
*/
if (node_ar.nid == nid) {
dbg("reserve_bootmem %lx %lx nid=%d\n",
physbase, reserve_size, node_ar.nid);
reserve_bootmem_node(NODE_DATA(node_ar.nid),
physbase, reserve_size,
BOOTMEM_DEFAULT);
}
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
/*
* if reserved region is contained in the active region
* then done.
*/
if (end_pfn <= node_ar.end_pfn)
break;
/*
* reserved region extends past the active region
* get next active region that contains this
* reserved region
*/
start_pfn = node_ar.end_pfn;
physbase = start_pfn << PAGE_SHIFT;
size = size - reserve_size;
get_node_active_region(start_pfn, &node_ar);
}
}
}
void __init do_init_bootmem(void)
{
int nid;
min_low_pfn = 0;
max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
max_pfn = max_low_pfn;
if (parse_numa_properties())
setup_nonnuma();
else
dump_numa_memory_topology();
for_each_online_node(nid) {
unsigned long start_pfn, end_pfn;
void *bootmem_vaddr;
unsigned long bootmap_pages;
get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
/*
* Allocate the node structure node local if possible
*
* Be careful moving this around, as it relies on all
* previous nodes' bootmem to be initialized and have
* all reserved areas marked.
*/
NODE_DATA(nid) = careful_zallocation(nid,
sizeof(struct pglist_data),
SMP_CACHE_BYTES, end_pfn);
dbg("node %d\n", nid);
dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
NODE_DATA(nid)->node_start_pfn = start_pfn;
NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
if (NODE_DATA(nid)->node_spanned_pages == 0)
continue;
dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
bootmem_vaddr = careful_zallocation(nid,
bootmap_pages << PAGE_SHIFT,
PAGE_SIZE, end_pfn);
dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
init_bootmem_node(NODE_DATA(nid),
__pa(bootmem_vaddr) >> PAGE_SHIFT,
start_pfn, end_pfn);
free_bootmem_with_active_regions(nid, end_pfn);
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
/*
* Be very careful about moving this around. Future
* calls to careful_zallocation() depend on this getting
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
* done correctly.
*/
mark_reserved_regions_for_nid(nid);
powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes If there are multiple reserved memory blocks via lmb_reserve() that are contiguous addresses and on different NUMA nodes we are losing track of which address ranges to reserve in bootmem on which node. I discovered this when I recently got to try 16GB huge pages on a system with more then 2 nodes. When scanning the device tree in early boot we call lmb_reserve() with the addresses of the 16G pages that we find so that the memory doesn't get used for something else. For example the addresses for the pages could be 4000000000, 4400000000, 4800000000, 4C00000000, etc - 8 pages, one on each of eight nodes. In the lmb after all the pages have been reserved it will look something like the following: lmb_dump_all: memory.cnt = 0x2 memory.size = 0x3e80000000 memory.region[0x0].base = 0x0 .size = 0x1e80000000 memory.region[0x1].base = 0x4000000000 .size = 0x2000000000 reserved.cnt = 0x5 reserved.size = 0x3e80000000 reserved.region[0x0].base = 0x0 .size = 0x7b5000 reserved.region[0x1].base = 0x2a00000 .size = 0x78c000 reserved.region[0x2].base = 0x328c000 .size = 0x43000 reserved.region[0x3].base = 0xf4e8000 .size = 0xb18000 reserved.region[0x4].base = 0x4000000000 .size = 0x2000000000 The reserved.region[0x4] contains the 16G pages. In arch/powerpc/mm/num.c: do_init_bootmem() we loop through each of the node numbers looking for the reserved regions that belong to the particular node. It is not able to identify region 0x4 as being a part of each of the 8 nodes. It is assuming that a reserved region is only on a single node. This patch takes out the reserved region loop from inside the loop that goes over each node. It looks up the active region containing the start of the reserved region. If it extends past that active region then it adjusts the size and gets the next active region containing it. Signed-off-by: Jon Tollefson <kniht@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2008-10-09 12:18:40 +02:00
sparse_memory_present_with_active_regions(nid);
powerpc: Fix boot freeze on machine with empty memory node I got a bug report about a distro kernel not booting on a particular machine. It would freeze during boot: > ... > Could not find start_pfn for node 1 > [boot]0015 Setup Done > Built 2 zonelists in Node order, mobility grouping on. Total pages: 123783 > Policy zone: DMA > Kernel command line: > [boot]0020 XICS Init > [boot]0021 XICS Done > PID hash table entries: 4096 (order: 12, 32768 bytes) > clocksource: timebase mult[7d0000] shift[22] registered > Console: colour dummy device 80x25 > console handover: boot [udbg0] -> real [hvc0] > Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes) > Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes) > freeing bootmem node 0 I've reproduced this on 2.6.27.7. It is caused by commit 8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem lmb reserved regions that cross NUMA nodes"). The problem is that Jon took a loop which was (in pseudocode): for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); reserve_node_bootmem(nid); and broke it up into: for_each_node(nid) NODE_DATA(nid) = careful_alloc(nid); setup_bootmem(nid); for_each_node(nid) reserve_node_bootmem(nid); The issue comes in when the 'careful_alloc()' is called on a node with no memory. It falls back to using bootmem from a previously-initialized node. But, bootmem has not yet been reserved when Jon's patch is applied. It gives back bogus memory (0xc000000000000000) and pukes later in boot. The following patch collapses the loop back together. It also breaks the mark_reserved_regions_for_nid() code out into a function and adds some comments. I think a huge part of introducing this bug is because for loop was too long and hard to read. The actual bug fix here is the: + if (end_pfn <= node->node_start_pfn || + start_pfn >= node_end_pfn) + continue; Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-11-24 13:02:35 +01:00
}
init_bootmem_done = 1;
/*
* Now bootmem is initialised we can create the node to cpumask
* lookup tables and setup the cpu callback to populate them.
*/
setup_node_to_cpumask_map();
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
reset_numa_cpu_lookup_table();
register_cpu_notifier(&ppc64_numa_nb);
cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
(void *)(unsigned long)boot_cpuid);
}
void __init paging_init(void)
{
unsigned long max_zone_pfns[MAX_NR_ZONES];
memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT;
free_area_init_nodes(max_zone_pfns);
}
static int __init early_numa(char *p)
{
if (!p)
return 0;
if (strstr(p, "off"))
numa_enabled = 0;
if (strstr(p, "debug"))
numa_debug = 1;
p = strstr(p, "fake=");
if (p)
cmdline = p + strlen("fake=");
return 0;
}
early_param("numa", early_numa);
#ifdef CONFIG_MEMORY_HOTPLUG
/*
* Find the node associated with a hot added memory section for
* memory represented in the device tree by the property
* ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
*/
static int hot_add_drconf_scn_to_nid(struct device_node *memory,
unsigned long scn_addr)
{
const __be32 *dm;
unsigned int drconf_cell_cnt, rc;
unsigned long lmb_size;
struct assoc_arrays aa;
int nid = -1;
drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
if (!drconf_cell_cnt)
return -1;
lmb_size = of_get_lmb_size(memory);
if (!lmb_size)
return -1;
rc = of_get_assoc_arrays(memory, &aa);
if (rc)
return -1;
for (; drconf_cell_cnt != 0; --drconf_cell_cnt) {
struct of_drconf_cell drmem;
read_drconf_cell(&drmem, &dm);
/* skip this block if it is reserved or not assigned to
* this partition */
if ((drmem.flags & DRCONF_MEM_RESERVED)
|| !(drmem.flags & DRCONF_MEM_ASSIGNED))
continue;
if ((scn_addr < drmem.base_addr)
|| (scn_addr >= (drmem.base_addr + lmb_size)))
continue;
nid = of_drconf_to_nid_single(&drmem, &aa);
break;
}
return nid;
}
/*
* Find the node associated with a hot added memory section for memory
* represented in the device tree as a node (i.e. memory@XXXX) for
* each memblock.
*/
static int hot_add_node_scn_to_nid(unsigned long scn_addr)
{
struct device_node *memory;
int nid = -1;
for_each_node_by_type(memory, "memory") {
unsigned long start, size;
int ranges;
const __be32 *memcell_buf;
unsigned int len;
memcell_buf = of_get_property(memory, "reg", &len);
if (!memcell_buf || len <= 0)
continue;
/* ranges in cell */
ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
while (ranges--) {
start = read_n_cells(n_mem_addr_cells, &memcell_buf);
size = read_n_cells(n_mem_size_cells, &memcell_buf);
if ((scn_addr < start) || (scn_addr >= (start + size)))
continue;
nid = of_node_to_nid_single(memory);
break;
}
if (nid >= 0)
break;
}
of_node_put(memory);
return nid;
}
/*
* Find the node associated with a hot added memory section. Section
* corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that
* sections are fully contained within a single MEMBLOCK.
*/
int hot_add_scn_to_nid(unsigned long scn_addr)
{
struct device_node *memory = NULL;
int nid, found = 0;
if (!numa_enabled || (min_common_depth < 0))
return first_online_node;
memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
if (memory) {
nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
of_node_put(memory);
} else {
nid = hot_add_node_scn_to_nid(scn_addr);
}
if (nid < 0 || !node_online(nid))
nid = first_online_node;
if (NODE_DATA(nid)->node_spanned_pages)
return nid;
for_each_online_node(nid) {
if (NODE_DATA(nid)->node_spanned_pages) {
found = 1;
break;
}
}
BUG_ON(!found);
return nid;
}
static u64 hot_add_drconf_memory_max(void)
{
struct device_node *memory = NULL;
unsigned int drconf_cell_cnt = 0;
u64 lmb_size = 0;
const __be32 *dm = NULL;
memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
if (memory) {
drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
lmb_size = of_get_lmb_size(memory);
of_node_put(memory);
}
return lmb_size * drconf_cell_cnt;
}
/*
* memory_hotplug_max - return max address of memory that may be added
*
* This is currently only used on systems that support drconfig memory
* hotplug.
*/
u64 memory_hotplug_max(void)
{
return max(hot_add_drconf_memory_max(), memblock_end_of_DRAM());
}
#endif /* CONFIG_MEMORY_HOTPLUG */
/* Virtual Processor Home Node (VPHN) support */
#ifdef CONFIG_PPC_SPLPAR
struct topology_update_data {
struct topology_update_data *next;
unsigned int cpu;
int old_nid;
int new_nid;
};
static u8 vphn_cpu_change_counts[NR_CPUS][MAX_DISTANCE_REF_POINTS];
static cpumask_t cpu_associativity_changes_mask;
static int vphn_enabled;
static int prrn_enabled;
static void reset_topology_timer(void);
/*
* Store the current values of the associativity change counters in the
* hypervisor.
*/
static void setup_cpu_associativity_change_counters(void)
{
int cpu;
/* The VPHN feature supports a maximum of 8 reference points */
BUILD_BUG_ON(MAX_DISTANCE_REF_POINTS > 8);
for_each_possible_cpu(cpu) {
int i;
u8 *counts = vphn_cpu_change_counts[cpu];
volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts;
for (i = 0; i < distance_ref_points_depth; i++)
counts[i] = hypervisor_counts[i];
}
}
/*
* The hypervisor maintains a set of 8 associativity change counters in
* the VPA of each cpu that correspond to the associativity levels in the
* ibm,associativity-reference-points property. When an associativity
* level changes, the corresponding counter is incremented.
*
* Set a bit in cpu_associativity_changes_mask for each cpu whose home
* node associativity levels have changed.
*
* Returns the number of cpus with unhandled associativity changes.
*/
static int update_cpu_associativity_changes_mask(void)
{
int cpu;
cpumask_t *changes = &cpu_associativity_changes_mask;
for_each_possible_cpu(cpu) {
int i, changed = 0;
u8 *counts = vphn_cpu_change_counts[cpu];
volatile u8 *hypervisor_counts = lppaca[cpu].vphn_assoc_counts;
for (i = 0; i < distance_ref_points_depth; i++) {
if (hypervisor_counts[i] != counts[i]) {
counts[i] = hypervisor_counts[i];
changed = 1;
}
}
if (changed) {
cpumask_or(changes, changes, cpu_sibling_mask(cpu));
cpu = cpu_last_thread_sibling(cpu);
}
}
return cpumask_weight(changes);
}
/*
* 6 64-bit registers unpacked into 12 32-bit associativity values. To form
* the complete property we have to add the length in the first cell.
*/
#define VPHN_ASSOC_BUFSIZE (6*sizeof(u64)/sizeof(u32) + 1)
/*
* Convert the associativity domain numbers returned from the hypervisor
* to the sequence they would appear in the ibm,associativity property.
*/
static int vphn_unpack_associativity(const long *packed, __be32 *unpacked)
{
int i, nr_assoc_doms = 0;
const __be16 *field = (const __be16 *) packed;
#define VPHN_FIELD_UNUSED (0xffff)
#define VPHN_FIELD_MSB (0x8000)
#define VPHN_FIELD_MASK (~VPHN_FIELD_MSB)
for (i = 1; i < VPHN_ASSOC_BUFSIZE; i++) {
if (be16_to_cpup(field) == VPHN_FIELD_UNUSED) {
/* All significant fields processed, and remaining
* fields contain the reserved value of all 1's.
* Just store them.
*/
unpacked[i] = *((__be32 *)field);
field += 2;
} else if (be16_to_cpup(field) & VPHN_FIELD_MSB) {
/* Data is in the lower 15 bits of this field */
unpacked[i] = cpu_to_be32(
be16_to_cpup(field) & VPHN_FIELD_MASK);
field++;
nr_assoc_doms++;
} else {
/* Data is in the lower 15 bits of this field
* concatenated with the next 16 bit field
*/
unpacked[i] = *((__be32 *)field);
field += 2;
nr_assoc_doms++;
}
}
/* The first cell contains the length of the property */
unpacked[0] = cpu_to_be32(nr_assoc_doms);
return nr_assoc_doms;
}
/*
* Retrieve the new associativity information for a virtual processor's
* home node.
*/
static long hcall_vphn(unsigned long cpu, __be32 *associativity)
{
long rc;
long retbuf[PLPAR_HCALL9_BUFSIZE] = {0};
u64 flags = 1;
int hwcpu = get_hard_smp_processor_id(cpu);
rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu);
vphn_unpack_associativity(retbuf, associativity);
return rc;
}
static long vphn_get_associativity(unsigned long cpu,
__be32 *associativity)
{
long rc;
rc = hcall_vphn(cpu, associativity);
switch (rc) {
case H_FUNCTION:
printk(KERN_INFO
"VPHN is not supported. Disabling polling...\n");
stop_topology_update();
break;
case H_HARDWARE:
printk(KERN_ERR
"hcall_vphn() experienced a hardware fault "
"preventing VPHN. Disabling polling...\n");
stop_topology_update();
}
return rc;
}
/*
* Update the CPU maps and sysfs entries for a single CPU when its NUMA
* characteristics change. This function doesn't perform any locking and is
* only safe to call from stop_machine().
*/
static int update_cpu_topology(void *data)
{
struct topology_update_data *update;
unsigned long cpu;
if (!data)
return -EINVAL;
cpu = smp_processor_id();
for (update = data; update; update = update->next) {
if (cpu != update->cpu)
continue;
unmap_cpu_from_node(update->cpu);
map_cpu_to_node(update->cpu, update->new_nid);
vdso_getcpu_init();
}
return 0;
}
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
static int update_lookup_table(void *data)
{
struct topology_update_data *update;
if (!data)
return -EINVAL;
/*
* Upon topology update, the numa-cpu lookup table needs to be updated
* for all threads in the core, including offline CPUs, to ensure that
* future hotplug operations respect the cpu-to-node associativity
* properly.
*/
for (update = data; update; update = update->next) {
int nid, base, j;
nid = update->new_nid;
base = cpu_first_thread_sibling(update->cpu);
for (j = 0; j < threads_per_core; j++) {
update_numa_cpu_lookup_table(base + j, nid);
}
}
return 0;
}
/*
* Update the node maps and sysfs entries for each cpu whose home node
* has changed. Returns 1 when the topology has changed, and 0 otherwise.
*/
int arch_update_cpu_topology(void)
{
unsigned int cpu, sibling, changed = 0;
struct topology_update_data *updates, *ud;
__be32 associativity[VPHN_ASSOC_BUFSIZE] = {0};
cpumask_t updated_cpus;
cpu: convert 'cpu' and 'machinecheck' sysdev_class to a regular subsystem This moves the 'cpu sysdev_class' over to a regular 'cpu' subsystem and converts the devices to regular devices. The sysdev drivers are implemented as subsystem interfaces now. After all sysdev classes are ported to regular driver core entities, the sysdev implementation will be entirely removed from the kernel. Userspace relies on events and generic sysfs subsystem infrastructure from sysdev devices, which are made available with this conversion. Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Hans-Christian Egtvedt <egtvedt@samfundet.no> Cc: Tony Luck <tony.luck@intel.com> Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: "David S. Miller" <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Borislav Petkov <bp@amd64.org> Cc: Tigran Aivazian <tigran@aivazian.fsnet.co.uk> Cc: Len Brown <lenb@kernel.org> Cc: Zhang Rui <rui.zhang@intel.com> Cc: Dave Jones <davej@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Russell King <rmk+kernel@arm.linux.org.uk> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: "Rafael J. Wysocki" <rjw@sisk.pl> Cc: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Kay Sievers <kay.sievers@vrfy.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-12-21 23:29:42 +01:00
struct device *dev;
int weight, new_nid, i = 0;
weight = cpumask_weight(&cpu_associativity_changes_mask);
if (!weight)
return 0;
updates = kzalloc(weight * (sizeof(*updates)), GFP_KERNEL);
if (!updates)
return 0;
cpumask_clear(&updated_cpus);
for_each_cpu(cpu, &cpu_associativity_changes_mask) {
/*
* If siblings aren't flagged for changes, updates list
* will be too short. Skip on this update and set for next
* update.
*/
if (!cpumask_subset(cpu_sibling_mask(cpu),
&cpu_associativity_changes_mask)) {
pr_info("Sibling bits not set for associativity "
"change, cpu%d\n", cpu);
cpumask_or(&cpu_associativity_changes_mask,
&cpu_associativity_changes_mask,
cpu_sibling_mask(cpu));
cpu = cpu_last_thread_sibling(cpu);
continue;
}
/* Use associativity from first thread for all siblings */
vphn_get_associativity(cpu, associativity);
new_nid = associativity_to_nid(associativity);
if (new_nid < 0 || !node_online(new_nid))
new_nid = first_online_node;
if (new_nid == numa_cpu_lookup_table[cpu]) {
cpumask_andnot(&cpu_associativity_changes_mask,
&cpu_associativity_changes_mask,
cpu_sibling_mask(cpu));
cpu = cpu_last_thread_sibling(cpu);
continue;
}
for_each_cpu(sibling, cpu_sibling_mask(cpu)) {
ud = &updates[i++];
ud->cpu = sibling;
ud->new_nid = new_nid;
ud->old_nid = numa_cpu_lookup_table[sibling];
cpumask_set_cpu(sibling, &updated_cpus);
if (i < weight)
ud->next = &updates[i];
}
cpu = cpu_last_thread_sibling(cpu);
}
power, sched: stop updating inside arch_update_cpu_topology() when nothing to be update Since v1: Edited the comment according to Srivatsa's suggestion. During the testing, we encounter below WARN followed by Oops: WARNING: at kernel/sched/core.c:6218 ... NIP [c000000000101660] .build_sched_domains+0x11d0/0x1200 LR [c000000000101358] .build_sched_domains+0xec8/0x1200 PACATMSCRATCH [800000000000f032] Call Trace: [c00000001b103850] [c000000000101358] .build_sched_domains+0xec8/0x1200 [c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510 [c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0 [c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30 ... Oops: Kernel access of bad area, sig: 11 [#1] ... NIP [c00000000045c000] .__bitmap_weight+0x60/0xf0 LR [c00000000010132c] .build_sched_domains+0xe9c/0x1200 PACATMSCRATCH [8000000000029032] Call Trace: [c00000001b1037a0] [c000000000288ff4] .kmem_cache_alloc_node_trace+0x184/0x3a0 [c00000001b103850] [c00000000010132c] .build_sched_domains+0xe9c/0x1200 [c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510 [c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0 [c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30 ... This was caused by that 'sd->groups == NULL' after building groups, which was caused by the empty 'sd->span'. The cpu's domain contained nothing because the cpu was assigned to a wrong node, due to the following unfortunate sequence of events: 1. The hypervisor sent a topology update to the guest OS, to notify changes to the cpu-node mapping. However, the update was actually redundant - i.e., the "new" mapping was exactly the same as the old one. 2. Due to this, the 'updated_cpus' mask turned out to be empty after exiting the 'for-loop' in arch_update_cpu_topology(). 3. So we ended up calling stop-machine() with an empty cpumask list, which made stop-machine internally elect cpumask_first(cpu_online_mask), i.e., CPU0 as the cpu to run the payload (the update_cpu_topology() function). 4. This causes update_cpu_topology() to be run by CPU0. And since 'updates' is kzalloc()'ed inside arch_update_cpu_topology(), update_cpu_topology() finds update->cpu as well as update->new_nid to be 0. In other words, we end up assigning CPU0 (and eventually its siblings) to node 0, incorrectly. Along with the following wrong updating, it causes the sched-domain rebuild code to break and crash the system. Fix this by skipping the topology update in cases where we find that the topology has not actually changed in reality (ie., spurious updates). CC: Benjamin Herrenschmidt <benh@kernel.crashing.org> CC: Paul Mackerras <paulus@samba.org> CC: Nathan Fontenot <nfont@linux.vnet.ibm.com> CC: Stephen Rothwell <sfr@canb.auug.org.au> CC: Andrew Morton <akpm@linux-foundation.org> CC: Robert Jennings <rcj@linux.vnet.ibm.com> CC: Jesse Larrew <jlarrew@linux.vnet.ibm.com> CC: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> CC: Alistair Popple <alistair@popple.id.au> Suggested-by: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com> Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-08 05:19:36 +02:00
/*
* In cases where we have nothing to update (because the updates list
* is too short or because the new topology is same as the old one),
* skip invoking update_cpu_topology() via stop-machine(). This is
* necessary (and not just a fast-path optimization) since stop-machine
* can end up electing a random CPU to run update_cpu_topology(), and
* thus trick us into setting up incorrect cpu-node mappings (since
* 'updates' is kzalloc()'ed).
*
* And for the similar reason, we will skip all the following updating.
*/
if (!cpumask_weight(&updated_cpus))
goto out;
stop_machine(update_cpu_topology, &updates[0], &updated_cpus);
powerpc: Fix the setup of CPU-to-Node mappings during CPU online On POWER platforms, the hypervisor can notify the guest kernel about dynamic changes in the cpu-numa associativity (VPHN topology update). Hence the cpu-to-node mappings that we got from the firmware during boot, may no longer be valid after such updates. This is handled using the arch_update_cpu_topology() hook in the scheduler, and the sched-domains are rebuilt according to the new mappings. But unfortunately, at the moment, CPU hotplug ignores these updated mappings and instead queries the firmware for the cpu-to-numa relationships and uses them during CPU online. So the kernel can end up assigning wrong NUMA nodes to CPUs during subsequent CPU hotplug online operations (after booting). Further, a particularly problematic scenario can result from this bug: On POWER platforms, the SMT mode can be switched between 1, 2, 4 (and even 8) threads per core. The switch to Single-Threaded (ST) mode is performed by offlining all except the first CPU thread in each core. Switching back to SMT mode involves onlining those other threads back, in each core. Now consider this scenario: 1. During boot, the kernel gets the cpu-to-node mappings from the firmware and assigns the CPUs to NUMA nodes appropriately, during CPU online. 2. Later on, the hypervisor updates the cpu-to-node mappings dynamically and communicates this update to the kernel. The kernel in turn updates its cpu-to-node associations and rebuilds its sched domains. Everything is fine so far. 3. Now, the user switches the machine from SMT to ST mode (say, by running ppc64_cpu --smt=1). This involves offlining all except 1 thread in each core. 4. The user then tries to switch back from ST to SMT mode (say, by running ppc64_cpu --smt=4), and this involves onlining those threads back. Since CPU hotplug ignores the new mappings, it queries the firmware and tries to associate the newly onlined sibling threads to the old NUMA nodes. This results in sibling threads within the same core getting associated with different NUMA nodes, which is incorrect. The scheduler's build-sched-domains code gets thoroughly confused with this and enters an infinite loop and causes soft-lockups, as explained in detail in commit 3be7db6ab (powerpc: VPHN topology change updates all siblings). So to fix this, use the numa_cpu_lookup_table to remember the updated cpu-to-node mappings, and use them during CPU hotplug online operations. Further, we also need to ensure that all threads in a core are assigned to a common NUMA node, irrespective of whether all those threads were online during the topology update. To achieve this, we take care not to use cpu_sibling_mask() since it is not hotplug invariant. Instead, we use cpu_first_sibling_thread() and set up the mappings manually using the 'threads_per_core' value for that particular platform. This helps us ensure that we don't hit this bug with any combination of CPU hotplug and SMT mode switching. Cc: stable@vger.kernel.org Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-12-30 12:35:34 +01:00
/*
* Update the numa-cpu lookup table with the new mappings, even for
* offline CPUs. It is best to perform this update from the stop-
* machine context.
*/
stop_machine(update_lookup_table, &updates[0],
cpumask_of(raw_smp_processor_id()));
for (ud = &updates[0]; ud; ud = ud->next) {
unregister_cpu_under_node(ud->cpu, ud->old_nid);
register_cpu_under_node(ud->cpu, ud->new_nid);
dev = get_cpu_device(ud->cpu);
cpu: convert 'cpu' and 'machinecheck' sysdev_class to a regular subsystem This moves the 'cpu sysdev_class' over to a regular 'cpu' subsystem and converts the devices to regular devices. The sysdev drivers are implemented as subsystem interfaces now. After all sysdev classes are ported to regular driver core entities, the sysdev implementation will be entirely removed from the kernel. Userspace relies on events and generic sysfs subsystem infrastructure from sysdev devices, which are made available with this conversion. Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Hans-Christian Egtvedt <egtvedt@samfundet.no> Cc: Tony Luck <tony.luck@intel.com> Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: "David S. Miller" <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Borislav Petkov <bp@amd64.org> Cc: Tigran Aivazian <tigran@aivazian.fsnet.co.uk> Cc: Len Brown <lenb@kernel.org> Cc: Zhang Rui <rui.zhang@intel.com> Cc: Dave Jones <davej@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Russell King <rmk+kernel@arm.linux.org.uk> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: "Rafael J. Wysocki" <rjw@sisk.pl> Cc: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Kay Sievers <kay.sievers@vrfy.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2011-12-21 23:29:42 +01:00
if (dev)
kobject_uevent(&dev->kobj, KOBJ_CHANGE);
cpumask_clear_cpu(ud->cpu, &cpu_associativity_changes_mask);
changed = 1;
}
power, sched: stop updating inside arch_update_cpu_topology() when nothing to be update Since v1: Edited the comment according to Srivatsa's suggestion. During the testing, we encounter below WARN followed by Oops: WARNING: at kernel/sched/core.c:6218 ... NIP [c000000000101660] .build_sched_domains+0x11d0/0x1200 LR [c000000000101358] .build_sched_domains+0xec8/0x1200 PACATMSCRATCH [800000000000f032] Call Trace: [c00000001b103850] [c000000000101358] .build_sched_domains+0xec8/0x1200 [c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510 [c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0 [c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30 ... Oops: Kernel access of bad area, sig: 11 [#1] ... NIP [c00000000045c000] .__bitmap_weight+0x60/0xf0 LR [c00000000010132c] .build_sched_domains+0xe9c/0x1200 PACATMSCRATCH [8000000000029032] Call Trace: [c00000001b1037a0] [c000000000288ff4] .kmem_cache_alloc_node_trace+0x184/0x3a0 [c00000001b103850] [c00000000010132c] .build_sched_domains+0xe9c/0x1200 [c00000001b1039a0] [c00000000010aad4] .partition_sched_domains+0x484/0x510 [c00000001b103aa0] [c00000000016d0a8] .rebuild_sched_domains+0x68/0xa0 [c00000001b103b30] [c00000000005cbf0] .topology_work_fn+0x10/0x30 ... This was caused by that 'sd->groups == NULL' after building groups, which was caused by the empty 'sd->span'. The cpu's domain contained nothing because the cpu was assigned to a wrong node, due to the following unfortunate sequence of events: 1. The hypervisor sent a topology update to the guest OS, to notify changes to the cpu-node mapping. However, the update was actually redundant - i.e., the "new" mapping was exactly the same as the old one. 2. Due to this, the 'updated_cpus' mask turned out to be empty after exiting the 'for-loop' in arch_update_cpu_topology(). 3. So we ended up calling stop-machine() with an empty cpumask list, which made stop-machine internally elect cpumask_first(cpu_online_mask), i.e., CPU0 as the cpu to run the payload (the update_cpu_topology() function). 4. This causes update_cpu_topology() to be run by CPU0. And since 'updates' is kzalloc()'ed inside arch_update_cpu_topology(), update_cpu_topology() finds update->cpu as well as update->new_nid to be 0. In other words, we end up assigning CPU0 (and eventually its siblings) to node 0, incorrectly. Along with the following wrong updating, it causes the sched-domain rebuild code to break and crash the system. Fix this by skipping the topology update in cases where we find that the topology has not actually changed in reality (ie., spurious updates). CC: Benjamin Herrenschmidt <benh@kernel.crashing.org> CC: Paul Mackerras <paulus@samba.org> CC: Nathan Fontenot <nfont@linux.vnet.ibm.com> CC: Stephen Rothwell <sfr@canb.auug.org.au> CC: Andrew Morton <akpm@linux-foundation.org> CC: Robert Jennings <rcj@linux.vnet.ibm.com> CC: Jesse Larrew <jlarrew@linux.vnet.ibm.com> CC: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> CC: Alistair Popple <alistair@popple.id.au> Suggested-by: "Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com> Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-08 05:19:36 +02:00
out:
kfree(updates);
return changed;
}
static void topology_work_fn(struct work_struct *work)
{
rebuild_sched_domains();
}
static DECLARE_WORK(topology_work, topology_work_fn);
static void topology_schedule_update(void)
{
schedule_work(&topology_work);
}
static void topology_timer_fn(unsigned long ignored)
{
if (prrn_enabled && cpumask_weight(&cpu_associativity_changes_mask))
topology_schedule_update();
else if (vphn_enabled) {
if (update_cpu_associativity_changes_mask() > 0)
topology_schedule_update();
reset_topology_timer();
}
}
static struct timer_list topology_timer =
TIMER_INITIALIZER(topology_timer_fn, 0, 0);
static void reset_topology_timer(void)
{
topology_timer.data = 0;
topology_timer.expires = jiffies + 60 * HZ;
mod_timer(&topology_timer, topology_timer.expires);
}
#ifdef CONFIG_SMP
static void stage_topology_update(int core_id)
{
cpumask_or(&cpu_associativity_changes_mask,
&cpu_associativity_changes_mask, cpu_sibling_mask(core_id));
reset_topology_timer();
}
static int dt_update_callback(struct notifier_block *nb,
unsigned long action, void *data)
{
struct of_prop_reconfig *update;
int rc = NOTIFY_DONE;
switch (action) {
case OF_RECONFIG_UPDATE_PROPERTY:
update = (struct of_prop_reconfig *)data;
if (!of_prop_cmp(update->dn->type, "cpu") &&
!of_prop_cmp(update->prop->name, "ibm,associativity")) {
u32 core_id;
of_property_read_u32(update->dn, "reg", &core_id);
stage_topology_update(core_id);
rc = NOTIFY_OK;
}
break;
}
return rc;
}
static struct notifier_block dt_update_nb = {
.notifier_call = dt_update_callback,
};
#endif
/*
* Start polling for associativity changes.
*/
int start_topology_update(void)
{
int rc = 0;
if (firmware_has_feature(FW_FEATURE_PRRN)) {
if (!prrn_enabled) {
prrn_enabled = 1;
vphn_enabled = 0;
#ifdef CONFIG_SMP
rc = of_reconfig_notifier_register(&dt_update_nb);
#endif
}
} else if (firmware_has_feature(FW_FEATURE_VPHN) &&
lppaca_shared_proc(get_lppaca())) {
if (!vphn_enabled) {
prrn_enabled = 0;
vphn_enabled = 1;
setup_cpu_associativity_change_counters();
init_timer_deferrable(&topology_timer);
reset_topology_timer();
}
}
return rc;
}
/*
* Disable polling for VPHN associativity changes.
*/
int stop_topology_update(void)
{
int rc = 0;
if (prrn_enabled) {
prrn_enabled = 0;
#ifdef CONFIG_SMP
rc = of_reconfig_notifier_unregister(&dt_update_nb);
#endif
} else if (vphn_enabled) {
vphn_enabled = 0;
rc = del_timer_sync(&topology_timer);
}
return rc;
}
int prrn_is_enabled(void)
{
return prrn_enabled;
}
static int topology_read(struct seq_file *file, void *v)
{
if (vphn_enabled || prrn_enabled)
seq_puts(file, "on\n");
else
seq_puts(file, "off\n");
return 0;
}
static int topology_open(struct inode *inode, struct file *file)
{
return single_open(file, topology_read, NULL);
}
static ssize_t topology_write(struct file *file, const char __user *buf,
size_t count, loff_t *off)
{
char kbuf[4]; /* "on" or "off" plus null. */
int read_len;
read_len = count < 3 ? count : 3;
if (copy_from_user(kbuf, buf, read_len))
return -EINVAL;
kbuf[read_len] = '\0';
if (!strncmp(kbuf, "on", 2))
start_topology_update();
else if (!strncmp(kbuf, "off", 3))
stop_topology_update();
else
return -EINVAL;
return count;
}
static const struct file_operations topology_ops = {
.read = seq_read,
.write = topology_write,
.open = topology_open,
.release = single_release
};
static int topology_update_init(void)
{
start_topology_update();
proc_create("powerpc/topology_updates", 0644, NULL, &topology_ops);
return 0;
}
device_initcall(topology_update_init);
#endif /* CONFIG_PPC_SPLPAR */