On Mon, Jun 29, 2015 at 04:53:12PM +0200, Igor Mammedov wrote:
> On Mon, 29 Jun 2015 19:11:30 +0530
> Bharata B Rao <bhar...@linux.vnet.ibm.com> wrote:
>
> > On Mon, Jun 29, 2015 at 02:08:20PM +0200, Igor Mammedov wrote:
> > > On Mon, 29 Jun 2015 13:50:25 +0530
> > > Bharata B Rao <bhar...@linux.vnet.ibm.com> wrote:
> > >
> > > > Start storing the (start_addr, end_addr) of the pc-dimm memory
> > > > in corresponding numa_info[node] so that this information can be used
> > > > to lookup node by address.
> > > >
> > > > Signed-off-by: Bharata B Rao <bhar...@linux.vnet.ibm.com>
> > > Reviewed-by: Igor Mammedov <imamm...@redhat.com>
> > >
> > > > ---
> > > > hw/mem/pc-dimm.c | 4 ++++
> > > > include/sysemu/numa.h | 10 ++++++++++
> > > > numa.c | 26 ++++++++++++++++++++++++++
> > > > 3 files changed, 40 insertions(+)
> > > >
> > > > diff --git a/hw/mem/pc-dimm.c b/hw/mem/pc-dimm.c
> > > > index 98971b7..bb04862 100644
> > > > --- a/hw/mem/pc-dimm.c
> > > > +++ b/hw/mem/pc-dimm.c
> > > > @@ -97,6 +97,7 @@ void pc_dimm_memory_plug(DeviceState *dev,
> > > > MemoryHotplugState *hpms,
> > > >
> > > > memory_region_add_subregion(&hpms->mr, addr - hpms->base, mr);
> > > > vmstate_register_ram(mr, dev);
> > > > + numa_set_mem_node_id(addr, memory_region_size(mr), dimm->node);
> > > >
> > > > out:
> > > > error_propagate(errp, local_err);
> > > > @@ -105,6 +106,9 @@ out:
> > > > void pc_dimm_memory_unplug(DeviceState *dev, MemoryHotplugState *hpms,
> > > > MemoryRegion *mr)
> > > > {
> > > > + PCDIMMDevice *dimm = PC_DIMM(dev);
> > > > +
> > > > + numa_unset_mem_node_id(dimm->addr, memory_region_size(mr),
> > > > dimm->node);
> > > > memory_region_del_subregion(&hpms->mr, mr);
> > > > vmstate_unregister_ram(mr, dev);
> > > > }
> > > > diff --git a/include/sysemu/numa.h b/include/sysemu/numa.h
> > > > index 6523b4d..7176364 100644
> > > > --- a/include/sysemu/numa.h
> > > > +++ b/include/sysemu/numa.h
> > > > @@ -10,16 +10,26 @@
> > > >
> > > > extern int nb_numa_nodes; /* Number of NUMA nodes */
> > > >
> > > > +struct numa_addr_range {
> > > > + ram_addr_t mem_start;
> > > > + ram_addr_t mem_end;
> > > > + QLIST_ENTRY(numa_addr_range) entry;
> > > > +};
> > > > +
> > > > typedef struct node_info {
> > > > uint64_t node_mem;
> > > > DECLARE_BITMAP(node_cpu, MAX_CPUMASK_BITS);
> > > > struct HostMemoryBackend *node_memdev;
> > > > bool present;
> > > > + QLIST_HEAD(, numa_addr_range) addr; /* List to store address
> > > > ranges */
> > > > } NodeInfo;
> > > > +
> > > > extern NodeInfo numa_info[MAX_NODES];
> > > > void parse_numa_opts(MachineClass *mc);
> > > > void numa_post_machine_init(void);
> > > > void query_numa_node_mem(uint64_t node_mem[]);
> > > > extern QemuOptsList qemu_numa_opts;
> > > > +void numa_set_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t
> > > > node);
> > > > +void numa_unset_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t
> > > > node);
> > > >
> > > > #endif
> > > > diff --git a/numa.c b/numa.c
> > > > index 91fc6c1..116d1fb 100644
> > > > --- a/numa.c
> > > > +++ b/numa.c
> > > > @@ -52,6 +52,28 @@ static int max_numa_nodeid; /* Highest specified
> > > > NUMA node ID, plus one.
> > > > int nb_numa_nodes;
> > > > NodeInfo numa_info[MAX_NODES];
> > > >
> > > > +void numa_set_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t
> > > > node)
> > > > +{
> > > > + struct numa_addr_range *range = g_malloc0(sizeof(*range));
> > > > +
> > > > + range->mem_start = addr;
> > > > + range->mem_end = addr + size - 1;
> > > nit:
> > > as a patch on top of it, add asserts that check for overflow, pls
> >
> > You suggested g_assert(size) in the previous version.
> >
> > However size can be zero when this API is called for boot time memory
> > and I have taken care of that in the next patch (5/6).
> >
> > And for pc-dimm memory, the size can never be zero.
> >
> > So do you still think overflow is possible ?
> make build on 32-bit host and theoretically with ram_addr_t==32bits + size ==
> uint64_t it could be possible,
>
> overflow check here doesn't harm in any way and saves headache of debugging
> when overflow happens.
Something like below which applies on Eduardo's numa branch ?
numa: Check for overflow in numa_[un]set_mem_node_id()
From: Bharata B Rao <bhar...@linux.vnet.ibm.com>
Assert when (addr + size) calculation overflows.
Signed-off-by: Bharata B Rao <bharata....@linux.vnet.ibm.com>
---
numa.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/numa.c b/numa.c
index 3c80059..c2fc7aa 100644
--- a/numa.c
+++ b/numa.c
@@ -65,16 +65,21 @@ void numa_set_mem_node_id(ram_addr_t addr, uint64_t size,
uint32_t node)
}
range->mem_start = addr;
- range->mem_end = addr + size - 1;
+ range->mem_end = addr + size;
+ g_assert(range->mem_end >= MAX(addr, size));
+ range->mem_end -= 1;
QLIST_INSERT_HEAD(&numa_info[node].addr, range, entry);
}
void numa_unset_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t node)
{
struct numa_addr_range *range, *next;
+ ram_addr_t end = addr + size;
+ g_assert(end >= MAX(addr, size));
+ end -= 1;
QLIST_FOREACH_SAFE(range, &numa_info[node].addr, entry, next) {
- if (addr == range->mem_start && (addr + size - 1) == range->mem_end) {
+ if (addr == range->mem_start && end == range->mem_end) {
QLIST_REMOVE(range, entry);
g_free(range);
return;
