[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [RFC PATCH v1 07/21] ARM: NUMA: Parse memory NUMA information



Hello Vijay,

On 09/02/17 15:56, vijay.kilari@xxxxxxxxx wrote:
From: Vijaya Kumar K <Vijaya.Kumar@xxxxxxxxxx>

Parse memory node and fetch numa-node-id information.
For each memory range, store in node_memblk_range[]
along with node id.

Signed-off-by: Vijaya Kumar K <Vijaya.Kumar@xxxxxxxxxx>
---
 xen/arch/arm/bootfdt.c        |  4 +--
 xen/arch/arm/dt_numa.c        | 84 ++++++++++++++++++++++++++++++++++++++++++-
 xen/common/numa.c             |  8 +++++
 xen/include/xen/device_tree.h |  3 ++
 xen/include/xen/numa.h        |  1 +
 5 files changed, 97 insertions(+), 3 deletions(-)

diff --git a/xen/arch/arm/bootfdt.c b/xen/arch/arm/bootfdt.c
index d1122d8..5e2df92 100644
--- a/xen/arch/arm/bootfdt.c
+++ b/xen/arch/arm/bootfdt.c
@@ -56,8 +56,8 @@ static bool_t __init device_tree_node_compatible(const void 
*fdt, int node,
     return 0;
 }

-static void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                       u32 size_cells, u64 *start, u64 *size)
+void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
+                                u32 size_cells, u64 *start, u64 *size)
 {
     *start = dt_next_cell(address_cells, cell);
     *size = dt_next_cell(size_cells, cell);
diff --git a/xen/arch/arm/dt_numa.c b/xen/arch/arm/dt_numa.c
index 4b94c36..fce9e67 100644
--- a/xen/arch/arm/dt_numa.c
+++ b/xen/arch/arm/dt_numa.c
@@ -27,6 +27,7 @@
 #include <xen/numa.h>

 nodemask_t numa_nodes_parsed;
+extern struct node node_memblk_range[NR_NODE_MEMBLKS];

This should have been declared in an header (likely in patch #3).


 /*
  * Even though we connect cpus to numa domains later in SMP
@@ -48,11 +49,73 @@ static int __init dt_numa_process_cpu_node(const void *fdt, 
int node,
     return 0;
 }

+static int __init dt_numa_process_memory_node(const void *fdt, int node,
+                                              const char *name,
+                                              u32 address_cells,
+                                              u32 size_cells)

Rather than reimplementing the wheel, it might be better to hook the parsing in bootfdt.c. This would avoid an extra parsing of the device-tree, duplicate the code and expose primitive for early DT parsing.

If parsing NUMA information can be done after the DT has been unflattened, this will be even better.

+{
+    const struct fdt_property *prop;
+    int i, ret, banks;

Both banks and i should be unsigned.

+    const __be32 *cell;
+    paddr_t start, size;
+    u32 reg_cells = address_cells + size_cells;
+    u32 nid;
+
+    if ( address_cells < 1 || size_cells < 1 )
+    {
+        printk(XENLOG_WARNING
+               "fdt: node `%s': invalid #address-cells or #size-cells", name);
+        return -EINVAL;
+    }
+
+    nid = device_tree_get_u32(fdt, node, "numa-node-id", MAX_NUMNODES);
+    if ( nid >= MAX_NUMNODES) {

Coding style

if ( ... )
{

+        /*
+         * No node id found. Skip this memory node.
+         */

This could be a single line:

/* ..... */

So no warning if it is impossible to get the numa-node-id? Also, I don't think this is right to boot using NUMA on platform having a buggy DT. So we should probably return an error and disable NUMA.

+        return 0;
+    }
+
+    prop = fdt_get_property(fdt, node, "reg", NULL);
+    if ( !prop )
+    {
+        printk(XENLOG_WARNING "fdt: node `%s': missing `reg' property\n",
+               name);
+        return -EINVAL;
+    }
+
+    cell = (const __be32 *)prop->data;
+    banks = fdt32_to_cpu(prop->len) / (reg_cells * sizeof (u32));
+
+    for ( i = 0; i < banks; i++ )
+    {
+        device_tree_get_reg(&cell, address_cells, size_cells, &start, &size);
+        if ( !size )
+            continue;
+
+        /* It is fine to add this area to the nodes data it will be used 
later*/
+        ret = conflicting_memblks(start, start + size);
+        if (ret < 0)
+             numa_add_memblk(nid, start, size);

numa_add_memblk rely on the caller to check whether the array is not full. I think we should move the check in numa_add_memblk and return an error in this case.

+        else
+        {
+             printk(XENLOG_ERR
+                    "NUMA DT: node %u (%"PRIx64"-%"PRIx64") overlaps with ret %d 
(%"PRIx64"-%"PRIx64")\n",
+                    nid, start, start + size, ret,
+                    node_memblk_range[i].start, node_memblk_range[i].end);

i != ret. Please use the correct variable accordingly. However, I don't think the overlap range really matters here.

+             return -EINVAL;
+        }
+    }
+
+    node_set(nid, numa_nodes_parsed);
+
+    return 0;
+}
+
 static int __init dt_numa_scan_cpu_node(const void *fdt, int node,
                                         const char *name, int depth,
                                         u32 address_cells, u32 size_cells,
                                         void *data)
-

Spurious change. Please don't add the blank line at the first place (patch #6).

 {
     if ( device_tree_node_matches(fdt, node, "cpu") )
         return dt_numa_process_cpu_node(fdt, node, name, address_cells,
@@ -61,6 +124,18 @@ static int __init dt_numa_scan_cpu_node(const void *fdt, 
int node,
     return 0;
 }

+static int __init dt_numa_scan_memory_node(const void *fdt, int node,
+                                           const char *name, int depth,
+                                           u32 address_cells, u32 size_cells,
+                                           void *data)
+{
+    if ( device_tree_node_matches(fdt, node, "memory") )
+        return dt_numa_process_memory_node(fdt, node, name, address_cells,
+                                           size_cells);

Similarly to the CPUs, this code is wrong. You should check the type = "memory".

+
+    return 0;
+}
+
 int __init dt_numa_init(void)
 {
     int ret;
@@ -68,5 +143,12 @@ int __init dt_numa_init(void)
     nodes_clear(numa_nodes_parsed);
     ret = device_tree_for_each_node((void *)device_tree_flattened,
                                     dt_numa_scan_cpu_node, NULL);
+
+    if ( ret )
+        return ret;
+
+    ret = device_tree_for_each_node((void *)device_tree_flattened,
+                                    dt_numa_scan_memory_node, NULL);
+
     return ret;
 }
diff --git a/xen/common/numa.c b/xen/common/numa.c
index 9b9cf9c..62c76af 100644
--- a/xen/common/numa.c
+++ b/xen/common/numa.c
@@ -55,6 +55,14 @@ struct node node_memblk_range[NR_NODE_MEMBLKS];
 nodeid_t memblk_nodeid[NR_NODE_MEMBLKS];
 struct node nodes[MAX_NUMNODES] __initdata;

+void __init numa_add_memblk(nodeid_t nodeid, u64 start, u64 size)

Please replace u64 by paddr_t.

+{
+    node_memblk_range[num_node_memblks].start = start;
+    node_memblk_range[num_node_memblks].end = start + size;
+    memblk_nodeid[num_node_memblks] = nodeid;
+    num_node_memblks++;
+}

You probably want to factor the code in acpi_numa_memory_affinity_init to create this function.

Also, you don't check if the array is full.

+
 int valid_numa_range(u64 start, u64 end, nodeid_t node)
 {
 #ifdef CONFIG_NUMA
diff --git a/xen/include/xen/device_tree.h b/xen/include/xen/device_tree.h
index de6b351..d92e47e 100644
--- a/xen/include/xen/device_tree.h
+++ b/xen/include/xen/device_tree.h
@@ -192,6 +192,9 @@ bool_t device_tree_node_matches(const void *fdt, int node,
                                 const char *match);
 u32 device_tree_get_u32(const void *fdt, int node,
                         const char *prop_name, u32 dflt);
+void device_tree_get_reg(const __be32 **cell, u32 address_cells,
+                         u32 size_cells, u64 *start, u64 *size);
+

Same remark as on patch #6 for the position of the declaration.

 /**
  * dt_unflatten_host_device_tree - Unflatten the host device tree
  *
diff --git a/xen/include/xen/numa.h b/xen/include/xen/numa.h
index 77c5cfd..9392a89 100644
--- a/xen/include/xen/numa.h
+++ b/xen/include/xen/numa.h
@@ -67,6 +67,7 @@ static inline __attribute__((pure)) nodeid_t 
phys_to_nid(paddr_t addr)
 #define clear_node_cpumask(cpu) do {} while (0)
 #endif /* CONFIG_NUMA */

+extern void numa_add_memblk(nodeid_t nodeid, u64 start, u64 size);
 extern int valid_numa_range(u64 start, u64 end, nodeid_t node);
 extern int conflicting_memblks(u64 start, u64 end);
 extern void cutoff_node(int i, u64 start, u64 end);


Regards,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.