Re: [PATCH 2/3] of: Support hashtable lookups for phandles

[Pantelis Antoniou]

Hi Rob,

> On May 10, 2016, at 00:21 , Rob Herring <robherring2@xxxxxxxxx> wrote:
> 
> On Mon, May 9, 2016 at 1:11 PM, Pantelis Antoniou
> <pantelis.antoniou@xxxxxxxxxxxx> wrote:
> 
> Why this is needed goes here.
> 
> Any data on how much time the current code takes?
> 

Iâll get some numbers. The current algorithm is exhaustive search of
all nodes (not even nodes that have phandles). So itâs kind of obvious
bad on large blobs.

>> Signed-off-by: Pantelis Antoniou <pantelis.antoniou@xxxxxxxxxxxx>
>> ---
>> drivers/of/base.c       | 35 ++++++++++++++++++++++++++++++++---
>> drivers/of/dynamic.c    |  8 ++++++++
>> drivers/of/of_private.h | 31 +++++++++++++++++++++++++++++++
>> include/linux/of.h      |  2 ++
>> 4 files changed, 73 insertions(+), 3 deletions(-)
>> 
>> diff --git a/drivers/of/base.c b/drivers/of/base.c
>> index 20bbc2f..770cb95 100644
>> --- a/drivers/of/base.c
>> +++ b/drivers/of/base.c
>> @@ -27,6 +27,7 @@
>> #include <linux/slab.h>
>> #include <linux/string.h>
>> #include <linux/proc_fs.h>
>> +#include <linux/rhashtable.h>
>> 
>> #include "of_private.h"
>> 
>> @@ -41,6 +42,16 @@ static const char *of_stdout_options;
>> 
>> struct kset *of_kset;
>> 
>> +const struct rhashtable_params of_phandle_ht_params = {
>> +       .key_offset = offsetof(struct device_node, phandle), /* base offset */
>> +       .key_len = sizeof(phandle),
>> +       .head_offset = offsetof(struct device_node, ht_node),
>> +       .automatic_shrinking = true,
>> +};
>> +
>> +struct rhashtable of_phandle_ht;
>> +bool of_phandle_ht_initialized;
> 
> Get rid of this and of_phandle_ht_available. It should always be initialized.
> 

Not that simple. We have to support the case where calls are made very early
in the boot sequence before the initialization.

>> +
>> /*
>>  * Used to protect the of_aliases, to hold off addition of nodes to sysfs.
>>  * This mutex must be held whenever modifications are being made to the
>> @@ -162,6 +173,12 @@ int __of_attach_node_post(struct device_node *np)
>>        struct property *pp;
>>        int rc;
>> 
>> +       if (of_phandle_ht_available()) {
>> +               rc = of_phandle_ht_insert(np);
>> +               WARN(rc, "insert to phandle hash fail @%s\n",
>> +                               of_node_full_name(np));
>> +       }
>> +
>>        if (!IS_ENABLED(CONFIG_SYSFS))
>>                return 0;
>> 
>> @@ -194,6 +211,13 @@ void __init of_core_init(void)
>>        struct device_node *np;
>>        int ret;
>> 
>> +       ret = rhashtable_init(&of_phandle_ht, &of_phandle_ht_params);
>> +       if (ret) {
>> +               pr_warn("devicetree: Failed to initialize hashtable\n");
>> +               return;
>> +       }
>> +       of_phandle_ht_initialized = 1;
>> +
>>        /* Create the kset, and register existing nodes */
>>        mutex_lock(&of_mutex);
>>        of_kset = kset_create_and_add("devicetree", NULL, firmware_kobj);
>> @@ -1073,9 +1097,14 @@ struct device_node *of_find_node_by_phandle(phandle handle)
>>                return NULL;
>> 
>>        raw_spin_lock_irqsave(&devtree_lock, flags);
>> -       for_each_of_allnodes(np)
>> -               if (np->phandle == handle)
>> -                       break;
>> +       /* when we're ready use the hash table */
>> +       if (of_phandle_ht_available() && !in_interrupt())
>> +               np = of_phandle_ht_lookup(handle);
>> +       else { /* fallback */
>> +               for_each_of_allnodes(np)
>> +                       if (np->phandle == handle)
>> +                               break;
>> +       }
>>        of_node_get(np);
>>        raw_spin_unlock_irqrestore(&devtree_lock, flags);
>>        return np;
>> diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
>> index eb31a55..bdebdf5 100644
>> --- a/drivers/of/dynamic.c
>> +++ b/drivers/of/dynamic.c
>> @@ -11,6 +11,7 @@
>> #include <linux/slab.h>
>> #include <linux/string.h>
>> #include <linux/proc_fs.h>
>> +#include <linux/rhashtable.h>
>> 
>> #include "of_private.h"
>> 
>> @@ -44,6 +45,13 @@ EXPORT_SYMBOL(of_node_put);
>> void __of_detach_node_post(struct device_node *np)
>> {
>>        struct property *pp;
>> +       int rc;
>> +
>> +       if (of_phandle_ht_available()) {
>> +               rc = of_phandle_ht_remove(np);
>> +               WARN(rc, "remove from phandle hash fail @%s\n",
>> +                               of_node_full_name(np));
> 
> Can this really fail?
> 

No, unless the structure is corrupted. I can move it under DEBUG.

> Rob
> â

Regards

â Pantelis