Date:   Tue, 16 Oct 2018 15:04:11 -0500
From:   Alan Tull <atull@...nel.org>
To:     Frank Rowand <frowand.list@...il.com>
Cc:     Rob Herring <robh+dt@...nel.org>,
        Pantelis Antoniou <pantelis.antoniou@...sulko.com>,
        Pantelis Antoniou <panto@...oniou-consulting.com>,
        "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" 
        <devicetree@...r.kernel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Geert Uytterhoeven <geert@...ux-m68k.org>
Subject: Re: [PATCH v2] of: overlay: user space synchronization

On Mon, Oct 15, 2018 at 7:28 PM <frowand.list@...il.com> wrote:

Hi Frank,

Thanks for all your work on this!

> From: Frank Rowand <frank.rowand@...y.com>
>
> When an overlay is applied or removed, the live devicetree visible in
> /proc/device-tree/, aka /sys/firmware/devicetree/base/, reflects the
> changes.  There is no method for user space to determine whether the
> live devicetree was modified by overlay actions.
>
> Provide a sysfs file, /sys/firmware/devicetree/tree_version,  to allow
> user space to determine if the live devicetree has remained unchanged
> while a series of one or more accesses of /proc/device-tree/ occur.
>
> The use of both dynamic devicetree modifications and overlay apply and
> removal are not supported during the same boot cycle.  Thus non-overlay
> dynamic modifications are not reflected in the value of tree_version.
>
> Example shell use:
>
>         done=1
>

Suggest a few comments, such as:

+ # Keep trying until we can make it through the loop without live
tree being changed
+ # by an overlay during the 'critical region'

>         while [ $done = 1 ] ; do

Nit: doesn't $done mean 'not done' in this script (assuming True is 1)?

>
>                 pre_version=$(cat /sys/firmware/devicetree/tree_version)

Suggest:
+ # loop while live tree is locked for overlay changes

>                 while [ $(( ${pre_version} & 1 )) != 0 ] ; do
>                         # echo is optional, sleep value can be tuned
>                         echo "${pre_version}  locked, sleeping 2 seconds"
>                         sleep 2;
>                         pre_version=$(cat /sys/firmware/devicetree/tree_version)
>                         version=${pre_version}

Unused 'version' variable

>                 done
>
>                 # 'critical region'
>                 # access /proc/device-tree/ one or more times
>
>                 post_version=$(cat /sys/firmware/devicetree/tree_version)
>

+ # Final check that DT wasn't possibly overlaid during the critical region

>                 if [ ${post_version} = ${pre_version} ] ; then
>                         done=0

This threw me off for a moment as I was figuring at first that setting
done = 0 meant that we weren't done, so keep looping.

>                 fi
>
>         done
>
> Signed-off-by: Frank Rowand <frank.rowand@...y.com>
> ---
>
> updates since v1:
>   - removed unneeded variable "version" from shell script
>   - explicitly state that an odd value of tree_version means the
>     devicetree is locked for an overlay update not just in the
>     source, but also in Documentation/ABI/testing/sysfs-firmware-ofw
>   - document that tree_version is reported as an ascii number, the
>     internal representation, and the resultant wrap around behavior
>
>  Documentation/ABI/testing/sysfs-firmware-ofw | 67 +++++++++++++++++++++++++---
>  drivers/of/base.c                            | 66 +++++++++++++++++++++++++++
>  drivers/of/of_private.h                      |  2 +
>  drivers/of/overlay.c                         | 12 +++++
>  4 files changed, 140 insertions(+), 7 deletions(-)
>
> diff --git a/Documentation/ABI/testing/sysfs-firmware-ofw b/Documentation/ABI/testing/sysfs-firmware-ofw
> index edcab3ccfcc0..d2346eb61924 100644
> --- a/Documentation/ABI/testing/sysfs-firmware-ofw
> +++ b/Documentation/ABI/testing/sysfs-firmware-ofw
> @@ -1,4 +1,10 @@
> -What:          /sys/firmware/devicetree/*
> +What:          /sys/firmware/devicetree/
> +Date:          November 2013
> +Contact:       Frank Rowand <frowand.list@...il.com>, devicetree@...r.kernel.org
> +Description:
> +               Top level Open Firmware, aka devicetree, sysfs directory.
> +
> +What:          /sys/firmware/devicetree/base
>  Date:          November 2013
>  Contact:       Grant Likely <grant.likely@....com>, devicetree@...r.kernel.org
>  Description:
> @@ -6,11 +12,6 @@ Description:
>                 hardware, the device tree structure will be exposed in this
>                 directory.
>
> -               It is possible for multiple device-tree directories to exist.
> -               Some device drivers use a separate detached device tree which
> -               have no attachment to the system tree and will appear in a
> -               different subdirectory under /sys/firmware/devicetree.
> -
>                 Userspace must not use the /sys/firmware/devicetree/base
>                 path directly, but instead should follow /proc/device-tree
>                 symlink. It is possible that the absolute path will change
> @@ -20,13 +21,65 @@ Description:
>                 filesystem support, and has largely the same semantics and
>                 should be compatible with existing userspace.
>
> -               The contents of /sys/firmware/devicetree/ is a
> +               The /sys/firmware/devicetree/base directory is the root
> +               node of the devicetree.
> +
> +               The contents of /sys/firmware/devicetree/base is a
>                 hierarchy of directories, one per device tree node. The
>                 directory name is the resolved path component name (node
>                 name plus address). Properties are represented as files
>                 in the directory. The contents of each file is the exact
>                 binary data from the device tree.
>
> +What:          /sys/firmware/devicetree/tree_version
> +Date:          October 2018
> +KernelVersion: 4.20
> +Contact:       Frank Rowand <frowand.list@...il.com>, devicetree@...r.kernel.org
> +Description:
> +               When an overlay is applied or removed, the live devicetree
> +               visible in /proc/device-tree/, aka
> +               /sys/firmware/devicetree/base/, reflects the changes.
> +
> +               tree_version provides a way for user space to determine if the
> +               live devicetree has remained unchanged while a series of one
> +               or more accesses of /proc/device-tree/ occur.  If tree_version
> +               is odd then the devicetree is locked for an overlay update.

This explains the intention very clearly and directly.

I think it would help to give more of an idea of its behavior.  I
initially had to look at the code to understand what kind of values to
expect and whether the tree_version somehow actually reflected some
version # that I hadn't known about, hidden in DT somewhere.  It would
be helpful to have it spelled out what granularity this counter has
and some other things about its behavior, somehow explaining that
* tree_version is a counter that is incremented and never decremented
* tree_version is incremented twice per change
* that the granularity is an overlay, tree_version is not counting
changeset changes or something else
* tree_version is updated (incremented) for both applied and for
removed overlays

Such as:

tree_version is incremented twice for each overlay that is applied or
removed (and never decremented).  When the tree is locked for
processing an overlay, tree_version is incremented once and its value
becomes odd.  Then when the overlay is done being applied or removed,
the tree is unlocked and tree_version is incremented again, becoming
an even value.

> +
> +               The contents of tree_version is the ascii representation of
> +               a kernel unsigned 32 bit int variable, thus when incremented
> +               from 4294967295 it will wrap around to 0.
> +
> +               The use of both dynamic devicetree modifications and overlay
> +               apply and removal are not supported during the same boot
> +               cycle.  Thus non-overlay dynamic modifications are not
> +               reflected in the value of tree_version.
> +
> +               Example shell use of tree_version:
> +

Same comments as in the header about adding some comments and possibly
reconsidering the 'done' variable.

> +               done=1
> +
> +               while [ $done = 1 ] ; do
> +
> +                  pre_version=$(cat /sys/firmware/devicetree/tree_version)
> +                  while [ $(( ${pre_version} & 1 )) != 0 ] ; do
> +                     # echo is optional, sleep value can be tuned
> +                     echo "${pre_version}  tree locked, sleeping 2 seconds"
> +                     sleep 2;
> +                     pre_version=$(cat /sys/firmware/devicetree/tree_version)
> +                  done
> +
> +                  # 'critical region'
> +                  # access /proc/device-tree/ one or more times
> +
> +                  post_version=$(cat /sys/firmware/devicetree/tree_version)
> +
> +                  if [ ${post_version} = ${pre_version} ] ; then
> +                     done=0
> +                  fi
> +
> +               done
> +
> +
>  What:          /sys/firmware/fdt
>  Date:          February 2015
>  KernelVersion: 3.19
> diff --git a/drivers/of/base.c b/drivers/of/base.c
> index 466e3c8582f0..ec0428e47577 100644
> --- a/drivers/of/base.c
> +++ b/drivers/of/base.c
> @@ -151,9 +151,71 @@ int of_free_phandle_cache(void)
>  late_initcall_sync(of_free_phandle_cache);
>  #endif
>
> +#define show_one(object)                                               \
> +       static ssize_t show_##object                                    \
> +       (struct kobject *kobj, struct attribute *attr, char *buf)       \
> +       {                                                               \
> +               return sprintf(buf, "%u\n", object);                    \
> +       }
> +
> +struct global_attr {
> +       struct attribute attr;
> +       ssize_t (*show)(struct kobject *kobj,
> +                       struct attribute *attr, char *buf);
> +       ssize_t (*store)(struct kobject *a, struct attribute *b,
> +                        const char *c, size_t count);
> +};
> +
> +#define define_one_global_ro(_name)            \
> +static struct global_attr attr_##_name =       \
> +__ATTR(_name, 0444, show_##_name, NULL)
> +
> +/*
> + * tree_version must be unsigned so that at maximum value it wraps
> + * from an odd value (4294967295) to an even value (0).
> + */
> +static u32 tree_version;
> +
> +show_one(tree_version);
> +
> +define_one_global_ro(tree_version);
> +
> +static struct attribute *of_attributes[] = {
> +       &attr_tree_version.attr,
> +       NULL
> +};
> +
> +static const struct attribute_group of_attr_group = {
> +       .attrs = of_attributes,
> +};
> +
> +/*
> + * internal documentation
> + * tree_version_increment() - increment base version
> + *
> + * Before starting overlay apply or overlay remove, call this function.
> + * After completing overlay apply or overlay remove, call this function.
> + *
> + * caller must hold of_mutex
> + *
> + * Userspace can use the value of this variable to determine whether the
> + * devicetree has changed while accessing the devicetree.  An odd value
> + * means a modification is underway.  An even value means no modification
> + * is underway.
> + *
> + * The use of both (1) dynamic devicetree modifications and (2) overlay apply
> + * and removal are not supported during the same boot cycle.  Thus non-overlay
> + * dynamic modifications are not reflected in the value of tree_version.
> + */
> +void tree_version_increment(void)
> +{
> +       tree_version++;
> +}
> +
>  void __init of_core_init(void)
>  {
>         struct device_node *np;
> +       int ret;
>
>         of_populate_phandle_cache();
>
> @@ -172,6 +234,10 @@ void __init of_core_init(void)
>         /* Symlink in /proc as required by userspace ABI */
>         if (of_root)
>                 proc_symlink("device-tree", NULL, "/sys/firmware/devicetree/base");
> +
> +       ret = sysfs_create_group(&of_kset->kobj, &of_attr_group);
> +       if (ret)
> +               pr_err("sysfs_create_group of_attr_group failed: %d\n", ret);
>  }
>
>  static struct property *__of_find_property(const struct device_node *np,
> diff --git a/drivers/of/of_private.h b/drivers/of/of_private.h
> index 216175d11d3d..adf89f6bad81 100644
> --- a/drivers/of/of_private.h
> +++ b/drivers/of/of_private.h
> @@ -31,6 +31,8 @@ struct alias_prop {
>  extern struct list_head aliases_lookup;
>  extern struct kset *of_kset;
>
> +void tree_version_increment(void);
> +
>  #if defined(CONFIG_OF_DYNAMIC)
>  extern int of_property_notify(int action, struct device_node *np,
>                               struct property *prop, struct property *old_prop);
> diff --git a/drivers/of/overlay.c b/drivers/of/overlay.c
> index eda57ef12fd0..53b1810b1e03 100644
> --- a/drivers/of/overlay.c
> +++ b/drivers/of/overlay.c
> @@ -770,6 +770,9 @@ static int of_overlay_apply(const void *fdt, struct device_node *tree,
>         of_overlay_mutex_lock();
>         mutex_lock(&of_mutex);
>
> +       /* live tree may change after this point, user space synchronization */
> +       tree_version_increment();
> +
>         ret = of_resolve_phandles(tree);
>         if (ret)
>                 goto err_free_tree;
> @@ -832,6 +835,9 @@ static int of_overlay_apply(const void *fdt, struct device_node *tree,
>         free_overlay_changeset(ovcs);
>
>  out_unlock:
> +       /* live tree change complete, user space synchronization */
> +       tree_version_increment();
> +
>         mutex_unlock(&of_mutex);
>         of_overlay_mutex_unlock();
>
> @@ -1028,6 +1034,9 @@ int of_overlay_remove(int *ovcs_id)
>
>         mutex_lock(&of_mutex);
>
> +       /* live tree may change after this point, user space synchronization */
> +       tree_version_increment();
> +
>         ovcs = idr_find(&ovcs_idr, *ovcs_id);
>         if (!ovcs) {
>                 ret = -ENODEV;
> @@ -1083,6 +1092,9 @@ int of_overlay_remove(int *ovcs_id)
>         free_overlay_changeset(ovcs);
>
>  out_unlock:
> +       /* live tree change complete, user space synchronization */
> +       tree_version_increment();
> +
>         mutex_unlock(&of_mutex);
>
>  out:
> --

Thanks!
Alan

> Frank Rowand <frank.rowand@...y.com>
>

Powered by blists - more mailing lists