[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20180927163141.GA7481@xps15>
Date: Thu, 27 Sep 2018 10:31:41 -0600
From: Mathieu Poirier <mathieu.poirier@...aro.org>
To: Alexander Shishkin <alexander.shishkin@...ux.intel.com>
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
linux-kernel@...r.kernel.org
Subject: Re: [QUEUED v20180920 01/16] stm class: Rework policy node fallback
Hi Alex,
On Thu, Sep 20, 2018 at 03:45:38PM +0300, Alexander Shishkin wrote:
> Currently, if no matching policy node can be found for a trace source,
> we'll try to use "default" policy node, then, if that doesn't exist,
> we'll pick the first node, in order of creation. If that also fails,
> we'll allocate M/C range from the beginning of the device's M/C range.
>
> This makes it difficult to know which node (if any) was used in any
> particular case.
>
> In order to make things more deterministic, the new order is as follows:
> * if they supply ID string, use that and nothing else,
> * if they are a task, use their task name (comm),
> * use "default", if it exists,
> * return failure, to let them know there is no suitable rule.
I am good with the "default" catch all clause but the change in behaviour has to
be documented. Otherwise people can be mislead in thinking that a regression
has happened.
Before this patch:
root@...o:~# mkdir /sys/kernel/config/stp-policy/20100000.stm.test
root@...o:~# echo 20100000.stm > /sys/class/stm_source/ftrace/stm_source_link
root@...aro-nano:~# cat /sys/class/stm_source/ftrace/stm_source_link
20100000.stm
After this patch:
root@...o:~# mkdir /sys/kernel/config/stp-policy/20100000.stm.test
root@...o:~# echo 20100000.stm > /sys/class/stm_source/ftrace/stm_source_link
bash: echo: write error: Invalid argument
root@...o:~# mkdir /sys/kernel/config/stp-policy/20100000.stm.test/default
root@...o:~# echo 20100000.stm > /sys/class/stm_source/ftrace/stm_source_link
root@...aro-nano:~# cat /sys/class/stm_source/ftrace/stm_source_link
20100000.stm
>
> This should provide enough convenience with the "default" catch-all node,
> while not leaving *everything* to chance. As a side effect, this relaxes
> the requirement of using ioctl() for identification with the possibility of
> using task names as policy nodes.
>
> Signed-off-by: Alexander Shishkin <alexander.shishkin@...ux.intel.com>
> ---
> drivers/hwtracing/stm/core.c | 89 ++++++++++++++++++++--------------
> drivers/hwtracing/stm/policy.c | 12 ++---
> drivers/hwtracing/stm/stm.h | 2 -
> 3 files changed, 58 insertions(+), 45 deletions(-)
>
> diff --git a/drivers/hwtracing/stm/core.c b/drivers/hwtracing/stm/core.c
> index 10bcb5d73f90..7b1c549d6320 100644
> --- a/drivers/hwtracing/stm/core.c
> +++ b/drivers/hwtracing/stm/core.c
> @@ -293,15 +293,15 @@ static int stm_output_assign(struct stm_device *stm, unsigned int width,
> if (width > stm->data->sw_nchannels)
> return -EINVAL;
>
> - if (policy_node) {
> - stp_policy_node_get_ranges(policy_node,
> - &midx, &mend, &cidx, &cend);
> - } else {
> - midx = stm->data->sw_start;
> - cidx = 0;
> - mend = stm->data->sw_end;
> - cend = stm->data->sw_nchannels - 1;
> - }
> + /* We no longer accept policy_node==NULL here */
> + if (WARN_ON_ONCE(!policy_node))
> + return -EINVAL;
> +
> + /*
> + * Also, the caller holds reference to policy_node, so it won't
> + * disappear on us.
> + */
> + stp_policy_node_get_ranges(policy_node, &midx, &mend, &cidx, &cend);
>
> spin_lock(&stm->mc_lock);
> spin_lock(&output->lock);
> @@ -405,19 +405,30 @@ static int stm_char_release(struct inode *inode, struct file *file)
> return 0;
> }
>
> -static int stm_file_assign(struct stm_file *stmf, char *id, unsigned int width)
> +static int
> +stm_assign_first_policy(struct stm_device *stm, struct stm_output *output,
> + char **ids, unsigned int width)
> {
> - struct stm_device *stm = stmf->stm;
> - int ret;
> + struct stp_policy_node *pn;
> + int err, n;
>
> - stmf->policy_node = stp_policy_node_lookup(stm, id);
> + /*
> + * On success, stp_policy_node_lookup() will return holding the
> + * configfs subsystem mutex, which is then released in
> + * stp_policy_node_put(). This allows the pdrv->output_open() in
> + * stm_output_assign() to serialize against the attribute accessors.
> + */
> + for (n = 0, pn = NULL; ids[n] && !pn; n++)
> + pn = stp_policy_node_lookup(stm, ids[n]);
>
> - ret = stm_output_assign(stm, width, stmf->policy_node, &stmf->output);
> + if (!pn)
> + return -EINVAL;
>
> - if (stmf->policy_node)
> - stp_policy_node_put(stmf->policy_node);
> + err = stm_output_assign(stm, width, pn, output);
>
> - return ret;
> + stp_policy_node_put(pn);
> +
> + return err;
> }
>
> static ssize_t notrace stm_write(struct stm_data *data, unsigned int master,
> @@ -455,16 +466,23 @@ static ssize_t stm_char_write(struct file *file, const char __user *buf,
> count = PAGE_SIZE - 1;
>
> /*
> - * if no m/c have been assigned to this writer up to this
> - * point, use "default" policy entry
> + * If no m/c have been assigned to this writer up to this
> + * point, try to use the task name and "default" policy entries.
> */
> if (!stmf->output.nr_chans) {
> - err = stm_file_assign(stmf, "default", 1);
> + char comm[sizeof(current->comm)];
> + char *ids[] = { comm, "default", NULL };
> +
> + get_task_comm(comm, current);
> +
> + err = stm_assign_first_policy(stmf->stm, &stmf->output, ids, 1);
> /*
> * EBUSY means that somebody else just assigned this
> - * output, which is just fine for write()
> + * output, which is just fine for write(), except for (XXX)
> + * it doesn't actually return -EBUSY, but -EINVAL and WARNs
> + * in this case. Hrrr.
> */
> - if (err && err != -EBUSY)
> + if (err)
> return err;
> }
>
> @@ -550,6 +568,7 @@ static int stm_char_policy_set_ioctl(struct stm_file *stmf, void __user *arg)
> {
> struct stm_device *stm = stmf->stm;
> struct stp_policy_id *id;
> + char *ids[] = { NULL, NULL };
> int ret = -EINVAL;
> u32 size;
>
> @@ -582,7 +601,9 @@ static int stm_char_policy_set_ioctl(struct stm_file *stmf, void __user *arg)
> id->width > PAGE_SIZE / stm->data->sw_mmiosz)
> goto err_free;
>
> - ret = stm_file_assign(stmf, id->id, id->width);
> + ids[0] = id->id;
> + ret = stm_assign_first_policy(stmf->stm, &stmf->output, ids,
> + id->width);
> if (ret)
> goto err_free;
>
> @@ -818,8 +839,8 @@ EXPORT_SYMBOL_GPL(stm_unregister_device);
> static int stm_source_link_add(struct stm_source_device *src,
> struct stm_device *stm)
> {
> - char *id;
> - int err;
> + char *ids[] = { NULL, "default", NULL };
> + int err = -ENOMEM;
>
> mutex_lock(&stm->link_mutex);
> spin_lock(&stm->link_lock);
> @@ -833,19 +854,13 @@ static int stm_source_link_add(struct stm_source_device *src,
> spin_unlock(&stm->link_lock);
> mutex_unlock(&stm->link_mutex);
>
> - id = kstrdup(src->data->name, GFP_KERNEL);
> - if (id) {
> - src->policy_node =
> - stp_policy_node_lookup(stm, id);
> -
> - kfree(id);
> - }
> -
> - err = stm_output_assign(stm, src->data->nr_chans,
> - src->policy_node, &src->output);
> + ids[0] = kstrdup(src->data->name, GFP_KERNEL);
> + if (!ids[0])
> + goto fail_detach;
>
> - if (src->policy_node)
> - stp_policy_node_put(src->policy_node);
> + err = stm_assign_first_policy(stm, &src->output, ids,
> + src->data->nr_chans);
> + kfree(ids[0]);
>
> if (err)
> goto fail_detach;
> diff --git a/drivers/hwtracing/stm/policy.c b/drivers/hwtracing/stm/policy.c
> index 3fd07e275b34..15d35d891643 100644
> --- a/drivers/hwtracing/stm/policy.c
> +++ b/drivers/hwtracing/stm/policy.c
> @@ -392,7 +392,7 @@ static struct configfs_subsystem stp_policy_subsys = {
> static struct stp_policy_node *
> __stp_policy_node_lookup(struct stp_policy *policy, char *s)
> {
> - struct stp_policy_node *policy_node, *ret;
> + struct stp_policy_node *policy_node, *ret = NULL;
> struct list_head *head = &policy->group.cg_children;
> struct config_item *item;
> char *start, *end = s;
> @@ -400,10 +400,6 @@ __stp_policy_node_lookup(struct stp_policy *policy, char *s)
> if (list_empty(head))
> return NULL;
>
> - /* return the first entry if everything else fails */
> - item = list_entry(head->next, struct config_item, ci_entry);
> - ret = to_stp_policy_node(item);
> -
> next:
> for (;;) {
> start = strsep(&end, "/");
> @@ -449,13 +445,17 @@ stp_policy_node_lookup(struct stm_device *stm, char *s)
>
> if (policy_node)
> config_item_get(&policy_node->group.cg_item);
> - mutex_unlock(&stp_policy_subsys.su_mutex);
> + else
> + mutex_unlock(&stp_policy_subsys.su_mutex);
>
> return policy_node;
> }
>
> void stp_policy_node_put(struct stp_policy_node *policy_node)
> {
> + lockdep_assert_held(&stp_policy_subsys.su_mutex);
> +
> + mutex_unlock(&stp_policy_subsys.su_mutex);
> config_item_put(&policy_node->group.cg_item);
> }
>
> diff --git a/drivers/hwtracing/stm/stm.h b/drivers/hwtracing/stm/stm.h
> index 923571adc6f4..e5df08ae59cf 100644
> --- a/drivers/hwtracing/stm/stm.h
> +++ b/drivers/hwtracing/stm/stm.h
> @@ -57,7 +57,6 @@ struct stm_output {
>
> struct stm_file {
> struct stm_device *stm;
> - struct stp_policy_node *policy_node;
> struct stm_output output;
> };
>
> @@ -71,7 +70,6 @@ struct stm_source_device {
> struct stm_device __rcu *link;
> struct list_head link_entry;
> /* one output per stm_source device */
> - struct stp_policy_node *policy_node;
> struct stm_output output;
> };
>
> --
> 2.18.0
>
Powered by blists - more mailing lists