Re: [PATCH v4] platform/x86/intel/pmc: core: Add support to show LTR-ignored components

[Rajat Khandelwal]

Hi David,
Please find the comments inline.

On 2/5/2023 12:49 AM, Box, David E wrote:
> Hi Rajat,
>
> On Sun, 2023-02-05 at 23:14 +0530, Rajat Khandelwal wrote:
> > Currently, 'ltr_ignore' sysfs attribute, when read, returns nothing, even
> > if there are components whose LTR values have been ignored.
> >
> > This patch adds the feature to print out such components, if they exist.
> >
> > Signed-off-by: Rajat Khandelwal <rajat.khandelwal@xxxxxxxxxxxxxxx>
> > ---
> >
> > v4: Mutex unlock during error conditions
> >
> > v3: Incorporated a mutex lock for accessing 'ltr_ignore_list'
> >
> > v2: kmalloc -> devm_kmalloc
> >
> >   drivers/platform/x86/intel/pmc/core.c | 59 ++++++++++++++++++++++-----
> >   drivers/platform/x86/intel/pmc/core.h |  2 +-
> >   2 files changed, 49 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/platform/x86/intel/pmc/core.c
> > b/drivers/platform/x86/intel/pmc/core.c
> > index 3a15d32d7644..f9d4b2865b03 100644
> > --- a/drivers/platform/x86/intel/pmc/core.c
> > +++ b/drivers/platform/x86/intel/pmc/core.c
> > @@ -53,6 +53,17 @@ const struct pmc_bit_map msr_map[] = {
> >          {}
> >   };
> >   
> > +/* Mutual exclusion to access the list of LTR-ignored components */
> > +static DEFINE_MUTEX(ltr_entry_mutex);
> > +
> > +struct ltr_entry {
> > +       u32 comp_index;
> > +       const char *comp_name;
> > +       struct list_head node;
> > +};
> > +
> > +static LIST_HEAD(ltr_ignore_list);
> > +
> >   static inline u32 pmc_core_reg_read(struct pmc_dev *pmcdev, int reg_offset)
> >   {
> >          return readl(pmcdev->regbase + reg_offset);
> > @@ -435,27 +446,18 @@ static int pmc_core_pll_show(struct seq_file *s, void
> > *unused)
> >   }
> >   DEFINE_SHOW_ATTRIBUTE(pmc_core_pll);
> >   
> > -int pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value)
> > +void pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value)
> >   {
> >          const struct pmc_reg_map *map = pmcdev->map;
> >          u32 reg;
> > -       int err = 0;
> >   
> >          mutex_lock(&pmcdev->lock);
> >   
> > -       if (value > map->ltr_ignore_max) {
> > -               err = -EINVAL;
> > -               goto out_unlock;
> > -       }
> > -
> >          reg = pmc_core_reg_read(pmcdev, map->ltr_ignore_offset);
> >          reg |= BIT(value);
> >          pmc_core_reg_write(pmcdev, map->ltr_ignore_offset, reg);
> >   
> > -out_unlock:
> >          mutex_unlock(&pmcdev->lock);
> > -
> > -       return err;
> >   }
> >   
> >   static ssize_t pmc_core_ltr_ignore_write(struct file *file,
> > @@ -464,6 +466,8 @@ static ssize_t pmc_core_ltr_ignore_write(struct file
> > *file,
> >   {
> >          struct seq_file *s = file->private_data;
> >          struct pmc_dev *pmcdev = s->private;
> > +       const struct pmc_reg_map *map = pmcdev->map;
> > +       struct ltr_entry *entry;
> >          u32 buf_size, value;
> >          int err;
> >   
> > @@ -473,13 +477,46 @@ static ssize_t pmc_core_ltr_ignore_write(struct file
> > *file,
> >          if (err)
> >                  return err;
> >   
> > -       err = pmc_core_send_ltr_ignore(pmcdev, value);
> > +       if (value > map->ltr_ignore_max)
> > +               return -EINVAL;
> > +
> > +       mutex_lock(&ltr_entry_mutex);
> > +
> > +       list_for_each_entry(entry, &ltr_ignore_list, node) {
> > +               if (entry->comp_index == value) {
> > +                       err = -EEXIST;
> Do we need to return an error here? We don't offer a way to undo the ignore and
> rewriting it doesn't hurt anything. I'm okay with ignoring this.

Surely, it won't hurt to just write the value again. It does provide a sense of notion
to the user that "this component was already set" (something like that).
Not that big a deal, but I would like to keep it that way, if that's okay? :)

> > +                       goto out_unlock;
> > +               }
> > +       }
> > +
> > +       entry = devm_kmalloc(&pmcdev->pdev->dev, sizeof(*entry), GFP_KERNEL);
> > +       if (!entry) {
> > +               err = -ENOMEM;
> > +               goto out_unlock;
> > +       }
> > +
> > +       entry->comp_name = map->ltr_show_sts[value].name;
> > +       entry->comp_index = value;
> > +       list_add_tail(&entry->node, &ltr_ignore_list);
> > +
> > +       pmc_core_send_ltr_ignore(pmcdev, value);
> > +
> > +out_unlock:
> > +       mutex_unlock(&ltr_entry_mutex);
> You can allocate your entry and do the assignment before you take the list lock.
> If the allocation fails, return immediately without a goto.
>
> You can also move pmc_core_send_ltr_ignore() after the unlock.

Ok, so I allocate it only after I see that the list doesn't already has the value.
That is why I take the lock and proceed.
pmc_core_send_ltr_ignore() can be moved after the unlock.

Please let me know your comments for v5.

Thanks
Rajat

> David
>
> >   
> >          return err == 0 ? count : err;
> >   }
> >   
> >   static int pmc_core_ltr_ignore_show(struct seq_file *s, void *unused)
> >   {
> > +       struct ltr_entry *entry;
> > +
> > +       mutex_lock(&ltr_entry_mutex);
> > +       list_for_each_entry(entry, &ltr_ignore_list, node) {
> > +               seq_printf(s, "%s\n", entry->comp_name);
> > +       }
> > +       mutex_unlock(&ltr_entry_mutex);
> > +
> >          return 0;
> >   }
> >   
> > diff --git a/drivers/platform/x86/intel/pmc/core.h
> > b/drivers/platform/x86/intel/pmc/core.h
> > index 810204d758ab..da35b0fcbe6e 100644
> > --- a/drivers/platform/x86/intel/pmc/core.h
> > +++ b/drivers/platform/x86/intel/pmc/core.h
> > @@ -396,7 +396,7 @@ extern const struct pmc_reg_map adl_reg_map;
> >   extern const struct pmc_reg_map mtl_reg_map;
> >   
> >   extern void pmc_core_get_tgl_lpm_reqs(struct platform_device *pdev);
> > -extern int pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value);
> > +extern void pmc_core_send_ltr_ignore(struct pmc_dev *pmcdev, u32 value);
> >   
> >   void spt_core_init(struct pmc_dev *pmcdev);
> >   void cnp_core_init(struct pmc_dev *pmcdev);
> > --
> > 2.34.1