Re: [PATCH v3] platform/x86: intel_pmc_core: Prevent possibile overflow

From: Andy Shevchenko
Date: Sat Aug 14 2021 - 05:51:11 EST

On Sat, Aug 14, 2021 at 4:49 AM David E. Box
<david.e.box@xxxxxxxxxxxxxxx> wrote:
>
> Substate priority levels are encoded in 4 bits in the LPM_PRI register.
> This value was used as an index to an array whose element size was less
> than 16, leading to the possibility of overflow should we read a larger
> than expected priority. In addition to the overflow, bad values could lead
> to incorrect state reporting. So rework the priority code to prevent the
> overflow and perform some validation of the register. Use the priority
> register values if they give an ordering of unique numbers between 0 and
> the maximum number of states. Otherwise, use a default ordering instead.

Thanks! Looks much better now. A couple of minor comments below. With
or without them being addressed
Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>

> Reported-by: Evgeny Novikov <novikov@xxxxxxxxx>
> Signed-off-by: David E. Box <david.e.box@xxxxxxxxxxxxxxx>
> ---
> v3: Modifying Andy's suggestion, just place the entire verification
> in a separate function. If it fails, then keep the default
> ordering. If it passes, overwrite with the verified ordering.
>
> Fix error in default order array.
>
> Also fix spelling noted by Andy drop the size comment since
> the array size is set when declared.
>
> v2: Remove lpm_priority size increase. Instead, remove that array and
> create 2 new local arrays, one to save priority levels in mode order,
> and one to save modes in priority order. Use the mode_order list to
> validate that no priority level is above the maximum and to validate
> that they are all unique values. Then we can safely create a
> priority_order list that will be the basis of how we report substate
> information.
>
> drivers/platform/x86/intel_pmc_core.c | 65 +++++++++++++++++++++------
> drivers/platform/x86/intel_pmc_core.h | 2 +
> 2 files changed, 53 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/platform/x86/intel_pmc_core.c b/drivers/platform/x86/intel_pmc_core.c
> index b0e486a6bdfb..ae410a358ffe 100644
> --- a/drivers/platform/x86/intel_pmc_core.c
> +++ b/drivers/platform/x86/intel_pmc_core.c
> @@ -1449,9 +1449,42 @@ static int pmc_core_pkgc_show(struct seq_file *s, void *unused)
> }
> DEFINE_SHOW_ATTRIBUTE(pmc_core_pkgc);
>
> -static void pmc_core_get_low_power_modes(struct pmc_dev *pmcdev)
> +static bool pmc_core_pri_verify(u32 lpm_pri, u8 *mode_order)
> {
> - u8 lpm_priority[LPM_MAX_NUM_MODES];
> + int i, j;
> +
> + if (!lpm_pri)
> + return false;
> + /*
> + * Each byte contains the priority level for 2 modes (7:4 and 3:0).
> + * In a 32 bit register this allows for describing 8 modes. Store the
> + * levels and look for values out of range.
> + */
> + for (i = 0; i < 8; i++) {
> + int level = lpm_pri & GENMASK(3, 0);
> +
> + if (level >= LPM_MAX_NUM_MODES)
> + return false;
> +
> + mode_order[i] = level;
> + lpm_pri >>= 4;
> + }

> + /* Check that we have unique values */
> + for (i = 0; i < LPM_MAX_NUM_MODES - 1; i++)
> + for (j = i + 1; j < LPM_MAX_NUM_MODES; j++)
> + if (mode_order[i] == mode_order[j])
> + return false;

There are ways to optimize this from O(n^2) to O(n), but it's not critical here.

> + return true;
> +}
> +
> +static void pmc_core_get_low_power_modes(struct platform_device *pdev)
> +{
> + struct pmc_dev *pmcdev = platform_get_drvdata(pdev);
> + u8 pri_order[LPM_MAX_NUM_MODES] = LPM_DEFAULT_PRI;
> + u8 mode_order[LPM_MAX_NUM_MODES];
> + u32 lpm_pri;
> u32 lpm_en;
> int mode, i, p;
>
> @@ -1462,24 +1495,28 @@ static void pmc_core_get_low_power_modes(struct pmc_dev *pmcdev)
> lpm_en = pmc_core_reg_read(pmcdev, pmcdev->map->lpm_en_offset);
> pmcdev->num_lpm_modes = hweight32(lpm_en);
>
> - /* Each byte contains information for 2 modes (7:4 and 3:0) */
> - for (mode = 0; mode < LPM_MAX_NUM_MODES; mode += 2) {
> - u8 priority = pmc_core_reg_read_byte(pmcdev,
> - pmcdev->map->lpm_priority_offset + (mode / 2));
> - int pri0 = GENMASK(3, 0) & priority;
> - int pri1 = (GENMASK(7, 4) & priority) >> 4;
> + /* Read 32 bit LPM_PRI register */
> + lpm_pri = pmc_core_reg_read(pmcdev, pmcdev->map->lpm_priority_offset);
>
> - lpm_priority[pri0] = mode;
> - lpm_priority[pri1] = mode + 1;
> - }
>
> /*
> - * Loop though all modes from lowest to highest priority,
> + * If lpm_pri value passes verification, then override the default
> + * modes here. Otherwise stick with the default.
> + */
> + if (pmc_core_pri_verify(lpm_pri, mode_order))
> + /* Get list of modes in priority order */
> + for (mode = 0; mode < LPM_MAX_NUM_MODES; mode++)
> + pri_order[mode_order[mode]] = mode;
> + else
> + dev_warn(&pdev->dev, "Assuming a default substate order for this platform\n");
> +
> + /*
> + * Loop through all modes from lowest to highest priority,
> * and capture all enabled modes in order
> */
> i = 0;
> for (p = LPM_MAX_NUM_MODES - 1; p >= 0; p--) {
> - int mode = lpm_priority[p];
> + int mode = pri_order[p];
>
> if (!(BIT(mode) & lpm_en))
> continue;
> @@ -1675,7 +1712,7 @@ static int pmc_core_probe(struct platform_device *pdev)
> mutex_init(&pmcdev->lock);
>
> pmcdev->pmc_xram_read_bit = pmc_core_check_read_lock_bit(pmcdev);
> - pmc_core_get_low_power_modes(pmcdev);
> + pmc_core_get_low_power_modes(pdev);
> pmc_core_do_dmi_quirks(pmcdev);
>
> if (pmcdev->map == &tgl_reg_map)
> diff --git a/drivers/platform/x86/intel_pmc_core.h b/drivers/platform/x86/intel_pmc_core.h
> index e8dae9c6c45f..b9bf3d3d6f7a 100644
> --- a/drivers/platform/x86/intel_pmc_core.h
> +++ b/drivers/platform/x86/intel_pmc_core.h
> @@ -188,6 +188,8 @@ enum ppfear_regs {
> #define ICL_PMC_SLP_S0_RES_COUNTER_STEP 0x64
>
> #define LPM_MAX_NUM_MODES 8
> +#define LPM_DEFAULT_PRI { 7, 6, 2, 5, 4, 1, 3, 0 }

What I meant here is to add static_assert() to avoid ARRAY_SIZE() != NUM_MODES.

> #define GET_X2_COUNTER(v) ((v) >> 1)
> #define LPM_STS_LATCH_MODE BIT(31)


--
With Best Regards,
Andy Shevchenko