[PATCH] drm/msm/a6xx: check pm_runtime_resume_and_get() during resume

[PATCH] drm/msm/a6xx: check pm_runtime_resume_and_get() during resume

[Roman Demidov]

The return values of pm_runtime_resume_and_get() calls in a6xx_pm_resume()
are not checked, which can lead to hardware access on suspended devices
and PM reference underflows.

Fix this by checking the return value of each pm_runtime_resume_and_get()
call and properly unwinding the previously acquired resources on failure.

Found by Linux Verification Center (linuxtesting.org) with SVACE.

Fixes: 5a903a44a984 ("drm/msm/a6xx: Introduce GMU wrapper support")
Signed-off-by: Roman Demidov <roman.demidov.nn@gmail.com>
---
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 38 +++++++++++++++------------
 1 file changed, 21 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index f9de9329dee3..0cf205ea744b 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -2175,44 +2175,48 @@ static int a6xx_pm_resume(struct msm_gpu *gpu)
 	opp = dev_pm_opp_find_freq_ceil(&gpu->pdev->dev, &freq);
 	if (IS_ERR(opp)) {
 		ret = PTR_ERR(opp);
-		goto err_set_opp;
+		goto err_unlock;
 	}
 	dev_pm_opp_put(opp);
 
 	/* Set the core clock and bus bw, having VDD scaling in mind */
 	dev_pm_opp_set_opp(&gpu->pdev->dev, opp);
 
-	pm_runtime_resume_and_get(gmu->dev);
-	pm_runtime_resume_and_get(gmu->gxpd);
+	ret = pm_runtime_resume_and_get(gmu->dev);
+	if (ret < 0)
+		goto err_opp_clear;
+	ret = pm_runtime_resume_and_get(gmu->gxpd);
+	if (ret < 0)
+		goto err_put_dev;
 
 	ret = clk_bulk_prepare_enable(gpu->nr_clocks, gpu->grp_clks);
 	if (ret)
-		goto err_bulk_clk;
+		goto err_put_gxpd;
 
 	ret = clk_bulk_prepare_enable(gmu->nr_clocks, gmu->clocks);
 	if (ret) {
 		clk_bulk_disable_unprepare(gpu->nr_clocks, gpu->grp_clks);
-		goto err_bulk_clk;
+		goto err_put_gxpd;
 	}
 
 	if (adreno_is_a619_holi(adreno_gpu))
 		a6xx_sptprac_enable(gmu);
 
-	/* If anything goes south, tear the GPU down piece by piece.. */
-	if (ret) {
-err_bulk_clk:
-		pm_runtime_put(gmu->gxpd);
-		pm_runtime_put(gmu->dev);
-		dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
-	}
-err_set_opp:
 	mutex_unlock(&a6xx_gpu->gmu.lock);
+	msm_devfreq_resume(gpu);
+	a6xx_llc_activate(a6xx_gpu);
 
-	if (!ret) {
-		msm_devfreq_resume(gpu);
-		a6xx_llc_activate(a6xx_gpu);
-	}
+	return 0;
 
+	/* If anything goes south, tear the GPU down piece by piece.. */
+err_put_gxpd:
+	pm_runtime_put(gmu->gxpd);
+err_put_dev:
+	pm_runtime_put(gmu->dev);
+err_opp_clear:
+	dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
+err_unlock:
+	mutex_unlock(&a6xx_gpu->gmu.lock);
 	return ret;
 }
 
-- 
2.53.0

Re: [PATCH] drm/msm/a6xx: check pm_runtime_resume_and_get() during resume

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Unconditional `pm_runtime_resume_and_get()` on `gmu->gxpd` causes a NULL pointer dereference if the gxpd power domain is absent.
- [High] The `opp` pointer is used after its reference count has been dropped, creating a Use-After-Free vulnerability.
--

commit 8f5f79a7915a90cb43eb05a5eb535e070ae669bd
Author: Roman Demidov <roman.demidov.nn@gmail.com>

drm/msm/a6xx: check pm_runtime_resume_and_get() during resume

This commit adds return value checks to pm_runtime_resume_and_get() calls
during a6xx GPU resume to prevent hardware access on suspended devices. It
also unwinds acquired resources properly if these calls fail.

> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> index f9de9329dee39..0cf205ea744be 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> @@ -2175,44 +2175,48 @@ static int a6xx_pm_resume(struct msm_gpu *gpu)
>  	opp = dev_pm_opp_find_freq_ceil(&gpu->pdev->dev, &freq);
>  	if (IS_ERR(opp)) {
>  		ret = PTR_ERR(opp);
> -		goto err_set_opp;
> +		goto err_unlock;
>  	}
>  	dev_pm_opp_put(opp);
>  
>  	/* Set the core clock and bus bw, having VDD scaling in mind */
>  	dev_pm_opp_set_opp(&gpu->pdev->dev, opp);

[Severity: High]
This is a pre-existing issue, but does this code cause a use-after-free of
the opp pointer?

The dev_pm_opp_find_freq_ceil() function returns an OPP object with an
incremented reference count. Dropping this reference via dev_pm_opp_put()
could cause the object to be freed if it is concurrently removed from the
OPP table. Can dev_pm_opp_set_opp() then dereference the freed memory?

>  
> -	pm_runtime_resume_and_get(gmu->dev);
> -	pm_runtime_resume_and_get(gmu->gxpd);
> +	ret = pm_runtime_resume_and_get(gmu->dev);
> +	if (ret < 0)
> +		goto err_opp_clear;
> +	ret = pm_runtime_resume_and_get(gmu->gxpd);

[Severity: High]
This isn't a bug introduced by this patch, but can this cause a NULL pointer
dereference if the gxpd power domain is absent?

If the driver initializes a GMU wrapper device on a platform where the
device tree does not provide a 'gx' power domain, gmu->gxpd remains NULL.
Calling pm_runtime_resume_and_get(NULL) will immediately dereference the
pointer inside __pm_runtime_resume().

> +	if (ret < 0)
> +		goto err_put_dev;

[ ... ]

> -	/* If anything goes south, tear the GPU down piece by piece.. */
> -	if (ret) {
> -err_bulk_clk:
> -		pm_runtime_put(gmu->gxpd);
> -		pm_runtime_put(gmu->dev);
> -		dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
> -	}
> -err_set_opp:
>  	mutex_unlock(&a6xx_gpu->gmu.lock);
> +	msm_devfreq_resume(gpu);
> +	a6xx_llc_activate(a6xx_gpu);
>  
> -	if (!ret) {
> -		msm_devfreq_resume(gpu);
> -		a6xx_llc_activate(a6xx_gpu);
> -	}
> +	return 0;
>  
> +	/* If anything goes south, tear the GPU down piece by piece.. */
> +err_put_gxpd:
> +	pm_runtime_put(gmu->gxpd);

[Severity: High]
Can this error path also crash by unconditionally dereferencing a NULL gxpd
pointer?

While the initial pm_runtime_resume_and_get() issue is pre-existing, this
patch introduces an additional crash point here. If gmu->gxpd is NULL, this
newly added pm_runtime_put() will trigger a kernel panic.

> +err_put_dev:
> +	pm_runtime_put(gmu->dev);
> +err_opp_clear:
> +	dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
> +err_unlock:
> +	mutex_unlock(&a6xx_gpu->gmu.lock);
>  	return ret;
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260904125523.43813-1-roman.demidov.nn@gmail.com?part=1