Skip to contentSkip to conversation

ASoC: mediatek: clear afe->regmap on devm_regmap_init_mmio() error

PATCH New Elkin Cruz 2 participants 4 messages Nov 20

Subsystems (3)
lore
Share & tools
lore

Elkin Cruz

author
#0 lore raw

cover ASoC: mediatek: clear afe->regmap on devm_regmap_init_mmio() error

I am trying out a new Clang static analysis linter for ERR_PTR/NULL handling, and it flagged this code as possibly buggy. The findings were triaged and fixed with an AI coding assistant, then checked against linux-next (not runtime-tested on hardware).

Thanks,
Elkin

Elkin Cruz (2):
ASoC: mediatek: mt8196: clear afe->regmap on devm_regmap_init_mmio()
error
ASoC: mediatek: mt8189: clear afe->regmap on devm_regmap_init_mmio()
error

 sound/soc/mediatek/mt8189/mt8189-afe-pcm.c | 1 +
 sound/soc/mediatek/mt8196/mt8196-afe-pcm.c | 1 +
 2 files changed, 2 insertions(+)
Signature · 1 line
2.56.1.750.gf70dc682cf-goog
author
#1 lore raw

PATCH 1/2 ASoC: mediatek: mt8196: clear afe->regmap on devm_regmap_init_mmio() error

When devm_regmap_init_mmio() fails in mt8196_afe_pcm_dev_probe(), afe->regmap holds an ERR_PTR() while unwinding through err_pm_put -> pm_runtime_put_sync(), which invokes mt8196_afe_runtime_suspend(). Because mt8196_afe_runtime_suspend() only guards if (!afe->regmap), it passes the error pointer into regcache_cache_only() and crashes.

Clear afe->regmap to NULL before jumping to err_pm_put.

Fixes e38353138a09 ("ASoC: mediatek: mt8196: Fix probe resource cleanup")
---
 sound/soc/mediatek/mt8196/mt8196-afe-pcm.c | 1 +
 1 file changed, 1 insertion(+)
sound/soc/mediatek/mt8196/mt8196-afe-pcm.c +1 -0
index 82b5ae91751a..3028aa82c4a8 100644
@@ -2450,6 +2450,7 @@ static int mt8196_afe_pcm_dev_probe(struct platform_device *pdev)
 					    &mt8196_afe_regmap_config);
 	if (IS_ERR(afe->regmap)) {
 		ret = PTR_ERR(afe->regmap);
+		afe->regmap = NULL;
 		goto err_pm_put;
 	}
 
Signature · 1 line
2.56.1.750.gf70dc682cf-goog
author
#2 lore raw

PATCH 2/2 ASoC: mediatek: mt8189: clear afe->regmap on devm_regmap_init_mmio() error

In reply to Elkin Cruz

When devm_regmap_init_mmio() fails in mt8189_afe_pcm_dev_probe(), afe->regmap holds an ERR_PTR() while unwinding through err_pm_put -> pm_runtime_put_sync(), which invokes mt8189_afe_runtime_suspend(). Because mt8189_afe_runtime_suspend() only checks if (!afe->regmap), the ERR_PTR() value is passed to regmap_write() and regcache_cache_only().

Clear afe->regmap to NULL before jumping to err_pm_put.

Fixes 5404599c3292 ("ASoC: mediatek: mt8189: Fix probe resource cleanup")
---
 sound/soc/mediatek/mt8189/mt8189-afe-pcm.c | 1 +
 1 file changed, 1 insertion(+)
sound/soc/mediatek/mt8189/mt8189-afe-pcm.c +1 -0
index f39b01590f71..85b5be6d45ec 100644
@@ -2563,6 +2563,7 @@ static int mt8189_afe_pcm_dev_probe(struct platform_device *pdev)
 					    &mt8189_afe_regmap_config);
 	if (IS_ERR(afe->regmap)) {
 		ret = PTR_ERR(afe->regmap);
+		afe->regmap = NULL;
 		goto err_pm_put;
 	}
 
Signature · 1 line
2.56.1.750.gf70dc682cf-goog
#3 lore raw

Re: [PATCH 2/2] ASoC: mediatek: mt8189: clear afe->regmap on devm_regmap_init_mmio() error

Thank you for your contribution! Sashiko AI review found no new issues in this patch, but noticed 1 related pre-existing issue it does not introduce: - [High] ASoC: mediatek: mt8189: clock disable…

Thank you for your contribution! Sashiko AI review found no new issues in this patch, but noticed 1 related pre-existing issue it does not introduce:
- [High] ASoC: mediatek: mt8189: clock disable underflow in mt8189_afe_pcm_dev_remove() [1]
--

commit d8496ec5beb278c4f15ca503ecbd50eb2598abc9
Author: Elkin Cruz <[email protected]>

ASoC: mediatek: mt8189: clear afe->regmap on devm_regmap_init_mmio() error

This commit clears afe->regmap to NULL when devm_regmap_init_mmio() fails in mt8189_afe_pcm_dev_probe() so that mt8189_afe_runtime_suspend() does not pass an ERR_PTR() to regmap functions during error unwinding.

sound/soc/mediatek/mt8189/mt8189-afe-pcm.c static int mt8189_afe_pcm_dev_probe(struct platform_device *pdev) · line 2568

Quoted 11 lines
sound/soc/mediatek/mt8189/mt8189-afe-pcm.c +1 -0
index f39b01590f714..85b5be6d45ec6 100644
@@ -2563,6 +2563,7 @@ static int mt8189_afe_pcm_dev_probe(struct platform_device *pdev)
 					    &mt8189_afe_regmap_config);
 	if (IS_ERR(afe->regmap)) {
 		ret = PTR_ERR(afe->regmap);
+		afe->regmap = NULL;
 		goto err_pm_put;
 	}

[Severity: High]
This problem wasn't introduced by this patch, but does mt8189_afe_pcm_dev_remove() [1] underflow the clock enable counts by disabling the AFE main and register read/write clocks after mt8189_afe_runtime_suspend()?

sound/soc/mediatek/mt8189/mt8189-afe-pcm.c:mt8189_afe_pcm_dev_remove() {
    ...
	if (!pm_runtime_status_suspended(dev))
		mt8189_afe_runtime_suspend(dev);

	mt8189_afe_disable_main_clock(afe);
	/* disable afe clock */
	mt8189_afe_disable_reg_rw_clk(afe);
}

mt8189_afe_runtime_suspend() already calls mt8189_afe_disable_main_clock() and mt8189_afe_disable_reg_rw_clk(). Unconditionally disabling both clocks a second time on device unbind corrupts the Common Clock Framework reference counts and can shut down shared parent clocks still in use by other peripherals.

[1] https://sashiko.dev/#/bug/linux-7209342b-8b8a-4947-a765-7c521cb675e9

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2