[PATCH] platform/x86: intel/pmc: Fix iounmap call for valid addresses

[PATCH] platform/x86: intel/pmc: Fix iounmap call for valid addresses

[Xi Pardee]

pmc_core_clean_structure() is called when generic_core_init() fails.
generic_core_init() could fail before ioremap() is called to get
a valid regbase for pmc structure. The current code does not check
regbase before calling iounmap(). Add a check to fix it.

Fixes: 1b8c7b843c00 ("platform/x86:intel/pmc: Discover PMC devices")
Signed-off-by: Xi Pardee <xi.pardee@linux.intel.com>
---
 drivers/platform/x86/intel/pmc/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c
index 7a1d11f2914f..de5fc06232e5 100644
--- a/drivers/platform/x86/intel/pmc/core.c
+++ b/drivers/platform/x86/intel/pmc/core.c
@@ -1471,7 +1471,7 @@ static void pmc_core_clean_structure(struct platform_device *pdev)
 	for (i = 0; i < ARRAY_SIZE(pmcdev->pmcs); ++i) {
 		struct pmc *pmc = pmcdev->pmcs[i];
 
-		if (pmc)
+		if (pmc && pmc->regbase)
 			iounmap(pmc->regbase);
 	}
 
-- 
2.43.0

Re: [PATCH] platform/x86: intel/pmc: Fix iounmap call for valid addresses

[Ilpo Järvinen]

On Wed, 19 Mar 2025, Xi Pardee wrote:

> pmc_core_clean_structure() is called when generic_core_init() fails.
> generic_core_init() could fail before ioremap() is called to get
> a valid regbase for pmc structure. The current code does not check
> regbase before calling iounmap(). Add a check to fix it.

Hi,

The approach that calls the same "full cleanup" function as deinit uses 
when init function fails midway is very error prone as once again is 
demonstrated. Is this the only error handling problem? Are you 100% sure?

Think about it, init is x% (<100%) done when it fails, then it calls a 
function that tries to undo 100%. One needs to add lots of special logic 
to handle 0-100% rollback into that cleanup function. The init function, 
on the other hand, knows exactly where it was so it can rollback just what 
is needed and not even try to rollback for more.

It's also very inconsistent to rollback ssram_pcidev in this file as ssram 
code was moved into core_ssram so I think the ssram deinit should be moved 
there too.

I think these init functions should be converted to do proper rollback 
within the init function(s) to avoid very hard to track error handling.
I tried to check the error handling now in the pmc driver and after I 
would have needed to jump between the files, I gave up.

> Fixes: 1b8c7b843c00 ("platform/x86:intel/pmc: Discover PMC devices")
> Signed-off-by: Xi Pardee <xi.pardee@linux.intel.com>
> ---
>  drivers/platform/x86/intel/pmc/core.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c
> index 7a1d11f2914f..de5fc06232e5 100644
> --- a/drivers/platform/x86/intel/pmc/core.c
> +++ b/drivers/platform/x86/intel/pmc/core.c
> @@ -1471,7 +1471,7 @@ static void pmc_core_clean_structure(struct platform_device *pdev)
>  	for (i = 0; i < ARRAY_SIZE(pmcdev->pmcs); ++i) {
>  		struct pmc *pmc = pmcdev->pmcs[i];
>  
> -		if (pmc)
> +		if (pmc && pmc->regbase)
>  			iounmap(pmc->regbase);

-- 
 i.

Re: [PATCH] platform/x86: intel/pmc: Fix iounmap call for valid addresses

[Xi Pardee]

On 3/21/2025 7:49 AM, Ilpo Järvinen wrote:
> On Wed, 19 Mar 2025, Xi Pardee wrote:
>
>> pmc_core_clean_structure() is called when generic_core_init() fails.
>> generic_core_init() could fail before ioremap() is called to get
>> a valid regbase for pmc structure. The current code does not check
>> regbase before calling iounmap(). Add a check to fix it.
> Hi,
>
> The approach that calls the same "full cleanup" function as deinit uses
> when init function fails midway is very error prone as once again is
> demonstrated. Is this the only error handling problem? Are you 100% sure?
>
> Think about it, init is x% (<100%) done when it fails, then it calls a
> function that tries to undo 100%. One needs to add lots of special logic
> to handle 0-100% rollback into that cleanup function. The init function,
> on the other hand, knows exactly where it was so it can rollback just what
> is needed and not even try to rollback for more.
>
> It's also very inconsistent to rollback ssram_pcidev in this file as ssram
> code was moved into core_ssram so I think the ssram deinit should be moved
> there too.
>
> I think these init functions should be converted to do proper rollback
> within the init function(s) to avoid very hard to track error handling.
> I tried to check the error handling now in the pmc driver and after I
> would have needed to jump between the files, I gave up.

Hi,

Thanks for the comments! Will move the error handing code for init 
function to init function in next cycle.

I will send out a new version to separate SSRAM device handling code to 
a new PCI driver and remove ssram_pcidev field from pmcdev in next cycle.


Thanks!

Xi

>> Fixes: 1b8c7b843c00 ("platform/x86:intel/pmc: Discover PMC devices")
>> Signed-off-by: Xi Pardee <xi.pardee@linux.intel.com>
>> ---
>>   drivers/platform/x86/intel/pmc/core.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/platform/x86/intel/pmc/core.c b/drivers/platform/x86/intel/pmc/core.c
>> index 7a1d11f2914f..de5fc06232e5 100644
>> --- a/drivers/platform/x86/intel/pmc/core.c
>> +++ b/drivers/platform/x86/intel/pmc/core.c
>> @@ -1471,7 +1471,7 @@ static void pmc_core_clean_structure(struct platform_device *pdev)
>>   	for (i = 0; i < ARRAY_SIZE(pmcdev->pmcs); ++i) {
>>   		struct pmc *pmc = pmcdev->pmcs[i];
>>   
>> -		if (pmc)
>> +		if (pmc && pmc->regbase)
>>   			iounmap(pmc->regbase);