[PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[Sanket Goswami]

Store the root port information in amd_pmc_probe() so that the
information can be used across multiple routines.

Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
---
Changes in v2:
- Store the rdev info in amd_pmc_probe() as suggested by Hans.

 drivers/platform/x86/amd-pmc.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/platform/x86/amd-pmc.c b/drivers/platform/x86/amd-pmc.c
index 55f14bdfdbfd..502f37eaba1f 100644
--- a/drivers/platform/x86/amd-pmc.c
+++ b/drivers/platform/x86/amd-pmc.c
@@ -119,6 +119,7 @@ struct amd_pmc_dev {
 	u16 minor;
 	u16 rev;
 	struct device *dev;
+	struct pci_dev *rdev;
 	struct mutex lock; /* generic mutex lock */
 #if IS_ENABLED(CONFIG_DEBUG_FS)
 	struct dentry *dbgfs_dir;
@@ -482,6 +483,7 @@ static int amd_pmc_probe(struct platform_device *pdev)
 		return -ENODEV;
 	}
 
+	dev->rdev = rdev;
 	dev->cpu_id = rdev->device;
 	err = pci_write_config_dword(rdev, AMD_PMC_SMU_INDEX_ADDRESS, AMD_PMC_BASE_ADDR_LO);
 	if (err) {
@@ -512,7 +514,6 @@ static int amd_pmc_probe(struct platform_device *pdev)
 	}
 
 	base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
-	pci_dev_put(dev->rdev);
 	base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
 
 	dev->regbase = devm_ioremap(dev->dev, base_addr + AMD_PMC_BASE_ADDR_OFFSET,
@@ -546,6 +547,7 @@ static int amd_pmc_remove(struct platform_device *pdev)
 	struct amd_pmc_dev *dev = platform_get_drvdata(pdev);
 
 	amd_pmc_dbgfs_unregister(dev);
+	pci_dev_put(dev->rdev);
 	mutex_destroy(&dev->lock);
 	return 0;
 }
-- 
2.25.1

Re: [PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[Hans de Goede]

Hi,

On 10/21/21 11:29, Sanket Goswami wrote:
> Store the root port information in amd_pmc_probe() so that the
> information can be used across multiple routines.
> 
> Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
> ---
> Changes in v2:
> - Store the rdev info in amd_pmc_probe() as suggested by Hans.

Thank you, but there are still some issues, see below.


>  drivers/platform/x86/amd-pmc.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/platform/x86/amd-pmc.c b/drivers/platform/x86/amd-pmc.c
> index 55f14bdfdbfd..502f37eaba1f 100644
> --- a/drivers/platform/x86/amd-pmc.c
> +++ b/drivers/platform/x86/amd-pmc.c
> @@ -119,6 +119,7 @@ struct amd_pmc_dev {
>  	u16 minor;
>  	u16 rev;
>  	struct device *dev;
> +	struct pci_dev *rdev;
>  	struct mutex lock; /* generic mutex lock */
>  #if IS_ENABLED(CONFIG_DEBUG_FS)
>  	struct dentry *dbgfs_dir;
> @@ -482,6 +483,7 @@ static int amd_pmc_probe(struct platform_device *pdev)
>  		return -ENODEV;
>  	}
>  
> +	dev->rdev = rdev;
>  	dev->cpu_id = rdev->device;
>  	err = pci_write_config_dword(rdev, AMD_PMC_SMU_INDEX_ADDRESS, AMD_PMC_BASE_ADDR_LO);
>  	if (err) {
> @@ -512,7 +514,6 @@ static int amd_pmc_probe(struct platform_device *pdev)
>  	}
>  
>  	base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
> -	pci_dev_put(dev->rdev);

The current code here actually reads:

	pci_dev_put(rdev);

Note (rdev) not (dev->rdev). I don't know what you based this on, this is weird.

Also there are a bunch of error-exits from amd_pmc_probe() which not all
need a "pci_dev_put(rdev)" added to them before there "return ERROR;"
statement.

It would be best to add:

err_pci_dev_put:
	pci_dev_put(rdev);
	return err;

Add the end off the function (after the return 0;) and replace all
"return FOO" error-exits with:

		err = <FOO>;
		goto err_pci_dev_put;
	}



Regards,

Hans




>  	base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
>  
>  	dev->regbase = devm_ioremap(dev->dev, base_addr + AMD_PMC_BASE_ADDR_OFFSET,
> @@ -546,6 +547,7 @@ static int amd_pmc_remove(struct platform_device *pdev)
>  	struct amd_pmc_dev *dev = platform_get_drvdata(pdev);
>  
>  	amd_pmc_dbgfs_unregister(dev);
> +	pci_dev_put(dev->rdev);
>  	mutex_destroy(&dev->lock);
>  	return 0;
>  }
> 

Re: [PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[Goswami, Sanket]

Hi Hans,

On 21-Oct-21 23:48, Hans de Goede wrote:
> [CAUTION: External Email]
> 
> Hi,
> 
> On 10/21/21 11:29, Sanket Goswami wrote:
>> Store the root port information in amd_pmc_probe() so that the
>> information can be used across multiple routines.
>>
>> Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
>> ---
>> Changes in v2:
>> - Store the rdev info in amd_pmc_probe() as suggested by Hans.
> 
> Thank you, but there are still some issues, see below.
> 
> 
>>  drivers/platform/x86/amd-pmc.c | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/platform/x86/amd-pmc.c b/drivers/platform/x86/amd-pmc.c
>> index 55f14bdfdbfd..502f37eaba1f 100644
>> --- a/drivers/platform/x86/amd-pmc.c
>> +++ b/drivers/platform/x86/amd-pmc.c
>> @@ -119,6 +119,7 @@ struct amd_pmc_dev {
>>       u16 minor;
>>       u16 rev;
>>       struct device *dev;
>> +     struct pci_dev *rdev;
>>       struct mutex lock; /* generic mutex lock */
>>  #if IS_ENABLED(CONFIG_DEBUG_FS)
>>       struct dentry *dbgfs_dir;
>> @@ -482,6 +483,7 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>               return -ENODEV;
>>       }
>>
>> +     dev->rdev = rdev;
>>       dev->cpu_id = rdev->device;
>>       err = pci_write_config_dword(rdev, AMD_PMC_SMU_INDEX_ADDRESS, AMD_PMC_BASE_ADDR_LO);
>>       if (err) {
>> @@ -512,7 +514,6 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>       }
>>
>>       base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
>> -     pci_dev_put(dev->rdev);
> 
> The current code here actually reads:
> 
>         pci_dev_put(rdev);
> 
> Note (rdev) not (dev->rdev). I don't know what you based this on, this is weird.

rdev is already retrieved before doing this:
	     pci_dev_put(dev->rdev);

i.e.
in amd_pmc_probe()

rdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0, 0));
	if (!rdev || !pci_match_id(pmc_pci_ids, rdev)) {
		pci_dev_put(rdev);
		return -ENODEV;
	}

after this I am storing rdev in "dev->rdev"
i.e.
dev->rdev = rdev;

after this I am using "dev->rdev" at places where "rdev" was getting used earlier.
Do you see any problem?

> 
> Also there are a bunch of error-exits from amd_pmc_probe() which not all
> need a "pci_dev_put(rdev)" added to them before there "return ERROR;"
> statement.
> 
> It would be best to add:
> 
> err_pci_dev_put:
>         pci_dev_put(rdev);
>         return err;
> 
> Add the end off the function (after the return 0;) and replace all
> "return FOO" error-exits with:
> 
>                 err = <FOO>;
>                 goto err_pci_dev_put;
>         }
> 
Thank you, will take it as a separate patch in v3.


Thanks,
Sanket

Re: [PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[Hans de Goede]

Hi Sanket,

On 10/22/21 08:55, Goswami, Sanket wrote:
> Hi Hans,
> 
> On 21-Oct-21 23:48, Hans de Goede wrote:
>> [CAUTION: External Email]
>>
>> Hi,
>>
>> On 10/21/21 11:29, Sanket Goswami wrote:
>>> Store the root port information in amd_pmc_probe() so that the
>>> information can be used across multiple routines.
>>>
>>> Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
>>> ---
>>> Changes in v2:
>>> - Store the rdev info in amd_pmc_probe() as suggested by Hans.
>>
>> Thank you, but there are still some issues, see below.
>>
>>
>>>  drivers/platform/x86/amd-pmc.c | 4 +++-
>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/platform/x86/amd-pmc.c b/drivers/platform/x86/amd-pmc.c
>>> index 55f14bdfdbfd..502f37eaba1f 100644
>>> --- a/drivers/platform/x86/amd-pmc.c
>>> +++ b/drivers/platform/x86/amd-pmc.c
>>> @@ -119,6 +119,7 @@ struct amd_pmc_dev {
>>>       u16 minor;
>>>       u16 rev;
>>>       struct device *dev;
>>> +     struct pci_dev *rdev;
>>>       struct mutex lock; /* generic mutex lock */
>>>  #if IS_ENABLED(CONFIG_DEBUG_FS)
>>>       struct dentry *dbgfs_dir;
>>> @@ -482,6 +483,7 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>>               return -ENODEV;
>>>       }
>>>
>>> +     dev->rdev = rdev;
>>>       dev->cpu_id = rdev->device;
>>>       err = pci_write_config_dword(rdev, AMD_PMC_SMU_INDEX_ADDRESS, AMD_PMC_BASE_ADDR_LO);
>>>       if (err) {
>>> @@ -512,7 +514,6 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>>       }
>>>
>>>       base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
>>> -     pci_dev_put(dev->rdev);
>>
>> The current code here actually reads:
>>
>>         pci_dev_put(rdev);
>>
>> Note (rdev) not (dev->rdev). I don't know what you based this on, this is weird.
> 
> rdev is already retrieved before doing this:
> 	     pci_dev_put(dev->rdev);
> 
> i.e.
> in amd_pmc_probe()
> 
> rdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0, 0));
> 	if (!rdev || !pci_match_id(pmc_pci_ids, rdev)) {
> 		pci_dev_put(rdev);
> 		return -ENODEV;
> 	}
> 
> after this I am storing rdev in "dev->rdev"
> i.e.
> dev->rdev = rdev;
> 
> after this I am using "dev->rdev" at places where "rdev" was getting used earlier.
> Do you see any problem?

What I was trying to say is that the patch does not apply, because it is
trying to remove the pci_put_dev() line from a block of code like this:

	base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
	pci_dev_put(dev->rdev);
	base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);

But the actual code in platform-drivers-x86/review-hans (and for-next too) has:

	base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
	pci_dev_put(rdev);
	base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);



After your patch using dev->rdev instead of just rdev is fine
(but please be consistent, which would mean use just rdev everywhere).

But your patch is removing a line which does not exist in that form,
IOW it is based on some intermediate version of amd-pmc.c and not
on the HEAD of platform-drivers-x86/review-hans.

Regards,

Hans






> 
>>
>> Also there are a bunch of error-exits from amd_pmc_probe() which not all
>> need a "pci_dev_put(rdev)" added to them before there "return ERROR;"
>> statement.
>>
>> It would be best to add:
>>
>> err_pci_dev_put:
>>         pci_dev_put(rdev);
>>         return err;
>>
>> Add the end off the function (after the return 0;) and replace all
>> "return FOO" error-exits with:
>>
>>                 err = <FOO>;
>>                 goto err_pci_dev_put;
>>         }
>>
> Thank you, will take it as a separate patch in v3.
> 
> 
> Thanks,
> Sanket
> 

Re: [PATCH v2 1/2] platform/x86: amd-pmc: Store the pci_dev instance inside struct amd_pmc_dev

[Goswami, Sanket]

Hi Hans,

On 22-Oct-21 14:21, Hans de Goede wrote:
> [CAUTION: External Email]
> 
> Hi Sanket,
> 
> On 10/22/21 08:55, Goswami, Sanket wrote:
>> Hi Hans,
>>
>> On 21-Oct-21 23:48, Hans de Goede wrote:
>>> [CAUTION: External Email]
>>>
>>> Hi,
>>>
>>> On 10/21/21 11:29, Sanket Goswami wrote:
>>>> Store the root port information in amd_pmc_probe() so that the
>>>> information can be used across multiple routines.
>>>>
>>>> Signed-off-by: Sanket Goswami <Sanket.Goswami@amd.com>
>>>> ---
>>>> Changes in v2:
>>>> - Store the rdev info in amd_pmc_probe() as suggested by Hans.
>>>
>>> Thank you, but there are still some issues, see below.
>>>
>>>
>>>>  drivers/platform/x86/amd-pmc.c | 4 +++-
>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/platform/x86/amd-pmc.c b/drivers/platform/x86/amd-pmc.c
>>>> index 55f14bdfdbfd..502f37eaba1f 100644
>>>> --- a/drivers/platform/x86/amd-pmc.c
>>>> +++ b/drivers/platform/x86/amd-pmc.c
>>>> @@ -119,6 +119,7 @@ struct amd_pmc_dev {
>>>>       u16 minor;
>>>>       u16 rev;
>>>>       struct device *dev;
>>>> +     struct pci_dev *rdev;
>>>>       struct mutex lock; /* generic mutex lock */
>>>>  #if IS_ENABLED(CONFIG_DEBUG_FS)
>>>>       struct dentry *dbgfs_dir;
>>>> @@ -482,6 +483,7 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>>>               return -ENODEV;
>>>>       }
>>>>
>>>> +     dev->rdev = rdev;
>>>>       dev->cpu_id = rdev->device;
>>>>       err = pci_write_config_dword(rdev, AMD_PMC_SMU_INDEX_ADDRESS, AMD_PMC_BASE_ADDR_LO);
>>>>       if (err) {
>>>> @@ -512,7 +514,6 @@ static int amd_pmc_probe(struct platform_device *pdev)
>>>>       }
>>>>
>>>>       base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
>>>> -     pci_dev_put(dev->rdev);
>>>
>>> The current code here actually reads:
>>>
>>>         pci_dev_put(rdev);
>>>
>>> Note (rdev) not (dev->rdev). I don't know what you based this on, this is weird.
>>
>> rdev is already retrieved before doing this:
>>            pci_dev_put(dev->rdev);
>>
>> i.e.
>> in amd_pmc_probe()
>>
>> rdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0, 0));
>>       if (!rdev || !pci_match_id(pmc_pci_ids, rdev)) {
>>               pci_dev_put(rdev);
>>               return -ENODEV;
>>       }
>>
>> after this I am storing rdev in "dev->rdev"
>> i.e.
>> dev->rdev = rdev;
>>
>> after this I am using "dev->rdev" at places where "rdev" was getting used earlier.
>> Do you see any problem?
> 
> What I was trying to say is that the patch does not apply, because it is
> trying to remove the pci_put_dev() line from a block of code like this:
> 
>         base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
>         pci_dev_put(dev->rdev);
>         base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
> 
> But the actual code in platform-drivers-x86/review-hans (and for-next too) has:
> 
>         base_addr_hi = val & AMD_PMC_BASE_ADDR_LO_MASK;
>         pci_dev_put(rdev);
>         base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
> 
> 
> 
> After your patch using dev->rdev instead of just rdev is fine
> (but please be consistent, which would mean use just rdev everywhere).
> 
> But your patch is removing a line which does not exist in that form,
> IOW it is based on some intermediate version of amd-pmc.c and not
> on the HEAD of platform-drivers-x86/review-hans.

I will rebase it to review-hans branch and will respin a new version.

Thanks,
Sanket