diff mbox series

[V1,net,2/4] net: ena: set initial DMA width to avoid intel iommu issue

Message ID 20201118215947.8970-3-shayagr@amazon.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series Fixes for ENA driver | expand

Checks

Context Check Description
netdev/cover_letter success Link
netdev/fixes_present success Link
netdev/patch_count success Link
netdev/tree_selection success Clearly marked for net
netdev/subject_prefix success Link
netdev/source_inline success Was 0 now: 0
netdev/verify_signedoff success Link
netdev/module_param success Was 0 now: 0
netdev/build_32bit success Errors and warnings before: 0 this patch: 0
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/verify_fixes success Link
netdev/checkpatch warning WARNING: line length of 90 exceeds 80 columns
netdev/build_allmodconfig_warn success Errors and warnings before: 0 this patch: 0
netdev/header_inline success Link
netdev/stable success Stable not CCed

Commit Message

Shay Agroskin Nov. 18, 2020, 9:59 p.m. UTC
The ENA driver uses the readless mechanism, which uses DMA, to find
out what the DMA mask is supposed to be.

If DMA is used without setting the dma_mask first, it causes the
Intel IOMMU driver to think that ENA is a 32-bit device and therefore
disables IOMMU passthrough permanently.

This patch sets the dma_mask to be ENA_MAX_PHYS_ADDR_SIZE_BITS=48
before readless initialization in
ena_device_init()->ena_com_mmio_reg_read_request_init(),
which is large enough to workaround the intel_iommu issue.

DMA mask is set again to the correct value after it's received from the
device after readless is initialized.

Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Mike Cui <mikecui@amazon.com>
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: Shay Agroskin <shayagr@amazon.com>
---
 drivers/net/ethernet/amazon/ena/ena_netdev.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

Comments

Heiner Kallweit Nov. 18, 2020, 10:35 p.m. UTC | #1
Am 18.11.2020 um 22:59 schrieb Shay Agroskin:
> The ENA driver uses the readless mechanism, which uses DMA, to find
> out what the DMA mask is supposed to be.
> 
> If DMA is used without setting the dma_mask first, it causes the
> Intel IOMMU driver to think that ENA is a 32-bit device and therefore
> disables IOMMU passthrough permanently.
> 
> This patch sets the dma_mask to be ENA_MAX_PHYS_ADDR_SIZE_BITS=48
> before readless initialization in
> ena_device_init()->ena_com_mmio_reg_read_request_init(),
> which is large enough to workaround the intel_iommu issue.
> 
> DMA mask is set again to the correct value after it's received from the
> device after readless is initialized.
> 
> Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
> Signed-off-by: Mike Cui <mikecui@amazon.com>
> Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
> Signed-off-by: Shay Agroskin <shayagr@amazon.com>
> ---
>  drivers/net/ethernet/amazon/ena/ena_netdev.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c
> index 574c2b5ba21e..854a22e692bf 100644
> --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c
> +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c
> @@ -4146,6 +4146,19 @@ static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>  		return rc;
>  	}
>  
> +	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
> +	if (rc) {
> +		dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", rc);
> +		goto err_disable_device;
> +	}
> +
> +	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
> +	if (rc) {
> +		dev_err(&pdev->dev, "err_pci_set_consistent_dma_mask failed %d\n",
> +			rc);
> +		goto err_disable_device;
> +	}
> +
>  	pci_set_master(pdev);
>  
>  	ena_dev = vzalloc(sizeof(*ena_dev));
> 

The old pci_ dma wrappers are being phased out and shouldn't be used in
new code. See e.g. e059c6f340f6 ("tulip: switch from 'pci_' to 'dma_' API").
So better use:
dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
Heiner Kallweit Nov. 18, 2020, 10:41 p.m. UTC | #2
Am 18.11.2020 um 23:35 schrieb Heiner Kallweit:
> Am 18.11.2020 um 22:59 schrieb Shay Agroskin:
>> The ENA driver uses the readless mechanism, which uses DMA, to find
>> out what the DMA mask is supposed to be.
>>
>> If DMA is used without setting the dma_mask first, it causes the
>> Intel IOMMU driver to think that ENA is a 32-bit device and therefore
>> disables IOMMU passthrough permanently.
>>
>> This patch sets the dma_mask to be ENA_MAX_PHYS_ADDR_SIZE_BITS=48
>> before readless initialization in
>> ena_device_init()->ena_com_mmio_reg_read_request_init(),
>> which is large enough to workaround the intel_iommu issue.
>>
>> DMA mask is set again to the correct value after it's received from the
>> device after readless is initialized.
>>
>> Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
>> Signed-off-by: Mike Cui <mikecui@amazon.com>
>> Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
>> Signed-off-by: Shay Agroskin <shayagr@amazon.com>
>> ---
>>  drivers/net/ethernet/amazon/ena/ena_netdev.c | 13 +++++++++++++
>>  1 file changed, 13 insertions(+)
>>
>> diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>> index 574c2b5ba21e..854a22e692bf 100644
>> --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c
>> +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>> @@ -4146,6 +4146,19 @@ static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>>  		return rc;
>>  	}
>>  
>> +	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>> +	if (rc) {
>> +		dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", rc);
>> +		goto err_disable_device;
>> +	}
>> +
>> +	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>> +	if (rc) {
>> +		dev_err(&pdev->dev, "err_pci_set_consistent_dma_mask failed %d\n",
>> +			rc);
>> +		goto err_disable_device;
>> +	}
>> +
>>  	pci_set_master(pdev);
>>  
>>  	ena_dev = vzalloc(sizeof(*ena_dev));
>>
> 
> The old pci_ dma wrappers are being phased out and shouldn't be used in
> new code. See e.g. e059c6f340f6 ("tulip: switch from 'pci_' to 'dma_' API").
> So better use:
> dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
> 

However instead of dev_err(&pdev->dev, ..) you could use pci_err(pdev, ..).
Shay Agroskin Nov. 19, 2020, 7:18 p.m. UTC | #3
Heiner Kallweit <hkallweit1@gmail.com> writes:

> Am 18.11.2020 um 23:35 schrieb Heiner Kallweit:
>> Am 18.11.2020 um 22:59 schrieb Shay Agroskin:
>>> The ENA driver uses the readless mechanism, which uses DMA, to 
>>> find
>>> out what the DMA mask is supposed to be.
>>>
>>> If DMA is used without setting the dma_mask first, it causes 
>>> the
>>> Intel IOMMU driver to think that ENA is a 32-bit device and 
>>> therefore
>>> disables IOMMU passthrough permanently.
>>>
>>> This patch sets the dma_mask to be 
>>> ENA_MAX_PHYS_ADDR_SIZE_BITS=48
>>> before readless initialization in
>>> ena_device_init()->ena_com_mmio_reg_read_request_init(),
>>> which is large enough to workaround the intel_iommu issue.
>>>
>>> DMA mask is set again to the correct value after it's received 
>>> from the
>>> device after readless is initialized.
>>>
>>> Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon 
>>> Elastic Network Adapters (ENA)")
>>> Signed-off-by: Mike Cui <mikecui@amazon.com>
>>> Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
>>> Signed-off-by: Shay Agroskin <shayagr@amazon.com>
>>> ---
>>>  drivers/net/ethernet/amazon/ena/ena_netdev.c | 13 
>>>  +++++++++++++
>>>  1 file changed, 13 insertions(+)
>>>
>>> diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c 
>>> b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>> index 574c2b5ba21e..854a22e692bf 100644
>>> --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>> +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>> @@ -4146,6 +4146,19 @@ static int ena_probe(struct pci_dev 
>>> *pdev, const struct pci_device_id *ent)
>>>  		return rc;
>>>  	}
>>>  
>>> +	rc = pci_set_dma_mask(pdev, 
>>> DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>>> +	if (rc) {
>>> +		dev_err(&pdev->dev, "pci_set_dma_mask failed 
>>> %d\n", rc);
>>> +		goto err_disable_device;
>>> +	}
>>> +
>>> +	rc = pci_set_consistent_dma_mask(pdev, 
>>> DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>>> +	if (rc) {
>>> +		dev_err(&pdev->dev, 
>>> "err_pci_set_consistent_dma_mask failed %d\n",
>>> +			rc);
>>> +		goto err_disable_device;
>>> +	}
>>> +
>>>  	pci_set_master(pdev);
>>>  
>>>  	ena_dev = vzalloc(sizeof(*ena_dev));
>>>
>> 
>> The old pci_ dma wrappers are being phased out and shouldn't be 
>> used in
>> new code. See e.g. e059c6f340f6 ("tulip: switch from 'pci_' to 
>> 'dma_' API").
>> So better use:
>> dma_set_mask_and_coherent(&pdev->dev, 
>> DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));

Thank you for reviewing these patches. We will switch to using 
dma_set_...() instead

>> 
>
> However instead of dev_err(&pdev->dev, ..) you could use 
> pci_err(pdev, ..).

I see that pci_err evaluates to dev_err. While I see how using 
pci_* log function helps code readability, I prefer using
dev_err here to keep the code consistent with the rest of the 
driver. We'll discuss changing all log functions in future patches 
to net-next if that's okay.
Heiner Kallweit Nov. 19, 2020, 7:36 p.m. UTC | #4
Am 19.11.2020 um 20:18 schrieb Shay Agroskin:
> 
> Heiner Kallweit <hkallweit1@gmail.com> writes:
> 
>> Am 18.11.2020 um 23:35 schrieb Heiner Kallweit:
>>> Am 18.11.2020 um 22:59 schrieb Shay Agroskin:
>>>> The ENA driver uses the readless mechanism, which uses DMA, to find
>>>> out what the DMA mask is supposed to be.
>>>>
>>>> If DMA is used without setting the dma_mask first, it causes the
>>>> Intel IOMMU driver to think that ENA is a 32-bit device and therefore
>>>> disables IOMMU passthrough permanently.
>>>>
>>>> This patch sets the dma_mask to be ENA_MAX_PHYS_ADDR_SIZE_BITS=48
>>>> before readless initialization in
>>>> ena_device_init()->ena_com_mmio_reg_read_request_init(),
>>>> which is large enough to workaround the intel_iommu issue.
>>>>
>>>> DMA mask is set again to the correct value after it's received from the
>>>> device after readless is initialized.
>>>>
>>>> Fixes: 1738cd3ed342 ("net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)")
>>>> Signed-off-by: Mike Cui <mikecui@amazon.com>
>>>> Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
>>>> Signed-off-by: Shay Agroskin <shayagr@amazon.com>
>>>> ---
>>>>  drivers/net/ethernet/amazon/ena/ena_netdev.c | 13  +++++++++++++
>>>>  1 file changed, 13 insertions(+)
>>>>
>>>> diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>>> index 574c2b5ba21e..854a22e692bf 100644
>>>> --- a/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>>> +++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c
>>>> @@ -4146,6 +4146,19 @@ static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>>>>          return rc;
>>>>      }
>>>>  
>>>> +    rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>>>> +    if (rc) {
>>>> +        dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", rc);
>>>> +        goto err_disable_device;
>>>> +    }
>>>> +
>>>> +    rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
>>>> +    if (rc) {
>>>> +        dev_err(&pdev->dev, "err_pci_set_consistent_dma_mask failed %d\n",
>>>> +            rc);
>>>> +        goto err_disable_device;
>>>> +    }
>>>> +
>>>>      pci_set_master(pdev);
>>>>  
>>>>      ena_dev = vzalloc(sizeof(*ena_dev));
>>>>
>>>
>>> The old pci_ dma wrappers are being phased out and shouldn't be used in
>>> new code. See e.g. e059c6f340f6 ("tulip: switch from 'pci_' to 'dma_' API").
>>> So better use:
>>> dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
> 
> Thank you for reviewing these patches. We will switch to using dma_set_...() instead
> 
>>>
>>
>> However instead of dev_err(&pdev->dev, ..) you could use pci_err(pdev, ..).
> 
> I see that pci_err evaluates to dev_err. While I see how using pci_* log function helps code readability, I prefer using
> dev_err here to keep the code consistent with the rest of the driver. We'll discuss changing all log functions in future patches to net-next if that's okay.
> 
The comment was just meant to make you aware of pci_err(), there's no need
to switch to it.
diff mbox series

Patch

diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c
index 574c2b5ba21e..854a22e692bf 100644
--- a/drivers/net/ethernet/amazon/ena/ena_netdev.c
+++ b/drivers/net/ethernet/amazon/ena/ena_netdev.c
@@ -4146,6 +4146,19 @@  static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		return rc;
 	}
 
+	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
+	if (rc) {
+		dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", rc);
+		goto err_disable_device;
+	}
+
+	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
+	if (rc) {
+		dev_err(&pdev->dev, "err_pci_set_consistent_dma_mask failed %d\n",
+			rc);
+		goto err_disable_device;
+	}
+
 	pci_set_master(pdev);
 
 	ena_dev = vzalloc(sizeof(*ena_dev));