r8169: set coherent DMA mask as well as streaming DMA mask

Message ID 1476445170-4496-1-git-send-email-ard.biesheuvel@linaro.org
State New
Headers show

Commit Message

Ard Biesheuvel Oct. 14, 2016, 11:39 a.m.
PCI devices that are 64-bit DMA capable should set the coherent
DMA mask as well as the streaming DMA mask. On some architectures,
these are managed separately, and so the coherent DMA mask will be
left at its default value of 32 if it is not set explicitly. This
results in errors such as

     r8169 Gigabit Ethernet driver 2.3LK-NAPI loaded
     hwdev DMA mask = 0x00000000ffffffff, dev_addr = 0x00000080fbfff000
     swiotlb: coherent allocation failed for device 0000:02:00.0 size=4096
     CPU: 0 PID: 1062 Comm: systemd-udevd Not tainted 4.8.0+ #35
     Hardware name: AMD Seattle/Seattle, BIOS 10:53:24 Oct 13 2016

on systems without memory that is 32-bit addressable by PCI devices.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

---
 drivers/net/ethernet/realtek/r8169.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

-- 
2.7.4

Comments

David Miller Oct. 14, 2016, 1:31 p.m. | #1
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>

Date: Fri, 14 Oct 2016 12:39:30 +0100

> PCI devices that are 64-bit DMA capable should set the coherent

> DMA mask as well as the streaming DMA mask. On some architectures,

> these are managed separately, and so the coherent DMA mask will be

> left at its default value of 32 if it is not set explicitly. This

> results in errors such as

> 

>      r8169 Gigabit Ethernet driver 2.3LK-NAPI loaded

>      hwdev DMA mask = 0x00000000ffffffff, dev_addr = 0x00000080fbfff000

>      swiotlb: coherent allocation failed for device 0000:02:00.0 size=4096

>      CPU: 0 PID: 1062 Comm: systemd-udevd Not tainted 4.8.0+ #35

>      Hardware name: AMD Seattle/Seattle, BIOS 10:53:24 Oct 13 2016

> 

> on systems without memory that is 32-bit addressable by PCI devices.

> 

> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

 ...
> @@ -8281,6 +8282,8 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)

>  		dev->features |= NETIF_F_HIGHDMA;

>  	} else {

>  		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));

> +		if (!rc)

> +			rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));


As you state 32-bit is the default, therefore this part of your patch is unnecessary.
Ard Biesheuvel Oct. 14, 2016, 1:36 p.m. | #2
On 14 October 2016 at 14:34, David Miller <davem@davemloft.net> wrote:
> From: Ard Biesheuvel <ard.biesheuvel@linaro.org>

> Date: Fri, 14 Oct 2016 14:32:24 +0100

>

>> On 14 October 2016 at 14:31, David Miller <davem@davemloft.net> wrote:

>>> From: Ard Biesheuvel <ard.biesheuvel@linaro.org>

>>> Date: Fri, 14 Oct 2016 12:39:30 +0100

>>>

>>>> PCI devices that are 64-bit DMA capable should set the coherent

>>>> DMA mask as well as the streaming DMA mask. On some architectures,

>>>> these are managed separately, and so the coherent DMA mask will be

>>>> left at its default value of 32 if it is not set explicitly. This

>>>> results in errors such as

>>>>

>>>>      r8169 Gigabit Ethernet driver 2.3LK-NAPI loaded

>>>>      hwdev DMA mask = 0x00000000ffffffff, dev_addr = 0x00000080fbfff000

>>>>      swiotlb: coherent allocation failed for device 0000:02:00.0 size=4096

>>>>      CPU: 0 PID: 1062 Comm: systemd-udevd Not tainted 4.8.0+ #35

>>>>      Hardware name: AMD Seattle/Seattle, BIOS 10:53:24 Oct 13 2016

>>>>

>>>> on systems without memory that is 32-bit addressable by PCI devices.

>>>>

>>>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

>>>  ...

>>>> @@ -8281,6 +8282,8 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)

>>>>               dev->features |= NETIF_F_HIGHDMA;

>>>>       } else {

>>>>               rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));

>>>> +             if (!rc)

>>>> +                     rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));

>>>

>>> As you state 32-bit is the default, therefore this part of your patch is unnecessary.

>>

>> Perhaps, but the original code did not assume that either. Should I

>> remove the other call in a subsequent patch as well?

>

> I simply want you to respin this with the above hunk removed.

>

> Your code changes and your commit message must be consistent.


OK, fair enough

Patch

diff --git a/drivers/net/ethernet/realtek/r8169.c b/drivers/net/ethernet/realtek/r8169.c
index e55638c7505a..04957a36b11f 100644
--- a/drivers/net/ethernet/realtek/r8169.c
+++ b/drivers/net/ethernet/realtek/r8169.c
@@ -8273,7 +8273,8 @@  static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	if ((sizeof(dma_addr_t) > 4) &&
 	    (use_dac == 1 || (use_dac == -1 && pci_is_pcie(pdev) &&
 			      tp->mac_version >= RTL_GIGA_MAC_VER_18)) &&
-	    !pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
+	    !pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
+	    !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
 
 		/* CPlusCmd Dual Access Cycle is only needed for non-PCIe */
 		if (!pci_is_pcie(pdev))
@@ -8281,6 +8282,8 @@  static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 		dev->features |= NETIF_F_HIGHDMA;
 	} else {
 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
+		if (!rc)
+			rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
 		if (rc < 0) {
 			netif_err(tp, probe, dev, "DMA configuration failed\n");
 			goto err_out_unmap_4;