diff mbox series

[v2,1/3] PCI: mediatek: Allocate MSI address with dmam_alloc_coherent()

Message ID 20231211085256.31292-2-jianjun.wang@mediatek.com (mailing list archive)
State New, archived
Headers show
Series PCI: mediatek: Allocate MSI address with dmam_alloc_coherent() | expand

Commit Message

Jianjun Wang (王建军) Dec. 11, 2023, 8:52 a.m. UTC 
Use dmam_alloc_coherent() to allocate the MSI address, instead of using
virt_to_phys().

Signed-off-by: Jianjun Wang <jianjun.wang@mediatek.com>
---
 drivers/pci/controller/pcie-mediatek.c | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)

Comments

Manivannan Sadhasivam June 8, 2024, 9:01 a.m. UTC | #1 
On Mon, Dec 11, 2023 at 04:52:54PM +0800, Jianjun Wang wrote:
> Use dmam_alloc_coherent() to allocate the MSI address, instead of using
> virt_to_phys().
> 

What is the reason for this change? So now PCIE_MSI_VECTOR becomes unused?

- Mani

> Signed-off-by: Jianjun Wang <jianjun.wang@mediatek.com>
> ---
>  drivers/pci/controller/pcie-mediatek.c | 24 ++++++++++++++++--------
>  1 file changed, 16 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/pci/controller/pcie-mediatek.c b/drivers/pci/controller/pcie-mediatek.c
> index 66a8f73296fc..2fb9e44369f8 100644
> --- a/drivers/pci/controller/pcie-mediatek.c
> +++ b/drivers/pci/controller/pcie-mediatek.c
> @@ -178,6 +178,7 @@ struct mtk_pcie_soc {
>   * @phy: pointer to PHY control block
>   * @slot: port slot
>   * @irq: GIC irq
> + * @msg_addr: MSI message address
>   * @irq_domain: legacy INTx IRQ domain
>   * @inner_domain: inner IRQ domain
>   * @msi_domain: MSI IRQ domain
> @@ -198,6 +199,7 @@ struct mtk_pcie_port {
>  	struct phy *phy;
>  	u32 slot;
>  	int irq;
> +	dma_addr_t msg_addr;
>  	struct irq_domain *irq_domain;
>  	struct irq_domain *inner_domain;
>  	struct irq_domain *msi_domain;
> @@ -394,12 +396,10 @@ static struct pci_ops mtk_pcie_ops_v2 = {
>  static void mtk_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
>  {
>  	struct mtk_pcie_port *port = irq_data_get_irq_chip_data(data);
> -	phys_addr_t addr;
>  
>  	/* MT2712/MT7622 only support 32-bit MSI addresses */
> -	addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
>  	msg->address_hi = 0;
> -	msg->address_lo = lower_32_bits(addr);
> +	msg->address_lo = lower_32_bits(port->msg_addr);
>  
>  	msg->data = data->hwirq;
>  
> @@ -494,6 +494,14 @@ static struct msi_domain_info mtk_msi_domain_info = {
>  static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
>  {
>  	struct fwnode_handle *fwnode = of_node_to_fwnode(port->pcie->dev->of_node);
> +	void *msi_vaddr;
> +
> +	msi_vaddr = dmam_alloc_coherent(port->pcie->dev, sizeof(dma_addr_t), &port->msg_addr,
> +					GFP_KERNEL);
> +	if (!msi_vaddr) {
> +		dev_err(port->pcie->dev, "failed to alloc and map MSI address\n");
> +		return -ENOMEM;
> +	}
>  
>  	mutex_init(&port->lock);
>  
> @@ -501,6 +509,7 @@ static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
>  						      &msi_domain_ops, port);
>  	if (!port->inner_domain) {
>  		dev_err(port->pcie->dev, "failed to create IRQ domain\n");
> +		dmam_free_coherent(port->pcie->dev, sizeof(dma_addr_t), msi_vaddr, port->msg_addr);
>  		return -ENOMEM;
>  	}
>  
> @@ -508,6 +517,7 @@ static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
>  						     port->inner_domain);
>  	if (!port->msi_domain) {
>  		dev_err(port->pcie->dev, "failed to create MSI domain\n");
> +		dmam_free_coherent(port->pcie->dev, sizeof(dma_addr_t), msi_vaddr, port->msg_addr);
>  		irq_domain_remove(port->inner_domain);
>  		return -ENOMEM;
>  	}
> @@ -518,10 +528,8 @@ static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
>  static void mtk_pcie_enable_msi(struct mtk_pcie_port *port)
>  {
>  	u32 val;
> -	phys_addr_t msg_addr;
>  
> -	msg_addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
> -	val = lower_32_bits(msg_addr);
> +	val = lower_32_bits(port->msg_addr);
>  	writel(val, port->base + PCIE_IMSI_ADDR);
>  
>  	val = readl(port->base + PCIE_INT_MASK);
> @@ -588,7 +596,7 @@ static int mtk_pcie_init_irq_domain(struct mtk_pcie_port *port,
>  	if (IS_ENABLED(CONFIG_PCI_MSI)) {
>  		ret = mtk_pcie_allocate_msi_domains(port);
>  		if (ret)
> -			return ret;
> +			dev_warn(dev, "no MSI supported, only INTx available\n");
>  	}
>  
>  	return 0;
> @@ -732,7 +740,7 @@ static int mtk_pcie_startup_port_v2(struct mtk_pcie_port *port)
>  	val &= ~INTX_MASK;
>  	writel(val, port->base + PCIE_INT_MASK);
>  
> -	if (IS_ENABLED(CONFIG_PCI_MSI))
> +	if (IS_ENABLED(CONFIG_PCI_MSI) && port->msi_domain)
>  		mtk_pcie_enable_msi(port);
>  
>  	/* Set AHB to PCIe translation windows */
> -- 
> 2.18.0
> 
>
Marc Zyngier June 9, 2024, 12:32 p.m. UTC | #2 
On Sat, 08 Jun 2024 10:01:52 +0100,
Manivannan Sadhasivam <mani@kernel.org> wrote:
> 
> On Mon, Dec 11, 2023 at 04:52:54PM +0800, Jianjun Wang wrote:
> > Use dmam_alloc_coherent() to allocate the MSI address, instead of using
> > virt_to_phys().
> > 
> 
> What is the reason for this change? So now PCIE_MSI_VECTOR becomes unused?

More importantly, this is yet another example of the DW reference
driver nonsense, where memory is allocated for *MSI*, while the whole
point of MSIs is that it is a write that doesn't target memory, making
any form of RAM allocation absolutely pointless.

This silly approach has been cargo-culted for years, and while I
caught a few in my time, you can't beat copy-paste.

IMO, this patch is only making things worse instead of fixing things.

	M.
Bjorn Helgaas June 11, 2024, 5:21 p.m. UTC | #3 
On Sun, Jun 09, 2024 at 01:32:38PM +0100, Marc Zyngier wrote:
> On Sat, 08 Jun 2024 10:01:52 +0100,
> Manivannan Sadhasivam <mani@kernel.org> wrote:
> > 
> > On Mon, Dec 11, 2023 at 04:52:54PM +0800, Jianjun Wang wrote:
> > > Use dmam_alloc_coherent() to allocate the MSI address, instead of using
> > > virt_to_phys().
> > 
> > What is the reason for this change? So now PCIE_MSI_VECTOR becomes unused?
> 
> More importantly, this is yet another example of the DW reference
> driver nonsense, where memory is allocated for *MSI*, while the whole
> point of MSIs is that it is a write that doesn't target memory, making
> any form of RAM allocation absolutely pointless.
> 
> This silly approach has been cargo-culted for years, and while I
> caught a few in my time, you can't beat copy-paste.
> 
> IMO, this patch is only making things worse instead of fixing things.

Probably partly my fault.  I think there are two pieces here:

  1) allocating the MSI address
  2) computing the PCI bus address

I don't know how to do 1), but I do encourage people not to use
virt_to_phys() for 2), since (in general) CPU physical addresses are
not the same as PCI bus addresses.
diff mbox series

Patch

diff --git a/drivers/pci/controller/pcie-mediatek.c b/drivers/pci/controller/pcie-mediatek.c
index 66a8f73296fc..2fb9e44369f8 100644
--- a/drivers/pci/controller/pcie-mediatek.c
+++ b/drivers/pci/controller/pcie-mediatek.c
@@ -178,6 +178,7 @@  struct mtk_pcie_soc {
  * @phy: pointer to PHY control block
  * @slot: port slot
  * @irq: GIC irq
+ * @msg_addr: MSI message address
  * @irq_domain: legacy INTx IRQ domain
  * @inner_domain: inner IRQ domain
  * @msi_domain: MSI IRQ domain
@@ -198,6 +199,7 @@  struct mtk_pcie_port {
 	struct phy *phy;
 	u32 slot;
 	int irq;
+	dma_addr_t msg_addr;
 	struct irq_domain *irq_domain;
 	struct irq_domain *inner_domain;
 	struct irq_domain *msi_domain;
@@ -394,12 +396,10 @@  static struct pci_ops mtk_pcie_ops_v2 = {
 static void mtk_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 {
 	struct mtk_pcie_port *port = irq_data_get_irq_chip_data(data);
-	phys_addr_t addr;
 
 	/* MT2712/MT7622 only support 32-bit MSI addresses */
-	addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
 	msg->address_hi = 0;
-	msg->address_lo = lower_32_bits(addr);
+	msg->address_lo = lower_32_bits(port->msg_addr);
 
 	msg->data = data->hwirq;
 
@@ -494,6 +494,14 @@  static struct msi_domain_info mtk_msi_domain_info = {
 static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
 {
 	struct fwnode_handle *fwnode = of_node_to_fwnode(port->pcie->dev->of_node);
+	void *msi_vaddr;
+
+	msi_vaddr = dmam_alloc_coherent(port->pcie->dev, sizeof(dma_addr_t), &port->msg_addr,
+					GFP_KERNEL);
+	if (!msi_vaddr) {
+		dev_err(port->pcie->dev, "failed to alloc and map MSI address\n");
+		return -ENOMEM;
+	}
 
 	mutex_init(&port->lock);
 
@@ -501,6 +509,7 @@  static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
 						      &msi_domain_ops, port);
 	if (!port->inner_domain) {
 		dev_err(port->pcie->dev, "failed to create IRQ domain\n");
+		dmam_free_coherent(port->pcie->dev, sizeof(dma_addr_t), msi_vaddr, port->msg_addr);
 		return -ENOMEM;
 	}
 
@@ -508,6 +517,7 @@  static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
 						     port->inner_domain);
 	if (!port->msi_domain) {
 		dev_err(port->pcie->dev, "failed to create MSI domain\n");
+		dmam_free_coherent(port->pcie->dev, sizeof(dma_addr_t), msi_vaddr, port->msg_addr);
 		irq_domain_remove(port->inner_domain);
 		return -ENOMEM;
 	}
@@ -518,10 +528,8 @@  static int mtk_pcie_allocate_msi_domains(struct mtk_pcie_port *port)
 static void mtk_pcie_enable_msi(struct mtk_pcie_port *port)
 {
 	u32 val;
-	phys_addr_t msg_addr;
 
-	msg_addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
-	val = lower_32_bits(msg_addr);
+	val = lower_32_bits(port->msg_addr);
 	writel(val, port->base + PCIE_IMSI_ADDR);
 
 	val = readl(port->base + PCIE_INT_MASK);
@@ -588,7 +596,7 @@  static int mtk_pcie_init_irq_domain(struct mtk_pcie_port *port,
 	if (IS_ENABLED(CONFIG_PCI_MSI)) {
 		ret = mtk_pcie_allocate_msi_domains(port);
 		if (ret)
-			return ret;
+			dev_warn(dev, "no MSI supported, only INTx available\n");
 	}
 
 	return 0;
@@ -732,7 +740,7 @@  static int mtk_pcie_startup_port_v2(struct mtk_pcie_port *port)
 	val &= ~INTX_MASK;
 	writel(val, port->base + PCIE_INT_MASK);
 
-	if (IS_ENABLED(CONFIG_PCI_MSI))
+	if (IS_ENABLED(CONFIG_PCI_MSI) && port->msi_domain)
 		mtk_pcie_enable_msi(port);
 
 	/* Set AHB to PCIe translation windows */