diff mbox

[v6,3/3] PCI: Add tango MSI controller support

Message ID 82d6bf37-96b9-cd9d-134b-f01638fa2b1b@sigmadesigns.com
State Superseded
Headers show

Commit Message

Marc Gonzalez June 13, 2017, 2:01 p.m. UTC
The MSI controller in Tango supports 256 message-signaled interrupts,
and a single doorbell address.

Signed-off-by: Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
---
Changes from v5 to v6
o Rename 'used' bitmap to 'used_msi'
o Rename 'lock' spinlock to 'used_msi_lock'
o Take lock in interrupt handler
o Remove irq_dom in error path
---
 drivers/pci/host/pcie-tango.c | 225 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 225 insertions(+)

Comments

Marc Zyngier June 13, 2017, 2:22 p.m. UTC | #1
On 13/06/17 15:01, Marc Gonzalez wrote:
> The MSI controller in Tango supports 256 message-signaled interrupts,
> and a single doorbell address.
> 
> Signed-off-by: Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
> ---
> Changes from v5 to v6
> o Rename 'used' bitmap to 'used_msi'
> o Rename 'lock' spinlock to 'used_msi_lock'
> o Take lock in interrupt handler
> o Remove irq_dom in error path
> ---
>  drivers/pci/host/pcie-tango.c | 225 ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 225 insertions(+)
> 
> diff --git a/drivers/pci/host/pcie-tango.c b/drivers/pci/host/pcie-tango.c
> index 67aaadcc1c5e..b06446b23bc8 100644
> --- a/drivers/pci/host/pcie-tango.c
> +++ b/drivers/pci/host/pcie-tango.c
> @@ -1,16 +1,228 @@
> +#include <linux/irqchip/chained_irq.h>
> +#include <linux/irqdomain.h>
>  #include <linux/pci-ecam.h>
>  #include <linux/delay.h>
> +#include <linux/msi.h>
>  #include <linux/of.h>
>  
>  #define MSI_MAX 256
>  
>  #define SMP8759_MUX		0x48
>  #define SMP8759_TEST_OUT	0x74
> +#define SMP8759_STATUS		0x80
> +#define SMP8759_ENABLE		0xa0
> +#define SMP8759_DOORBELL	0xa002e07c
>  
>  struct tango_pcie {
> +	DECLARE_BITMAP(used_msi, MSI_MAX);
> +	spinlock_t used_msi_lock;
>  	void __iomem *mux;
> +	void __iomem *msi_status;
> +	void __iomem *msi_enable;
> +	phys_addr_t msi_doorbell;
> +	struct irq_domain *irq_dom;
> +	struct irq_domain *msi_dom;
> +	int irq;
>  };
>  
> +/*** MSI CONTROLLER SUPPORT ***/
> +
> +static void tango_msi_isr(struct irq_desc *desc)
> +{
> +	struct irq_chip *chip = irq_desc_get_chip(desc);
> +	struct tango_pcie *pcie = irq_desc_get_handler_data(desc);
> +	unsigned long flags, status, base, virq, idx, pos = 0;
> +
> +	chained_irq_enter(chip, desc);
> +	spin_lock_irqsave(&pcie->used_msi_lock, flags);

You're already in interrupt context, so there is no need to disable
interrupts any further. spin_lock() should do the trick

Thanks,

	M.
Marc Gonzalez June 13, 2017, 2:47 p.m. UTC | #2
On 13/06/2017 16:22, Marc Zyngier wrote:
> On 13/06/17 15:01, Marc Gonzalez wrote:
>> The MSI controller in Tango supports 256 message-signaled interrupts,
>> and a single doorbell address.
>>
>> Signed-off-by: Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
>> ---
>> Changes from v5 to v6
>> o Rename 'used' bitmap to 'used_msi'
>> o Rename 'lock' spinlock to 'used_msi_lock'
>> o Take lock in interrupt handler
>> o Remove irq_dom in error path
>> ---
>>  drivers/pci/host/pcie-tango.c | 225 ++++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 225 insertions(+)
>>
>> diff --git a/drivers/pci/host/pcie-tango.c b/drivers/pci/host/pcie-tango.c
>> index 67aaadcc1c5e..b06446b23bc8 100644
>> --- a/drivers/pci/host/pcie-tango.c
>> +++ b/drivers/pci/host/pcie-tango.c
>> @@ -1,16 +1,228 @@
>> +#include <linux/irqchip/chained_irq.h>
>> +#include <linux/irqdomain.h>
>>  #include <linux/pci-ecam.h>
>>  #include <linux/delay.h>
>> +#include <linux/msi.h>
>>  #include <linux/of.h>
>>  
>>  #define MSI_MAX 256
>>  
>>  #define SMP8759_MUX		0x48
>>  #define SMP8759_TEST_OUT	0x74
>> +#define SMP8759_STATUS		0x80
>> +#define SMP8759_ENABLE		0xa0
>> +#define SMP8759_DOORBELL	0xa002e07c
>>  
>>  struct tango_pcie {
>> +	DECLARE_BITMAP(used_msi, MSI_MAX);
>> +	spinlock_t used_msi_lock;
>>  	void __iomem *mux;
>> +	void __iomem *msi_status;
>> +	void __iomem *msi_enable;
>> +	phys_addr_t msi_doorbell;
>> +	struct irq_domain *irq_dom;
>> +	struct irq_domain *msi_dom;
>> +	int irq;
>>  };
>>  
>> +/*** MSI CONTROLLER SUPPORT ***/
>> +
>> +static void tango_msi_isr(struct irq_desc *desc)
>> +{
>> +	struct irq_chip *chip = irq_desc_get_chip(desc);
>> +	struct tango_pcie *pcie = irq_desc_get_handler_data(desc);
>> +	unsigned long flags, status, base, virq, idx, pos = 0;
>> +
>> +	chained_irq_enter(chip, desc);
>> +	spin_lock_irqsave(&pcie->used_msi_lock, flags);
> 
> You're already in interrupt context, so there is no need to disable
> interrupts any further. spin_lock() should do the trick

Thanks for the hint.

I am confused, because Documentation/locking/spinlocks.txt states:

> If you have a case where you have to protect a data structure across
> several CPU's and you want to use spinlocks you can potentially use
> cheaper versions of the spinlocks. IFF you know that the spinlocks are
> never used in interrupt handlers, you can use the non-irq versions:
> 
> 	spin_lock(&lock);
> 	...
> 	spin_unlock(&lock);
> 
> (and the equivalent read-write versions too, of course). The spinlock will
> guarantee the same kind of exclusive access, and it will be much faster.
> This is useful if you know that the data in question is only ever
> manipulated from a "process context", ie no interrupts involved.
> 
> The reasons you mustn't use these versions if you have interrupts that
> play with the spinlock is that you can get deadlocks:
> 
> 	spin_lock(&lock);
> 	...
> 		<- interrupt comes in:
> 			spin_lock(&lock);
> 
> where an interrupt tries to lock an already locked variable. This is ok if
> the other interrupt happens on another CPU, but it is _not_ ok if the
> interrupt happens on the same CPU that already holds the lock, because the
> lock will obviously never be released (because the interrupt is waiting
> for the lock, and the lock-holder is interrupted by the interrupt and will
> not continue until the interrupt has been processed).
> 
> (This is also the reason why the irq-versions of the spinlocks only need
> to disable the _local_ interrupts - it's ok to use spinlocks in interrupts
> on other CPU's, because an interrupt on another CPU doesn't interrupt the
> CPU that holds the lock, so the lock-holder can continue and eventually
> releases the lock).

Isn't this saying that it is not safe to call spin_lock() from
the interrupt handler? (Sorry if I misunderstood.)

Regards.
Marc Zyngier June 13, 2017, 4:53 p.m. UTC | #3
On 13/06/17 15:47, Marc Gonzalez wrote:
> On 13/06/2017 16:22, Marc Zyngier wrote:
>> On 13/06/17 15:01, Marc Gonzalez wrote:
>>> The MSI controller in Tango supports 256 message-signaled interrupts,
>>> and a single doorbell address.
>>>
>>> Signed-off-by: Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
>>> ---
>>> Changes from v5 to v6
>>> o Rename 'used' bitmap to 'used_msi'
>>> o Rename 'lock' spinlock to 'used_msi_lock'
>>> o Take lock in interrupt handler
>>> o Remove irq_dom in error path
>>> ---
>>>  drivers/pci/host/pcie-tango.c | 225 ++++++++++++++++++++++++++++++++++++++++++
>>>  1 file changed, 225 insertions(+)
>>>
>>> diff --git a/drivers/pci/host/pcie-tango.c b/drivers/pci/host/pcie-tango.c
>>> index 67aaadcc1c5e..b06446b23bc8 100644
>>> --- a/drivers/pci/host/pcie-tango.c
>>> +++ b/drivers/pci/host/pcie-tango.c
>>> @@ -1,16 +1,228 @@
>>> +#include <linux/irqchip/chained_irq.h>
>>> +#include <linux/irqdomain.h>
>>>  #include <linux/pci-ecam.h>
>>>  #include <linux/delay.h>
>>> +#include <linux/msi.h>
>>>  #include <linux/of.h>
>>>  
>>>  #define MSI_MAX 256
>>>  
>>>  #define SMP8759_MUX		0x48
>>>  #define SMP8759_TEST_OUT	0x74
>>> +#define SMP8759_STATUS		0x80
>>> +#define SMP8759_ENABLE		0xa0
>>> +#define SMP8759_DOORBELL	0xa002e07c
>>>  
>>>  struct tango_pcie {
>>> +	DECLARE_BITMAP(used_msi, MSI_MAX);
>>> +	spinlock_t used_msi_lock;
>>>  	void __iomem *mux;
>>> +	void __iomem *msi_status;
>>> +	void __iomem *msi_enable;
>>> +	phys_addr_t msi_doorbell;
>>> +	struct irq_domain *irq_dom;
>>> +	struct irq_domain *msi_dom;
>>> +	int irq;
>>>  };
>>>  
>>> +/*** MSI CONTROLLER SUPPORT ***/
>>> +
>>> +static void tango_msi_isr(struct irq_desc *desc)
>>> +{
>>> +	struct irq_chip *chip = irq_desc_get_chip(desc);
>>> +	struct tango_pcie *pcie = irq_desc_get_handler_data(desc);
>>> +	unsigned long flags, status, base, virq, idx, pos = 0;
>>> +
>>> +	chained_irq_enter(chip, desc);
>>> +	spin_lock_irqsave(&pcie->used_msi_lock, flags);
>>
>> You're already in interrupt context, so there is no need to disable
>> interrupts any further. spin_lock() should do the trick
> 
> Thanks for the hint.
> 
> I am confused, because Documentation/locking/spinlocks.txt states:
> 
>> If you have a case where you have to protect a data structure across
>> several CPU's and you want to use spinlocks you can potentially use
>> cheaper versions of the spinlocks. IFF you know that the spinlocks are
>> never used in interrupt handlers, you can use the non-irq versions:
>>
>> 	spin_lock(&lock);
>> 	...
>> 	spin_unlock(&lock);
>>
>> (and the equivalent read-write versions too, of course). The spinlock will
>> guarantee the same kind of exclusive access, and it will be much faster.
>> This is useful if you know that the data in question is only ever
>> manipulated from a "process context", ie no interrupts involved.
>>
>> The reasons you mustn't use these versions if you have interrupts that
>> play with the spinlock is that you can get deadlocks:
>>
>> 	spin_lock(&lock);
>> 	...
>> 		<- interrupt comes in:
>> 			spin_lock(&lock);
>>
>> where an interrupt tries to lock an already locked variable. This is ok if
>> the other interrupt happens on another CPU, but it is _not_ ok if the
>> interrupt happens on the same CPU that already holds the lock, because the
>> lock will obviously never be released (because the interrupt is waiting
>> for the lock, and the lock-holder is interrupted by the interrupt and will
>> not continue until the interrupt has been processed).
>>
>> (This is also the reason why the irq-versions of the spinlocks only need
>> to disable the _local_ interrupts - it's ok to use spinlocks in interrupts
>> on other CPU's, because an interrupt on another CPU doesn't interrupt the
>> CPU that holds the lock, so the lock-holder can continue and eventually
>> releases the lock).
> 
> Isn't this saying that it is not safe to call spin_lock() from
> the interrupt handler? (Sorry if I misunderstood.)

It is saying exactly the opposite.

If you take a spinlock and can be interrupted by an interrupt that takes
the same spinlock, then you must use the irq-safe version *outside of
the interrupt handler*. That's because Linux interrupts are not
preemptible (well, in general -- it is different with RT, but let's not
get there). If you're guaranteed that no interrupt handler will take
this spinlock, then you don't have to use the irq-safe version.

	M.
diff mbox

Patch

diff --git a/drivers/pci/host/pcie-tango.c b/drivers/pci/host/pcie-tango.c
index 67aaadcc1c5e..b06446b23bc8 100644
--- a/drivers/pci/host/pcie-tango.c
+++ b/drivers/pci/host/pcie-tango.c
@@ -1,16 +1,228 @@ 
+#include <linux/irqchip/chained_irq.h>
+#include <linux/irqdomain.h>
 #include <linux/pci-ecam.h>
 #include <linux/delay.h>
+#include <linux/msi.h>
 #include <linux/of.h>
 
 #define MSI_MAX 256
 
 #define SMP8759_MUX		0x48
 #define SMP8759_TEST_OUT	0x74
+#define SMP8759_STATUS		0x80
+#define SMP8759_ENABLE		0xa0
+#define SMP8759_DOORBELL	0xa002e07c
 
 struct tango_pcie {
+	DECLARE_BITMAP(used_msi, MSI_MAX);
+	spinlock_t used_msi_lock;
 	void __iomem *mux;
+	void __iomem *msi_status;
+	void __iomem *msi_enable;
+	phys_addr_t msi_doorbell;
+	struct irq_domain *irq_dom;
+	struct irq_domain *msi_dom;
+	int irq;
 };
 
+/*** MSI CONTROLLER SUPPORT ***/
+
+static void tango_msi_isr(struct irq_desc *desc)
+{
+	struct irq_chip *chip = irq_desc_get_chip(desc);
+	struct tango_pcie *pcie = irq_desc_get_handler_data(desc);
+	unsigned long flags, status, base, virq, idx, pos = 0;
+
+	chained_irq_enter(chip, desc);
+	spin_lock_irqsave(&pcie->used_msi_lock, flags);
+
+	while ((pos = find_next_bit(pcie->used_msi, MSI_MAX, pos)) < MSI_MAX) {
+		base = round_down(pos, 32);
+		status = readl_relaxed(pcie->msi_status + base / 8);
+		for_each_set_bit(idx, &status, 32) {
+			virq = irq_find_mapping(pcie->irq_dom, base + idx);
+			generic_handle_irq(virq);
+		}
+		pos = base + 32;
+	}
+
+	spin_unlock_irqrestore(&pcie->used_msi_lock, flags);
+	chained_irq_exit(chip, desc);
+}
+
+static void tango_ack(struct irq_data *d)
+{
+	struct tango_pcie *pcie = d->chip_data;
+	u32 offset = (d->hwirq / 32) * 4;
+	u32 bit = BIT(d->hwirq % 32);
+
+	writel_relaxed(bit, pcie->msi_status + offset);
+}
+
+static void update_msi_enable(struct irq_data *d, bool unmask)
+{
+	unsigned long flags;
+	struct tango_pcie *pcie = d->chip_data;
+	u32 offset = (d->hwirq / 32) * 4;
+	u32 bit = BIT(d->hwirq % 32);
+	u32 val;
+
+	spin_lock_irqsave(&pcie->used_msi_lock, flags);
+	val = readl_relaxed(pcie->msi_enable + offset);
+	val = unmask ? val | bit : val & ~bit;
+	writel_relaxed(val, pcie->msi_enable + offset);
+	spin_unlock_irqrestore(&pcie->used_msi_lock, flags);
+}
+
+static void tango_mask(struct irq_data *d)
+{
+	update_msi_enable(d, false);
+}
+
+static void tango_unmask(struct irq_data *d)
+{
+	update_msi_enable(d, true);
+}
+
+static int tango_set_affinity(struct irq_data *d,
+		const struct cpumask *mask, bool force)
+{
+	return -EINVAL;
+}
+
+static void tango_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
+{
+	struct tango_pcie *pcie = d->chip_data;
+	msg->address_lo = lower_32_bits(pcie->msi_doorbell);
+	msg->address_hi = upper_32_bits(pcie->msi_doorbell);
+	msg->data = d->hwirq;
+}
+
+static struct irq_chip tango_chip = {
+	.irq_ack		= tango_ack,
+	.irq_mask		= tango_mask,
+	.irq_unmask		= tango_unmask,
+	.irq_set_affinity	= tango_set_affinity,
+	.irq_compose_msi_msg	= tango_compose_msi_msg,
+};
+
+static void msi_ack(struct irq_data *d)
+{
+	irq_chip_ack_parent(d);
+}
+
+static void msi_mask(struct irq_data *d)
+{
+	pci_msi_mask_irq(d);
+	irq_chip_mask_parent(d);
+}
+
+static void msi_unmask(struct irq_data *d)
+{
+	pci_msi_unmask_irq(d);
+	irq_chip_unmask_parent(d);
+}
+
+static struct irq_chip msi_chip = {
+	.name = "MSI",
+	.irq_ack = msi_ack,
+	.irq_mask = msi_mask,
+	.irq_unmask = msi_unmask,
+};
+
+static struct msi_domain_info msi_dom_info = {
+	.flags	= MSI_FLAG_PCI_MSIX
+		| MSI_FLAG_USE_DEF_DOM_OPS
+		| MSI_FLAG_USE_DEF_CHIP_OPS,
+	.chip	= &msi_chip,
+};
+
+static int tango_irq_domain_alloc(struct irq_domain *dom,
+		unsigned int virq, unsigned int nr_irqs, void *args)
+{
+	unsigned long flags;
+	int pos, err = -ENOSPC;
+	struct tango_pcie *pcie = dom->host_data;
+
+	spin_lock_irqsave(&pcie->used_msi_lock, flags);
+	pos = find_first_zero_bit(pcie->used_msi, MSI_MAX);
+	if (pos < MSI_MAX) {
+		err = 0;
+		__set_bit(pos, pcie->used_msi);
+		irq_domain_set_info(dom, virq, pos,
+				&tango_chip, pcie, handle_edge_irq, NULL, NULL);
+	}
+	spin_unlock_irqrestore(&pcie->used_msi_lock, flags);
+
+	return err;
+}
+
+static void tango_irq_domain_free(struct irq_domain *dom,
+		unsigned int virq, unsigned int nr_irqs)
+{
+	unsigned long flags;
+	struct irq_data *d = irq_domain_get_irq_data(dom, virq);
+	struct tango_pcie *pcie = d->chip_data;
+
+	spin_lock_irqsave(&pcie->used_msi_lock, flags);
+	__clear_bit(d->hwirq, pcie->used_msi);
+	spin_unlock_irqrestore(&pcie->used_msi_lock, flags);
+}
+
+static const struct irq_domain_ops irq_dom_ops = {
+	.alloc	= tango_irq_domain_alloc,
+	.free	= tango_irq_domain_free,
+};
+
+static int tango_msi_remove(struct platform_device *pdev)
+{
+	struct tango_pcie *pcie = platform_get_drvdata(pdev);
+
+	irq_set_chained_handler_and_data(pcie->irq, NULL, NULL);
+	irq_domain_remove(pcie->msi_dom);
+	irq_domain_remove(pcie->irq_dom);
+
+	return 0;
+}
+
+static int tango_msi_probe(struct platform_device *pdev, struct tango_pcie *pcie)
+{
+	int i, virq;
+	struct irq_domain *msi_dom, *irq_dom;
+	struct fwnode_handle *fwnode = of_node_to_fwnode(pdev->dev.of_node);
+
+	spin_lock_init(&pcie->used_msi_lock);
+	for (i = 0; i < MSI_MAX / 32; ++i)
+		writel_relaxed(0, pcie->msi_enable + i * 4);
+
+	virq = platform_get_irq(pdev, 1);
+	if (virq <= 0) {
+		pr_err("Failed to map IRQ\n");
+		return -ENXIO;
+	}
+
+	irq_dom = irq_domain_create_linear(fwnode, MSI_MAX, &irq_dom_ops, pcie);
+	if (!irq_dom) {
+		pr_err("Failed to create IRQ domain\n");
+		return -ENOMEM;
+	}
+
+	msi_dom = pci_msi_create_irq_domain(fwnode, &msi_dom_info, irq_dom);
+	if (!msi_dom) {
+		pr_err("Failed to create MSI domain\n");
+		irq_domain_remove(irq_dom);
+		return -ENOMEM;
+	}
+
+	pcie->irq_dom = irq_dom;
+	pcie->msi_dom = msi_dom;
+	pcie->irq = virq;
+
+	irq_set_chained_handler_and_data(virq, tango_msi_isr, pcie);
+
+	return 0;
+}
+
 /*** HOST BRIDGE SUPPORT ***/
 
 static int smp8759_config_read(struct pci_bus *bus,
@@ -88,6 +300,9 @@  static int tango_check_pcie_link(void __iomem *test_out)
 static int smp8759_init(struct tango_pcie *pcie, void __iomem *base)
 {
 	pcie->mux		= base + SMP8759_MUX;
+	pcie->msi_status	= base + SMP8759_STATUS;
+	pcie->msi_enable	= base + SMP8759_ENABLE;
+	pcie->msi_doorbell	= SMP8759_DOORBELL;
 
 	return tango_check_pcie_link(base + SMP8759_TEST_OUT);
 }
@@ -121,11 +336,21 @@  static int tango_pcie_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	ret = tango_msi_probe(pdev, pcie);
+	if (ret)
+		return ret;
+
 	return pci_host_common_probe(pdev, &smp8759_ecam_ops);
 }
 
+static int tango_pcie_remove(struct platform_device *pdev)
+{
+	return tango_msi_remove(pdev);
+}
+
 static struct platform_driver tango_pcie_driver = {
 	.probe	= tango_pcie_probe,
+	.remove	= tango_pcie_remove,
 	.driver	= {
 		.name = KBUILD_MODNAME,
 		.of_match_table = tango_pcie_ids,