Skip to contentSkip to conversation

pmi: initial version

PATCH In review Christian Krafft 7 participants 17 messages Feb 9, 2007 – Feb 15, 2007 PowerPC

lore
Share & tools
lore

Christian Krafft

author
#0 lore raw

cover pmi: initial version

This patch adds a driver for the=20
"Platform Management Interrupt" interface.
PMI is basically yet another way to communicate with the BMC. It provides small latency and thus can be used=20 to request frequency changes from the BMC.
PMI allows bidirectional communication.=20
The cpufreq driver can be notified about a new limit for=20 frequency changes for instance.
PMI can be found on future IBM products.

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

author
#1 lore raw

This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>
arch/powerpc/Kconfig +9 -0
Index: linux/arch/powerpc/Kconfig
===================================================================
@@ -524,6 +524,7 @@ config PPC_IBM_CELL_BLADE
 	select MMIO_NVRAM
 	select PPC_UDBG_16550
 	select UDBG_RTAS_CONSOLE
+#	select PPC_PMI
 
 config PPC_PS3
 	bool "Sony PS3 (incomplete)"

@@ -577,6 +578,14 @@ config RTAS_FLASH
 	tristate "Firmware flash interface"
 	depends on PPC64 && RTAS_PROC
 
+config PPC_PMI
+	tristate "Support for PMI"
+	depends PPC_IBM_CELL_BLADE
+	help
+	  PMI is a way to communicate with the board mangement controller.
+	  It is used in some IBM Cell blades.
+	default m
+
 config MMIO_NVRAM
 	bool
 	default n
arch/powerpc/sysdev/Makefile +1 -0
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
@@ -7,6 +7,7 @@ obj-$(CONFIG_PPC_INDIRECT_PCI)	+= indire
 obj-$(CONFIG_PPC_MPC106)	+= grackle.o
 obj-$(CONFIG_PPC_DCR)		+= dcr.o
 obj-$(CONFIG_PPC_DCR_NATIVE)	+= dcr-low.o
+obj-$(CONFIG_PPC_PMI)		+= pmi.o
 obj-$(CONFIG_U3_DART)		+= dart_iommu.o
 obj-$(CONFIG_MMIO_NVRAM)	+= mmio_nvram.o
 obj-$(CONFIG_FSL_SOC)		+= fsl_soc.o
arch/powerpc/sysdev/pmi.c new file +275 -0

275 lines not on this page · raw message

include/asm-powerpc/pmi.h new file +63 -0

63 lines not on this page · raw message

#2 lore raw

Hi Christian-

Some minor comments.

Quoted 11 lines · Christian Krafft

This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>

Index: linux/arch/powerpc/Kconfig
===================================================================

arch/powerpc/Kconfig config PPC_IBM_CELL_BLADE · line 527

Quoted 4 lines
arch/powerpc/Kconfig +1 -0
@@ -524,6 +524,7 @@ config PPC_IBM_CELL_BLADE
 	select MMIO_NVRAM
 	select PPC_UDBG_16550
 	select UDBG_RTAS_CONSOLE
+#	select PPC_PMI

Did you mean for this to be commented out?

<snip>

arch/powerpc/Kconfig config PPC_IBM_CELL_BLADE

Quoted 9 lines
arch/powerpc/Kconfig +9 -0
+static void __iomem *of_iomap(struct device_node *np)
+{
+	struct resource res;
+
+	if (of_address_to_resource(np, 0, &res))
+		return NULL;
+
+	pr_debug("Resource start: 0x%lx\n", res.start);
+        pr_debug("Resource end: 0x%lx\n", res.end);

You need a tab instead of spaces.

arch/powerpc/Kconfig

Quoted 13 lines
arch/powerpc/Kconfig +13 -0
+
+	return ioremap(res.start, 1 + res.end - res.start);
+}
+
+static int pmi_irq_handler(int irq, void *dev_id)
+{
+	struct pmi_data *data;
+	int type;
+	struct pmi_handler *handler;
+
+	data = dev_id;
+
+	BUG_ON(!data);

Not necessary, you'll oops in two lines anyway if data is NULL.

Quoted 7 lines
arch/powerpc/Kconfig +7 -0
+
+	pr_debug("pmi: got a PMI message\n");
+
+	spin_lock(&data->pmi_spinlock);
+	type = ioread8(data->pmi_reg + PMI_READ_TYPE);
+	spin_unlock(&data->pmi_spinlock);
+

arch/powerpc/Kconfig

Quoted 10 lines
arch/powerpc/Kconfig +10 -0
+	pr_debug("pmi: message type is %d\n", type);
+
+	if (type & PMI_ACK) {
+		BUG_ON(!data->msg);
+		BUG_ON(!data->completion);
+		pr_debug("pmi: got an ACK\n");
+		data->msg->type = type;
+		data->msg->data0 = ioread8(data->pmi_reg + PMI_READ_DATA0);
+		data->msg->data1 = ioread8(data->pmi_reg + PMI_READ_DATA1);
+		data->msg->data2 = ioread8(data->pmi_reg + PMI_READ_DATA2);

Should these accesses to data->pmi_reg be performed while holding data->pmi_spinlock as well?

Quoted 29 lines
arch/powerpc/Kconfig +29 -0
+		complete(data->completion);
+		return IRQ_HANDLED;
+	}
+
+	spin_lock(&data->handler_spinlock);
+	list_for_each_entry(handler, &data->handler, node) {
+		pr_debug("pmi: notifying handlers\n");
+		if (handler->type == type) {
+			pr_debug("pmi: notify handler %p\n", handler);
+			handler->handle_pmi_message(data->dev, data->msg);
+		}
+	}
+	spin_unlock(&data->handler_spinlock);
+
+	return IRQ_HANDLED;
+}
+
+
+static struct of_device_id pmi_match[] = {
+	{ .type = "ibm,pmi", .name = "pmi" },
+	{},
+};
+
+MODULE_DEVICE_TABLE(of, pmi_match);
+
+static int pmi_of_probe(struct of_device *dev,
+			const struct of_device_id *match)
+{
+	DEFINE_SPINLOCK(handler_spinlock);

arch/powerpc/Kconfig

Quoted 10 lines
arch/powerpc/Kconfig +10 -0
+	DEFINE_SPINLOCK(pmi_spinlock);
+
+	struct device_node *np = dev->node;
+	struct pmi_data *data;
+	int rc;
+
+	data = kzalloc(sizeof(struct pmi_data), GFP_KERNEL);
+	if (!data) {
+		printk(KERN_ERR "pmi: could not allocate memory.\n");
+		return -EFAULT;

-ENOMEM is the appropriate error code here.

Quoted 20 lines
arch/powerpc/Kconfig +20 -0
+	}
+
+	data->pmi_reg = of_iomap(np);
+
+	if (!data->pmi_reg) {
+		printk(KERN_ERR "pmi: invalid register address.\n");
+		kfree(data);
+		return -EFAULT;
+	}
+
+	INIT_LIST_HEAD(&data->handler);
+
+	data->irq = irq_of_parse_and_map(np, 0);
+	if (data->irq == NO_IRQ) {
+		printk(KERN_ERR "pmi: invalid interrupt.\n");
+		iounmap(data->pmi_reg);
+		kfree(data);
+		return -EFAULT;
+	}
+

arch/powerpc/Kconfig

Quoted 10 lines
arch/powerpc/Kconfig +10 -0
+	data->handler_spinlock = handler_spinlock;
+	data->pmi_spinlock = pmi_spinlock;
+
+	rc = request_irq(data->irq, pmi_irq_handler,
+			IRQF_SHARED, "pmi", data);
+	if (rc) {
+		printk(KERN_ERR "pmi: can't request IRQ %d: returned %d\n",
+			data->irq, rc);
+		iounmap(data->pmi_reg);
+		kfree(data);

Consider using goto to reduce duplication of the iounmap, kfree sequence.

arch/powerpc/Kconfig

Quoted 1 line
arch/powerpc/Kconfig +1 -0
+		return -EFAULT;

Just return rc; request_irq returns standard error codes.

<snip>

arch/powerpc/Kconfig

Quoted 7 lines
arch/powerpc/Kconfig +7 -0
+void pmi_send_message(struct of_device *device, struct pmi_message *msg)
+{
+	struct pmi_data *data;
+	unsigned long flags;
+	DECLARE_COMPLETION_ONSTACK(completion);
+
+	BUG_ON(!device || !msg);

This BUG_ON is also unnecessary, and could actually make debugging more difficult since it wouldn't be immediately apparent which condition triggered it.

Quoted 29 lines
arch/powerpc/Kconfig +29 -0
+
+	data = device->dev.driver_data;
+	pr_debug("pmi_send_message: data is %p\n", data);
+
+	mutex_lock(&data->msg_mutex);
+
+	data->msg = msg;
+
+	pr_debug("pmi_send_message: msg is %p\n", msg);
+
+	data->completion = &completion;
+
+	spin_lock_irqsave(&data->pmi_spinlock, flags);
+	iowrite8(msg->data0, data->pmi_reg + PMI_WRITE_DATA0);
+	iowrite8(msg->data1, data->pmi_reg + PMI_WRITE_DATA1);
+	iowrite8(msg->data2, data->pmi_reg + PMI_WRITE_DATA2);
+	iowrite8(msg->type, data->pmi_reg + PMI_WRITE_TYPE);
+	spin_unlock_irqrestore(&data->pmi_spinlock, flags);
+
+	pr_debug(KERN_INFO "pmi_send_message: wait for completion %p\n", data->completion);
+
+	wait_for_completion_interruptible_timeout(data->completion, PMI_TIMEOUT);
+
+	data->msg = NULL;
+	data->completion = NULL;
+
+	mutex_unlock(&data->msg_mutex);
+}
+EXPORT_SYMBOL_GPL(pmi_send_message);

...

Quoted 1 line

+EXPORT_SYMBOL_GPL(pmi_register_handler);

...

Quoted 1 line

+EXPORT_SYMBOL_GPL(pmi_unregister_handler);

Are the symbol exports necessary?

<snip>

arch/powerpc/Kconfig

Quoted 4 lines
arch/powerpc/Kconfig +4 -0
+void pmi_register_handler(struct of_device *, struct pmi_handler *);
+void pmi_unregister_handler(struct of_device *, struct pmi_handler *);
+
+void pmi_send_message(struct of_device *, struct pmi_message *);

Are there users of these functions?

author
#3 lore raw

Hi Nathan,

thanks for your comments.
Will send an updated version.

Quoted 6 lines · On Fri, 9 Feb 2007 16:29:56 -0600 Nathan Lynch <[email protected]> wrote:

Hi Christian-
=20
Some minor comments.
=20

Quoted 1 line · Christian Krafft

This patch adds driver code for the PMI device found in future IBM prod=

ucts.

Quoted 1 line
Quoted 1 line

PMI stands for "Platform Management Interrupt" and is a way to communic=

ate

Quoted 1 line
Quoted 1 line

with the BMC. It provides bidirectional communication with a low latenc=

y.

Quoted 5 lines
Quoted 5 lines

=20

Signed-off-by Christian Krafft <[email protected]>

=20
Index: linux/arch/powerpc/Kconfig =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=

=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

Quoted 9 lines

arch/powerpc/Kconfig config PPC_IBM_CELL_BLADE · line 527

Quoted 7 lines
arch/powerpc/Kconfig +1 -0
@@ -524,6 +524,7 @@ config PPC_IBM_CELL_BLADE
 	select MMIO_NVRAM
 	select PPC_UDBG_16550
 	select UDBG_RTAS_CONSOLE
+#	select PPC_PMI

=20
Did you mean for this to be commented out?

no, good point ;-)

Quoted 14 lines

=20
<snip>
=20

arch/powerpc/Kconfig config PPC_IBM_CELL_BLADE

Quoted 9 lines
arch/powerpc/Kconfig +9 -0
+static void __iomem *of_iomap(struct device_node *np)
+{
+	struct resource res;
+
+	if (of_address_to_resource(np, 0, &res))
+		return NULL;
+
+	pr_debug("Resource start: 0x%lx\n", res.start);
+        pr_debug("Resource end: 0x%lx\n", res.end);

=20
You need a tab instead of spaces.

copy and past bug,

Quoted 17 lines

=20
=20

arch/powerpc/Kconfig

Quoted 13 lines
arch/powerpc/Kconfig +13 -0
+
+	return ioremap(res.start, 1 + res.end - res.start);
+}
+
+static int pmi_irq_handler(int irq, void *dev_id)
+{
+	struct pmi_data *data;
+	int type;
+	struct pmi_handler *handler;
+
+	data =3D dev_id;
+
+	BUG_ON(!data);

=20
Not necessary, you'll oops in two lines anyway if data is NULL.

ok,

Quoted 14 lines

=20
=20

Quoted 12 lines
arch/powerpc/Kconfig +12 -0
+
+	pr_debug("pmi: got a PMI message\n");
+
+	spin_lock(&data->pmi_spinlock);
+	type =3D ioread8(data->pmi_reg + PMI_READ_TYPE);
+	spin_unlock(&data->pmi_spinlock);
+
+	pr_debug("pmi: message type is %d\n", type);
+
+	if (type & PMI_ACK) {
+		BUG_ON(!data->msg);
+		BUG_ON(!data->completion);
Quoted 8 lines

arch/powerpc/Kconfig

Quoted 5 lines
arch/powerpc/Kconfig +5 -0
+		pr_debug("pmi: got an ACK\n");
+		data->msg->type =3D type;
+		data->msg->data0 =3D ioread8(data->pmi_reg + PMI_READ_DATA0);
+		data->msg->data1 =3D ioread8(data->pmi_reg + PMI_READ_DATA1);
+		data->msg->data2 =3D ioread8(data->pmi_reg + PMI_READ_DATA2);

=20
Should these accesses to data->pmi_reg be performed while holding data->pmi_spinlock as well?

yep, good catch.

Quoted 35 lines

=20
=20

Quoted 33 lines
arch/powerpc/Kconfig +33 -0
+		complete(data->completion);
+		return IRQ_HANDLED;
+	}
+
+	spin_lock(&data->handler_spinlock);
+	list_for_each_entry(handler, &data->handler, node) {
+		pr_debug("pmi: notifying handlers\n");
+		if (handler->type =3D=3D type) {
+			pr_debug("pmi: notify handler %p\n", handler);
+			handler->handle_pmi_message(data->dev, data->msg);
+		}
+	}
+	spin_unlock(&data->handler_spinlock);
+
+	return IRQ_HANDLED;
+}
+
+
+static struct of_device_id pmi_match[] =3D {
+	{ .type =3D "ibm,pmi", .name =3D "pmi" },
+	{},
+};
+
+MODULE_DEVICE_TABLE(of, pmi_match);
+
+static int pmi_of_probe(struct of_device *dev,
+			const struct of_device_id *match)
+{
+	DEFINE_SPINLOCK(handler_spinlock);
+	DEFINE_SPINLOCK(pmi_spinlock);
+
+	struct device_node *np =3D dev->node;
+	struct pmi_data *data;
Quoted 8 lines

arch/powerpc/Kconfig

Quoted 6 lines
arch/powerpc/Kconfig +6 -0
+	int rc;
+
+	data =3D kzalloc(sizeof(struct pmi_data), GFP_KERNEL);
+	if (!data) {
+		printk(KERN_ERR "pmi: could not allocate memory.\n");
+		return -EFAULT;

=20
-ENOMEM is the appropriate error code here.

will be updated.

Quoted 35 lines

=20
=20

Quoted 20 lines
arch/powerpc/Kconfig +20 -0
+	}
+
+	data->pmi_reg =3D of_iomap(np);
+
+	if (!data->pmi_reg) {
+		printk(KERN_ERR "pmi: invalid register address.\n");
+		kfree(data);
+		return -EFAULT;
+	}
+
+	INIT_LIST_HEAD(&data->handler);
+
+	data->irq =3D irq_of_parse_and_map(np, 0);
+	if (data->irq =3D=3D NO_IRQ) {
+		printk(KERN_ERR "pmi: invalid interrupt.\n");
+		iounmap(data->pmi_reg);
+		kfree(data);
+		return -EFAULT;
+	}
+

arch/powerpc/Kconfig

Quoted 10 lines
arch/powerpc/Kconfig +10 -0
+	data->handler_spinlock =3D handler_spinlock;
+	data->pmi_spinlock =3D pmi_spinlock;
+
+	rc =3D request_irq(data->irq, pmi_irq_handler,
+			IRQF_SHARED, "pmi", data);
+	if (rc) {
+		printk(KERN_ERR "pmi: can't request IRQ %d: returned %d\n",
+			data->irq, rc);
+		iounmap(data->pmi_reg);
+		kfree(data);

=20
Consider using goto to reduce duplication of the iounmap, kfree sequence. =20

Quoted 3 lines

arch/powerpc/Kconfig

Quoted 1 line
arch/powerpc/Kconfig +1 -0
+		return -EFAULT;

=20
Just return rc; request_irq returns standard error codes.

ok,

Quoted 4 lines

=20
<snip>
=20

Quoted 1 line

+void pmi_send_message(struct of_device *device, struct pmi_message *ms=

g)

Quoted 10 lines

arch/powerpc/Kconfig

Quoted 6 lines
arch/powerpc/Kconfig +6 -0
+{
+	struct pmi_data *data;
+	unsigned long flags;
+	DECLARE_COMPLETION_ONSTACK(completion);
+
+	BUG_ON(!device || !msg);

=20
This BUG_ON is also unnecessary, and could actually make debugging more difficult since it wouldn't be immediately apparent which condition triggered it.

right,

Quoted 13 lines

=20

Quoted 12 lines
arch/powerpc/Kconfig +12 -0
+
+	data =3D device->dev.driver_data;
+	pr_debug("pmi_send_message: data is %p\n", data);
+
+	mutex_lock(&data->msg_mutex);
+
+	data->msg =3D msg;
+
+	pr_debug("pmi_send_message: msg is %p\n", msg);
+
+	data->completion =3D &completion;
+
Quoted 8 lines

arch/powerpc/Kconfig

Quoted 8 lines
arch/powerpc/Kconfig +8 -0
+	spin_lock_irqsave(&data->pmi_spinlock, flags);
+	iowrite8(msg->data0, data->pmi_reg + PMI_WRITE_DATA0);
+	iowrite8(msg->data1, data->pmi_reg + PMI_WRITE_DATA1);
+	iowrite8(msg->data2, data->pmi_reg + PMI_WRITE_DATA2);
+	iowrite8(msg->type, data->pmi_reg + PMI_WRITE_TYPE);
+	spin_unlock_irqrestore(&data->pmi_spinlock, flags);
+
+	pr_debug(KERN_INFO "pmi_send_message: wait for completion %p\n", data=

->completion);

Quoted 2 lines

arch/powerpc/Kconfig

Quoted 2 lines
arch/powerpc/Kconfig +2 -0
+
+	wait_for_completion_interruptible_timeout(data->completion, PMI_TIMEO=

UT);

Quoted 16 lines
Quoted 7 lines
arch/powerpc/Kconfig +7 -0
+
+	data->msg =3D NULL;
+	data->completion =3D NULL;
+
+	mutex_unlock(&data->msg_mutex);
+}
+EXPORT_SYMBOL_GPL(pmi_send_message);

...

Quoted 1 line

+EXPORT_SYMBOL_GPL(pmi_register_handler);

...

Quoted 1 line

+EXPORT_SYMBOL_GPL(pmi_unregister_handler);

=20
Are the symbol exports necessary?
=20
<snip>
=20

Quoted 7 lines

arch/powerpc/Kconfig

Quoted 4 lines
arch/powerpc/Kconfig +4 -0
+void pmi_register_handler(struct of_device *, struct pmi_handler *);
+void pmi_unregister_handler(struct of_device *, struct pmi_handler *);
+
+void pmi_send_message(struct of_device *, struct pmi_message *);

=20
Are there users of these functions?
=20

yes, they will be used by the cbe_cpufreq driver.
I will send a patch for it, as soon as i was able to test it, hopefully today or tomorrow.

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

author
#4 lore raw

Subject: powerpc: add PMI driver for cell blade

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>

---
Index: linux/arch/powerpc/Kconfig =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

arch/powerpc/Kconfig +0 -0
@@ -577,6 +577,14 @@ config RTAS_FLASH
 	tristate "Firmware flash interface"
 	depends on PPC64 && RTAS_PROC

=20
+config PPC_PMI
+ tristate "Support for PMI"
+ depends PPC_IBM_CELL_BLADE
+ help
+ PMI is a way to communicate with the board mangement controller.
+ It is used in some IBM Cell blades.
+ default m
+
config MMIO_NVRAM
bool
default n
Index: linux/arch/powerpc/sysdev/Makefile =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

arch/powerpc/sysdev/Makefile +1 -0
@@ -7,6 +7,7 @@ obj-$(CONFIG_PPC_INDIRECT_PCI)	+=3D indire
 obj-$(CONFIG_PPC_MPC106)	+=3D grackle.o
 obj-$(CONFIG_PPC_DCR)		+=3D dcr.o
 obj-$(CONFIG_PPC_DCR_NATIVE)	+=3D dcr-low.o
+obj-$(CONFIG_PPC_PMI)		+=3D pmi.o
 obj-$(CONFIG_U3_DART)		+=3D dart_iommu.o
 obj-$(CONFIG_MMIO_NVRAM)	+=3D mmio_nvram.o
 obj-$(CONFIG_FSL_SOC)		+=3D fsl_soc.o

Index: linux/arch/powerpc/sysdev/pmi.c =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

arch/powerpc/sysdev/pmi.c new file +291 -0

291 lines not on this page · raw message

Index: linux/include/asm-powerpc/pmi.h =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

include/asm-powerpc/pmi.h new file +63 -0

63 lines not on this page · raw message

Index: linux/arch/powerpc/configs/cell_defconfig =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

arch/powerpc/configs/cell_defconfig +1 -0
@@ -147,6 +147,7 @@ CONFIG_PPC_RTAS=3Dy
 # CONFIG_RTAS_ERROR_LOGGING is not set
 CONFIG_RTAS_PROC=3Dy
 CONFIG_RTAS_FLASH=3Dy
+CONFIG_PPC_PMI=3Dm
 CONFIG_MMIO_NVRAM=3Dy
 # CONFIG_PPC_MPC106 is not set
 # CONFIG_PPC_970_NAP is not set

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

author
#5 lore raw

Hi,

sorry for resending this patch again.
I added a work queue to the interrupt handler to shorten the time we run with interrupts off.

Cheers,
ck

author
#6 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Subject: powerpc: add PMI driver for cell blade

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>

---

arch/powerpc/Kconfig +8 -0
Index: linux/arch/powerpc/Kconfig
===================================================================
@@ -577,6 +577,14 @@ config RTAS_FLASH
 	tristate "Firmware flash interface"
 	depends on PPC64 && RTAS_PROC
 
+config PPC_PMI
+	tristate "Support for PMI"
+	depends PPC_IBM_CELL_BLADE
+	help
+	  PMI is a way to communicate with the board mangement controller.
+	  It is used in some IBM Cell blades.
+	default m
+
 config MMIO_NVRAM
 	bool
 	default n
arch/powerpc/sysdev/Makefile +1 -0
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
@@ -7,6 +7,7 @@ obj-$(CONFIG_PPC_INDIRECT_PCI)	+= indire
 obj-$(CONFIG_PPC_MPC106)	+= grackle.o
 obj-$(CONFIG_PPC_DCR)		+= dcr.o
 obj-$(CONFIG_PPC_DCR_NATIVE)	+= dcr-low.o
+obj-$(CONFIG_PPC_PMI)		+= pmi.o
 obj-$(CONFIG_U3_DART)		+= dart_iommu.o
 obj-$(CONFIG_MMIO_NVRAM)	+= mmio_nvram.o
 obj-$(CONFIG_FSL_SOC)		+= fsl_soc.o
arch/powerpc/sysdev/pmi.c new file +312 -0

312 lines not on this page · raw message

include/asm-powerpc/pmi.h new file +63 -0

63 lines not on this page · raw message

arch/powerpc/configs/cell_defconfig +1 -0
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
@@ -147,6 +147,7 @@ CONFIG_PPC_RTAS=y
 # CONFIG_RTAS_ERROR_LOGGING is not set
 CONFIG_RTAS_PROC=y
 CONFIG_RTAS_FLASH=y
+CONFIG_PPC_PMI=m
 CONFIG_MMIO_NVRAM=y
 # CONFIG_PPC_MPC106 is not set
 # CONFIG_PPC_970_NAP is not set
author
#7 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

arch/powerpc/sysdev/pmi.c line 171

Quoted 4 lines · On Tue, 13 Feb 2007 19:48:06 +0100 Christian Krafft <[email protected]> wrote:
arch/powerpc/sysdev/pmi.c new file +4 -0
+
+	data->work_queue =3D create_workqueue("pmi");
+	INIT_WORK(&data->work, pmi_notify_handlers);
+

That's not what I wanted!
Will send an update.

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

author
#8 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

In reply to Christian Krafft

Subject: powerpc: add PMI driver for cell blade

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>

---

arch/powerpc/Kconfig +8 -0
Index: linux/arch/powerpc/Kconfig
===================================================================
@@ -577,6 +577,14 @@ config RTAS_FLASH
 	tristate "Firmware flash interface"
 	depends on PPC64 && RTAS_PROC
 
+config PPC_PMI
+	tristate "Support for PMI"
+	depends PPC_IBM_CELL_BLADE
+	help
+	  PMI is a way to communicate with the board mangement controller.
+	  It is used in some IBM Cell blades.
+	default m
+
 config MMIO_NVRAM
 	bool
 	default n
arch/powerpc/sysdev/Makefile +1 -0
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
@@ -7,6 +7,7 @@ obj-$(CONFIG_PPC_INDIRECT_PCI)	+= indire
 obj-$(CONFIG_PPC_MPC106)	+= grackle.o
 obj-$(CONFIG_PPC_DCR)		+= dcr.o
 obj-$(CONFIG_PPC_DCR_NATIVE)	+= dcr-low.o
+obj-$(CONFIG_PPC_PMI)		+= pmi.o
 obj-$(CONFIG_U3_DART)		+= dart_iommu.o
 obj-$(CONFIG_MMIO_NVRAM)	+= mmio_nvram.o
 obj-$(CONFIG_FSL_SOC)		+= fsl_soc.o
arch/powerpc/sysdev/pmi.c new file +307 -0

307 lines not on this page · raw message

include/asm-powerpc/pmi.h new file +63 -0

63 lines not on this page · raw message

arch/powerpc/configs/cell_defconfig +1 -0
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
@@ -147,6 +147,7 @@ CONFIG_PPC_RTAS=y
 # CONFIG_RTAS_ERROR_LOGGING is not set
 CONFIG_RTAS_PROC=y
 CONFIG_RTAS_FLASH=y
+CONFIG_PPC_PMI=m
 CONFIG_MMIO_NVRAM=y
 # CONFIG_PPC_MPC106 is not set
 # CONFIG_PPC_970_NAP is not set
#9 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 4 lines · Christian Krafft

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC. It provides bidirectional communication with a low latency.

Thanks for telling us what PMI is, but what's a BMC?

Paul.

author
#10 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 4 lines · On Wed, 14 Feb 2007 11:30:51 +1100 Paul Mackerras <[email protected]> wrote:

Christian Krafft writes:
=20

Quoted 2 lines

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM prod=

ucts.

Quoted 1 line
Quoted 1 line

PMI stands for "Platform Management Interrupt" and is a way to communic=

ate

Quoted 1 line
Quoted 1 line

with the BMC. It provides bidirectional communication with a low latenc=

y.

Quoted 4 lines

=20
Thanks for telling us what PMI is, but what's a BMC? =20
Paul.

BMC stands for Baseboard Management Controller:
http://en.wikipedia.org/wiki/Baseboard_management_controller

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

#11 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 2 lines · Christian Krafft

BMC stands for Baseboard Management Controller:
http://en.wikipedia.org/wiki/Baseboard_management_controller

Thanks.

I was actually hinting that it would be good to explain that acronym as well as PMI in the description of the patch. I should have said so explicitly.

Paul.

author
#12 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 4 lines · On Wed, 14 Feb 2007 21:22:16 +1100 Paul Mackerras <[email protected]> wrote:

Christian Krafft writes:
=20

Quoted 2 lines

BMC stands for Baseboard Management Controller:
http://en.wikipedia.org/wiki/Baseboard_management_controller

Quoted 8 lines

=20
Thanks.
=20
I was actually hinting that it would be good to explain that acronym as well as PMI in the description of the patch. I should have said so explicitly.
=20
Paul.

No need to, I also don't like TLA's ;-)
Will send an update.=20

--=20
Mit freundlichen Gr=FCssen,
kind regards,

Christian Krafft
IBM Systems & Technology Group,=20
Linux Kernel Development
IT Specialist

#13 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Subject: powerpc: add PMI driver for cell blade

From: Christian Krafft <[email protected]>
This patch adds driver code for the PMI device found in future IBM products. PMI stands for "Platform Management Interrupt" and is a way to communicate with the BMC (Baseboard Management Controller).
It provides bidirectional communication with a low latency.

Signed-off-by Christian Krafft <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>

---

arch/powerpc/sysdev/pmi.c new file +305 -0
Index: linux/arch/powerpc/sysdev/pmi.c
===================================================================
@@ -0,0 +1,305 @@
+/*
+ * pmi driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * PMI (Platform Management Interrupt) is a way to communicate
+ * with the BMC (Baseboard Management Controller) via interrupts.
+ * Unlike IPMI it is bidirectional and has a low latency.
+ *
+ * Author: Christian Krafft <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/completion.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+
+#include <asm/of_device.h>
+#include <asm/of_platform.h>
+#include <asm/io.h>
+#include <asm/pmi.h>
+
+
+struct pmi_data {
+	struct list_head	handler;
+	spinlock_t		handler_spinlock;
+	spinlock_t		pmi_spinlock;
+	struct mutex		msg_mutex;
+	pmi_message_t		msg;
+	struct completion	*completion;
+	struct of_device	*dev;
+	int			irq;
+	u8 __iomem		*pmi_reg;
+	struct work_struct	work;
+};
+
+
+
+static void __iomem *of_iomap(struct device_node *np)
+{
+	struct resource res;
+
+	if (of_address_to_resource(np, 0, &res))
+		return NULL;
+
+	pr_debug("Resource start: 0x%lx\n", res.start);
+	pr_debug("Resource end: 0x%lx\n", res.end);
+
+	return ioremap(res.start, 1 + res.end - res.start);
+}
+
+
+static int pmi_irq_handler(int irq, void *dev_id)
+{
+	struct pmi_data *data;
+	u8 type;
+	int rc;
+
+	data = dev_id;
+
+	spin_lock(&data->pmi_spinlock);
+
+	type = ioread8(data->pmi_reg + PMI_READ_TYPE);
+	pr_debug("pmi: got message of type %d\n", type);
+
+	if (type & PMI_ACK && !data->completion) {
+		printk(KERN_WARNING "pmi: got unexpected ACK message.\n");
+		rc = -EIO;
+		goto unlock;
+	}
+
+	if (data->completion && !(type & PMI_ACK)) {
+		printk(KERN_WARNING "pmi: expected ACK, but got %d\n", type);
+		rc = -EIO;
+		goto unlock;
+	}
+
+	data->msg.type = type;
+	data->msg.data0 = ioread8(data->pmi_reg + PMI_READ_DATA0);
+	data->msg.data1 = ioread8(data->pmi_reg + PMI_READ_DATA1);
+	data->msg.data2 = ioread8(data->pmi_reg + PMI_READ_DATA2);
+	rc = 0;
+unlock:
+	spin_unlock(&data->pmi_spinlock);
+
+	if (rc == -EIO) {
+		rc = IRQ_HANDLED;
+		goto out;
+	}
+
+	if (data->msg.type & PMI_ACK) {
+		complete(data->completion);
+		rc = IRQ_HANDLED;
+		goto out;
+	}
+
+	schedule_work(&data->work);
+
+	rc = IRQ_HANDLED;
+out:
+	return rc;
+}
+
+
+static struct of_device_id pmi_match[] = {
+	{ .type = "ibm,pmi", .name = "ibm,pmi" },
+	{},
+};
+
+MODULE_DEVICE_TABLE(of, pmi_match);
+
+static void pmi_notify_handlers(struct work_struct *work)
+{
+	struct pmi_data *data;
+	struct pmi_handler *handler;
+
+	data = container_of(work, struct pmi_data, work);
+
+	spin_lock(&data->handler_spinlock);
+	list_for_each_entry(handler, &data->handler, node) {
+		pr_debug(KERN_INFO "pmi: notifying handler %p\n", handler);
+		if (handler->type == data->msg.type)
+			handler->handle_pmi_message(data->dev, data->msg);
+	}
+	spin_unlock(&data->handler_spinlock);
+}
+
+static int pmi_of_probe(struct of_device *dev,
+			const struct of_device_id *match)
+{
+	struct device_node *np = dev->node;
+	struct pmi_data *data;
+	int rc;
+
+	data = kzalloc(sizeof(struct pmi_data), GFP_KERNEL);
+	if (!data) {
+		printk(KERN_ERR "pmi: could not allocate memory.\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	data->pmi_reg = of_iomap(np);
+	if (!data->pmi_reg) {
+		printk(KERN_ERR "pmi: invalid register address.\n");
+		rc = -EFAULT;
+		goto error_cleanup_data;
+	}
+
+	INIT_LIST_HEAD(&data->handler);
+
+	mutex_init(&data->msg_mutex);
+	spin_lock_init(&data->pmi_spinlock);
+	spin_lock_init(&data->handler_spinlock);
+
+	INIT_WORK(&data->work, pmi_notify_handlers);
+
+	dev->dev.driver_data = data;
+	data->dev = dev;
+
+	data->irq = irq_of_parse_and_map(np, 0);
+	if (data->irq == NO_IRQ) {
+		printk(KERN_ERR "pmi: invalid interrupt.\n");
+		rc = -EFAULT;
+		goto error_cleanup_iomap;
+	}
+
+	rc = request_irq(data->irq, pmi_irq_handler, 0, "pmi", data);
+	if (rc) {
+		printk(KERN_ERR "pmi: can't request IRQ %d: returned %d\n",
+				data->irq, rc);
+		goto error_cleanup_iomap;
+	}
+
+	printk(KERN_INFO "pmi: found pmi device at addr %p.\n", data->pmi_reg);
+
+	goto out;
+
+error_cleanup_iomap:
+	iounmap(data->pmi_reg);
+
+error_cleanup_data:
+	kfree(data);
+
+out:
+	return rc;
+}
+
+static int pmi_of_remove(struct of_device *dev)
+{
+	struct pmi_data *data;
+	struct pmi_handler *handler, *tmp;
+
+	data = dev->dev.driver_data;
+
+	free_irq(data->irq, data);
+	iounmap(data->pmi_reg);
+
+	spin_lock(&data->handler_spinlock);
+
+	list_for_each_entry_safe(handler, tmp, &data->handler, node)
+		list_del(&handler->node);
+
+	spin_unlock(&data->handler_spinlock);
+
+	kfree(dev->dev.driver_data);
+
+	return 0;
+}
+
+static struct of_platform_driver pmi_of_platform_driver = {
+	.name		= "pmi",
+	.match_table	= pmi_match,
+	.probe		= pmi_of_probe,
+	.remove		= pmi_of_remove
+};
+
+static int __init pmi_module_init(void)
+{
+	return of_register_platform_driver(&pmi_of_platform_driver);
+}
+module_init(pmi_module_init);
+
+static void __exit pmi_module_exit(void)
+{
+	of_unregister_platform_driver(&pmi_of_platform_driver);
+}
+module_exit(pmi_module_exit);
+
+void pmi_send_message(struct of_device *device, pmi_message_t msg)
+{
+	struct pmi_data *data;
+	unsigned long flags;
+	DECLARE_COMPLETION_ONSTACK(completion);
+
+	data = device->dev.driver_data;
+
+	mutex_lock(&data->msg_mutex);
+
+	data->msg = msg;
+	pr_debug("pmi_send_message: msg is %08x\n", *(u32*)&msg);
+
+	data->completion = &completion;
+
+	spin_lock_irqsave(&data->pmi_spinlock, flags);
+	iowrite8(msg.data0, data->pmi_reg + PMI_WRITE_DATA0);
+	iowrite8(msg.data1, data->pmi_reg + PMI_WRITE_DATA1);
+	iowrite8(msg.data2, data->pmi_reg + PMI_WRITE_DATA2);
+	iowrite8(msg.type, data->pmi_reg + PMI_WRITE_TYPE);
+	spin_unlock_irqrestore(&data->pmi_spinlock, flags);
+
+	pr_debug("pmi_send_message: wait for completion\n");
+
+	wait_for_completion_interruptible_timeout(data->completion,
+						  PMI_TIMEOUT);
+
+	data->completion = NULL;
+
+	mutex_unlock(&data->msg_mutex);
+}
+EXPORT_SYMBOL_GPL(pmi_send_message);
+
+void pmi_register_handler(struct of_device *device,
+			  struct pmi_handler *handler)
+{
+	struct pmi_data *data;
+	data = device->dev.driver_data;
+
+	spin_lock(&data->handler_spinlock);
+	list_add_tail(&handler->node, &data->handler);
+	spin_unlock(&data->handler_spinlock);
+}
+EXPORT_SYMBOL_GPL(pmi_register_handler);
+
+void pmi_unregister_handler(struct of_device *device,
+			    struct pmi_handler *handler)
+{
+	struct pmi_data *data;
+
+	pr_debug("pmi: unregistering handler %p\n", handler);
+
+	data = device->dev.driver_data;
+
+	spin_lock(&data->handler_spinlock);
+	list_del(&handler->node);
+	spin_unlock(&data->handler_spinlock);
+}
+EXPORT_SYMBOL_GPL(pmi_unregister_handler);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Christian Krafft <[email protected]>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
include/asm-powerpc/pmi.h new file +67 -0

67 lines not on this page · raw message

arch/powerpc/Kconfig +9 -0
Index: linux/arch/powerpc/Kconfig
===================================================================
@@ -577,6 +577,15 @@ config RTAS_FLASH
 	tristate "Firmware flash interface"
 	depends on PPC64 && RTAS_PROC
 
+config PPC_PMI
+	tristate "Support for PMI"
+	depends PPC_IBM_CELL_BLADE
+	help
+	  PMI (Platform Management Interrupt) is a way to
+	  communicate with the BMC (Baseboard Mangement Controller).
+	  It is used in some IBM Cell blades.
+	default m
+
 config MMIO_NVRAM
 	bool
 	default n
arch/powerpc/configs/cell_defconfig +1 -0
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
@@ -147,6 +147,7 @@ CONFIG_PPC_RTAS=y
 # CONFIG_RTAS_ERROR_LOGGING is not set
 CONFIG_RTAS_PROC=y
 CONFIG_RTAS_FLASH=y
+CONFIG_PPC_PMI=m
 CONFIG_MMIO_NVRAM=y
 # CONFIG_PPC_MPC106 is not set
 # CONFIG_PPC_970_NAP is not set
arch/powerpc/sysdev/Makefile +1 -0
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
@@ -7,6 +7,7 @@ obj-$(CONFIG_PPC_INDIRECT_PCI)	+= indire
 obj-$(CONFIG_PPC_MPC106)	+= grackle.o
 obj-$(CONFIG_PPC_DCR)		+= dcr.o
 obj-$(CONFIG_PPC_DCR_NATIVE)	+= dcr-low.o
+obj-$(CONFIG_PPC_PMI)		+= pmi.o
 obj-$(CONFIG_U3_DART)		+= dart_iommu.o
 obj-$(CONFIG_MMIO_NVRAM)	+= mmio_nvram.o
 obj-$(CONFIG_FSL_SOC)		+= fsl_soc.o
Acked-by
#14 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Hello Christian,

arch/powerpc/sysdev/pmi.c line 120

Quoted 1 line · Christian Krafft
arch/powerpc/sysdev/pmi.c new file +1 -0
+	{ .type = "ibm,pmi", .name = "ibm,pmi" },

In the FW the device name is still "pmi" - only the type is "ibm,pmi".

I think it is better to rename it to "ibm,pmi", since PMI is a quite common abbreviation.

Beside this:

Acked-by Heiko J Schick <[email protected]>

Regards,
Heiko

#15 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 10 lines · Heiko J Schick

Hello Christian,

arch/powerpc/sysdev/pmi.c line 120

Quoted 1 line · Christian Krafft
arch/powerpc/sysdev/pmi.c new file +1 -0
+	{ .type = "ibm,pmi", .name = "ibm,pmi" },

In the FW the device name is still "pmi" - only the type is "ibm,pmi".

I think it is better to rename it to "ibm,pmi", since PMI is a quite common abbreviation.

Nah, the name doesn't matter that much as long as it's unique at a given level, just keep it to "pmi"

Ben.

#16 lore raw

Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade

Quoted 7 lines
Quoted 4 lines

In the FW the device name is still "pmi" - only the type is "ibm,pmi".

I think it is better to rename it to "ibm,pmi", since PMI is a quite common abbreviation.

Nah, the name doesn't matter that much as long as it's unique at a given level, just keep it to "pmi"

It doesn't have to be unique either, not even unique among siblings.

Segher