* [patch 0/1] pmi: initial version
@ 2007-02-09 17:35 Christian Krafft
2007-02-09 17:45 ` [patch 1/1] " Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-09 17:35 UTC (permalink / raw)
To: linuxppc-dev, krafft, cbe-oss-dev
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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [patch 1/1] pmi: initial version
2007-02-09 17:35 [patch 0/1] pmi: initial version Christian Krafft
@ 2007-02-09 17:45 ` Christian Krafft
2007-02-09 22:29 ` Nathan Lynch
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-09 17:45 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, cbe-oss-dev
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 <krafft@de.ibm.com>
Index: linux/arch/powerpc/Kconfig
===================================================================
--- linux.orig/arch/powerpc/Kconfig
+++ 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
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
--- linux.orig/arch/powerpc/sysdev/Makefile
+++ 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
Index: linux/arch/powerpc/sysdev/pmi.c
===================================================================
--- /dev/null
+++ linux/arch/powerpc/sysdev/pmi.c
@@ -0,0 +1,275 @@
+/*
+ * pmi driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * PMI is a way to communicate with the BMC via interrupts.
+ * Unlike IPMI it is bidirectional and has a low latency.
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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 <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;
+ struct pmi_message *msg;
+ struct completion *completion;
+ struct of_device *dev;
+ int irq;
+ u8 __iomem *pmi_reg;
+};
+
+
+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;
+ int type;
+ struct pmi_handler *handler;
+
+ data = dev_id;
+
+ BUG_ON(!data);
+
+ 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);
+
+ 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);
+ 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);
+ 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;
+ }
+
+ 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;
+ }
+
+ 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);
+ return -EFAULT;
+ }
+
+ dev->dev.driver_data = data;
+ data->dev = dev;
+
+ printk(KERN_INFO "pmi: found pmi device at address %p.\n", data->pmi_reg);
+
+ return 0;
+}
+
+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_irq(&data->handler_spinlock);
+
+ list_for_each_entry_safe(handler, tmp, &data->handler, node)
+ list_del(&handler->node);
+
+ spin_unlock_irq(&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, struct pmi_message *msg)
+{
+ struct pmi_data *data;
+ unsigned long flags;
+ DECLARE_COMPLETION_ONSTACK(completion);
+
+ BUG_ON(!device || !msg);
+
+ 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);
+
+void pmi_register_handler(struct of_device *device, struct pmi_handler *handler)
+{
+ struct pmi_data *data;
+
+ pr_debug("pmi: registering handler %p\n", handler);
+
+ data = device->dev.driver_data;
+
+ spin_lock_irq(&data->handler_spinlock);
+ list_add_tail(&handler->node, &data->handler);
+ spin_unlock_irq(&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_irq(&data->handler_spinlock);
+ list_del(&handler->node);
+ spin_unlock_irq(&data->handler_spinlock);
+}
+EXPORT_SYMBOL_GPL(pmi_unregister_handler);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Christian Krafft <krafft@de.ibm.com>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
Index: linux/include/asm-powerpc/pmi.h
===================================================================
--- /dev/null
+++ linux/include/asm-powerpc/pmi.h
@@ -0,0 +1,63 @@
+#ifndef _POWERPC_PMI_H
+#define _POWERPC_PMI_H
+
+/*
+ * Definitions for talking with PMI device on PowerPC
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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.
+ */
+
+#ifdef __KERNEL__
+
+#include <asm/of_device.h>
+
+#define PMI_TYPE_FREQ_CHANGE 0x01
+#define PMI_READ_TYPE 0
+#define PMI_READ_DATA0 1
+#define PMI_READ_DATA1 2
+#define PMI_READ_DATA2 3
+#define PMI_WRITE_TYPE 4
+#define PMI_WRITE_DATA0 5
+#define PMI_WRITE_DATA1 6
+#define PMI_WRITE_DATA2 7
+
+#define PMI_ACK 0x80
+
+#define PMI_TIMEOUT 100
+
+struct pmi_message {
+ u8 type;
+ u8 data0;
+ u8 data1;
+ u8 data2;
+};
+
+struct pmi_handler {
+ struct list_head node;
+ u8 type;
+ void (*handle_pmi_message) (struct of_device *, struct pmi_message *);
+};
+
+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 *);
+
+#endif /* __KERNEL__ */
+#endif /* _POWERPC_PMI_H */
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [patch 1/1] pmi: initial version
2007-02-09 17:45 ` [patch 1/1] " Christian Krafft
@ 2007-02-09 22:29 ` Nathan Lynch
2007-02-12 9:12 ` Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Nathan Lynch @ 2007-02-09 22:29 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, cbe-oss-dev
Hi Christian-
Some minor comments.
Christian Krafft wrote:
> 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 <krafft@de.ibm.com>
>
> Index: linux/arch/powerpc/Kconfig
> ===================================================================
> --- linux.orig/arch/powerpc/Kconfig
> +++ 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
Did you mean for this to be commented out?
<snip>
> +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.
> +
> + 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.
> +
> + 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);
> +
> + 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?
> + 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);
> + 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.
> + }
> +
> + 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;
> + }
> +
> + 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.
> + return -EFAULT;
Just return rc; request_irq returns standard error codes.
<snip>
> +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.
> +
> + 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);
...
> +EXPORT_SYMBOL_GPL(pmi_register_handler);
...
> +EXPORT_SYMBOL_GPL(pmi_unregister_handler);
Are the symbol exports necessary?
<snip>
> +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?
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [patch 1/1] pmi: initial version
2007-02-09 22:29 ` Nathan Lynch
@ 2007-02-12 9:12 ` Christian Krafft
2007-02-13 13:34 ` [Cbe-oss-dev] [patch v2] " Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-12 9:12 UTC (permalink / raw)
To: Nathan Lynch; +Cc: linuxppc-dev, cbe-oss-dev
Hi Nathan,
thanks for your comments.
Will send an updated version.
On Fri, 9 Feb 2007 16:29:56 -0600
Nathan Lynch <ntl@pobox.com> wrote:
> Hi Christian-
>=20
> Some minor comments.
>=20
> Christian Krafft wrote:
> > This patch adds driver code for the PMI device found in future IBM prod=
ucts.
> > PMI stands for "Platform Management Interrupt" and is a way to communic=
ate
> > with the BMC. It provides bidirectional communication with a low latenc=
y.
> >=20
> > Signed-off-by: Christian Krafft <krafft@de.ibm.com>
> >=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
> > --- linux.orig/arch/powerpc/Kconfig
> > +++ 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
>=20
> Did you mean for this to be commented out?
no, good point ;-)
>=20
> <snip>
>=20
> > +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,
>=20
>=20
> > +
> > + 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,
>=20
>=20
> > +
> > + 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);
> > + 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.
>=20
>=20
> > + 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;
> > + 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.
>=20
>=20
> > + }
> > +
> > + 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;
> > + }
> > +
> > + 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
> > + return -EFAULT;
>=20
> Just return rc; request_irq returns standard error codes.
ok,
>=20
> <snip>
>=20
> > +void pmi_send_message(struct of_device *device, struct pmi_message *ms=
g)
> > +{
> > + 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,
>=20
> > +
> > + 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;
> > +
> > + 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_TIMEO=
UT);
> > +
> > + data->msg =3D NULL;
> > + data->completion =3D NULL;
> > +
> > + mutex_unlock(&data->msg_mutex);
> > +}
> > +EXPORT_SYMBOL_GPL(pmi_send_message);
> ...
> > +EXPORT_SYMBOL_GPL(pmi_register_handler);
> ...
> > +EXPORT_SYMBOL_GPL(pmi_unregister_handler);
>=20
> Are the symbol exports necessary?
>=20
> <snip>
>=20
> > +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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v2] pmi: initial version
2007-02-12 9:12 ` Christian Krafft
@ 2007-02-13 13:34 ` Christian Krafft
2007-02-13 18:44 ` [Cbe-oss-dev] [patch v3] " Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-13 13:34 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
Subject: powerpc: add PMI driver for cell blade
From: Christian Krafft <krafft@de.ibm.com>
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 <krafft@de.ibm.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
---
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
--- linux.orig/arch/powerpc/Kconfig
+++ linux/arch/powerpc/Kconfig
@@ -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
--- linux.orig/arch/powerpc/sysdev/Makefile
+++ linux/arch/powerpc/sysdev/Makefile
@@ -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
--- /dev/null
+++ linux/arch/powerpc/sysdev/pmi.c
@@ -0,0 +1,291 @@
+/*
+ * pmi driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * PMI is a way to communicate with the BMC via interrupts.
+ * Unlike IPMI it is bidirectional and has a low latency.
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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 <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;
+};
+
+
+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;
+ struct pmi_handler *handler;
+ u8 type;
+ int rc;
+
+ data =3D dev_id;
+
+ spin_lock(&data->pmi_spinlock);
+
+ type =3D 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 =3D -EIO;
+ goto unlock;
+ }
+
+ if (data->completion && !(type & PMI_ACK)) {
+ printk(KERN_WARNING "pmi: expected ACK, but got %d\n", type);
+ rc =3D -EIO;
+ goto unlock;
+ }
+
+ 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);
+ rc =3D 0;
+unlock:
+ spin_unlock(&data->pmi_spinlock);
+
+ if (rc =3D=3D -EIO) {
+ rc =3D IRQ_HANDLED;
+ goto out;
+ }
+
+ if (data->msg.type & PMI_ACK) {
+ complete(data->completion);
+ rc =3D IRQ_HANDLED;
+ goto out;
+ }
+
+ list_for_each_entry(handler, &data->handler, node) {
+ pr_debug(KERN_INFO "pmi: notifying handler %p\n", handler);
+ if (handler->type =3D=3D data->msg.type)
+ handler->handle_pmi_message(data->dev, data->msg);
+ }
+
+ rc =3D IRQ_HANDLED;
+out:
+ return rc;
+}
+
+
+static struct of_device_id pmi_match[] =3D {
+ { .type =3D "ibm,pmi", .name =3D "ibm,pmi" },
+ {},
+};
+
+MODULE_DEVICE_TABLE(of, pmi_match);
+
+static int pmi_of_probe(struct of_device *dev,
+ const struct of_device_id *match)
+{
+ struct device_node *np =3D dev->node;
+ struct pmi_data *data;
+ int rc;
+
+ data =3D kzalloc(sizeof(struct pmi_data), GFP_KERNEL);
+ if (!data) {
+ printk(KERN_ERR "pmi: could not allocate memory.\n");
+ rc =3D -ENOMEM;
+ goto out;
+ }
+
+ data->pmi_reg =3D of_iomap(np);
+ if (!data->pmi_reg) {
+ printk(KERN_ERR "pmi: invalid register address.\n");
+ rc =3D -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);
+
+ dev->dev.driver_data =3D data;
+ data->dev =3D dev;
+
+ data->irq =3D irq_of_parse_and_map(np, 0);
+ if (data->irq =3D=3D NO_IRQ) {
+ printk(KERN_ERR "pmi: invalid interrupt.\n");
+ rc =3D -EFAULT;
+ goto error_cleanup_iomap;
+ }
+
+ rc =3D 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 =3D dev->dev.driver_data;
+
+ free_irq(data->irq, data);
+ iounmap(data->pmi_reg);
+
+ spin_lock_irq(&data->handler_spinlock);
+
+ list_for_each_entry_safe(handler, tmp, &data->handler, node)
+ list_del(&handler->node);
+
+ spin_unlock_irq(&data->handler_spinlock);
+
+ kfree(dev->dev.driver_data);
+
+ return 0;
+}
+
+static struct of_platform_driver pmi_of_platform_driver =3D {
+ .name =3D "pmi",
+ .match_table =3D pmi_match,
+ .probe =3D pmi_of_probe,
+ .remove =3D 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 =3D device->dev.driver_data;
+
+ mutex_lock(&data->msg_mutex);
+
+ data->msg =3D msg;
+ pr_debug("pmi_send_message: msg is %08x\n", *(u32*)&msg);
+
+ data->completion =3D &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 =3D 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;
+
+ pr_debug("pmi: registering handler %p\n", handler);
+
+ data =3D device->dev.driver_data;
+
+ spin_lock_irq(&data->handler_spinlock);
+ list_add_tail(&handler->node, &data->handler);
+ spin_unlock_irq(&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 =3D device->dev.driver_data;
+
+ spin_lock_irq(&data->handler_spinlock);
+ list_del(&handler->node);
+ spin_unlock_irq(&data->handler_spinlock);
+}
+EXPORT_SYMBOL_GPL(pmi_unregister_handler);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Christian Krafft <krafft@de.ibm.com>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
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
--- /dev/null
+++ linux/include/asm-powerpc/pmi.h
@@ -0,0 +1,63 @@
+#ifndef _POWERPC_PMI_H
+#define _POWERPC_PMI_H
+
+/*
+ * Definitions for talking with PMI device on PowerPC
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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.
+ */
+
+#ifdef __KERNEL__
+
+#include <asm/of_device.h>
+
+#define PMI_TYPE_FREQ_CHANGE 0x01
+#define PMI_READ_TYPE 0
+#define PMI_READ_DATA0 1
+#define PMI_READ_DATA1 2
+#define PMI_READ_DATA2 3
+#define PMI_WRITE_TYPE 4
+#define PMI_WRITE_DATA0 5
+#define PMI_WRITE_DATA1 6
+#define PMI_WRITE_DATA2 7
+
+#define PMI_ACK 0x80
+
+#define PMI_TIMEOUT 100
+
+typedef struct {
+ u8 type;
+ u8 data0;
+ u8 data1;
+ u8 data2;
+} pmi_message_t;
+
+struct pmi_handler {
+ struct list_head node;
+ u8 type;
+ void (*handle_pmi_message) (struct of_device *, pmi_message_t);
+};
+
+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 *, pmi_message_t);
+
+#endif /* __KERNEL__ */
+#endif /* _POWERPC_PMI_H */
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
--- linux.orig/arch/powerpc/configs/cell_defconfig
+++ linux/arch/powerpc/configs/cell_defconfig
@@ -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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] pmi: initial version
2007-02-13 13:34 ` [Cbe-oss-dev] [patch v2] " Christian Krafft
@ 2007-02-13 18:44 ` Christian Krafft
2007-02-13 18:48 ` [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-13 18:44 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-13 18:44 ` [Cbe-oss-dev] [patch v3] " Christian Krafft
@ 2007-02-13 18:48 ` Christian Krafft
2007-02-13 19:20 ` Christian Krafft
2007-02-13 19:28 ` Christian Krafft
0 siblings, 2 replies; 17+ messages in thread
From: Christian Krafft @ 2007-02-13 18:48 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
Subject: powerpc: add PMI driver for cell blade
From: Christian Krafft <krafft@de.ibm.com>
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 <krafft@de.ibm.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
---
Index: linux/arch/powerpc/Kconfig
===================================================================
--- linux.orig/arch/powerpc/Kconfig
+++ 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
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
--- linux.orig/arch/powerpc/sysdev/Makefile
+++ 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
Index: linux/arch/powerpc/sysdev/pmi.c
===================================================================
--- /dev/null
+++ linux/arch/powerpc/sysdev/pmi.c
@@ -0,0 +1,312 @@
+/*
+ * pmi driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * PMI is a way to communicate with the BMC via interrupts.
+ * Unlike IPMI it is bidirectional and has a low latency.
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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 workqueue_struct *work_queue;
+ 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;
+ }
+
+ queue_work(data->work_queue, &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);
+
+ data->work_queue = create_workqueue("pmi");
+ 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;
+
+ flush_workqueue(data->work_queue);
+ destroy_workqueue(data->work_queue);
+
+ 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;
+
+ pr_debug("pmi: registering handler %p\n", handler);
+
+ 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 <krafft@de.ibm.com>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
Index: linux/include/asm-powerpc/pmi.h
===================================================================
--- /dev/null
+++ linux/include/asm-powerpc/pmi.h
@@ -0,0 +1,63 @@
+#ifndef _POWERPC_PMI_H
+#define _POWERPC_PMI_H
+
+/*
+ * Definitions for talking with PMI device on PowerPC
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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.
+ */
+
+#ifdef __KERNEL__
+
+#include <asm/of_device.h>
+
+#define PMI_TYPE_FREQ_CHANGE 0x01
+#define PMI_READ_TYPE 0
+#define PMI_READ_DATA0 1
+#define PMI_READ_DATA1 2
+#define PMI_READ_DATA2 3
+#define PMI_WRITE_TYPE 4
+#define PMI_WRITE_DATA0 5
+#define PMI_WRITE_DATA1 6
+#define PMI_WRITE_DATA2 7
+
+#define PMI_ACK 0x80
+
+#define PMI_TIMEOUT 100
+
+typedef struct {
+ u8 type;
+ u8 data0;
+ u8 data1;
+ u8 data2;
+} pmi_message_t;
+
+struct pmi_handler {
+ struct list_head node;
+ u8 type;
+ void (*handle_pmi_message) (struct of_device *, pmi_message_t);
+};
+
+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 *, pmi_message_t);
+
+#endif /* __KERNEL__ */
+#endif /* _POWERPC_PMI_H */
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
--- linux.orig/arch/powerpc/configs/cell_defconfig
+++ 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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-13 18:48 ` [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade Christian Krafft
@ 2007-02-13 19:20 ` Christian Krafft
2007-02-13 19:28 ` Christian Krafft
1 sibling, 0 replies; 17+ messages in thread
From: Christian Krafft @ 2007-02-13 19:20 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
On Tue, 13 Feb 2007 19:48:06 +0100
Christian Krafft <krafft@de.ibm.com> wrote:
> +
> + 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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-13 18:48 ` [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade Christian Krafft
2007-02-13 19:20 ` Christian Krafft
@ 2007-02-13 19:28 ` Christian Krafft
2007-02-14 0:30 ` Paul Mackerras
1 sibling, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-13 19:28 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
Subject: powerpc: add PMI driver for cell blade
From: Christian Krafft <krafft@de.ibm.com>
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 <krafft@de.ibm.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
---
Index: linux/arch/powerpc/Kconfig
===================================================================
--- linux.orig/arch/powerpc/Kconfig
+++ 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
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
--- linux.orig/arch/powerpc/sysdev/Makefile
+++ 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
Index: linux/arch/powerpc/sysdev/pmi.c
===================================================================
--- /dev/null
+++ linux/arch/powerpc/sysdev/pmi.c
@@ -0,0 +1,307 @@
+/*
+ * pmi driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * PMI is a way to communicate with the BMC via interrupts.
+ * Unlike IPMI it is bidirectional and has a low latency.
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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;
+
+ pr_debug("pmi: registering handler %p\n", handler);
+
+ 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 <krafft@de.ibm.com>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
Index: linux/include/asm-powerpc/pmi.h
===================================================================
--- /dev/null
+++ linux/include/asm-powerpc/pmi.h
@@ -0,0 +1,63 @@
+#ifndef _POWERPC_PMI_H
+#define _POWERPC_PMI_H
+
+/*
+ * Definitions for talking with PMI device on PowerPC
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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.
+ */
+
+#ifdef __KERNEL__
+
+#include <asm/of_device.h>
+
+#define PMI_TYPE_FREQ_CHANGE 0x01
+#define PMI_READ_TYPE 0
+#define PMI_READ_DATA0 1
+#define PMI_READ_DATA1 2
+#define PMI_READ_DATA2 3
+#define PMI_WRITE_TYPE 4
+#define PMI_WRITE_DATA0 5
+#define PMI_WRITE_DATA1 6
+#define PMI_WRITE_DATA2 7
+
+#define PMI_ACK 0x80
+
+#define PMI_TIMEOUT 100
+
+typedef struct {
+ u8 type;
+ u8 data0;
+ u8 data1;
+ u8 data2;
+} pmi_message_t;
+
+struct pmi_handler {
+ struct list_head node;
+ u8 type;
+ void (*handle_pmi_message) (struct of_device *, pmi_message_t);
+};
+
+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 *, pmi_message_t);
+
+#endif /* __KERNEL__ */
+#endif /* _POWERPC_PMI_H */
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
--- linux.orig/arch/powerpc/configs/cell_defconfig
+++ 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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-13 19:28 ` Christian Krafft
@ 2007-02-14 0:30 ` Paul Mackerras
2007-02-14 8:39 ` Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Paul Mackerras @ 2007-02-14 0:30 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
Christian Krafft writes:
> From: Christian Krafft <krafft@de.ibm.com>
> 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.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-14 0:30 ` Paul Mackerras
@ 2007-02-14 8:39 ` Christian Krafft
2007-02-14 10:22 ` Paul Mackerras
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-14 8:39 UTC (permalink / raw)
To: Paul Mackerras; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
On Wed, 14 Feb 2007 11:30:51 +1100
Paul Mackerras <paulus@samba.org> wrote:
> Christian Krafft writes:
>=20
> > From: Christian Krafft <krafft@de.ibm.com>
> > This patch adds driver code for the PMI device found in future IBM prod=
ucts.
> > PMI stands for "Platform Management Interrupt" and is a way to communic=
ate
> > with the BMC. It provides bidirectional communication with a low latenc=
y.
>=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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-14 8:39 ` Christian Krafft
@ 2007-02-14 10:22 ` Paul Mackerras
2007-02-14 13:07 ` Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Paul Mackerras @ 2007-02-14 10:22 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
Christian Krafft writes:
> 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.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-14 10:22 ` Paul Mackerras
@ 2007-02-14 13:07 ` Christian Krafft
2007-02-14 13:09 ` Christian Krafft
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-14 13:07 UTC (permalink / raw)
To: Paul Mackerras; +Cc: linuxppc-dev, Nathan Lynch, cbe-oss-dev
On Wed, 14 Feb 2007 21:22:16 +1100
Paul Mackerras <paulus@samba.org> wrote:
> Christian Krafft writes:
>=20
> > BMC stands for Baseboard Management Controller:
> > http://en.wikipedia.org/wiki/Baseboard_management_controller
>=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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-14 13:07 ` Christian Krafft
@ 2007-02-14 13:09 ` Christian Krafft
2007-02-15 7:06 ` Heiko J Schick
0 siblings, 1 reply; 17+ messages in thread
From: Christian Krafft @ 2007-02-14 13:09 UTC (permalink / raw)
To: Christian Krafft; +Cc: linuxppc-dev, Nathan Lynch, Paul Mackerras, cbe-oss-dev
Subject: powerpc: add PMI driver for cell blade
From: Christian Krafft <krafft@de.ibm.com>
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 <krafft@de.ibm.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
---
Index: linux/arch/powerpc/sysdev/pmi.c
===================================================================
--- /dev/null
+++ 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 <krafft@de.ibm.com>
+ *
+ * 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 <krafft@de.ibm.com>");
+MODULE_DESCRIPTION("IBM Platform Management Interrupt driver");
Index: linux/include/asm-powerpc/pmi.h
===================================================================
--- /dev/null
+++ linux/include/asm-powerpc/pmi.h
@@ -0,0 +1,67 @@
+#ifndef _POWERPC_PMI_H
+#define _POWERPC_PMI_H
+
+/*
+ * Definitions for talking with PMI device on PowerPC
+ *
+ * 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.
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * 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.
+ */
+
+#ifdef __KERNEL__
+
+#include <asm/of_device.h>
+
+#define PMI_TYPE_FREQ_CHANGE 0x01
+#define PMI_READ_TYPE 0
+#define PMI_READ_DATA0 1
+#define PMI_READ_DATA1 2
+#define PMI_READ_DATA2 3
+#define PMI_WRITE_TYPE 4
+#define PMI_WRITE_DATA0 5
+#define PMI_WRITE_DATA1 6
+#define PMI_WRITE_DATA2 7
+
+#define PMI_ACK 0x80
+
+#define PMI_TIMEOUT 100
+
+typedef struct {
+ u8 type;
+ u8 data0;
+ u8 data1;
+ u8 data2;
+} pmi_message_t;
+
+struct pmi_handler {
+ struct list_head node;
+ u8 type;
+ void (*handle_pmi_message) (struct of_device *, pmi_message_t);
+};
+
+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 *, pmi_message_t);
+
+#endif /* __KERNEL__ */
+#endif /* _POWERPC_PMI_H */
Index: linux/arch/powerpc/Kconfig
===================================================================
--- linux.orig/arch/powerpc/Kconfig
+++ 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
Index: linux/arch/powerpc/configs/cell_defconfig
===================================================================
--- linux.orig/arch/powerpc/configs/cell_defconfig
+++ 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
Index: linux/arch/powerpc/sysdev/Makefile
===================================================================
--- linux.orig/arch/powerpc/sysdev/Makefile
+++ 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
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-14 13:09 ` Christian Krafft
@ 2007-02-15 7:06 ` Heiko J Schick
2007-02-15 10:50 ` Benjamin Herrenschmidt
0 siblings, 1 reply; 17+ messages in thread
From: Heiko J Schick @ 2007-02-15 7:06 UTC (permalink / raw)
To: Christian Krafft
Cc: Christian Krafft, Nathan Lynch, cbe-oss-dev, Paul Mackerras,
linuxppc-dev
Hello Christian,
On 14.02.2007, at 14:09, Christian Krafft wrote:
> + { .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 <schickhj@de.ibm.com>
Regards,
Heiko
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-15 7:06 ` Heiko J Schick
@ 2007-02-15 10:50 ` Benjamin Herrenschmidt
2007-02-15 19:19 ` Segher Boessenkool
0 siblings, 1 reply; 17+ messages in thread
From: Benjamin Herrenschmidt @ 2007-02-15 10:50 UTC (permalink / raw)
To: Heiko J Schick
Cc: Christian Krafft, linuxppc-dev, Nathan Lynch, Christian Krafft,
cbe-oss-dev, Paul Mackerras
On Thu, 2007-02-15 at 08:06 +0100, Heiko J Schick wrote:
> Hello Christian,
>
> On 14.02.2007, at 14:09, Christian Krafft wrote:
>
> > + { .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.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade
2007-02-15 10:50 ` Benjamin Herrenschmidt
@ 2007-02-15 19:19 ` Segher Boessenkool
0 siblings, 0 replies; 17+ messages in thread
From: Segher Boessenkool @ 2007-02-15 19:19 UTC (permalink / raw)
To: Benjamin Herrenschmidt
Cc: Christian Krafft, linuxppc-dev, Nathan Lynch, cbe-oss-dev
>> 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
^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2007-02-15 19:19 UTC | newest]
Thread overview: 17+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-02-09 17:35 [patch 0/1] pmi: initial version Christian Krafft
2007-02-09 17:45 ` [patch 1/1] " Christian Krafft
2007-02-09 22:29 ` Nathan Lynch
2007-02-12 9:12 ` Christian Krafft
2007-02-13 13:34 ` [Cbe-oss-dev] [patch v2] " Christian Krafft
2007-02-13 18:44 ` [Cbe-oss-dev] [patch v3] " Christian Krafft
2007-02-13 18:48 ` [Cbe-oss-dev] [patch v3] powerpc: add PMI driver for cell blade Christian Krafft
2007-02-13 19:20 ` Christian Krafft
2007-02-13 19:28 ` Christian Krafft
2007-02-14 0:30 ` Paul Mackerras
2007-02-14 8:39 ` Christian Krafft
2007-02-14 10:22 ` Paul Mackerras
2007-02-14 13:07 ` Christian Krafft
2007-02-14 13:09 ` Christian Krafft
2007-02-15 7:06 ` Heiko J Schick
2007-02-15 10:50 ` Benjamin Herrenschmidt
2007-02-15 19:19 ` Segher Boessenkool
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.