* [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.