From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from e32.co.us.ibm.com (e32.co.us.ibm.com [32.97.110.150]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client CN "e32.co.us.ibm.com", Issuer "Equifax" (verified OK)) by ozlabs.org (Postfix) with ESMTPS id 7C909DDF24 for ; Tue, 8 Jul 2008 12:58:19 +1000 (EST) Received: from d03relay02.boulder.ibm.com (d03relay02.boulder.ibm.com [9.17.195.227]) by e32.co.us.ibm.com (8.13.8/8.13.8) with ESMTP id m682r0sn014672 for ; Mon, 7 Jul 2008 22:53:00 -0400 Received: from d03av01.boulder.ibm.com (d03av01.boulder.ibm.com [9.17.195.167]) by d03relay02.boulder.ibm.com (8.13.8/8.13.8/NCO v9.0) with ESMTP id m682wDRh162892 for ; Mon, 7 Jul 2008 20:58:13 -0600 Received: from d03av01.boulder.ibm.com (loopback [127.0.0.1]) by d03av01.boulder.ibm.com (8.12.11.20060308/8.13.3) with ESMTP id m682wDo5005688 for ; Mon, 7 Jul 2008 20:58:13 -0600 Date: Mon, 7 Jul 2008 21:58:12 -0500 From: Sonny Rao To: Benjamin Herrenschmidt Subject: [PATCHv3] Power5,Power6 BSR driver Message-ID: <20080708025812.GC15821@us.ibm.com> References: <20080616185344.GB16192@gamma> <20080617223952.GA9594@localdomain> <20080617224443.GD7552@localhost.localdomain> <20080618065123.GB13318@localhost.localdomain> <1215406775.8970.57.camel@pasglop> <20080707211740.GA15821@us.ibm.com> <1215469617.8970.146.camel@pasglop> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1215469617.8970.146.camel@pasglop> Cc: sonnyrao@linux.vnet.ibm.com, paulus@samba.org, jschopp@us.ibm.com, Nathan Lynch , linuxppc-dev@ozlabs.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Tue, Jul 08, 2008 at 08:26:57AM +1000, Benjamin Herrenschmidt wrote: > On Mon, 2008-07-07 at 16:17 -0500, Sonny Rao wrote: > > On Mon, Jul 07, 2008 at 02:59:35PM +1000, Benjamin Herrenschmidt wrote: > > > > > > > + cur->bsr_addr = reg[i * 2]; > > > > + cur->bsr_len = reg[i * 2 + 1]; > > > > > > That's fishy... hand-reading of "reg" property without taking > > > into account the parent's #size-cells/#address-cells... can't you > > > use of_address_to_resource or something similar and carry a struct > > > resource around instead ? > > > > So, with this suggestion I looked at the resource API... not very well > > documented, and I get the feeling like it's more for carving up a PCI > > memory address range. In the case of the BSR, everything is already > > partitioned (by hardware) so I don't see the point of using this API > > here. Or am I missing something about it? > > It's about properly parsing a "reg" property in OF. The format of the > reg property depends on the parent #address-cells and #size-cells, and > the addresses in there may need remapping (or not) depending on parent > "ranges" properties. The special cases in prom_parse.c for PCI and ISA > are purely to deal with the additional address space flags those add to > addresses, but you shouldn't hit that with BSR. > > > > In fact, same goes with the way you do num_bsr_devs = reg_len / 16. > > > > > > You should rather use -another- property of well known lenght, or > > > get the #address/#size-cells of the parent and use those appropriately. > > > > Well, I check to make sure the lengths are consistent with each other > > right above there so we shouldn't walk off the end of anything, but I > > will take a look at using #size-cells / #address-cells instead. > > The code in prom_parse.c will do it for you :-) From: Sonny Rao Adds a character driver for BSR support on IBM POWER systems including Power5 and Power6. The BSR is an optional processor facility not currently implemented by any other processors. It's primary purpose is fast large SMP synchronization. More details on the BSR are in comments to the code which follows. Patch adds BSR to pseries_defconfig Signed-off-by: Sonny Rao Signed-off-by: Joel Schopp Index: linux-dev/drivers/char/bsr.c =================================================================== --- /dev/null 1970-01-01 00:00:00.000000000 +0000 +++ linux-dev/drivers/char/bsr.c 2008-07-07 21:49:06.000000000 -0500 @@ -0,0 +1,313 @@ +/* IBM POWER Barrier Synchronization Register Driver + * + * Copyright IBM Corporation 2008 + * + * Author: Sonny Rao + * + * 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 of the License, 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* + This driver exposes a special register which can be used for fast + synchronization across a large SMP machine. The hardware is exposed + as an array of bytes where each process will write to one of the bytes to + indicate it has finished the current stage and this update is broadcast to + all processors without having to bounce a cacheline between them. In + POWER5 and POWER6 there is one of these registers per SMP, but it is + presented in two forms; first, it is given as a whole and then as a number + of smaller registers which alias to parts of the single whole register. + This can potentially allow multiple groups of processes to each have their + own private synchronization device. + + Note that this hardware *must* be written to using *only* single byte writes. + It may be read using 1, 2, 4, or 8 byte loads which must be aligned since + this region is treated as cache-inhibited processes should also use a + full sync before and after writing to the BSR to ensure all stores and + the BSR update have made it to all chips in the system +*/ + +/* This is arbitrary number, up to Power6 it's been 17 or fewer */ +#define BSR_MAX_DEVS (32) + +struct bsr_dev { + u64 bsr_addr; /* Real address */ + u64 bsr_len; /* length of mem region we can map */ + unsigned bsr_bytes; /* size of the BSR reg itself */ + unsigned bsr_stride; /* interval at which BSR repeats in the page */ + unsigned bsr_type; /* maps to enum below */ + unsigned bsr_num; /* bsr id number for its type */ + int bsr_minor; + + dev_t bsr_dev; + struct cdev bsr_cdev; + struct device *bsr_device; + char bsr_name[32]; + +}; + +static unsigned num_bsr_devs; +static struct bsr_dev *bsr_devs; +static struct class *bsr_class; +static int bsr_major; + +enum { + BSR_8 = 0, + BSR_16 = 1, + BSR_64 = 2, + BSR_128 = 3, + BSR_UNKNOWN = 4, + BSR_MAX = 5, +}; + +static unsigned bsr_types[BSR_MAX]; + +static ssize_t +bsr_size_show(struct device *dev, struct device_attribute *attr, char *buf) +{ + struct bsr_dev *bsr_dev = dev_get_drvdata(dev); + return sprintf(buf, "%u\n", bsr_dev->bsr_bytes); +} + +static ssize_t +bsr_stride_show(struct device *dev, struct device_attribute *attr, char *buf) +{ + struct bsr_dev *bsr_dev = dev_get_drvdata(dev); + return sprintf(buf, "%u\n", bsr_dev->bsr_stride); +} + +static ssize_t +bsr_len_show(struct device *dev, struct device_attribute *attr, char *buf) +{ + struct bsr_dev *bsr_dev = dev_get_drvdata(dev); + return sprintf(buf, "%lu\n", bsr_dev->bsr_len); +} + +static struct device_attribute bsr_dev_attrs[] = { + __ATTR(bsr_size, S_IRUGO, bsr_size_show, NULL), + __ATTR(bsr_stride, S_IRUGO, bsr_stride_show, NULL), + __ATTR(bsr_length, S_IRUGO, bsr_len_show, NULL), + __ATTR_NULL +}; + +static int bsr_mmap(struct file *filp, struct vm_area_struct *vma) +{ + unsigned long size = vma->vm_end - vma->vm_start; + struct bsr_dev *dev = filp->private_data; + + if (size > dev->bsr_len || (size & (PAGE_SIZE-1))) + return -EINVAL; + + vma->vm_flags |= (VM_IO | VM_DONTEXPAND); + vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); + + if (io_remap_pfn_range(vma, vma->vm_start, dev->bsr_addr >> PAGE_SHIFT, + size, vma->vm_page_prot)) + return -EAGAIN; + + return 0; +} + +static int bsr_open(struct inode * inode, struct file * filp) +{ + struct cdev *cdev = inode->i_cdev; + struct bsr_dev *dev = container_of(cdev, struct bsr_dev, bsr_cdev); + + filp->private_data = dev; + return 0; +} + +const static struct file_operations bsr_fops = { + .owner = THIS_MODULE, + .mmap = bsr_mmap, + .open = bsr_open, +}; + +static void bsr_cleanup_devs(void) +{ + int i; + for (i=0 ; i < num_bsr_devs; i++) { + struct bsr_dev *cur = bsr_devs + i; + if (cur->bsr_device) { + cdev_del(&cur->bsr_cdev); + device_del(cur->bsr_device); + } + } + + kfree(bsr_devs); +} + +static int bsr_create_devs(struct device_node *bn) +{ + int bsr_stride_len, bsr_bytes_len; + const u32 *bsr_stride; + const u32 *bsr_bytes; + unsigned i; + + bsr_stride = of_get_property(bn, "ibm,lock-stride", &bsr_stride_len); + bsr_bytes = of_get_property(bn, "ibm,#lock-bytes", &bsr_bytes_len); + + if (!bsr_stride || !bsr_bytes || + (bsr_stride_len != bsr_bytes_len)) { + printk(KERN_ERR "bsr of-node has missing/incorrect property\n"); + return -ENODEV; + } + + num_bsr_devs = bsr_bytes_len / sizeof(u32); + + /* only a warning, its informational since we'll fail and exit */ + WARN_ON(num_bsr_devs > BSR_MAX_DEVS); + + bsr_devs = kzalloc(sizeof(struct bsr_dev) * num_bsr_devs, GFP_KERNEL); + if (!bsr_devs) + return -ENOMEM; + + for (i = 0 ; i < num_bsr_devs; i++) { + struct bsr_dev *cur = bsr_devs + i; + struct resource res; + int result; + + result = of_address_to_resource(bn, i, &res); + if (result < 0) { + printk(KERN_ERR "bsr of-node has invalid reg property\n"); + goto out_err; + } + + cur->bsr_minor = i; + cur->bsr_addr = res.start; + cur->bsr_len = res.end - res.start + 1; + cur->bsr_bytes = bsr_bytes[i]; + cur->bsr_stride = bsr_stride[i]; + cur->bsr_dev = MKDEV(bsr_major, i); + + switch(cur->bsr_bytes) { + case 8: + cur->bsr_type = BSR_8; + break; + case 16: + cur->bsr_type = BSR_16; + break; + case 64: + cur->bsr_type = BSR_64; + break; + case 128: + cur->bsr_type = BSR_128; + break; + default: + cur->bsr_type = BSR_UNKNOWN; + printk(KERN_INFO "unknown BSR size %d\n",cur->bsr_bytes); + } + + cur->bsr_num = bsr_types[cur->bsr_type]; + bsr_types[cur->bsr_type] = cur->bsr_num + 1; + snprintf(cur->bsr_name, 32, "bsr%d_%d", + cur->bsr_bytes, cur->bsr_num); + + cdev_init(&cur->bsr_cdev, &bsr_fops); + result = cdev_add(&cur->bsr_cdev, cur->bsr_dev, 1); + if (result) + goto out_err; + + cur->bsr_device = device_create(bsr_class, NULL, + cur->bsr_dev, + cur->bsr_name); + if (!cur->bsr_device) { + printk(KERN_ERR "device_create failed for %s\n", + cur->bsr_name); + cdev_del(&cur->bsr_cdev); + goto out_err; + } + dev_set_drvdata(cur->bsr_device, cur); + } + + return 0; + + out_err: + + bsr_cleanup_devs(); + return -ENODEV; +} + +static int __init bsr_init(void) +{ + struct device_node *np; + dev_t bsr_dev = MKDEV(bsr_major, 0); + int ret = -ENODEV; + int result; + + np = of_find_compatible_node(NULL, "ibm,bsr", "ibm,bsr"); + if (!np) + goto out_err; + + bsr_class = class_create(THIS_MODULE, "bsr"); + if (IS_ERR(bsr_class)) { + printk(KERN_ERR "class_create() failed for bsr_class\n"); + goto out_err_1; + } + bsr_class->dev_attrs = bsr_dev_attrs; + + result = alloc_chrdev_region(&bsr_dev, 0, BSR_MAX_DEVS, "bsr"); + bsr_major = MAJOR(bsr_dev); + if (result < 0) { + printk(KERN_ERR "alloc_chrdev_region() failed for bsr\n"); + goto out_err_2; + } + + if ((ret = bsr_create_devs(np)) < 0) + goto out_err_3; + + of_node_put(np); + + return 0; + + out_err_3: + unregister_chrdev_region(bsr_dev, BSR_MAX_DEVS); + + out_err_2: + class_destroy(bsr_class); + + out_err_1: + of_node_put(np); + + out_err: + + return ret; +} + +static void __exit bsr_exit(void) +{ + + bsr_cleanup_devs(); + + if (bsr_class) + class_destroy(bsr_class); + + if (bsr_major) + unregister_chrdev_region(MKDEV(bsr_major, 0), BSR_MAX_DEVS); +} + +module_init(bsr_init); +module_exit(bsr_exit); +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Sonny Rao "); Index: linux-dev/drivers/char/Makefile =================================================================== --- linux-dev.orig/drivers/char/Makefile 2008-06-18 01:24:04.000000000 -0500 +++ linux-dev/drivers/char/Makefile 2008-06-18 01:38:30.000000000 -0500 @@ -57,6 +57,7 @@ obj-$(CONFIG_VIOCONS) += viocons.o obj-$(CONFIG_VIOTAPE) += viotape.o obj-$(CONFIG_HVCS) += hvcs.o +obj-$(CONFIG_IBM_BSR) += bsr.o obj-$(CONFIG_SGI_MBCS) += mbcs.o obj-$(CONFIG_BRIQ_PANEL) += briq_panel.o obj-$(CONFIG_BFIN_OTP) += bfin-otp.o Index: linux-dev/drivers/char/Kconfig =================================================================== --- linux-dev.orig/drivers/char/Kconfig 2008-06-18 01:24:04.000000000 -0500 +++ linux-dev/drivers/char/Kconfig 2008-06-18 01:38:30.000000000 -0500 @@ -649,6 +649,14 @@ which will also be compiled when this driver is built as a module. +config IBM_BSR + tristate "IBM POWER Barrier Synchronization Register support" + depends on PPC_PSERIES + help + This devices exposes a hardware mechanism for fast synchronization + of threads across a large system which avoids bouncing a cacheline + between several cores on a system + source "drivers/char/ipmi/Kconfig" config DS1620 Index: linux-dev/arch/powerpc/configs/pseries_defconfig =================================================================== --- linux-dev.orig/arch/powerpc/configs/pseries_defconfig 2008-07-07 20:44:42.000000000 -0500 +++ linux-dev/arch/powerpc/configs/pseries_defconfig 2008-07-07 20:45:21.000000000 -0500 @@ -946,6 +946,7 @@ CONFIG_HVC_CONSOLE=y CONFIG_HVC_RTAS=y CONFIG_HVCS=m +CONFIG_IBM_BSR=m # CONFIG_IPMI_HANDLER is not set # CONFIG_HW_RANDOM is not set CONFIG_GEN_RTC=y