From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from tidalnetworks.net (mail.consentry.com [75.35.230.10]) by ozlabs.org (Postfix) with ESMTP id E9369DDDFF for ; Sat, 18 Oct 2008 13:14:10 +1100 (EST) Message-ID: <48F9466D.8060603@consentry.com> Date: Fri, 17 Oct 2008 19:14:05 -0700 From: Mike Ditto MIME-Version: 1.0 To: Mike Ditto , linuxppc-dev@ozlabs.org Subject: Re: device tree variations References: <48F7E1E6.4040701@consentry.com> <20081017012604.GA22213@yookeroo.seuss> In-Reply-To: <20081017012604.GA22213@yookeroo.seuss> Content-Type: text/plain; charset=ISO-8859-1 List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , David Gibson wrote: > Deleting the irrelevant parts or picking a device tree to pass to > fdt_init() are both reasonable solutions. libfdt which is included in > the bootwrapper has functions for removing unwanted nodes: either > fdt_nop_node() or fdt_del_node() will suffice. There isn't currently > a dt_ops hook to call though to those functions though. You could > either add one, or (knowing that your platform always has a flat dt) > bypass the dt_ops hooks and call libfdt directly. Thanks. The fdt_del_node approach works pretty nicely. I added a dt_ops hook since fdt is static in libfdt-wrapper.c. At first I tried fdt_nop_node, fearing that find_node_by_prop_value() and fdt_del_node() would interact badly when deleting in mid-traversal, but it turns out that fdt_nop_node() upsets find_node_by_prop_value() anyway. Plus, the kernel prints harmless but strange messages when there are NOPs in the tree. So I use fdt_del_node() and rescan from the top each time I delete a node and it works fine. /* Find all product-dependent nodes and delete inapplicable ones. */ dev = NULL; while ((dev = find_node_by_prop_value(dev, "product-dependent", "", 0)) != NULL) { u32 mask; int len; len = getprop(dev, "productmask", &mask, sizeof mask); if (len == sizeof mask) { if ((mask & (1 << product_id)) == 0) { del_node(dev); /* Have to restart from the beginning. */ dev = NULL; } } } I had to fix the memcmp bug I mentioned in an earlier posting to allow searching for a boolean (empty) property. If anyone cares, below is a trivial patch to expose the del_node() operation via dt_ops. Thanks again, -=] Mike [=- Index: arch/powerpc/boot/ops.h =================================================================== retrieving revision 1.1.1.1 diff -u -r1.1.1.1 ops.h --- arch/powerpc/boot/ops.h 11 Oct 2008 02:51:35 -0000 1.1.1.1 +++ arch/powerpc/boot/ops.h 18 Oct 2008 02:06:45 -0000 @@ -40,6 +40,7 @@ const int buflen); int (*setprop)(const void *phandle, const char *name, const void *buf, const int buflen); + int (*del_node)(const void *phandle); void *(*get_parent)(const void *phandle); /* The node must not already exist. */ void *(*create_node)(const void *parent, const char *name); @@ -124,6 +125,11 @@ return dt_ops.setprop(devp, name, buf, strlen(buf) + 1); return -1; +} + +static inline int del_node(const void *devp) +{ + return dt_ops.del_node ? dt_ops.del_node(devp) : -1; } static inline void *get_parent(const char *devp) Index: arch/powerpc/boot/libfdt-wrapper.c =================================================================== retrieving revision 1.1.1.1 diff -u -r1.1.1.1 libfdt-wrapper.c --- arch/powerpc/boot/libfdt-wrapper.c 11 Oct 2008 02:51:35 -0000 1.1.1.1 +++ arch/powerpc/boot/libfdt-wrapper.c 17 Oct 2008 22:08:44 -0000 @@ -105,6 +105,11 @@ return check_err(rc); } +static int fdt_wrapper_del_node(const void *devp) +{ + return fdt_del_node(fdt, devp_offset(devp)); +} + static void *fdt_wrapper_get_parent(const void *devp) { return offset_devp(fdt_parent_offset(fdt, devp_offset(devp))); @@ -173,6 +178,7 @@ dt_ops.create_node = fdt_wrapper_create_node; dt_ops.find_node_by_prop_value = fdt_wrapper_find_node_by_prop_value; dt_ops.find_node_by_compatible = fdt_wrapper_find_node_by_compatible; + dt_ops.del_node = fdt_wrapper_del_node; dt_ops.get_path = fdt_wrapper_get_path; dt_ops.finalize = fdt_wrapper_finalize;