Re: [PATCH] of/fdt: Add FDT address translation
From: Grant Likely
Date: Sat Jun 15 2013 - 09:55:00 EST
On Fri, 24 May 2013 17:36:30 +0800, Ley Foon Tan <lftan@xxxxxxxxxx> wrote:
> This patch adds address translation to fdt. It is needed when the early
> console is connected to a simple-bus (bridge) that has address translation
> enabled.
>
> Walter Goossens have submitted first version of patch previously. This
> patch resolved the feedback from first submission and some enhancements
> on translation functions.
>
> Reviewed-by: Walter Goossens <waltergoossens@xxxxxxx>
> Signed-off-by: Ley Foon Tan <lftan@xxxxxxxxxx>
Hi Ley,
Thanks for looking at this. Comments below. Also, I would like to see
the patch that uses this new API.
> ---
> drivers/of/fdt.c | 188 ++++++++++++++++++++++++++++++++++++++++++++++++
> include/linux/of_fdt.h | 2 +
> 2 files changed, 190 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
> index 808be06..74cc1bc 100644
> --- a/drivers/of/fdt.c
> +++ b/drivers/of/fdt.c
> @@ -695,6 +695,194 @@ int __init early_init_dt_scan_chosen(unsigned long node, const char *uname,
> /* break now */
> return 1;
> }
> +/**
> + * flat_dt_translate_address - Translate an address using the ranges property
> + *
> + * This function converts address from "node address-space" to "parent address-
> + * space"
> + */
> +static int __init flat_dt_translate_address(unsigned long node,
> + unsigned long parent, u64 *address)
> +{
> + unsigned long size = 0;
> + __be32 *prop;
> + __be32 *ranges;
> + int size_cells = 0;
> + int addr_cells = 0;
> + int paddr_cells = OF_ROOT_NODE_ADDR_CELLS_DEFAULT;
> +
> + ranges = of_get_flat_dt_prop(node, "ranges", &size);
> +
> + if (!ranges) {
> + pr_warn("Address cannot be translated\n");
> + return -EINVAL;
> + }
> +
> + if (!size) {
> + pr_debug("No translation possible/necessary\n");
> + return 0;
This debug statement is misleading. An empty ranges property means
identity mapping. "Identity mapping, no translation necessary" would be
more accurate.
> + }
> +
> + prop = of_get_flat_dt_prop(node, "#size-cells", NULL);
> + if (!prop)
> + return -EINVAL;
> + size_cells = be32_to_cpup(prop);
> +
> + prop = of_get_flat_dt_prop(node, "#address-cells", NULL);
> + if (!prop)
> + return -EINVAL;
> + addr_cells = be32_to_cpup(prop);
> +
> + if (parent) {
> + prop = of_get_flat_dt_prop(parent, "#address-cells", NULL);
> + if (prop)
> + paddr_cells = be32_to_cpup(prop);
> + }
> + if ((addr_cells <= 0) || (size_cells <= 0) ||
> + (addr_cells > 2) || (size_cells > 2) || (paddr_cells > 2)) {
> + pr_warn("Translation not possible in fdt. Invalid address.\n");
addr_cells or size_cells > 2 isn't necessarily an error, it just means
this simplistic function cannot parse them. The message is misleading.
> + *address = 0;
> + return -1;
Return a real error code here. -EOVERFLOW would be appropriate.
> + }
> +
> + while (size > 0) {
should be:
while (size >= (addr_cells + size_cells + paddr_cells) * sizeof(u32))
Otherwise you risk overflowing the property.
> + u64 from, to, tsize;
> + from = be32_to_cpup(ranges++);
> + size -= 4;
> + if (addr_cells == 2) {
> + from += (((u64)be32_to_cpup(ranges++)) << 32);
> + size -= 4;
> + }
> + to = be32_to_cpup(ranges++);
> + size -= 4;
> + if (paddr_cells == 2) {
> + to += (((u64)be32_to_cpup(ranges++)) << 32);
> + size -= 4;
> + }
> + tsize = be32_to_cpup(ranges++);
> + size -= 4;
> + if (size_cells == 2) {
> + tsize += (((u64)be32_to_cpup(ranges++)) << 32);
> + size -= 4;
> + }
All of the above looks backwards. The first cell is the MSB, not the
second. The parser needs to swap the cells around.
The following is cleaner and shorter:
u64 from = 0, to = 0, tsize = 0;
int i;
for (i = 0; i < addr_cells; i++, size -= 4)
from = (from << 32) | be32_to_cpup(ranges++);
for (i = 0; i < paddr_cells; i++, size -= 4)
to = (from << 32) | be32_to_cpup(ranges++);
for (i = 0; i < size_cells; i++, size -= 4)
tsize = (from << 32) | be32_to_cpup(ranges++);
> + pr_debug(" From %llX To %llX Size %llX\n", from, to, tsize);
> + if ((*address >= from) && (*address < (from + tsize)))
> + *address += (to - from);
> + }
> + return 1;
> +}
> +
> +static int __init of_scan_flat_dt_ranges(unsigned long *pnode,
> + unsigned long parent, unsigned long target,
> + u64 *address, int ignore)
> +{
> + int rc = 0;
> + int depth = -1;
> + const char *pathp;
> + unsigned long p = *pnode;
> + do {
> + u32 tag = be32_to_cpup((__be32 *)p);
> +
> + p += 4;
> + if (tag == OF_DT_END_NODE) {
> + if (depth--)
> + break;
> + else
> + continue;
> + }
> + if (tag == OF_DT_NOP)
> + continue;
> + if (tag == OF_DT_END)
> + break;
> + if (tag == OF_DT_PROP) {
> + u32 sz = be32_to_cpup((__be32 *)p);
> + p += 8;
> + if (be32_to_cpu(initial_boot_params->version) < 0x10)
> + p = ALIGN(p, sz >= 8 ? 8 : 4);
> + p += sz;
> + p = ALIGN(p, 4);
> + continue;
> + }
> + if (tag != OF_DT_BEGIN_NODE) {
> + pr_err("Invalid tag %x in flat device tree!\n", tag);
> + return -EINVAL;
> + }
> + pathp = (char *)p;
> + p = ALIGN(p + strlen(pathp) + 1, 4);
> + if ((*pathp) == '/')
> + pathp = kbasename(pathp);
> + if ((ignore == 0) && (p == target)) {
> + rc = 1;
> + ignore++;
> + pr_debug("Found target. Start address translation\n");
> + }
> + if (depth) {
> + int res;
> + *pnode = p;
> + res = of_scan_flat_dt_ranges(pnode, p, target,
> + address, ignore);
> + if (res < 0) {
> + /* Something bad happened */
> + return -1;
> + } else if (res > 0) {
> + /* Something good happened. Translate. */
> + rc++;
> + pr_debug("translate %08lX %s\n", p, pathp);
> + if (flat_dt_translate_address(p, parent,
> + address) < 0)
> + return -1;
> + }
> + p = *pnode;
> + } else
> + depth++;
> + } while (1);
This is quite a complicated function, and it is intermingling the
parsing of ranges with merely finding the parent of a node. Finding the
parent node is quite a useful activity, so I think it would be much
better to split that bit out into a separate function. Something like
"of_fdt_find_parent()".
Rather than reimplementing the entire flat tree parser,
of_fdt_find_parent() should be a wrapper around of_scan_flat_dt(). Use a
callback function and a state variable to keep track of each node.
Something like the following (incomplete; you'll need to work out the
details):
struct find_parent_data {
unsigned long child;
unsigned long parents[64]; /* arbitrary limit of 64 nodes deep */
unsigned long result;
};
int check_parent(unsigned long node, const char *uname, int depth, void *__data)
{
struct find_parent_data *data = __data;
if (depth >= 64)
return -EOVERFLOW;
data->parents[depth] = node;
if (node == data->child && depth > 0)) {
data->result = parents[depth-1];
return 1;
}
return 0;
}
unsigned long of_flat_dt_find_parent(unsigned long node)
{
struct find_parent_data data;
data.child = node;
if (of_scan_flat_dt(check_parent, &data) == 1)
return data.result;
return 0;
}
> +
> + *pnode = p;
> + return rc;
> +}
> +/**
> + * of_get_flat_dt_address - get address from a node
> + * @node: targeted node
> + *
> + * Returns address or OF_BAD_ADDR if error.
> + */
> +u64 __init of_get_flat_dt_address(unsigned long node)
> +{
> + __be32 *valp;
> + u64 addr = OF_BAD_ADDR;
> + unsigned long sz;
> +
> + valp = of_get_flat_dt_prop(node, "reg", &sz);
> + if (valp) {
> + if (sz == 1)
This is buggy. The code cannot know the size of the address portion of
the reg property without reading #address-cells of the parent node. Plus
a reg property may have more than one address range.
> + addr = be32_to_cpu(valp[0]);
> + else if (sz == 2)
> + addr = (((u64)be32_to_cpu(valp[0]) << 32)) |
> + be32_to_cpu(valp[1]);
> + }
> + return addr;
> +}
> +
> +/**
> + * of_get_flat_dt_translated_address - get translated address from a node
> + * @node: targeted node
> + *
> + * Returns translated address or OF_BAD_ADDR if error.
> + */
> +u64 __init of_get_flat_dt_translated_address(unsigned long node)
> +{
> + unsigned long p = of_get_flat_dt_root();
> + u64 addr64;
> + int rc;
> +
> + addr64 = of_get_flat_dt_address(node);
> + if (addr64 != OF_BAD_ADDR) {
> + rc = of_scan_flat_dt_ranges(&p, 0, node, &addr64, 0);
> + if (rc > 0)
> + return addr64;
> + }
> + return OF_BAD_ADDR;
> +}
>
> /**
> * unflatten_device_tree - create tree of device_nodes from flat blob
> diff --git a/include/linux/of_fdt.h b/include/linux/of_fdt.h
> index ed136ad..48046ed 100644
> --- a/include/linux/of_fdt.h
> +++ b/include/linux/of_fdt.h
> @@ -100,6 +100,8 @@ extern void early_init_dt_add_memory_arch(u64 base, u64 size);
> extern void * early_init_dt_alloc_memory_arch(u64 size, u64 align);
> extern u64 dt_mem_next_cell(int s, __be32 **cellp);
>
> +extern u64 __init of_get_flat_dt_address(unsigned long node);
> +extern u64 __init of_get_flat_dt_translated_address(unsigned long node);
> /*
> * If BLK_DEV_INITRD, the fdt early init code will call this function,
> * to be provided by the arch code. start and end are specified as
> --
> 1.7.7.4
>
>
--
email sent from notmuch.vim plugin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/