[PATCH] ARM: mach-shmobile: Add memchunk support V2

[PATCH] ARM: mach-shmobile: Add memchunk support V2

[Magnus Damm]

From: Magnus Damm <damm@opensource.se>

Add SH-Mobile ARM support for "memchunk" V2. The code is based
on the SH implementation but has been slightly modified to use 
memblock for physical memory reservation.

The function shmobile_memchunk_setup() should be called during
boot from the ->reserve() callback in struct machine_desc.

Signed-off-by: Magnus Damm <damm@opensource.se>
---

 arch/arm/mach-shmobile/Makefile              |    1 
 arch/arm/mach-shmobile/include/mach/common.h |    3 +
 arch/arm/mach-shmobile/memchunk.c            |   76 ++++++++++++++++++++++++++
 3 files changed, 80 insertions(+)

--- 0008/arch/arm/mach-shmobile/Makefile
+++ work/arch/arm/mach-shmobile/Makefile	2010-10-14 17:44:33.000000000 +0900
@@ -4,6 +4,7 @@
 
 # Common objects
 obj-y				:= timer.o console.o clock.o pm_runtime.o
+obj-y				+= memchunk.o
 
 # CPU objects
 obj-$(CONFIG_ARCH_AP45)		+= head-ap45.o
--- 0001/arch/arm/mach-shmobile/include/mach/common.h
+++ work/arch/arm/mach-shmobile/include/mach/common.h	2010-10-14 17:53:04.000000000 +0900
@@ -3,6 +3,9 @@
 
 extern struct sys_timer shmobile_timer;
 extern void shmobile_setup_console(void);
+struct platform_device;
+extern int shmobile_memchunk_setup(struct platform_device *pdev,
+				   char *name, unsigned long memsize);
 struct clk;
 extern int clk_init(void);
 
--- /dev/null
+++ work/arch/arm/mach-shmobile/memchunk.c	2010-10-14 17:53:22.000000000 +0900
@@ -0,0 +1,76 @@
+/*
+ * SH-Mobile "memchunk" - Physically contiguous memory reservation code
+ *
+ * Copyright (C) 2010  Magnus Damm
+ *
+ * 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; version 2 of the License.
+ *
+ * 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., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ */
+#include <linux/init.h>
+#include <linux/platform_device.h>
+#include <linux/memblock.h>
+#include <mach/common.h>
+
+static int __init memchunk_setup(char *str)
+{
+	return 1; /* accept anything that begins with "memchunk." */
+}
+__setup("memchunk.", memchunk_setup);
+
+static void __init memchunk_cmdline_override(char *name, unsigned long *sizep)
+{
+	char *p = boot_command_line;
+	int k = strlen(name);
+
+	while ((p = strstr(p, "memchunk."))) {
+		p += 9; /* strlen("memchunk.") */
+		if (!strncmp(name, p, k) && p[k] = '=') {
+			p += k + 1;
+			*sizep = memparse(p, NULL);
+			pr_info("%s: forcing memory chunk size to 0x%08lx\n",
+				name, *sizep);
+			break;
+		}
+	}
+}
+
+int __init shmobile_memchunk_setup(struct platform_device *pdev,
+				   char *name, unsigned long memsize)
+{
+	struct resource *r;
+	u64 addr;
+
+	r = pdev->resource + pdev->num_resources - 1;
+	if (r->flags) {
+		pr_warning("%s: unable to find empty space for resource\n",
+			name);
+		return -EINVAL;
+	}
+
+	memchunk_cmdline_override(name, &memsize);
+	if (!memsize)
+		return 0;
+
+	addr = memblock_alloc(memsize, PAGE_SIZE);
+	if (addr = ~(u64)0) {
+		pr_warning("%s: unable to allocate memory\n", name);
+		return -ENOMEM;
+	}
+
+	r->flags = IORESOURCE_MEM;
+	r->start = addr;
+	r->end = r->start + memsize - 1;
+	r->name = name;
+	return 0;
+}

Re: [PATCH] ARM: mach-shmobile: Add memchunk support V2

[Paul Mundt]

On Thu, Oct 14, 2010 at 06:21:52PM +0900, Magnus Damm wrote:
> From: Magnus Damm <damm@opensource.se>
> 
> Add SH-Mobile ARM support for "memchunk" V2. The code is based
> on the SH implementation but has been slightly modified to use 
> memblock for physical memory reservation.
> 
There are two issues here..

- first:

Is there any reason not to use the same approach on SH? I rather dislike
having wholly generic code split between the architectures for no reason,
as seems to be the trend lately.

All of this started out abusing the platform_ namespace because it
related to platform devices, as it still does. You were told previously
to run this by gregkh and make a drivers/base/platform.c push for it
the last time you posted these patches, and so far I haven't seen any
progress in that department.

Now you've moved on to memblock allocation/reservation, which while not
as generic as the DMA allocator, is still common between all the
platforms that care about this. The only stuff that really belongs in the
architecture subdirectories are the reservation callbacks, which in this
case we can quite easily wrap in to a generic callback.

Now that memblock is common enough, the alternative you have is simply to
do the reservations from the ->reserve() callback and then have the
driver code grovel for a mapping with memblock_find(). You can of course
use ATAGs to aid in this if you wish, as OMAP does for its framebuffer
memory.

- second, and more fundamental:

My main concern however is that this basically won't work for the common
platform device case on ARM given that all of the memblock allocations
have valid backing PFNs. The UIO case "works" by virtue of blindsiding
the remap path and calling in to remap_pfn_range() directly, which is
unaware of any architectural limitations. Any non-UIO driver using this
scheme today will just error out on the first ioremap() once the
pfn_valid() check returns.

In order to support this properly you will need to punch out memblock
region holes, not unlike the ARCH_HAS_HOLES_MEMORYMODEL behaviour, albeit
with a bit more flexibility. I vaguely recall having mentioned this
before, too.