Skip to contentSkip to conversation

physmap: make map_info customizable

PATCH In review [email protected] 3 participants 4 messages Dec 1, 2008 – Jan 20, 2009

Subsystems (2)
Memory technology devices (MTD) MIPS
lore
Share & tools
lore

[email protected]

author
#0 lore raw

PATCH 1/6 physmap: make map_info customizable

From: Atsushi Nemoto <[email protected]>

Add a hook to physmap_flash_data for map_info initialization. This makes platform code can customize map operations.

Background: I wrote this patch to support RBTX4939 board, which requires custom map->{read,write,copy_from} function. I think extending the physmap driver is better than adding a new map driver.

Signed-off-by Atsushi Nemoto <[email protected]>
Signed-off-by Andrew Morton <[email protected]>

---

 drivers/mtd/maps/physmap.c  |    5 ++++-
 include/linux/mtd/physmap.h |    1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
drivers/mtd/maps/physmap.c renamed from drivers/mtd/maps/physmap.c~physmap-make-map_info-customizable +4 -1
diff -puN drivers/mtd/maps/physmap.c~physmap-make-map_info-customizable drivers/mtd/maps/physmap.c
@@ -128,7 +128,10 @@ static int physmap_flash_probe(struct pl
 			goto err_out;
 		}
 
-		simple_map_init(&info->map[i]);
+		if (physmap_data->map_init)
+			(*physmap_data->map_init)(&info->map[i]);
+		else
+			simple_map_init(&info->map[i]);
 
 		probe_type = rom_probe_types;
 		for (; info->mtd[i] == NULL && *probe_type != NULL; probe_type++)
include/linux/mtd/physmap.h renamed from include/linux/mtd/physmap.h~physmap-make-map_info-customizable +1 -0
diff -puN include/linux/mtd/physmap.h~physmap-make-map_info-customizable include/linux/mtd/physmap.h
@@ -25,6 +25,7 @@ struct physmap_flash_data {
 	void			(*set_vpp)(struct map_info *, int);
 	unsigned int		nr_parts;
 	struct mtd_partition	*parts;
+	void			(*map_init)(struct map_info *map);
 };
 
 /*

_

#1 lore raw
Quoted 6 lines · Atsushi Nemoto

Add a hook to physmap_flash_data for map_info initialization. This makes platform code can customize map operations.

Background: I wrote this patch to support RBTX4939 board, which requires custom map->{read,write,copy_from} function. I think extending the physmap driver is better than adding a new map driver.

Hm. I'm unconvinced -- surely you're calling into a function which does everything that a 'new map driver' would have to do anyway?

Would you advocate that we remove _all_ the dedicated map drivers and replace them with this method?

Can't you enable physmap without CONFIG_MTD_COMPLEX_MAPPINGS being set anyway? The 'special' functions can't be used in that config, and we just assume that direct access works.

Signature · 2 lines
David Woodhouse                            Open Source Technology Centre
[email protected]                              Intel Corporation
#2 lore raw
Quoted 9 lines · David Woodhouse
Quoted 6 lines

Add a hook to physmap_flash_data for map_info initialization. This makes platform code can customize map operations.

Background: I wrote this patch to support RBTX4939 board, which requires custom map->{read,write,copy_from} function. I think extending the physmap driver is better than adding a new map driver.

Hm. I'm unconvinced -- surely you're calling into a function which does everything that a 'new map driver' would have to do anyway?

Yes. Once I wrote a new map driver and realized most of it is duplication of physmap driver.

Quoted 2 lines

Would you advocate that we remove _all_ the dedicated map drivers and replace them with this method?

Well, not _all_, but perhaps some drivers can be converted to physmap with this method. (And I suppose some drivers can be converted even without this patch.)

Quoted 3 lines

Can't you enable physmap without CONFIG_MTD_COMPLEX_MAPPINGS being set anyway? The 'special' functions can't be used in that config, and we just assume that direct access works.

The board has four flash rom areas and only boot flash rom area needs special mapping. So I can enable physmap always, but the boot rom cannot be used without MTD_COMPLEX_MAPPINGS.

FYI, Here is a platform code I want to send if the physmap patch was merged. Note that this patch cannot be applied cleanly on any tree while it depends on some patches pending on linux-mips ML.

Subject: [PATCH] rbtx4939: mtd support

This board has complex flash mappings, controlled by its DIPSW setting.

Signed-off-by Atsushi Nemoto <[email protected]>
---
 arch/mips/txx9/rbtx4939/setup.c |  161 +++++++++++++++++++++++++++++++++++++++
 1 files changed, 161 insertions(+), 0 deletions(-)
arch/mips/txx9/rbtx4939/setup.c +161 -0
index 74839f2..b1a5949 100644
@@ -16,6 +16,8 @@
 #include <linux/leds.h>
 #include <linux/interrupt.h>
 #include <linux/smc91x.h>
+#include <linux/mtd/physmap.h>
+#include <linux/mtd/map.h>
 #include <asm/reboot.h>
 #include <asm/txx9/generic.h>
 #include <asm/txx9/pci.h>

@@ -282,6 +284,164 @@ static void rbtx4939_7segled_putc(unsigned int pos, unsigned char val)
 	__rbtx4939_7segled_putc(pos, val);
 }
 
+#ifdef CONFIG_MTD_COMPLEX_MAPPINGS
+/* special mapping for boot rom */
+static unsigned long rbtx4939_flash_fixup_ofs(unsigned long ofs)
+{
+	u8 bdipsw = readb(rbtx4939_bdipsw_addr) & 0x0f;
+	unsigned char shift;
+
+	if (bdipsw & 8) {
+		/* BOOT Mode: USER ROM1 / USER ROM2 */
+		shift = bdipsw & 3;
+		/* rotate A[23:22] */
+		return (ofs & ~0xc00000) | ((((ofs >> 22) + shift) & 3) << 22);
+	}
+#ifdef __BIG_ENDIAN
+	if (bdipsw == 0)
+		/* BOOT Mode: Monitor ROM */
+		ofs ^= 0x400000;	/* swap A[22] */
+#endif
+	return ofs;
+}
+
+static map_word rbtx4939_flash_read16(struct map_info *map, unsigned long ofs)
+{
+	map_word r;
+
+	ofs = rbtx4939_flash_fixup_ofs(ofs);
+	r.x[0] = __raw_readw(map->virt + ofs);
+	return r;
+}
+
+static void rbtx4939_flash_write16(struct map_info *map, const map_word datum,
+				   unsigned long ofs)
+{
+	ofs = rbtx4939_flash_fixup_ofs(ofs);
+	__raw_writew(datum.x[0], map->virt + ofs);
+	mb();	/* see inline_map_write() in mtd/map.h */
+}
+
+static void rbtx4939_flash_copy_from(struct map_info *map, void *to,
+				     unsigned long from, ssize_t len)
+{
+	u8 bdipsw = readb(rbtx4939_bdipsw_addr) & 0x0f;
+	unsigned char shift;
+	ssize_t curlen;
+
+	from += (unsigned long)map->virt;
+	if (bdipsw & 8) {
+		/* BOOT Mode: USER ROM1 / USER ROM2 */
+		shift = bdipsw & 3;
+		while (len) {
+			curlen = min((unsigned long)len,
+				     0x400000 -	(from & (0x400000 - 1)));
+			memcpy(to,
+			       (void *)((from & ~0xc00000) |
+					((((from >> 22) + shift) & 3) << 22)),
+			       curlen);
+			len -= curlen;
+			from += curlen;
+			to += curlen;
+		}
+		return;
+	}
+#ifdef __BIG_ENDIAN
+	if (bdipsw == 0) {
+		/* BOOT Mode: Monitor ROM */
+		while (len) {
+			curlen = min((unsigned long)len,
+				     0x400000 - (from & (0x400000 - 1)));
+			memcpy(to, (void *)(from ^ 0x400000), curlen);
+			len -= curlen;
+			from += curlen;
+			to += curlen;
+		}
+		return;
+	}
+#endif
+	memcpy(to, (void *)from, len);
+}
+
+static void rbtx4939_flash_map_init(struct map_info *map)
+{
+	map->read = rbtx4939_flash_read16;
+	map->write = rbtx4939_flash_write16;
+	map->copy_from = rbtx4939_flash_copy_from;
+}
+#endif /* CONFIG_MTD_COMPLEX_MAPPINGS */
+
+static void __init rbtx4939_mtd_init(void)
+{
+#if defined(CONFIG_MTD_PHYSMAP) || defined(CONFIG_MTD_PHYSMAP_MODULE)
+	static struct {
+		struct platform_device dev;
+		struct resource res;
+		struct physmap_flash_data data;
+	} pdevs[4];
+	int i;
+#ifdef CONFIG_MTD_COMPLEX_MAPPINGS
+	static char names[4][8];
+	static struct mtd_partition parts[4];
+	struct physmap_flash_data *boot_pdata = &pdevs[0].data;
+	u8 bdipsw = readb(rbtx4939_bdipsw_addr) & 0x0f;
+
+	if (bdipsw & 8) {
+		/* BOOT Mode: USER ROM1 / USER ROM2 */
+		boot_pdata->nr_parts = 4;
+		for (i = 0; i < boot_pdata->nr_parts; i++) {
+			sprintf(names[i], "img%d", 4 - i);
+			parts[i].name = names[i];
+			parts[i].size = 0x400000;
+			parts[i].offset = MTDPART_OFS_NXTBLK;
+		}
+	} else if (bdipsw == 0) {
+		/* BOOT Mode: Monitor ROM */
+		boot_pdata->nr_parts = 2;
+		strcpy(names[0], "big");
+		strcpy(names[1], "little");
+		for (i = 0; i < boot_pdata->nr_parts; i++) {
+			parts[i].name = names[i];
+			parts[i].size = 0x400000;
+			parts[i].offset = MTDPART_OFS_NXTBLK;
+		}
+	} else {
+		/* BOOT Mode: ROM Emulator */
+		boot_pdata->nr_parts = 2;
+		parts[0].name = "boot";
+		parts[0].offset = 0xc00000;
+		parts[0].size = 0x400000;
+		parts[1].name = "user";
+		parts[1].offset = 0;
+		parts[1].size = 0xc00000;
+	}
+	boot_pdata->parts = parts;
+	boot_pdata->map_init = rbtx4939_flash_map_init;
+#endif
+
+	for (i = 0; i < ARRAY_SIZE(pdevs); i++) {
+		struct resource *r = &pdevs[i].res;
+		struct platform_device *dev = &pdevs[i].dev;
+
+#ifndef CONFIG_MTD_COMPLEX_MAPPINGS
+		/* skip boot flash */
+		if (i == 0)
+			continue;
+#endif
+		r->start = 0x1f000000 - i * 0x1000000;
+		r->end = r->start + 0x1000000 - 1;
+		r->flags = IORESOURCE_MEM;
+		pdevs[i].data.width = 2;
+		dev->num_resources = 1;
+		dev->resource = r;
+		dev->id = i;
+		dev->name = "physmap-flash";
+		dev->dev.platform_data = &pdevs[i].data;
+		platform_device_register(dev);
+	}
+#endif
+}
+
 static void __init rbtx4939_arch_init(void)
 {
 	rbtx4939_pci_setup();

@@ -333,6 +493,7 @@ static void __init rbtx4939_device_init(void)
 	    platform_device_add_data(pdev, &smc_pdata, sizeof(smc_pdata)) ||
 	    platform_device_add(pdev))
 		platform_device_put(pdev);
+	rbtx4939_mtd_init();
 	/* TC58DVM82A1FT: tDH=10ns, tWP=tRP=tREADID=35ns */
 	tx4939_ndfmc_init(10, 35,
 			  (1 << 1) | (1 << 2),
#3 lore raw
Quoted 7 lines · Atsushi Nemoto
Quoted 7 lines
Quoted 6 lines

Add a hook to physmap_flash_data for map_info initialization. This makes platform code can customize map operations.

Background: I wrote this patch to support RBTX4939 board, which requires custom map->{read,write,copy_from} function. I think extending the physmap driver is better than adding a new map driver.

Quoted 5 lines
Quoted 2 lines

Hm. I'm unconvinced -- surely you're calling into a function which does everything that a 'new map driver' would have to do anyway?

Yes. Once I wrote a new map driver and realized most of it is duplication of physmap driver.

While this patch had NACKed a while ago, I have just posted a new map driver for the board. I hope this driver is acceptable.

---
Atsushi Nemoto