[net-next-2.6.git] / arch / powerpc / platforms / cell / io-workarounds.c

/*
 *  Copyright (C) 2006 Benjamin Herrenschmidt <benh@kernel.crashing.org>
 *		       IBM, Corp.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */
#undef DEBUG

#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/pci.h>
#include <asm/io.h>
#include <asm/machdep.h>
#include <asm/pci-bridge.h>
#include <asm/ppc-pci.h>


#define SPIDER_PCI_REG_BASE		0xd000
#define SPIDER_PCI_VCI_CNTL_STAT	0x0110
#define SPIDER_PCI_DUMMY_READ		0x0810
#define SPIDER_PCI_DUMMY_READ_BASE	0x0814

/* Undefine that to re-enable bogus prefetch
 *
 * Without that workaround, the chip will do bogus prefetch past
 * page boundary from system memory. This setting will disable that,
 * though the documentation is unclear as to the consequences of doing
 * so, either purely performances, or possible misbehaviour... It's not
 * clear wether the chip can handle unaligned accesses at all without
 * prefetching enabled.
 *
 * For now, things appear to be behaving properly with that prefetching
 * disabled and IDE, possibly because IDE isn't doing any unaligned
 * access.
 */
#define SPIDER_DISABLE_PREFETCH

#define MAX_SPIDERS	3

static struct spider_pci_bus {
	void __iomem	*regs;
	unsigned long	mmio_start;
	unsigned long	mmio_end;
	unsigned long	pio_vstart;
	unsigned long	pio_vend;
} spider_pci_busses[MAX_SPIDERS];
static int spider_pci_count;

static struct spider_pci_bus *spider_pci_find(unsigned long vaddr,
					      unsigned long paddr)
{
	int i;

	for (i = 0; i < spider_pci_count; i++) {
		struct spider_pci_bus *bus = &spider_pci_busses[i];
		if (paddr && paddr >= bus->mmio_start && paddr < bus->mmio_end)
			return bus;
		if (vaddr && vaddr >= bus->pio_vstart && vaddr < bus->pio_vend)
			return bus;
	}
	return NULL;
}

static void spider_io_flush(const volatile void __iomem *addr)
{
	struct spider_pci_bus *bus;
	int token;

	/* Get platform token (set by ioremap) from address */
	token = PCI_GET_ADDR_TOKEN(addr);

	/* Fast path if we have a non-0 token, it indicates which bus we
	 * are on.
	 *
	 * If the token is 0, that means either that the ioremap was done
	 * before we initialized this layer, or it's a PIO operation. We
	 * fallback to a low path in this case. Hopefully, internal devices
	 * which are ioremap'ed early should use in_XX/out_XX functions
	 * instead of the PCI ones and thus not suffer from the slowdown.
	 *
	 * Also note that currently, the workaround will not work for areas
	 * that are not mapped with PTEs (bolted in the hash table). This
	 * is the case for ioremaps done very early at boot (before
	 * mem_init_done) and includes the mapping of the ISA IO space.
	 *
	 * Fortunately, none of the affected devices is expected to do DMA
	 * and thus there should be no problem in practice.
	 *
	 * In order to improve performances, we only do the PTE search for
	 * addresses falling in the PHB IO space area. That means it will
	 * not work for hotplug'ed PHBs but those don't exist with Spider.
	 */
	if (token && token <= spider_pci_count)
		bus = &spider_pci_busses[token - 1];
	else {
		unsigned long vaddr, paddr;
		pte_t *ptep;

		/* Fixup physical address */
		vaddr = (unsigned long)PCI_FIX_ADDR(addr);

		/* Check if it's in allowed range for  PIO */
		if (vaddr < PHBS_IO_BASE || vaddr >= IMALLOC_BASE)
			return;

		/* Try to find a PTE. If not, clear the paddr, we'll do
		 * a vaddr only lookup (PIO only)
		 */
		ptep = find_linux_pte(init_mm.pgd, vaddr);
		if (ptep == NULL)
			paddr = 0;
		else
			paddr = pte_pfn(*ptep) << PAGE_SHIFT;

		bus = spider_pci_find(vaddr, paddr);
		if (bus == NULL)
			return;
	}

	/* Now do the workaround
	 */
	(void)in_be32(bus->regs + SPIDER_PCI_DUMMY_READ);
}

static u8 spider_readb(const volatile void __iomem *addr)
{
	u8 val = __do_readb(addr);
	spider_io_flush(addr);
	return val;
}

static u16 spider_readw(const volatile void __iomem *addr)
{
	u16 val = __do_readw(addr);
	spider_io_flush(addr);
	return val;
}

static u32 spider_readl(const volatile void __iomem *addr)
{
	u32 val = __do_readl(addr);
	spider_io_flush(addr);
	return val;
}

static u64 spider_readq(const volatile void __iomem *addr)
{
	u64 val = __do_readq(addr);
	spider_io_flush(addr);
	return val;
}

static u16 spider_readw_be(const volatile void __iomem *addr)
{
	u16 val = __do_readw_be(addr);
	spider_io_flush(addr);
	return val;
}

static u32 spider_readl_be(const volatile void __iomem *addr)
{
	u32 val = __do_readl_be(addr);
	spider_io_flush(addr);
	return val;
}

static u64 spider_readq_be(const volatile void __iomem *addr)
{
	u64 val = __do_readq_be(addr);
	spider_io_flush(addr);
	return val;
}

static void spider_readsb(const volatile void __iomem *addr, void *buf,
			  unsigned long count)
{
	__do_readsb(addr, buf, count);
	spider_io_flush(addr);
}

static void spider_readsw(const volatile void __iomem *addr, void *buf,
			  unsigned long count)
{
	__do_readsw(addr, buf, count);
	spider_io_flush(addr);
}

static void spider_readsl(const volatile void __iomem *addr, void *buf,
			  unsigned long count)
{
	__do_readsl(addr, buf, count);
	spider_io_flush(addr);
}

static void spider_memcpy_fromio(void *dest, const volatile void __iomem *src,
				 unsigned long n)
{
	__do_memcpy_fromio(dest, src, n);
	spider_io_flush(src);
}


static void __iomem * spider_ioremap(unsigned long addr, unsigned long size,
				     unsigned long flags)
{
	struct spider_pci_bus *bus;
	void __iomem *res = __ioremap(addr, size, flags);
	int busno;

	pr_debug("spider_ioremap(0x%lx, 0x%lx, 0x%lx) -> 0x%p\n",
		 addr, size, flags, res);

	bus = spider_pci_find(0, addr);
	if (bus != NULL) {
		busno = bus - spider_pci_busses;
		pr_debug(" found bus %d, setting token\n", busno);
		PCI_SET_ADDR_TOKEN(res, busno + 1);
	}
	pr_debug(" result=0x%p\n", res);

	return res;
}

static void __init spider_pci_setup_chip(struct spider_pci_bus *bus)
{
#ifdef SPIDER_DISABLE_PREFETCH
	u32 val = in_be32(bus->regs + SPIDER_PCI_VCI_CNTL_STAT);
	pr_debug(" PVCI_Control_Status was 0x%08x\n", val);
	out_be32(bus->regs + SPIDER_PCI_VCI_CNTL_STAT, val | 0x8);
#endif

	/* Configure the dummy address for the workaround */
	out_be32(bus->regs + SPIDER_PCI_DUMMY_READ_BASE, 0x80000000);
}

static void __init spider_pci_add_one(struct pci_controller *phb)
{
	struct spider_pci_bus *bus = &spider_pci_busses[spider_pci_count];
	struct device_node *np = phb->arch_data;
	struct resource rsrc;
	void __iomem *regs;

	if (spider_pci_count >= MAX_SPIDERS) {
		printk(KERN_ERR "Too many spider bridges, workarounds"
		       " disabled for %s\n", np->full_name);
		return;
	}

	/* Get the registers for the beast */
	if (of_address_to_resource(np, 0, &rsrc)) {
		printk(KERN_ERR "Failed to get registers for spider %s"
		       " workarounds disabled\n", np->full_name);
		return;
	}

	/* Mask out some useless bits in there to get to the base of the
	 * spider chip
	 */
	rsrc.start &= ~0xfffffffful;

	/* Map them */
	regs = ioremap(rsrc.start + SPIDER_PCI_REG_BASE, 0x1000);
	if (regs == NULL) {
		printk(KERN_ERR "Failed to map registers for spider %s"
		       " workarounds disabled\n", np->full_name);
		return;
	}

	spider_pci_count++;

	/* We assume spiders only have one MMIO resource */
	bus->mmio_start = phb->mem_resources[0].start;
	bus->mmio_end = phb->mem_resources[0].end + 1;

	bus->pio_vstart = (unsigned long)phb->io_base_virt;
	bus->pio_vend = bus->pio_vstart + phb->pci_io_size;

	bus->regs = regs;

	printk(KERN_INFO "PCI: Spider MMIO workaround for %s\n",np->full_name);

	pr_debug(" mmio (P) = 0x%016lx..0x%016lx\n",
		 bus->mmio_start, bus->mmio_end);
	pr_debug("  pio (V) = 0x%016lx..0x%016lx\n",
		 bus->pio_vstart, bus->pio_vend);
	pr_debug(" regs (P) = 0x%016lx (V) = 0x%p\n",
		 rsrc.start + SPIDER_PCI_REG_BASE, bus->regs);

	spider_pci_setup_chip(bus);
}

static struct ppc_pci_io __initdata spider_pci_io = {
	.readb = spider_readb,
	.readw = spider_readw,
	.readl = spider_readl,
	.readq = spider_readq,
	.readw_be = spider_readw_be,
	.readl_be = spider_readl_be,
	.readq_be = spider_readq_be,
	.readsb = spider_readsb,
	.readsw = spider_readsw,
	.readsl = spider_readsl,
	.memcpy_fromio = spider_memcpy_fromio,
};

static int __init spider_pci_workaround_init(void)
{
	struct pci_controller *phb;

	if (!machine_is(cell))
		return 0;

	/* Find spider bridges. We assume they have been all probed
	 * in setup_arch(). If that was to change, we would need to
	 * update this code to cope with dynamically added busses
	 */
	list_for_each_entry(phb, &hose_list, list_node) {
		struct device_node *np = phb->arch_data;
		const char *model = of_get_property(np, "model", NULL);

		/* If no model property or name isn't exactly "pci", skip */
		if (model == NULL || strcmp(np->name, "pci"))
			continue;
		/* If model is not "Spider", skip */
		if (strcmp(model, "Spider"))
			continue;
		spider_pci_add_one(phb);
	}

	/* No Spider PCI found, exit */
	if (spider_pci_count == 0)
		return 0;

	/* Setup IO callbacks. We only setup MMIO reads. PIO reads will
	 * fallback to MMIO reads (though without a token, thus slower)
	 */
	ppc_pci_io = spider_pci_io;

	/* Setup ioremap callback */
	ppc_md.ioremap = spider_ioremap;

	return 0;
}
arch_initcall(spider_pci_workaround_init);
Commit	Line	Data
014da7ff BH	1	/*
	2	* Copyright (C) 2006 Benjamin Herrenschmidt <benh@kernel.crashing.org>
	3	* IBM, Corp.
	4	*
	5	* This program is free software; you can redistribute it and/or modify
	6	* it under the terms of the GNU General Public License version 2 as
	7	* published by the Free Software Foundation.
	8	*/
	9	#undef DEBUG
	10
	11	#include <linux/kernel.h>
	12	#include <linux/mm.h>
	13	#include <linux/pci.h>
	14	#include <asm/io.h>
	15	#include <asm/machdep.h>
	16	#include <asm/pci-bridge.h>
	17	#include <asm/ppc-pci.h>
	18
	19
	20	#define SPIDER_PCI_REG_BASE 0xd000
	21	#define SPIDER_PCI_VCI_CNTL_STAT 0x0110
	22	#define SPIDER_PCI_DUMMY_READ 0x0810
	23	#define SPIDER_PCI_DUMMY_READ_BASE 0x0814
	24
	25	/* Undefine that to re-enable bogus prefetch
	26	*
	27	* Without that workaround, the chip will do bogus prefetch past
	28	* page boundary from system memory. This setting will disable that,
	29	* though the documentation is unclear as to the consequences of doing
	30	* so, either purely performances, or possible misbehaviour... It's not
	31	* clear wether the chip can handle unaligned accesses at all without
	32	* prefetching enabled.
	33	*
	34	* For now, things appear to be behaving properly with that prefetching
	35	* disabled and IDE, possibly because IDE isn't doing any unaligned
	36	* access.
	37	*/
	38	#define SPIDER_DISABLE_PREFETCH
	39
a24e57be	40	#define MAX_SPIDERS 3
014da7ff BH	41
	42	static struct spider_pci_bus {
	43	void __iomem *regs;
	44	unsigned long mmio_start;
	45	unsigned long mmio_end;
	46	unsigned long pio_vstart;
	47	unsigned long pio_vend;
	48	} spider_pci_busses[MAX_SPIDERS];
	49	static int spider_pci_count;
	50
	51	static struct spider_pci_bus *spider_pci_find(unsigned long vaddr,
	52	unsigned long paddr)
	53	{
	54	int i;
	55
	56	for (i = 0; i < spider_pci_count; i++) {
	57	struct spider_pci_bus *bus = &spider_pci_busses[i];
	58	if (paddr && paddr >= bus->mmio_start && paddr < bus->mmio_end)
	59	return bus;
	60	if (vaddr && vaddr >= bus->pio_vstart && vaddr < bus->pio_vend)
	61	return bus;
	62	}
	63	return NULL;
	64	}
	65
	66	static void spider_io_flush(const volatile void __iomem *addr)
	67	{
	68	struct spider_pci_bus *bus;
	69	int token;
	70
	71	/* Get platform token (set by ioremap) from address */
	72	token = PCI_GET_ADDR_TOKEN(addr);
	73
	74	/* Fast path if we have a non-0 token, it indicates which bus we
	75	* are on.
	76	*
59c51591	77	* If the token is 0, that means either that the ioremap was done
014da7ff BH	78	* before we initialized this layer, or it's a PIO operation. We
	79	* fallback to a low path in this case. Hopefully, internal devices
	80	* which are ioremap'ed early should use in_XX/out_XX functions
	81	* instead of the PCI ones and thus not suffer from the slowdown.
	82	*
	83	* Also note that currently, the workaround will not work for areas
	84	* that are not mapped with PTEs (bolted in the hash table). This
	85	* is the case for ioremaps done very early at boot (before
	86	* mem_init_done) and includes the mapping of the ISA IO space.
	87	*
	88	* Fortunately, none of the affected devices is expected to do DMA
	89	* and thus there should be no problem in practice.
	90	*
	91	* In order to improve performances, we only do the PTE search for
	92	* addresses falling in the PHB IO space area. That means it will
	93	* not work for hotplug'ed PHBs but those don't exist with Spider.
	94	*/
	95	if (token && token <= spider_pci_count)
	96	bus = &spider_pci_busses[token - 1];
	97	else {
	98	unsigned long vaddr, paddr;
	99	pte_t *ptep;
	100
	101	/* Fixup physical address */
	102	vaddr = (unsigned long)PCI_FIX_ADDR(addr);
	103
	104	/* Check if it's in allowed range for PIO */
	105	if (vaddr < PHBS_IO_BASE \|\| vaddr >= IMALLOC_BASE)
	106	return;
	107
	108	/* Try to find a PTE. If not, clear the paddr, we'll do
	109	* a vaddr only lookup (PIO only)
	110	*/
	111	ptep = find_linux_pte(init_mm.pgd, vaddr);
	112	if (ptep == NULL)
	113	paddr = 0;
	114	else
	115	paddr = pte_pfn(*ptep) << PAGE_SHIFT;
	116
	117	bus = spider_pci_find(vaddr, paddr);
	118	if (bus == NULL)
	119	return;
	120	}
	121
	122	/* Now do the workaround
	123	*/
	124	(void)in_be32(bus->regs + SPIDER_PCI_DUMMY_READ);
	125	}
	126
	127	static u8 spider_readb(const volatile void __iomem *addr)
	128	{
	129	u8 val = __do_readb(addr);
	130	spider_io_flush(addr);
	131	return val;
	132	}
	133
	134	static u16 spider_readw(const volatile void __iomem *addr)
	135	{
	136	u16 val = __do_readw(addr);
	137	spider_io_flush(addr);
	138	return val;
	139	}
	140
	141	static u32 spider_readl(const volatile void __iomem *addr)
142	{
143	u32 val = __do_readl(addr);
144	spider_io_flush(addr);
145	return val;
146	}
147
148	static u64 spider_readq(const volatile void __iomem *addr)
149	{
150	u64 val = __do_readq(addr);
151	spider_io_flush(addr);
152	return val;
153	}
154
155	static u16 spider_readw_be(const volatile void __iomem *addr)
156	{
157	u16 val = __do_readw_be(addr);
158	spider_io_flush(addr);
159	return val;
160	}
161
162	static u32 spider_readl_be(const volatile void __iomem *addr)
163	{
164	u32 val = __do_readl_be(addr);
165	spider_io_flush(addr);
166	return val;
167	}
168
169	static u64 spider_readq_be(const volatile void __iomem *addr)
170	{
171	u64 val = __do_readq_be(addr);
172	spider_io_flush(addr);
173	return val;
174	}
175
176	static void spider_readsb(const volatile void __iomem addr, void buf,
177	unsigned long count)
178	{
179	__do_readsb(addr, buf, count);
180	spider_io_flush(addr);
181	}
182
183	static void spider_readsw(const volatile void __iomem addr, void buf,
184	unsigned long count)
185	{
186	__do_readsw(addr, buf, count);
187	spider_io_flush(addr);
188	}
189
190	static void spider_readsl(const volatile void __iomem addr, void buf,
191	unsigned long count)
192	{
193	__do_readsl(addr, buf, count);
194	spider_io_flush(addr);
195	}
196
197	static void spider_memcpy_fromio(void dest, const volatile void __iomem src,
198	unsigned long n)
199	{
200	__do_memcpy_fromio(dest, src, n);
201	spider_io_flush(src);
202	}
203
204
205	static void __iomem * spider_ioremap(unsigned long addr, unsigned long size,
206	unsigned long flags)
207	{
208	struct spider_pci_bus *bus;
209	void __iomem *res = __ioremap(addr, size, flags);
210	int busno;
211
212	pr_debug("spider_ioremap(0x%lx, 0x%lx, 0x%lx) -> 0x%p\n",
213	addr, size, flags, res);
214
215	bus = spider_pci_find(0, addr);
216	if (bus != NULL) {
217	busno = bus - spider_pci_busses;
218	pr_debug(" found bus %d, setting token\n", busno);
219	PCI_SET_ADDR_TOKEN(res, busno + 1);
220	}
221	pr_debug(" result=0x%p\n", res);
222
223	return res;
224	}
225
226	static void __init spider_pci_setup_chip(struct spider_pci_bus *bus)
227	{
228	#ifdef SPIDER_DISABLE_PREFETCH
229	u32 val = in_be32(bus->regs + SPIDER_PCI_VCI_CNTL_STAT);
230	pr_debug(" PVCI_Control_Status was 0x%08x\n", val);
231	out_be32(bus->regs + SPIDER_PCI_VCI_CNTL_STAT, val \| 0x8);
232	#endif
233
234	/* Configure the dummy address for the workaround */
235	out_be32(bus->regs + SPIDER_PCI_DUMMY_READ_BASE, 0x80000000);
236	}
237
238	static void __init spider_pci_add_one(struct pci_controller *phb)
239	{
240	struct spider_pci_bus *bus = &spider_pci_busses[spider_pci_count];
241	struct device_node *np = phb->arch_data;
242	struct resource rsrc;
243	void __iomem *regs;
244
245	if (spider_pci_count >= MAX_SPIDERS) {
246	printk(KERN_ERR "Too many spider bridges, workarounds"
247	" disabled for %s\n", np->full_name);
248	return;
249	}
250
251	/* Get the registers for the beast */
252	if (of_address_to_resource(np, 0, &rsrc)) {
253	printk(KERN_ERR "Failed to get registers for spider %s"
254	" workarounds disabled\n", np->full_name);
255	return;
256	}
257
258	/* Mask out some useless bits in there to get to the base of the
259	* spider chip
260	*/
261	rsrc.start &= ~0xfffffffful;
262
263	/* Map them */
264	regs = ioremap(rsrc.start + SPIDER_PCI_REG_BASE, 0x1000);
265	if (regs == NULL) {
266	printk(KERN_ERR "Failed to map registers for spider %s"
267	" workarounds disabled\n", np->full_name);
268	return;
269	}
270
271	spider_pci_count++;
272
273	/* We assume spiders only have one MMIO resource */
274	bus->mmio_start = phb->mem_resources[0].start;
275	bus->mmio_end = phb->mem_resources[0].end + 1;
276
277	bus->pio_vstart = (unsigned long)phb->io_base_virt;
278	bus->pio_vend = bus->pio_vstart + phb->pci_io_size;
279
280	bus->regs = regs;
281
282	printk(KERN_INFO "PCI: Spider MMIO workaround for %s\n",np->full_name);
283
284	pr_debug(" mmio (P) = 0x%016lx..0x%016lx\n",
285	bus->mmio_start, bus->mmio_end);
286	pr_debug(" pio (V) = 0x%016lx..0x%016lx\n",
287	bus->pio_vstart, bus->pio_vend);
288	pr_debug(" regs (P) = 0x%016lx (V) = 0x%p\n",
289	rsrc.start + SPIDER_PCI_REG_BASE, bus->regs);
290
291	spider_pci_setup_chip(bus);
292	}
293
294	static struct ppc_pci_io __initdata spider_pci_io = {
295	.readb = spider_readb,
296	.readw = spider_readw,
297	.readl = spider_readl,
298	.readq = spider_readq,
299	.readw_be = spider_readw_be,
300	.readl_be = spider_readl_be,
301	.readq_be = spider_readq_be,
302	.readsb = spider_readsb,
303	.readsw = spider_readsw,
304	.readsl = spider_readsl,
305	.memcpy_fromio = spider_memcpy_fromio,
306	};
307
308	static int __init spider_pci_workaround_init(void)
309	{
310	struct pci_controller *phb;
311
312	if (!machine_is(cell))
313	return 0;
314
315	/* Find spider bridges. We assume they have been all probed
316	* in setup_arch(). If that was to change, we would need to
317	* update this code to cope with dynamically added busses
318	*/
319	list_for_each_entry(phb, &hose_list, list_node) {
320	struct device_node *np = phb->arch_data;
e2eb6392	321	const char *model = of_get_property(np, "model", NULL);
014da7ff BH	322
	323	/* If no model property or name isn't exactly "pci", skip */
	324	if (model == NULL \|\| strcmp(np->name, "pci"))
	325	continue;
	326	/* If model is not "Spider", skip */
	327	if (strcmp(model, "Spider"))
	328	continue;
	329	spider_pci_add_one(phb);
	330	}
	331
	332	/* No Spider PCI found, exit */
	333	if (spider_pci_count == 0)
	334	return 0;
	335
	336	/* Setup IO callbacks. We only setup MMIO reads. PIO reads will
	337	* fallback to MMIO reads (though without a token, thus slower)
	338	*/
	339	ppc_pci_io = spider_pci_io;
	340
	341	/* Setup ioremap callback */
	342	ppc_md.ioremap = spider_ioremap;
	343
	344	return 0;
	345	}
	346	arch_initcall(spider_pci_workaround_init);