sym53c8xx.c

讲述linux的初始化过程

**
**	This wrapper allows to get rid of old kernel PCI interface 
**	and still allows to preserve linux-2.0 compatibilty.
**	In fact, it is mostly an incomplete emulation of the new 
**	PCI code for pre-2.2 kernels. When kernel-2.0 support 
**	will be dropped, we will just have to remove most of this 
**	code.
*/

#if LINUX_VERSION_CODE >= LinuxVersionCode(2,2,0)

typedef struct pci_dev *pcidev_t;
#define PCIDEV_NULL		(0)
#define PciBusNumber(d)		(d)->bus->number
#define PciDeviceFn(d)		(d)->devfn
#define PciVendorId(d)		(d)->vendor
#define PciDeviceId(d)		(d)->device
#define PciIrqLine(d)		(d)->irq

#if LINUX_VERSION_CODE > LinuxVersionCode(2,3,12)

static int __init 
pci_get_base_address(struct pci_dev *pdev, int index, u_long *base)
{
	*base = pdev->resource[index].start;
	if ((pdev->resource[index].flags & 0x7) == 0x4)
		++index;
	return ++index;
}
#else
static int __init 
pci_get_base_address(struct pci_dev *pdev, int index, u_long *base)
{
	*base = pdev->base_address[index++];
	if ((*base & 0x7) == 0x4) {
#if BITS_PER_LONG > 32
		*base |= (((u_long)pdev->base_address[index]) << 32);
#endif
		++index;
	}
	return index;
}
#endif

#else	/* Incomplete emulation of current PCI code for pre-2.2 kernels */

typedef unsigned int pcidev_t;
#define PCIDEV_NULL		(~0u)
#define PciBusNumber(d)		((d)>>8)
#define PciDeviceFn(d)		((d)&0xff)
#define __PciDev(busn, devfn)	(((busn)<<8)+(devfn))

#define pci_present pcibios_present

#define pci_read_config_byte(d, w, v) \
	pcibios_read_config_byte(PciBusNumber(d), PciDeviceFn(d), w, v)
#define pci_read_config_word(d, w, v) \
	pcibios_read_config_word(PciBusNumber(d), PciDeviceFn(d), w, v)
#define pci_read_config_dword(d, w, v) \
	pcibios_read_config_dword(PciBusNumber(d), PciDeviceFn(d), w, v)

#define pci_write_config_byte(d, w, v) \
	pcibios_write_config_byte(PciBusNumber(d), PciDeviceFn(d), w, v)
#define pci_write_config_word(d, w, v) \
	pcibios_write_config_word(PciBusNumber(d), PciDeviceFn(d), w, v)
#define pci_write_config_dword(d, w, v) \
	pcibios_write_config_dword(PciBusNumber(d), PciDeviceFn(d), w, v)

static pcidev_t __init
pci_find_device(unsigned int vendor, unsigned int device, pcidev_t prev)
{
	static unsigned short pci_index;
	int retv;
	unsigned char bus_number, device_fn;

	if (prev == PCIDEV_NULL)
		pci_index = 0;
	else
		++pci_index;
	retv = pcibios_find_device (vendor, device, pci_index,
				    &bus_number, &device_fn);
	return retv ? PCIDEV_NULL : __PciDev(bus_number, device_fn);
}

static u_short __init PciVendorId(pcidev_t dev)
{
	u_short vendor_id;
	pci_read_config_word(dev, PCI_VENDOR_ID, &vendor_id);
	return vendor_id;
}

static u_short __init PciDeviceId(pcidev_t dev)
{
	u_short device_id;
	pci_read_config_word(dev, PCI_DEVICE_ID, &device_id);
	return device_id;
}

static u_int __init PciIrqLine(pcidev_t dev)
{
	u_char irq;
	pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
	return irq;
}

static int __init 
pci_get_base_address(pcidev_t dev, int offset, u_long *base)
{
	u_int32 tmp;
	
	pci_read_config_dword(dev, PCI_BASE_ADDRESS_0 + offset, &tmp);
	*base = tmp;
	offset += sizeof(u_int32);
	if ((tmp & 0x7) == 0x4) {
#if BITS_PER_LONG > 32
		pci_read_config_dword(dev, PCI_BASE_ADDRESS_0 + offset, &tmp);
		*base |= (((u_long)tmp) << 32);
#endif
		offset += sizeof(u_int32);
	}
	return offset;
}

#endif	/* LINUX_VERSION_CODE >= LinuxVersionCode(2,2,0) */

/*==========================================================
**
**	Debugging tags
**
**==========================================================
*/

#define DEBUG_ALLOC    (0x0001)
#define DEBUG_PHASE    (0x0002)
#define DEBUG_QUEUE    (0x0008)
#define DEBUG_RESULT   (0x0010)
#define DEBUG_POINTER  (0x0020)
#define DEBUG_SCRIPT   (0x0040)
#define DEBUG_TINY     (0x0080)
#define DEBUG_TIMING   (0x0100)
#define DEBUG_NEGO     (0x0200)
#define DEBUG_TAGS     (0x0400)
#define DEBUG_IC       (0x0800)

/*
**    Enable/Disable debug messages.
**    Can be changed at runtime too.
*/

#ifdef SCSI_NCR_DEBUG_INFO_SUPPORT
static int ncr_debug = SCSI_NCR_DEBUG_FLAGS;
	#define DEBUG_FLAGS ncr_debug
#else
	#define DEBUG_FLAGS	SCSI_NCR_DEBUG_FLAGS
#endif

/*
**	SMP threading.
**
**	Assuming that SMP systems are generally high end systems and may 
**	use several SCSI adapters, we are using one lock per controller 
**	instead of some global one. For the moment (linux-2.1.95), driver's 
**	entry points are called with the 'io_request_lock' lock held, so:
**	- We are uselessly loosing a couple of micro-seconds to lock the 
**	  controller data structure.
**	- But the driver is not broken by design for SMP and so can be 
**	  more resistant to bugs or bad changes in the IO sub-system code.
**	- A small advantage could be that the interrupt code is grained as 
**	  wished (e.g.: threaded by controller).
*/

#if LINUX_VERSION_CODE >= LinuxVersionCode(2,1,93)

spinlock_t sym53c8xx_lock = SPIN_LOCK_UNLOCKED;
#define	NCR_LOCK_DRIVER(flags)     spin_lock_irqsave(&sym53c8xx_lock, flags)
#define	NCR_UNLOCK_DRIVER(flags)   spin_unlock_irqrestore(&sym53c8xx_lock,flags)

#define NCR_INIT_LOCK_NCB(np)      spin_lock_init(&np->smp_lock);
#define	NCR_LOCK_NCB(np, flags)    spin_lock_irqsave(&np->smp_lock, flags)
#define	NCR_UNLOCK_NCB(np, flags)  spin_unlock_irqrestore(&np->smp_lock, flags)

#define	NCR_LOCK_SCSI_DONE(np, flags) \
		spin_lock_irqsave(&io_request_lock, flags)
#define	NCR_UNLOCK_SCSI_DONE(np, flags) \
		spin_unlock_irqrestore(&io_request_lock, flags)

#else

#define	NCR_LOCK_DRIVER(flags)     do { save_flags(flags); cli(); } while (0)
#define	NCR_UNLOCK_DRIVER(flags)   do { restore_flags(flags); } while (0)

#define	NCR_INIT_LOCK_NCB(np)      do { } while (0)
#define	NCR_LOCK_NCB(np, flags)    do { save_flags(flags); cli(); } while (0)
#define	NCR_UNLOCK_NCB(np, flags)  do { restore_flags(flags); } while (0)

#define	NCR_LOCK_SCSI_DONE(np, flags)    do {;} while (0)
#define	NCR_UNLOCK_SCSI_DONE(np, flags)  do {;} while (0)

#endif

/*
**	Memory mapped IO
**
**	Since linux-2.1, we must use ioremap() to map the io memory space.
**	iounmap() to unmap it. That allows portability.
**	Linux 1.3.X and 2.0.X allow to remap physical pages addresses greater 
**	than the highest physical memory address to kernel virtual pages with 
**	vremap() / vfree(). That was not portable but worked with i386 
**	architecture.
*/

#if LINUX_VERSION_CODE < LinuxVersionCode(2,1,0)
#define ioremap vremap
#define iounmap vfree
#endif

#ifdef __sparc__
#  include <asm/irq.h>
#  define pcivtobus(p)			bus_dvma_to_mem(p)
#  define memcpy_to_pci(a, b, c)	memcpy_toio((a), (b), (c))
#elif defined(__alpha__)
#  define pcivtobus(p)			((p) & 0xfffffffful)
#  define memcpy_to_pci(a, b, c)	memcpy_toio((a), (b), (c))
#elif defined(CONFIG_PPC)
#  define pcivtobus(p)			phys_to_bus(p)
#  define memcpy_to_pci(a, b, c)	memcpy_toio((a), (b), (c))
#else	/* others */
#  define pcivtobus(p)			(p)
#  define memcpy_to_pci(a, b, c)	memcpy_toio((a), (b), (c))
#endif

#ifndef SCSI_NCR_PCI_MEM_NOT_SUPPORTED
static u_long __init remap_pci_mem(u_long base, u_long size)
{
	u_long page_base	= ((u_long) base) & PAGE_MASK;
	u_long page_offs	= ((u_long) base) - page_base;
	u_long page_remapped	= (u_long) ioremap(page_base, page_offs+size);

	return page_remapped? (page_remapped + page_offs) : 0UL;
}

static void __init unmap_pci_mem(u_long vaddr, u_long size)
{
	if (vaddr)
		iounmap((void *) (vaddr & PAGE_MASK));
}

#endif /* not def SCSI_NCR_PCI_MEM_NOT_SUPPORTED */

/*
**	Insert a delay in micro-seconds and milli-seconds.
**	-------------------------------------------------
**	Under Linux, udelay() is restricted to delay < 1 milli-second.
**	In fact, it generally works for up to 1 second delay.
**	Since 2.1.105, the mdelay() function is provided for delays 
**	in milli-seconds.
**	Under 2.0 kernels, udelay() is an inline function that is very 
**	inaccurate on Pentium processors.
*/

#if LINUX_VERSION_CODE >= LinuxVersionCode(2,1,105)
#define UDELAY udelay
#define MDELAY mdelay
#else
static void UDELAY(long us) { udelay(us); }
static void MDELAY(long ms) { while (ms--) UDELAY(1000); }
#endif

/*
**	Simple power of two buddy-like allocator
**	----------------------------------------
**	This simple code is not intended to be fast, but to provide 
**	power of 2 aligned memory allocations.
**	Since the SCRIPTS processor only supplies 8 bit arithmetic,
**	this allocator allows simple and fast address calculations  
**	from the SCRIPTS code. In addition, cache line alignment 
**	is guaranteed for power of 2 cache line size.
**	Enhanced in linux-2.3.44 to provide a memory pool per pcidev 
**	to support dynamic dma mapping. (I would have preferred a 
**	real bus astraction, btw).
*/

#if LINUX_VERSION_CODE >= LinuxVersionCode(2,1,0)
#define __GetFreePages(flags, order) __get_free_pages(flags, order)
#else
#define __GetFreePages(flags, order) __get_free_pages(flags, order, 0)
#endif

#define MEMO_SHIFT	4	/* 16 bytes minimum memory chunk */
#if PAGE_SIZE >= 8192
#define MEMO_PAGE_ORDER	0	/* 1 PAGE  maximum */
#else
#define MEMO_PAGE_ORDER	1	/* 2 PAGES maximum */
#endif
#define MEMO_FREE_UNUSED	/* Free unused pages immediately */
#define MEMO_WARN	1
#define MEMO_GFP_FLAGS	GFP_ATOMIC
#define MEMO_CLUSTER_SHIFT	(PAGE_SHIFT+MEMO_PAGE_ORDER)
#define MEMO_CLUSTER_SIZE	(1UL << MEMO_CLUSTER_SHIFT)
#define MEMO_CLUSTER_MASK	(MEMO_CLUSTER_SIZE-1)

typedef u_long m_addr_t;	/* Enough bits to bit-hack addresses */
typedef pcidev_t m_bush_t;	/* Something that addresses DMAable */

typedef struct m_link {		/* Link between free memory chunks */
	struct m_link *next;
} m_link_s;

#ifdef	SCSI_NCR_DYNAMIC_DMA_MAPPING
typedef struct m_vtob {		/* Virtual to Bus address translation */
	struct m_vtob *next;
	m_addr_t vaddr;
	m_addr_t baddr;
} m_vtob_s;
#define VTOB_HASH_SHIFT		5
#define VTOB_HASH_SIZE		(1UL << VTOB_HASH_SHIFT)
#define VTOB_HASH_MASK		(VTOB_HASH_SIZE-1)
#define VTOB_HASH_CODE(m)	\
	((((m_addr_t) (m)) >> MEMO_CLUSTER_SHIFT) & VTOB_HASH_MASK)
#endif

typedef struct m_pool {		/* Memory pool of a given kind */
#ifdef	SCSI_NCR_DYNAMIC_DMA_MAPPING
	m_bush_t bush;
	m_addr_t (*getp)(struct m_pool *);
	void (*freep)(struct m_pool *, m_addr_t);
#define M_GETP()		mp->getp(mp)
#define M_FREEP(p)		mp->freep(mp, p)
#define GetPages()		__GetFreePages(MEMO_GFP_FLAGS, MEMO_PAGE_ORDER)
#define FreePages(p)		free_pages(p, MEMO_PAGE_ORDER)
	int nump;
	m_vtob_s *(vtob[VTOB_HASH_SIZE]);
	struct m_pool *next;
#else
#define M_GETP()		__GetFreePages(MEMO_GFP_FLAGS, MEMO_PAGE_ORDER)
#define M_FREEP(p)		free_pages(p, MEMO_PAGE_ORDER)
#endif	/* SCSI_NCR_DYNAMIC_DMA_MAPPING */
	struct m_link h[PAGE_SHIFT-MEMO_SHIFT+MEMO_PAGE_ORDER+1];
} m_pool_s;

static void *___m_alloc(m_pool_s *mp, int size)
{
	int i = 0;
	int s = (1 << MEMO_SHIFT);
	int j;
	m_addr_t a;
	m_link_s *h = mp->h;

	if (size > (PAGE_SIZE << MEMO_PAGE_ORDER))
		return 0;

	while (size > s) {
		s <<= 1;
		++i;
	}

	j = i;
	while (!h[j].next) {
		if (s == (PAGE_SIZE << MEMO_PAGE_ORDER)) {
			h[j].next = (m_link_s *) M_GETP();
			if (h[j].next)
				h[j].next->next = 0;
			break;
		}
		++j;
		s <<= 1;
	}
	a = (m_addr_t) h[j].next;
	if (a) {
		h[j].next = h[j].next->next;
		while (j > i) {
			j -= 1;
			s >>= 1;
			h[j].next = (m_link_s *) (a+s);
			h[j].next->next = 0;
		}
	}
#ifdef DEBUG
	printk("___m_alloc(%d) = %p\n", size, (void *) a);
#endif
	return (void *) a;
}

static void ___m_free(m_pool_s *mp, void *ptr, int size)
{
	int i = 0;
	int s = (1 << MEMO_SHIFT);
	m_link_s *q;
	m_addr_t a, b;
	m_link_s *h = mp->h;

#ifdef DEBUG
	printk("___m_free(%p, %d)\n", ptr, size);
#endif

	if (size > (PAGE_SIZE << MEMO_PAGE_ORDER))
		return;

	while (size > s) {
		s <<= 1;
		++i;
	}

	a = (m_addr_t) ptr;

	while (1) {
#ifdef MEMO_FREE_UNUSED
		if (s == (PAGE_SIZE << MEMO_PAGE_ORDER)) {
			M_FREEP(a);
			break;
		}
#endif
		b = a ^ s;
		q = &h[i];
		while (q->next && q->next != (m_link_s *) b) {
			q = q->next;
		}
		if (!q->next) {
			((m_link_s *) a)->next = h[i].next;
			h[i].next = (m_link_s *) a;
			break;
		}
		q->next = q->next->next;
		a = a & b;
		s <<= 1;
		++i;
	}
}

static void *__m_calloc2(m_pool_s *mp, int size, char *name, int uflags)
{
	void *p;

	p = ___m_alloc(mp, size);

	if (DEBUG_FLAGS & DEBUG_ALLOC)
		printk ("new %-10s[%4d] @%p.\n", name, size, p);

	if (p)
		bzero(p, size);
	else if (uflags & MEMO_WARN)
		printk (NAME53C8XX ": failed to allocate %s[%d]\n", name, size);

	return p;
}

#define __m_calloc(mp, s, n)	__m_calloc2(mp, s, n, MEMO_WARN)

static void __m_free(m_pool_s *mp, void *ptr, int size, char *name)
{
	if (DEBUG_FLAGS & DEBUG_ALLOC)
		printk ("freeing %-10s[%4d] @%p.\n", name, size, ptr);

	___m_free(mp, ptr, size);

}

/*
 * With pci bus iommu support, we use a default pool of unmapped memory 
 * for memory we donnot need to DMA from/to and one pool per pcidev for 
 * memory accessed by the PCI chip. `mp0' is the default not DMAable pool.
 */

#ifndef	SCSI_NCR_DYNAMIC_DMA_MAPPING

static m_pool_s mp0;

#else

static m_addr_t ___mp0_getp(m_pool_s *mp)
{
	m_addr_t m = GetPages();
	if (m)
		++mp->nump;
	return m;
}

static void ___mp0_freep(m_pool_s *mp, m_addr_t m)