Merge pull request #77 from weety/at91sam9260

Fixed At91sam9260 context switch issue, enable mmu, update SDIO and EMAC drivers

bsp/at91sam9260/at91_mci.c

@@ -27,6 +27,12 @@
 #define mci_dbg(fmt, ...)
 #endif
 
+#define MMU_NOCACHE_ADDR(a)      	((rt_uint32_t)a | (1UL<<31))
+
+extern void mmu_clean_dcache(rt_uint32_t buffer, rt_uint32_t size);
+extern void mmu_invalidate_dcache(rt_uint32_t buffer, rt_uint32_t size);
+
+
 #define AT91_MCI_ERRORS	(AT91_MCI_RINDE | AT91_MCI_RDIRE | AT91_MCI_RCRCE	\
 		| AT91_MCI_RENDE | AT91_MCI_RTOE | AT91_MCI_DCRCE		\
 		| AT91_MCI_DTOE | AT91_MCI_OVRE | AT91_MCI_UNRE)
@@ -334,6 +340,7 @@ static void at91_mci_send_command(struct at91_mci *mci, struct rt_mmcsd_cmd *cmd
 				 * Handle a read
 				 */
 
+				mmu_invalidate_dcache((rt_uint32_t)data->buf, data->blksize*data->blks);
 				at91_mci_init_dma_read(mci);
 				ier = AT91_MCI_ENDRX /* | AT91_MCI_RXBUFF */;
 			}
@@ -359,13 +366,15 @@ static void at91_mci_send_command(struct at91_mci *mci, struct rt_mmcsd_cmd *cmd
 						return;
 					}
 					rt_memset(mci->buf, 0, 12);
-					rt_memcpy(mci->buf, data->buf, block_length * blocks);
+					rt_memcpy(mci->buf, data->buf, length);
+					mmu_clean_dcache((rt_uint32_t)mci->buf, length);
 					at91_mci_write(AT91_PDC_TPR, (rt_uint32_t)(mci->buf));
 					at91_mci_write(AT91_PDC_TCR, (data->blksize & 0x3) ?
 							length : length / 4);
 				}
 				else
 				{
+					mmu_clean_dcache((rt_uint32_t)data->buf, data->blksize*data->blks);
 					at91_mci_write(AT91_PDC_TPR, (rt_uint32_t)(data->buf));
 					at91_mci_write(AT91_PDC_TCR, (data->blksize & 0x3) ?
 							length : length / 4);
@@ -839,7 +848,8 @@ rt_int32_t at91_mci_init(void)
 	host->freq_min = 375000;
 	host->freq_max = 25000000;
 	host->valid_ocr = VDD_32_33 | VDD_33_34;
-	host->flags = MMCSD_BUSWIDTH_4 | MMCSD_MUTBLKWRITE;
+	host->flags = MMCSD_BUSWIDTH_4 | MMCSD_MUTBLKWRITE | \
+				MMCSD_SUP_HIGHSPEED | MMCSD_SUP_SDIO_IRQ;
 	host->max_seg_size = 65535;
 	host->max_dma_segs = 2;
 	host->max_blk_size = 512;

bsp/at91sam9260/board.c

@@ -341,7 +341,7 @@ void rt_hw_board_init()
 	rt_hw_uart_init();
 
 	/* initialize mmu */
-	//rt_hw_mmu_init();
+	rt_hw_mmu_init();
 
 	/* initialize timer0 */
 	rt_hw_timer_init();

bsp/at91sam9260/macb.c

@@ -18,6 +18,26 @@
 #include <at91sam926x.h>
 #include "macb.h"
 
+#define MMU_NOCACHE_ADDR(a)      	((rt_uint32_t)a | (1UL<<31))
+
+extern void mmu_clean_dcache(rt_uint32_t buffer, rt_uint32_t size);
+extern void mmu_invalidate_dcache(rt_uint32_t buffer, rt_uint32_t size);
+
+/* Cache macros - Packet buffers would be from pbuf pool which is cached */
+#define EMAC_VIRT_NOCACHE(addr) (addr)
+#define EMAC_CACHE_INVALIDATE(addr, size) \
+	mmu_invalidate_dcache((rt_uint32_t)addr, size)
+#define EMAC_CACHE_WRITEBACK(addr, size) \
+	mmu_clean_dcache((rt_uint32_t)addr, size)
+#define EMAC_CACHE_WRITEBACK_INVALIDATE(addr, size) \
+	mmu_clean_invalidated_dcache((rt_uint32_t)addr, size)
+
+/* EMAC has BD's in cached memory - so need cache functions */
+#define BD_CACHE_INVALIDATE(addr, size)
+#define BD_CACHE_WRITEBACK(addr, size)
+#define BD_CACHE_WRITEBACK_INVALIDATE(addr, size)
+
+
 #define CONFIG_RMII
 
 #define MACB_RX_BUFFER_SIZE		4096*4
@@ -378,6 +398,8 @@ static rt_err_t rt_macb_init(rt_device_t dev)
 	}
 	macb->rx_tail = macb->tx_head = macb->tx_tail = 0;
 
+	BD_CACHE_WRITEBACK_INVALIDATE(macb->rx_ring, MACB_RX_RING_SIZE * sizeof(struct macb_dma_desc));
+	BD_CACHE_WRITEBACK_INVALIDATE(macb->tx_ring, MACB_TX_RING_SIZE * sizeof(struct macb_dma_desc));
 	macb_writel(macb, RBQP, macb->rx_ring_dma);
 	macb_writel(macb, TBQP, macb->tx_ring_dma);
 	
@@ -470,28 +492,12 @@ static rt_err_t rt_macb_control(rt_device_t dev, rt_uint8_t cmd, void *args)
 /* transmit packet. */
 rt_err_t rt_macb_tx( rt_device_t dev, struct pbuf* p)
 {
-	struct pbuf* q;
-	rt_uint8_t* bufptr, *buf = RT_NULL;
 	unsigned long ctrl;
 	struct rt_macb_eth *macb = dev->user_data;
 	unsigned int tx_head = macb->tx_head;
-	
+
-	/* lock macb device */
 	rt_sem_take(&sem_lock, RT_WAITING_FOREVER);
-	buf = rt_malloc(p->tot_len);
+	EMAC_CACHE_WRITEBACK(p->payload, p->tot_len);
-	if (!buf) {
-		rt_kprintf("%s:alloc buf failed\n", __func__);
-		return -RT_ENOMEM;
-	}
-	bufptr = buf;
-
-
-	for (q = p; q != NULL; q = q->next)
-	{
-		memcpy(bufptr, q->payload, q->len);
-		bufptr += q->len;
-	}
-
 	ctrl = p->tot_len & TXBUF_FRMLEN_MASK;
 	ctrl |= TXBUF_FRAME_END;
 	if (tx_head == (MACB_TX_RING_SIZE - 1)) {
@@ -499,17 +505,14 @@ rt_err_t rt_macb_tx( rt_device_t dev, struct pbuf* p)
 		macb->tx_head = 0;
 	} else
 		macb->tx_head++;
-
 	macb->tx_ring[tx_head].ctrl = ctrl;
-	macb->tx_ring[tx_head].addr = (rt_uint32_t)buf;
+	macb->tx_ring[tx_head].addr = (rt_uint32_t)p->payload;
+	BD_CACHE_WRITEBACK_INVALIDATE(macb->tx_ring[tx_head], sizeof(struct macb_dma_desc));
 	macb_writel(macb, NCR, MACB_BIT(TE) | MACB_BIT(RE) | MACB_BIT(TSTART));
-	
-	/* unlock macb device */
-	rt_sem_release(&sem_lock);
 
+	rt_sem_release(&sem_lock);
 	/* wait ack */
 	rt_sem_take(&sem_ack, RT_WAITING_FOREVER);
-	rt_free(buf);
 
 	return RT_EOK;
 }
@@ -546,12 +549,7 @@ struct pbuf *rt_macb_rx(rt_device_t dev)
 	int wrapped = 0;
 	rt_uint32_t status;
 
-	struct pbuf* q;
-	rt_uint8_t *buf = RT_NULL;
-	
-	/* lock macb device */
 	rt_sem_take(&sem_lock, RT_WAITING_FOREVER);
-	
 	for (;;) {
 		if (!(macb->rx_ring[rx_tail].addr & RXADDR_USED))
 			break;
@@ -574,39 +572,18 @@ struct pbuf *rt_macb_rx(rt_device_t dev)
 			}
 			if (wrapped) {
 				unsigned int headlen, taillen;
-				buf = rt_malloc(len);
-				if (!buf)
-				{
-					rt_kprintf("%s:alloc memory failed\n", __func__);
-					pbuf_free(p);
-					p = RT_NULL;
-					break;
-				}
 
 				headlen = 128 * (MACB_RX_RING_SIZE
 						 - macb->rx_tail);
 				taillen = len - headlen;
-				memcpy((void *)buf,
+				EMAC_CACHE_INVALIDATE(buffer, headlen);
-				       buffer, headlen);
+				EMAC_CACHE_INVALIDATE(macb->rx_buffer, taillen);
-				memcpy((void *)((unsigned int)buf + headlen),
+				memcpy((void *)p->payload, buffer, headlen);
+				memcpy((void *)((unsigned int)p->payload + headlen),
 				       macb->rx_buffer, taillen);
-				buffer = (void *)buf;
-				for (q = p; q != RT_NULL; q= q->next)
-				{
-					/* read data from device */
-					memcpy((void *)q->payload, buffer, q->len);
-					buffer = (void *)((unsigned int)buffer + q->len);
-				}
-				rt_free(buf);
-				buf = RT_NULL;
 			} else {
-				for (q = p; q != RT_NULL; q= q->next)
+				EMAC_CACHE_INVALIDATE(buffer, len);
-				{
+				memcpy((void *)p->payload, buffer, p->len);
-					/* read data from device */
-					memcpy((void *)q->payload, buffer, q->len);
-					buffer = (void *)((unsigned int)buffer + q->len);
-				}
-			
 			}
 			
 			if (++rx_tail >= MACB_RX_RING_SIZE)
@@ -620,7 +597,7 @@ struct pbuf *rt_macb_rx(rt_device_t dev)
 			}
 		}
 	}
-	/* unlock macb device */
+
 	rt_sem_release(&sem_lock);
 
 	return p;
@@ -647,11 +624,17 @@ void macb_initialize()
 	macb->rx_buffer = rt_malloc(MACB_RX_BUFFER_SIZE);
 	macb->rx_ring = rt_malloc(MACB_RX_RING_SIZE * sizeof(struct macb_dma_desc));
 	macb->tx_ring = rt_malloc(MACB_TX_RING_SIZE * sizeof(struct macb_dma_desc));
-	
+
+	EMAC_CACHE_INVALIDATE(macb->rx_ring, MACB_RX_RING_SIZE * sizeof(struct macb_dma_desc));
+	EMAC_CACHE_INVALIDATE(macb->tx_ring, MACB_TX_RING_SIZE * sizeof(struct macb_dma_desc));
+
 	macb->rx_buffer_dma = (unsigned long)macb->rx_buffer;
 	macb->rx_ring_dma = (unsigned long)macb->rx_ring;
 	macb->tx_ring_dma = (unsigned long)macb->tx_ring;
 
+	macb->rx_ring = MMU_NOCACHE_ADDR(macb->rx_ring);
+	macb->tx_ring = MMU_NOCACHE_ADDR(macb->tx_ring);
+
 	macb->regs = AT91SAM9260_BASE_EMAC;
 	macb->phy_addr = 0x00;
 	

bsp/at91sam9260/rtconfig.h

@@ -141,8 +141,6 @@
 #define RT_USING_LWIP
 #define RT_LWIP_DNS
 
-#define LWIP_NETIF_LINK_CALLBACK 1
-
 /* Trace LwIP protocol */
 // #define RT_LWIP_DEBUG 
 

libcpu/arm/at91sam926x/context_gcc.S

@@ -59,9 +59,9 @@ rt_hw_context_switch:
 	ldmfd	sp!, {r4}		@ pop new task spsr
 	msr	spsr_cxsf, r4
 	ldmfd	sp!, {r4}		@ pop new task cpsr
-	msr	cpsr_cxsf, r4
+	msr	spsr_cxsf, r4
 
-	ldmfd	sp!, {r0-r12, lr, pc}	@ pop new task r0-r12, lr & pc
+	ldmfd	sp!, {r0-r12, lr, pc}^	@ pop new task r0-r12, lr & pc
 
 /*
  * void rt_hw_context_switch_to(rt_uint32 to);

libcpu/arm/at91sam926x/context_rvds.S

@@ -60,9 +60,9 @@ rt_hw_context_switch	PROC
 	LDMFD	sp!, {r4}			; pop new task spsr
 	MSR	spsr_cxsf, r4
 	LDMFD	sp!, {r4}			; pop new task cpsr
-	MSR	cpsr_cxsf, r4
+	MSR	spsr_cxsf, r4
 
-	LDMFD	sp!, {r0-r12, lr, pc}	; pop new task r0-r12, lr & pc
+	LDMFD	sp!, {r0-r12, lr, pc}^	; pop new task r0-r12, lr & pc
 	ENDP
 
 ;/*

libcpu/arm/at91sam926x/mmu.c

@@ -9,19 +9,17 @@
  *
  * Change Logs:
  * Date           Author       Notes
- * 2011-01-13     weety       modified from mini2440
  */
 
 #include <rtthread.h>
-#include "at91sam926x.h"
 
-#define _MMUTT_STARTADDRESS	 0x33FF0000
+#define CACHE_LINE_SIZE	32
 
 #define DESC_SEC		(0x2|(1<<4))
 #define CB				(3<<2)  //cache_on, write_back
 #define CNB				(2<<2)  //cache_on, write_through
 #define NCB				(1<<2)  //cache_off,WR_BUF on
-#define NCNB				(0<<2)  //cache_off,WR_BUF off
+#define NCNB			(0<<2)  //cache_off,WR_BUF off
 #define AP_RW			(3<<10) //supervisor=RW, user=RW
 #define AP_RO			(2<<10) //supervisor=RW, user=RO
 
@@ -34,147 +32,31 @@
 #define DOMAIN0_ATTR	(DOMAIN_CHK<<0)
 #define DOMAIN1_ATTR	(DOMAIN_FAULT<<2)
 
-#define RW_CB		(AP_RW|DOMAIN0|CB|DESC_SEC)
+#define RW_CB		(AP_RW|DOMAIN0|CB|DESC_SEC)		/* Read/Write, cache, write back */
-#define RW_CNB		(AP_RW|DOMAIN0|CNB|DESC_SEC)
+#define RW_CNB		(AP_RW|DOMAIN0|CNB|DESC_SEC)	/* Read/Write, cache, write through */
-#define RW_NCNB		(AP_RW|DOMAIN0|NCNB|DESC_SEC)
+#define RW_NCNB		(AP_RW|DOMAIN0|NCNB|DESC_SEC)	/* Read/Write without cache and write buffer */
-#define RW_FAULT	(AP_RW|DOMAIN1|NCNB|DESC_SEC)
+#define RW_FAULT	(AP_RW|DOMAIN1|NCNB|DESC_SEC)	/* Read/Write without cache and write buffer */
-
-#ifdef __GNUC__
-void mmu_setttbase(register rt_uint32_t i)
-{
-	asm ("mcr p15, 0, %0, c2, c0, 0": :"r" (i));
-}
-
-void mmu_set_domain(register rt_uint32_t i)
-{
-	asm ("mcr p15,0, %0, c3, c0,  0": :"r" (i));
-}
-
-void mmu_enable()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i |= 0x1;
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_disable()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i &= ~0x1;
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_enable_icache()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i |= (1 << 12);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_enable_dcache()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i |= (1 << 2);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_disable_icache()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i &= ~(1 << 12);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_disable_dcache()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i &= ~(1 << 2);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_enable_alignfault()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i |= (1 << 1);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_disable_alignfault()
-{
-	register rt_uint32_t i;
-
-	/* read control register */
-	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
-
-	i &= ~(1 << 1);
-
-	/* write back to control register */
-	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
-}
-
-void mmu_clean_invalidated_cache_index(int index)
-{
-	asm ("mcr p15, 0, %0, c7, c14, 2": :"r" (index));
-}
-
-void mmu_invalidate_tlb()
-{
-	asm ("mcr p15, 0, %0, c8, c7, 0": :"r" (0));
-}
-
-void mmu_invalidate_icache()
-{
-	asm ("mcr p15, 0, %0, c7, c5, 0": :"r" (0));
-}
-#endif
 
 #ifdef __CC_ARM
 void mmu_setttbase(rt_uint32_t i)
 {
+	register rt_uint32_t value;
+
+   /* Invalidates all TLBs.Domain access is selected as
+    * client by configuring domain access register,
+    * in that case access controlled by permission value
+    * set by page table entry
+    */
+	value = 0;
     __asm
     {
+        mcr p15, 0, value, c8, c7, 0
+	}
+
+	value = 0x55555555;
+	__asm
+	{
+        mcr p15, 0, value, c3, c0, 0
         mcr p15, 0, i, c2, c0, 0
     }
 }
@@ -291,6 +173,54 @@ void mmu_clean_invalidated_cache_index(int index)
     }
 }
 
+void mmu_clean_invalidated_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+    unsigned int ptr;
+
+    ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+    while(ptr < buffer + size)
+    {
+    	__asm
+    	{
+    		MCR p15, 0, ptr, c7, c14, 1
+    	}
+        ptr += CACHE_LINE_SIZE;
+    }
+}
+
+void mmu_clean_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+	unsigned int ptr;
+
+	ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+	while (ptr < buffer + size)
+	{
+		__asm
+		{
+			MCR p15, 0, ptr, c7, c10, 1
+		}
+		ptr += CACHE_LINE_SIZE;
+	}
+}
+
+void mmu_invalidate_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+	unsigned int ptr;
+
+	ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+	while (ptr < buffer + size)
+	{
+		__asm
+		{
+			MCR p15, 0, ptr, c7, c6, 1
+		}
+		ptr += CACHE_LINE_SIZE;
+	}
+}
+
 void mmu_invalidate_tlb()
 {
     register rt_uint32_t value;
@@ -313,13 +243,214 @@ void mmu_invalidate_icache()
         mcr p15, 0, value, c7, c5, 0
     }
 }
+
+
+void mmu_invalidate_dcache_all()
+{
+    register rt_uint32_t value;
+
+    value = 0;
+
+    __asm
+    {
+        mcr p15, 0, value, c7, c6, 0
+    }
+}
+#elif defined(__GNUC__)
+void mmu_setttbase(register rt_uint32_t i)
+{
+	register rt_uint32_t value;
+
+   /* Invalidates all TLBs.Domain access is selected as
+    * client by configuring domain access register,
+    * in that case access controlled by permission value
+    * set by page table entry
+    */
+	value = 0;
+	asm ("mcr p15, 0, %0, c8, c7, 0"::"r"(value));
+
+	value = 0x55555555;
+	asm ("mcr p15, 0, %0, c3, c0, 0"::"r"(value));
+	asm ("mcr p15, 0, %0, c2, c0, 0"::"r"(i));
+}
+
+void mmu_set_domain(register rt_uint32_t i)
+{
+	asm ("mcr p15,0, %0, c3, c0,  0": :"r" (i));
+}
+
+void mmu_enable()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i |= 0x1;
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_disable()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i &= ~0x1;
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_enable_icache()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i |= (1 << 12);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_enable_dcache()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i |= (1 << 2);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_disable_icache()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i &= ~(1 << 12);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_disable_dcache()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i &= ~(1 << 2);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_enable_alignfault()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i |= (1 << 1);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_disable_alignfault()
+{
+	register rt_uint32_t i;
+
+	/* read control register */
+	asm ("mrc p15, 0, %0, c1, c0, 0":"=r" (i));
+
+	i &= ~(1 << 1);
+
+	/* write back to control register */
+	asm ("mcr p15, 0, %0, c1, c0, 0": :"r" (i));
+}
+
+void mmu_clean_invalidated_cache_index(int index)
+{
+	asm ("mcr p15, 0, %0, c7, c14, 2": :"r" (index));
+}
+
+void mmu_clean_invalidated_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+    unsigned int ptr;
+
+    ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+    while(ptr < buffer + size)
+    {
+    	asm ("mcr p15, 0, %0, c7, c14, 1": :"r" (ptr));
+        ptr += CACHE_LINE_SIZE;
+    }
+}
+
+
+void mmu_clean_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+	unsigned int ptr;
+
+	ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+	while (ptr < buffer + size)
+	{
+		asm ("mcr p15, 0, %0, c7, c10, 1": :"r" (ptr));
+		ptr += CACHE_LINE_SIZE;
+	}
+}
+
+void mmu_invalidate_dcache(rt_uint32_t buffer, rt_uint32_t size)
+{
+	unsigned int ptr;
+
+	ptr = buffer & ~(CACHE_LINE_SIZE - 1);
+
+	while (ptr < buffer + size)
+	{
+		asm ("mcr p15, 0, %0, c7, c6, 1": :"r" (ptr));
+		ptr += CACHE_LINE_SIZE;
+	}
+}
+
+void mmu_invalidate_tlb()
+{
+	asm ("mcr p15, 0, %0, c8, c7, 0": :"r" (0));
+}
+
+void mmu_invalidate_icache()
+{
+	asm ("mcr p15, 0, %0, c7, c5, 0": :"r" (0));
+}
+
+void mmu_invalidate_dcache_all()
+{
+    asm ("mcr p15, 0, %0, c7, c6, 0": :"r" (0));
+}
 #endif
 
-void mmu_setmtt(int vaddrStart,int vaddrEnd,int paddrStart,int attr)
+/* level1 page table */
+static volatile unsigned int _page_table[4*1024] __attribute__((aligned(16*1024)));
+void mmu_setmtt(rt_uint32_t vaddrStart, rt_uint32_t vaddrEnd, rt_uint32_t paddrStart, rt_uint32_t attr)
 {
     volatile rt_uint32_t *pTT;
     volatile int i,nSec;
-    pTT=(rt_uint32_t *)_MMUTT_STARTADDRESS+(vaddrStart>>20);
+    pTT=(rt_uint32_t *)_page_table+(vaddrStart>>20);
     nSec=(vaddrEnd>>20)-(vaddrStart>>20);
     for(i=0;i<=nSec;i++)
     {
@@ -330,66 +461,32 @@ void mmu_setmtt(int vaddrStart,int vaddrEnd,int paddrStart,int attr)
 
 void rt_hw_mmu_init(void)
 {
-#if 0
+	/* disable I/D cache */
-	int i,j;
-	//========================== IMPORTANT NOTE =========================
-	//The current stack and code area can't be re-mapped in this routine.
-	//If you want memory map mapped freely, your own sophiscated mmu
-	//initialization code is needed.
-	//===================================================================
-
 	mmu_disable_dcache();
 	mmu_disable_icache();
-
-	//If write-back is used,the DCache should be cleared.
-	for(i=0;i<64;i++)
-		for(j=0;j<8;j++)
-			mmu_clean_invalidated_cache_index((i<<26)|(j<<5));
-
-	mmu_invalidate_icache();
-
-	//To complete mmu_Init() fast, Icache may be turned on here.
-	mmu_enable_icache();
-
 	mmu_disable();
 	mmu_invalidate_tlb();
 
-	//mmu_setmtt(int vaddrStart,int vaddrEnd,int paddrStart,int attr);
+	/* set page table */
-	mmu_setmtt(0x00000000,0x07f00000,0x00000000,RW_CNB);  //bank0
+	mmu_setmtt(0x00000000, 0xFFFFFFFF, 0x00000000, RW_NCNB); /* None cached for 4G memory */
-	mmu_setmtt(0x00000000,0x03f00000,(int)0x30000000,RW_CB);  //bank0
+  	mmu_setmtt(0x20000000, 0x24000000-1, 0x20000000, RW_CB); /* 64M cached SDRAM memory */
-	mmu_setmtt(0x04000000,0x07f00000,0,RW_NCNB);  			//bank0
+	mmu_setmtt(0x00000000, 0x100000, 0x20000000, RW_CB); /* isr vector table */
-	mmu_setmtt(0x08000000,0x0ff00000,0x08000000,RW_CNB);  //bank1
+	mmu_setmtt(0x90000000, 0x100000, 0x00000000, RW_CB); /* 4K cached internal memory */
-	mmu_setmtt(0x10000000,0x17f00000,0x10000000,RW_NCNB); //bank2
+	mmu_setmtt(0xA0000000, 0xA4000000-1, 0x20000000, RW_NCNB); /* 64M none-cached SDRAM memory */
-	mmu_setmtt(0x18000000,0x1ff00000,0x18000000,RW_NCNB); //bank3
-	//mmu_setmtt(0x20000000,0x27f00000,0x20000000,RW_CB); //bank4
-	mmu_setmtt(0x20000000,0x27f00000,0x20000000,RW_NCNB); //bank4 for  DM9000
-	mmu_setmtt(0x28000000,0x2ff00000,0x28000000,RW_NCNB); //bank5
-	//30f00000->30100000, 31000000->30200000
-	mmu_setmtt(0x30000000,0x30100000,0x30000000,RW_CB);	  //bank6-1
-	mmu_setmtt(0x30200000,0x33e00000,0x30200000,RW_CB); //bank6-2
 
-	mmu_setmtt(0x33f00000,0x34000000,0x33f00000,RW_NCNB);   //bank6-3
+	/* set MMU table address */
-	mmu_setmtt(0x38000000,0x3ff00000,0x38000000,RW_NCNB); //bank7
+	mmu_setttbase((rt_uint32_t)_page_table);
 
-	mmu_setmtt(0x40000000,0x47f00000,0x40000000,RW_NCNB); //SFR
+    /* enables MMU */
-	mmu_setmtt(0x48000000,0x5af00000,0x48000000,RW_NCNB); //SFR
+    mmu_enable();
-	mmu_setmtt(0x5b000000,0x5b000000,0x5b000000,RW_NCNB); //SFR
-	mmu_setmtt(0x5b100000,0xfff00000,0x5b100000,RW_FAULT);//not used
-	mmu_setmtt(0x60000000,0x67f00000,0x60000000,RW_NCNB); //SFR
 
-	mmu_setttbase(_MMUTT_STARTADDRESS);
+    /* enable Instruction Cache */
+    mmu_enable_icache();
 
-	/* DOMAIN1: no_access, DOMAIN0,2~15=client(AP is checked) */
+    /* enable Data Cache */
-	mmu_set_domain(0x55555550|DOMAIN1_ATTR|DOMAIN0_ATTR);
+    mmu_enable_dcache();
 
-	mmu_enable_alignfault();
+    mmu_invalidate_icache();
-
+    mmu_invalidate_dcache_all();
-	mmu_enable();
-
-	/* ICache enable */
-	mmu_enable_icache();
-	/* DCache should be turned on after mmu is turned on. */
-	mmu_enable_dcache();
-#endif
 }
 

libcpu/arm/at91sam926x/start_gcc.S

@@ -51,8 +51,8 @@
 #define		AT91_RSTC_PROCRST	(1 << 0)		/* Processor Reset */
 #define		AT91_RSTC_PERRST	(1 << 2)
 #define		AT91_RSTC_KEY		(0xa5 << 24)
-#define AT91_MATRIX_BASE	0xffffee00
+#define 	AT91_MATRIX_BASE	0xffffee00
-#define AT91_MATRIX_MRCR	(AT91_MATRIX_BASE + 0x100)	/* Master Remap Control Register */
+#define 	AT91_MATRIX_MRCR	(AT91_MATRIX_BASE + 0x100)	/* Master Remap Control Register */
 #define		AT91_MATRIX_RCB0		(1 << 0)	/* Remap Command for AHB Master 0 (ARM926EJ-S Instruction Master) */
 #define		AT91_MATRIX_RCB1		(1 << 1)	/* Remap Command for AHB Master 1 (ARM926EJ-S Data Master) */
 #define		AT91_AIC_BASE	0xfffff000
@@ -172,9 +172,7 @@ reset:
 	str	r1, [r0]
 
 	/* set interrupt vector 		*/
-#if 1
+	ldr 	r0, _TEXT_BASE
-	ldr 	r0, _TEXT_BASE//_load_address
-	//ldr		r1, =0x200000				/* target address    				*/
 	mov	r1, #0x00
 	add	r2, r0, #0x40			/* size, 32bytes         			*/
 
@@ -183,7 +181,6 @@ copy_loop:
 	stmia	r1!, {r3-r10}			/* copy to   target address [r1]    */
 	cmp		r0, r2					/* until source end addreee [r2]    */
 	ble		copy_loop
-#endif
 	
 	/* setup stack */
 	bl		stack_setup