]> git.dujemihanovic.xyz Git - u-boot.git/commitdiff
ppc4xx: Refactor ECC POST for AMCC Denali core
authorLarry Johnson <lrj@acm.org>
Tue, 15 Jan 2008 19:35:58 +0000 (14:35 -0500)
committerStefan Roese <sr@denx.de>
Wed, 16 Jan 2008 10:23:33 +0000 (11:23 +0100)
The ECC POST reported intermittent failures running after power-up on
the Korat PPC440EPx board.  Even when the test passed, the debugging
output occasionally reported additional unexpected ECC errors.

This refactoring has three main objectives: (1) minimize the code
executed with ECC enabled during the tests, (2) add more checking of the
results so any unexpected ECC errors would cause the test to fail, and
(3) use synchronization (only) where required by the processor.

Signed-off-by: Larry Johnson <lrj@acm.org>
post/cpu/ppc4xx/denali_ecc.c

index 77234838d9b806b1cc44e23623e1e63197bbd099..439f80d1f6586132799945e393a9f87ccaa85ec3 100644 (file)
@@ -49,7 +49,7 @@
 
 DECLARE_GLOBAL_DATA_PTR;
 
-const static unsigned char syndrome_codes[] = {
+const static uint8_t syndrome_codes[] = {
        0xF4, 0XF1, 0XEC, 0XEA, 0XE9, 0XE6, 0XE5, 0XE3,
        0XDC, 0XDA, 0XD9, 0XD6, 0XD5, 0XD3, 0XCE, 0XCB,
        0xB5, 0XB0, 0XAD, 0XAB, 0XA8, 0XA7, 0XA4, 0XA2,
@@ -65,174 +65,183 @@ const static unsigned char syndrome_codes[] = {
 #define ECC_STOP_ADDR          0x2000
 #define ECC_PATTERN            0x01010101
 #define ECC_PATTERN_CORR       0x11010101
-#define ECC_PATTERN_UNCORR     0xF1010101
+#define ECC_PATTERN_UNCORR     0x61010101
 
-static int test_ecc_error(void)
+inline static void disable_ecc(void)
 {
-       unsigned long value;
-       unsigned long hdata, ldata, haddr, laddr;
-       unsigned int bit;
+       uint32_t value;
 
-       int ret = 0;
-
-       mfsdram(DDR0_23, value);
+       sync(); /* Wait for any pending memory accesses to complete. */
+       mfsdram(DDR0_22, value);
+       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
+               | DDR0_22_CTRL_RAW_ECC_DISABLE);
+}
 
-       for (bit = 0; bit < sizeof(syndrome_codes); bit++)
-               if (syndrome_codes[bit] == ((value >> 16) & 0xff))
-                       break;
+inline static void clear_and_enable_ecc(void)
+{
+       uint32_t value;
 
+       sync(); /* Wait for any pending memory accesses to complete. */
        mfsdram(DDR0_00, value);
+       mtsdram(DDR0_00, value | DDR0_00_INT_ACK_ALL);
+       mfsdram(DDR0_22, value);
+       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
+               | DDR0_22_CTRL_RAW_ECC_ENABLE);
+}
+
+static uint32_t get_ecc_status(void)
+{
+       uint32_t int_status;
+#if defined(DEBUG)
+       uint8_t syndrome;
+       uint32_t hdata, ldata, haddr, laddr;
+       uint32_t value;
+#endif
+
+       mfsdram(DDR0_00, int_status);
+       int_status &= DDR0_00_INT_STATUS_MASK;
 
-       if (value & DDR0_00_INT_STATUS_BIT0) {
-               debug("Bit0. A single access outside the defined PHYSICAL"
-                     " memory space detected\n");
+#if defined(DEBUG)
+       if (int_status & (DDR0_00_INT_STATUS_BIT0 | DDR0_00_INT_STATUS_BIT1)) {
                mfsdram(DDR0_32, laddr);
                mfsdram(DDR0_33, haddr);
-               debug("        addr = 0x%08x%08x\n", haddr, laddr);
-               ret = 1;
-       }
-       if (value & DDR0_00_INT_STATUS_BIT1) {
-               debug("Bit1. Multiple accesses outside the defined PHYSICAL"
-                     " memory space detected\n");
-               ret = 2;
-       }
-       if (value & DDR0_00_INT_STATUS_BIT2) {
-               debug("Bit2. Single correctable ECC event detected\n");
-               mfsdram(DDR0_38, laddr);
-               mfsdram(DDR0_39, haddr);
-               mfsdram(DDR0_40, ldata);
-               mfsdram(DDR0_41, hdata);
-               debug("        0x%08x - 0x%08x%08x, bit - %d\n",
-                     laddr, hdata, ldata, bit);
-               ret = 3;
+               haddr &= 0x00000001;
+               if (int_status & DDR0_00_INT_STATUS_BIT1)
+                       debug("Multiple accesses");
+               else
+                       debug("A single access");
+
+               debug(" outside the defined physical memory space detected\n"
+                     "        addr = 0x%01x%08x\n", haddr, laddr);
        }
-       if (value & DDR0_00_INT_STATUS_BIT3) {
-               debug("Bit3. Multiple correctable ECC events detected\n");
+       if (int_status & (DDR0_00_INT_STATUS_BIT2 | DDR0_00_INT_STATUS_BIT3)) {
+               unsigned int bit;
+
+               mfsdram(DDR0_23, value);
+               syndrome = (value >> 16) & 0xff;
+               for (bit = 0; bit < sizeof(syndrome_codes); bit++)
+                       if (syndrome_codes[bit] == syndrome)
+                               break;
+
                mfsdram(DDR0_38, laddr);
                mfsdram(DDR0_39, haddr);
+               haddr &= 0x00000001;
                mfsdram(DDR0_40, ldata);
                mfsdram(DDR0_41, hdata);
-               debug("        0x%08x - 0x%08x%08x, bit - %d\n",
-                     laddr, hdata, ldata, bit);
-               ret = 4;
-       }
-       if (value & DDR0_00_INT_STATUS_BIT4) {
-               debug("Bit4. Single uncorrectable ECC event detected\n");
-               mfsdram(DDR0_34, laddr);
-               mfsdram(DDR0_35, haddr);
-               mfsdram(DDR0_36, ldata);
-               mfsdram(DDR0_37, hdata);
-               debug("        0x%08x - 0x%08x%08x, bit - %d\n",
-                     laddr, hdata, ldata, bit);
-               ret = 5;
+               if (int_status & DDR0_00_INT_STATUS_BIT3)
+                       debug("Multiple correctable ECC events");
+               else
+                       debug("Single correctable ECC event");
+
+               debug(" detected\n        0x%01x%08x - 0x%08x%08x, bit - %d\n",
+                     haddr, laddr, hdata, ldata, bit);
        }
-       if (value & DDR0_00_INT_STATUS_BIT5) {
-               debug("Bit5. Multiple uncorrectable ECC events detected\n");
+       if (int_status & (DDR0_00_INT_STATUS_BIT4 | DDR0_00_INT_STATUS_BIT5)) {
+               mfsdram(DDR0_23, value);
+               syndrome = (value >> 8) & 0xff;
                mfsdram(DDR0_34, laddr);
                mfsdram(DDR0_35, haddr);
+               haddr &= 0x00000001;
                mfsdram(DDR0_36, ldata);
                mfsdram(DDR0_37, hdata);
-               debug("        0x%08x - 0x%08x%08x, bit - %d\n",
-                     laddr, hdata, ldata, bit);
-               ret = 6;
-       }
-       if (value & DDR0_00_INT_STATUS_BIT6) {
-               debug("Bit6. DRAM initialization complete\n");
-               ret = 7;
+               if (int_status & DDR0_00_INT_STATUS_BIT5)
+                       debug("Multiple uncorrectable ECC events");
+               else
+                       debug("Single uncorrectable ECC event");
+
+               debug(" detected\n        0x%01x%08x - 0x%08x%08x, "
+                     "syndrome - 0x%02x\n",
+                     haddr, laddr, hdata, ldata, syndrome);
        }
+       if (int_status & DDR0_00_INT_STATUS_BIT6)
+               debug("DRAM initialization complete\n");
+#endif /* defined(DEBUG) */
 
-       /* error status cleared */
-       mfsdram(DDR0_00, value);
-       mtsdram(DDR0_00, value | DDR0_00_INT_ACK_ALL);
-
-       return ret;
+       return int_status;
 }
 
-static int test_ecc(unsigned long ecc_addr)
+static int test_ecc(uint32_t ecc_addr)
 {
-       unsigned long value;
-       volatile unsigned *const ecc_mem = (volatile unsigned *) ecc_addr;
-       int pret;
+       uint32_t value;
+       volatile uint32_t *const ecc_mem = (volatile uint32_t *)ecc_addr;
        int ret = 0;
 
-       sync();
-       eieio();
        WATCHDOG_RESET();
 
-       debug("Entering test_ecc(0x%08lX)\n", ecc_addr);
+       debug("Entering test_ecc(0x%08x)\n", ecc_addr);
+       /* Set up correct ECC in memory */
+       disable_ecc();
+       clear_and_enable_ecc();
        out_be32(ecc_mem, ECC_PATTERN);
        out_be32(ecc_mem + 1, ECC_PATTERN);
-       in_be32(ecc_mem);
-       pret = test_ecc_error();
-       if (pret != 0) {
-               debug("pret: expected 0, got %d\n", pret);
+
+       /* Verify no ECC error reading back */
+       value = in_be32(ecc_mem);
+       disable_ecc();
+       if (ECC_PATTERN != value) {
+               debug("Data read error (no-error case): "
+                     "expected 0x%08x, read 0x%08x\n", ECC_PATTERN, value);
+               ret = 1;
+       }
+       value = get_ecc_status();
+       if (0x00000000 != value) {
+               /* Expected no ECC status reported */
+               debug("get_ecc_status(): expected 0x%08x, got 0x%08x\n",
+                     0x00000000, value);
                ret = 1;
        }
-       /* test for correctable error */
-       /* disconnect from ecc storage */
-       mfsdram(DDR0_22, value);
-       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
-               | DDR0_22_CTRL_RAW_ECC_DISABLE);
 
-       /* creating (correctable) single-bit error */
+       /* Test for correctable error by creating a one-bit error */
        out_be32(ecc_mem, ECC_PATTERN_CORR);
-
-       /* enable ecc */
-       mfsdram(DDR0_22, value);
-       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
-               | DDR0_22_CTRL_RAW_ECC_ENABLE);
-       sync();
-       eieio();
-
-       in_be32(ecc_mem);
-       pret = test_ecc_error();
-       /* if read data ok, 1 correctable error must be fixed */
-       if (pret != 3) {
-               debug("pret: expected 3, got %d\n", pret);
+       clear_and_enable_ecc();
+       value = in_be32(ecc_mem);
+       disable_ecc();
+       /* Test that the corrected data was read */
+       if (ECC_PATTERN != value) {
+               debug("Data read error (correctable-error case): "
+                     "expected 0x%08x, read 0x%08x\n", ECC_PATTERN, value);
+               ret = 1;
+       }
+       value = get_ecc_status();
+       if ((DDR0_00_INT_STATUS_BIT2 | DDR0_00_INT_STATUS_BIT7) != value) {
+               /* Expected a single correctable error reported */
+               debug("get_ecc_status(): expected 0x%08x, got 0x%08x\n",
+                     DDR0_00_INT_STATUS_BIT2, value);
                ret = 1;
        }
-       /* test for uncorrectable error */
-       /* disconnect from ecc storage */
-       mfsdram(DDR0_22, value);
-       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
-               | DDR0_22_CTRL_RAW_NO_ECC_RAM);
 
-       /* creating (uncorrectable) multiple-bit error */
+       /* Test for uncorrectable error by creating a two-bit error */
        out_be32(ecc_mem, ECC_PATTERN_UNCORR);
-
-       /* enable ecc */
-       mfsdram(DDR0_22, value);
-       mtsdram(DDR0_22, (value & ~DDR0_22_CTRL_RAW_MASK)
-               | DDR0_22_CTRL_RAW_ECC_ENABLE);
-       sync();
-       eieio();
-
-       in_be32(ecc_mem);
-       pret = test_ecc_error();
-       /* info about uncorrectable error must appear */
-       if (pret != 5) {
-               debug("pret: expected 5, got %d\n", pret);
+       clear_and_enable_ecc();
+       value = in_be32(ecc_mem);
+       disable_ecc();
+       /* Test that the corrected data was read */
+       if (ECC_PATTERN_UNCORR != value) {
+               debug("Data read error (uncorrectable-error case): "
+                     "expected 0x%08x, read 0x%08x\n", ECC_PATTERN_UNCORR,
+                     value);
+               ret = 1;
+       }
+       value = get_ecc_status();
+       if ((DDR0_00_INT_STATUS_BIT4 | DDR0_00_INT_STATUS_BIT7) != value) {
+               /* Expected a single uncorrectable error reported */
+               debug("get_ecc_status(): expected 0x%08x, got 0x%08x\n",
+                     DDR0_00_INT_STATUS_BIT4, value);
                ret = 1;
        }
-       /* remove error from SDRAM */
+
+       /* Remove error from SDRAM and enable ECC. */
        out_be32(ecc_mem, ECC_PATTERN);
-       /* clear error caused by read-modify-write */
-       mfsdram(DDR0_00, value);
-       mtsdram(DDR0_00, value | DDR0_00_INT_ACK_ALL);
+       clear_and_enable_ecc();
 
-       sync();
-       eieio();
        return ret;
 }
 
-int ecc_post_test (int flags)
+int ecc_post_test(int flags)
 {
        int ret = 0;
-       unsigned long value;
-       unsigned long iaddr;
-
-       sync();
-       eieio();
+       uint32_t value;
+       uint32_t iaddr;
 
        mfsdram(DDR0_22, value);
        if (0x3 != DDR0_22_CTRL_RAW_DECODE(value)) {
@@ -240,28 +249,23 @@ int ecc_post_test (int flags)
                return 0;
        }
 
-       /* mask all int */
+       /* Mask all interrupts. */
        mfsdram(DDR0_01, value);
        mtsdram(DDR0_01, (value & ~DDR0_01_INT_MASK_MASK)
                | DDR0_01_INT_MASK_ALL_OFF);
 
-       /* clear error status */
-       mfsdram(DDR0_00, value);
-       mtsdram(DDR0_00, value | DDR0_00_INT_ACK_ALL);
-
        for (iaddr = ECC_START_ADDR; iaddr <= ECC_STOP_ADDR; iaddr += iaddr) {
                ret = test_ecc(iaddr);
                if (ret)
                        break;
        }
        /*
-        * Clear possible errors resulting from ECC testing.
-        * If not done, then we could get an interrupt later on when
-        * exceptions are enabled.
+        * Clear possible errors resulting from ECC testing.  (If not done, we
+        * we could get an interrupt later on when exceptions are enabled.)
         */
        set_mcsr(get_mcsr());
+       debug("ecc_post_test() returning %d\n", ret);
        return ret;
-
 }
 #endif /* CONFIG_POST & CFG_POST_ECC */
 #endif /* defined(CONFIG_POST) && ... */