nb/amd/{agesa,pi}: Avoid overflows during DRAM calculation
Do not use get_dram_base_mask to calculate system DRAM limits. Shift
operation around values operating on base and mask were causing
overflows and thus incorrect system DRAM limit. Another function
returning base and limit in KiB has been developed to avoid data loss.
Keep DRAM high base and limit in calculations only for Trinity where
the physical CPU address bits is 48. Although it is almost impossible
to have a non-zero value there, the platform would have to support
nearly 256GB of RAM.
TEST=boot PC Engines apu1 2GB, apu2 4GB and apu3 2GB and boot Debian
with Linux 4.14
Signed-off-by: Michał Żygowski <michal.zygowski@3mdeb.com>
Change-Id: I3b5c1df96c308ff50c8de104e213219a98f25e10
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52922
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
diff --git a/src/northbridge/amd/agesa/family14/northbridge.c b/src/northbridge/amd/agesa/family14/northbridge.c
index 020c9c6..4aed96b 100644
--- a/src/northbridge/amd/agesa/family14/northbridge.c
+++ b/src/northbridge/amd/agesa/family14/northbridge.c
@@ -26,29 +26,6 @@
static struct device *__f4_dev[FX_DEVS];
static unsigned int fx_devs = 0;
-struct dram_base_mask_t {
- u32 base; //[47:27] at [28:8]
- u32 mask; //[47:27] at [28:8] and enable at bit 0
-};
-
-static struct dram_base_mask_t get_dram_base_mask(u32 nodeid)
-{
- struct device *dev;
- struct dram_base_mask_t d;
- dev = __f1_dev[0];
-
- u32 temp;
- temp = pci_read_config32(dev, 0x44); //[39:24] at [31:16]
- d.mask = (temp & 0xffff0000); // mask out DramMask [26:24] too
-
- temp = pci_read_config32(dev, 0x40); //[35:24] at [27:16]
- d.mask |= (temp & 1); // read enable bit
-
- d.base = (temp & 0x0fff0000); // mask out DramBase [26:24) too
-
- return d;
-}
-
static u32 get_io_addr_index(u32 nodeid, u32 linkn)
{
return 0;
@@ -127,6 +104,33 @@
}
}
+static int get_dram_base_limit(u32 nodeid, resource_t *basek, resource_t *limitk)
+{
+ u32 temp;
+
+ if (fx_devs == 0)
+ get_fx_devs();
+
+
+ temp = pci_read_config32(__f1_dev[nodeid], 0x40 + (nodeid << 3)); //[39:24] at [31:16]
+ if (!(temp & 1))
+ return 0; // this memory range is not enabled
+ /*
+ * BKDG: {DramBase[35:24], 00_0000h} <= address[35:0] so shift left by 8 bits
+ * for physical address and the convert to KiB by shifting 10 bits left
+ */
+ *basek = ((temp & 0x0fff0000)) >> (10 - 8);
+ /*
+ * BKDG address[35:0] <= {DramLimit[35:24], FF_FFFFh} converted as above but
+ * ORed with 0xffff to get real limit before shifting.
+ */
+ temp = pci_read_config32(__f1_dev[nodeid], 0x44 + (nodeid << 3)); //[39:24] at [31:16]
+ *limitk = ((temp & 0x0fff0000) | 0xffff) >> (10 - 8);
+ *limitk += 1; // round up last byte
+
+ return 1;
+}
+
static u32 amdfam14_nodeid(struct device *dev)
{
return (dev->path.pci.devfn >> 3) - DEV_CDB;
@@ -305,10 +309,10 @@
mem_hole.hole_startk = CONFIG_HW_MEM_HOLE_SIZEK;
mem_hole.node_id = -1;
- struct dram_base_mask_t d;
+ resource_t basek, limitk;
u32 hole;
- d = get_dram_base_mask(0);
- if (d.mask & 1) {
+
+ if (get_dram_base_limit(0, &basek, &limitk)) {
hole = pci_read_config32(__f1_dev[0], 0xf0);
if (hole & 1) { // we find the hole
mem_hole.hole_startk = (hole & (0xff << 24)) >> 10;
@@ -542,27 +546,13 @@
#endif
idx = 0x10;
-
- struct dram_base_mask_t d;
resource_t basek, limitk, sizek; // 4 1T
- d = get_dram_base_mask(0);
+ if (get_dram_base_limit(0, &basek, &limitk)) {
+ sizek = limitk - basek;
- if (d.mask & 1) {
- basek = ((resource_t) ((u64) d.base)) << 8;
- limitk = (resource_t) (((u64) d.mask << 8) | 0xFFFFFF);
- printk(BIOS_DEBUG,
- "adsr: (before) basek = %llx, limitk = %llx.\n", basek,
- limitk);
-
- /* Convert these values to multiples of 1K for ease of math. */
- basek >>= 10;
- limitk >>= 10;
- sizek = limitk - basek + 1;
-
- printk(BIOS_DEBUG,
- "adsr: (after) basek = %llx, limitk = %llx, sizek = %llx.\n",
- basek, limitk, sizek);
+ printk(BIOS_DEBUG, "adsr: basek = %llx, limitk = %llx, sizek = %llx.\n",
+ basek, limitk, sizek);
/* see if we need a hole from 0xa0000 to 0xbffff */
if ((basek < 640) && (sizek > 768)) {
diff --git a/src/northbridge/amd/agesa/family15tn/northbridge.c b/src/northbridge/amd/agesa/family15tn/northbridge.c
index e345b50..fe56736 100644
--- a/src/northbridge/amd/agesa/family15tn/northbridge.c
+++ b/src/northbridge/amd/agesa/family15tn/northbridge.c
@@ -24,11 +24,6 @@
#define MAX_NODE_NUMS MAX_NODES
-typedef struct dram_base_mask {
- u32 base; //[47:27] at [28:8]
- u32 mask; //[47:27] at [28:8] and enable at bit 0
-} dram_base_mask_t;
-
static unsigned int node_nums;
static unsigned int sblink;
static struct device *__f0_dev[MAX_NODE_NUMS];
@@ -37,24 +32,6 @@
static struct device *__f4_dev[MAX_NODE_NUMS];
static unsigned int fx_devs = 0;
-static dram_base_mask_t get_dram_base_mask(u32 nodeid)
-{
- struct device *dev;
- dram_base_mask_t d;
- dev = __f1_dev[0];
- u32 temp;
- temp = pci_read_config32(dev, 0x44 + (nodeid << 3)); //[39:24] at [31:16]
- d.mask = ((temp & 0xfff80000)>>(8+3)); // mask out DramMask [26:24] too
- temp = pci_read_config32(dev, 0x144 + (nodeid <<3)) & 0xff; //[47:40] at [7:0]
- d.mask |= temp << 21;
- temp = pci_read_config32(dev, 0x40 + (nodeid << 3)); //[39:24] at [31:16]
- d.mask |= (temp & 1); // enable bit
- d.base = ((temp & 0xfff80000)>>(8+3)); // mask out DramBase [26:24) too
- temp = pci_read_config32(dev, 0x140 + (nodeid <<3)) & 0xff; //[47:40] at [7:0]
- d.base |= temp << 21;
- return d;
-}
-
static void set_io_addr_reg(struct device *dev, u32 nodeid, u32 linkn, u32 reg,
u32 io_min, u32 io_max)
{
@@ -125,6 +102,39 @@
}
}
+static int get_dram_base_limit(u32 nodeid, resource_t *basek, resource_t *limitk)
+{
+ u32 temp;
+
+ if (fx_devs == 0)
+ get_fx_devs();
+
+
+ temp = pci_read_config32(__f1_dev[nodeid], 0x40 + (nodeid << 3)); //[39:24] at [31:16]
+ if (!(temp & 1))
+ return 0; // this memory range is not enabled
+ /*
+ * BKDG: {DramBase[47:24], 00_0000h} <= address[47:0] so shift left by 8 bits
+ * for physical address and the convert to KiB by shifting 10 bits left
+ */
+ *basek = ((temp & 0xffff0000)) >> (10 - 8);
+ /* Now high bits [47:40] */
+ temp = pci_read_config32(__f1_dev[nodeid], 0x140 + (nodeid << 3)); //[47:40] at [7:0]
+ *basek = *basek | ((resource_t)temp << (40 - 10));
+ /*
+ * BKDG address[39:0] <= {DramLimit[47:24], FF_FFFFh} converted as above but
+ * ORed with 0xffff to get real limit before shifting.
+ */
+ temp = pci_read_config32(__f1_dev[nodeid], 0x44 + (nodeid << 3)); //[39:24] at [31:16]
+ *limitk = ((temp & 0xffff0000) | 0xffff) >> (10 - 8);
+ /* Now high bits [47:40] */
+ temp = pci_read_config32(__f1_dev[nodeid], 0x144 + (nodeid << 3)); //[47:40] at [7:0]
+ *limitk = *limitk | ((resource_t)temp << (40 - 10));
+ *limitk += 1; // round up last byte
+
+ return 1;
+}
+
static u32 amdfam15_nodeid(struct device *dev)
{
return (dev->path.pci.devfn >> 3) - DEV_CDB;
@@ -616,10 +626,10 @@
mem_hole.hole_startk = CONFIG_HW_MEM_HOLE_SIZEK;
mem_hole.node_id = -1;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
+ resource_t basek, limitk;
u32 hole;
- d = get_dram_base_mask(i);
- if (!(d.mask & 1)) continue; // no memory on this node
+ if (!get_dram_base_limit(i, &basek, &limitk))
+ continue; // no memory on this node
hole = pci_read_config32(__f1_dev[i], 0xf0);
if (hole & 1) { // we find the hole
mem_hole.hole_startk = (hole & (0xff << 24)) >> 10;
@@ -634,18 +644,15 @@
if (mem_hole.node_id == -1) {
resource_t limitk_pri = 0;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
resource_t base_k, limit_k;
- d = get_dram_base_mask(i);
- if (!(d.base & 1)) continue;
- base_k = ((resource_t)(d.base & 0x1fffff00)) <<9;
+ if (!get_dram_base_limit(i, &base_k, &limit_k))
+ continue; // no memory on this node
if (base_k > 4 *1024 * 1024) break; // don't need to go to check
if (limitk_pri != base_k) { // we find the hole
mem_hole.hole_startk = (unsigned int)limitk_pri; // must beblow 4G
mem_hole.node_id = i;
break; //only one hole
}
- limit_k = ((resource_t)(((d.mask & ~1) + 0x000FF) & 0x1fffff00)) << 9;
limitk_pri = limit_k;
}
}
@@ -698,14 +705,10 @@
idx = 0x10;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
resource_t basek, limitk, sizek; // 4 1T
- d = get_dram_base_mask(i);
-
- if (!(d.mask & 1)) continue;
- basek = ((resource_t)(d.base & 0x1fffff00)) << 9; // could overflow, we may lost 6 bit here
- limitk = ((resource_t)(((d.mask & ~1) + 0x000FF) & 0x1fffff00)) << 9;
+ if (!get_dram_base_limit(i, &basek, &limitk))
+ continue; // no memory on this node
sizek = limitk - basek;
diff --git a/src/northbridge/amd/agesa/family16kb/northbridge.c b/src/northbridge/amd/agesa/family16kb/northbridge.c
index b600bc9..21b4d69 100644
--- a/src/northbridge/amd/agesa/family16kb/northbridge.c
+++ b/src/northbridge/amd/agesa/family16kb/northbridge.c
@@ -24,11 +24,6 @@
#define MAX_NODE_NUMS MAX_NODES
-typedef struct dram_base_mask {
- u32 base; //[47:27] at [28:8]
- u32 mask; //[47:27] at [28:8] and enable at bit 0
-} dram_base_mask_t;
-
static unsigned int node_nums;
static unsigned int sblink;
static struct device *__f0_dev[MAX_NODE_NUMS];
@@ -37,24 +32,6 @@
static struct device *__f4_dev[MAX_NODE_NUMS];
static unsigned int fx_devs = 0;
-static dram_base_mask_t get_dram_base_mask(u32 nodeid)
-{
- struct device *dev;
- dram_base_mask_t d;
- dev = __f1_dev[0];
- u32 temp;
- temp = pci_read_config32(dev, 0x44 + (nodeid << 3)); //[39:24] at [31:16]
- d.mask = ((temp & 0xfff80000)>>(8+3)); // mask out DramMask [26:24] too
- temp = pci_read_config32(dev, 0x144 + (nodeid <<3)) & 0xff; //[47:40] at [7:0]
- d.mask |= temp << 21;
- temp = pci_read_config32(dev, 0x40 + (nodeid << 3)); //[39:24] at [31:16]
- d.mask |= (temp & 1); // enable bit
- d.base = ((temp & 0xfff80000)>>(8+3)); // mask out DramBase [26:24) too
- temp = pci_read_config32(dev, 0x140 + (nodeid <<3)) & 0xff; //[47:40] at [7:0]
- d.base |= temp << 21;
- return d;
-}
-
static void set_io_addr_reg(struct device *dev, u32 nodeid, u32 linkn, u32 reg,
u32 io_min, u32 io_max)
{
@@ -125,6 +102,33 @@
}
}
+static int get_dram_base_limit(u32 nodeid, resource_t *basek, resource_t *limitk)
+{
+ u32 temp;
+
+ if (fx_devs == 0)
+ get_fx_devs();
+
+
+ temp = pci_read_config32(__f1_dev[nodeid], 0x40 + (nodeid << 3)); //[39:24] at [31:16]
+ if (!(temp & 1))
+ return 0; // this memory range is not enabled
+ /*
+ * BKDG: {DramBase[39:24], 00_0000h} <= address[39:0] so shift left by 8 bits
+ * for physical address and the convert to KiB by shifting 10 bits left
+ */
+ *basek = ((temp & 0xffff0000)) >> (10 - 8);
+ /*
+ * BKDG address[39:0] <= {DramLimit[39:24], FF_FFFFh} converted as above but
+ * ORed with 0xffff to get real limit before shifting.
+ */
+ temp = pci_read_config32(__f1_dev[nodeid], 0x44 + (nodeid << 3)); //[39:24] at [31:16]
+ *limitk = ((temp & 0xffff0000) | 0xffff) >> (10 - 8);
+ *limitk += 1; // round up last byte
+
+ return 1;
+}
+
static u32 amdfam16_nodeid(struct device *dev)
{
return (dev->path.pci.devfn >> 3) - DEV_CDB;
@@ -634,10 +638,10 @@
mem_hole.hole_startk = CONFIG_HW_MEM_HOLE_SIZEK;
mem_hole.node_id = -1;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
+ resource_t basek, limitk;
u32 hole;
- d = get_dram_base_mask(i);
- if (!(d.mask & 1)) continue; // no memory on this node
+ if (!get_dram_base_limit(i, &basek, &limitk))
+ continue; // no memory on this node
hole = pci_read_config32(__f1_dev[i], 0xf0);
if (hole & 2) { // we find the hole
mem_hole.hole_startk = (hole & (0xff << 24)) >> 10;
@@ -652,18 +656,15 @@
if (mem_hole.node_id == -1) {
resource_t limitk_pri = 0;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
resource_t base_k, limit_k;
- d = get_dram_base_mask(i);
- if (!(d.base & 1)) continue;
- base_k = ((resource_t)(d.base & 0x1fffff00)) <<9;
+ if (!get_dram_base_limit(i, &base_k, &limit_k))
+ continue; // no memory on this node
if (base_k > 4 *1024 * 1024) break; // don't need to go to check
if (limitk_pri != base_k) { // we find the hole
mem_hole.hole_startk = (unsigned int)limitk_pri; // must beblow 4G
mem_hole.node_id = i;
break; //only one hole
}
- limit_k = ((resource_t)(((d.mask & ~1) + 0x000FF) & 0x1fffff00)) << 9;
limitk_pri = limit_k;
}
}
@@ -716,14 +717,10 @@
idx = 0x10;
for (i = 0; i < node_nums; i++) {
- dram_base_mask_t d;
resource_t basek, limitk, sizek; // 4 1T
- d = get_dram_base_mask(i);
-
- if (!(d.mask & 1)) continue;
- basek = ((resource_t)(d.base & 0x1fffff00)) << 9; // could overflow, we may lost 6 bit here
- limitk = ((resource_t)(((d.mask & ~1) + 0x000FF) & 0x1fffff00)) << 9;
+ if (!get_dram_base_limit(i, &basek, &limitk))
+ continue; // no memory on this node
sizek = limitk - basek;