[PATCH v4 06/13] EDAC/synopsys: Add data poisoning disable support

From: Serge Semin
Date: Wed Sep 20 2023 - 15:58:24 EST


Even though being a pure-debug feature currently the data poison can't be
disabled once it has been initialized and enabled. Irrespective to the way
the feature has been implemented it doesn't seem right since the system
may print false ECC errors in case if the poisoned address is accessed by
the kernel or by the user-space applications. It's possible since the
poisoned address isn't reserved in any kernel mm subsystems. Even though
that doesn't seem right either at least it's tolerable since the ECC data
poison is supposed to be utilized in the framework of the EDAC driver
debugging, but having the feature non-switchable can't be justified that
easy especially seeing it's not that hard to implement.

So in order to have the ECC data poison switchable define three possible
values acceptable by the "inject_data_poison" DebugFS node:
1. "CE" - emit correctable error (as before).
2. "UE" - emit uncorrectable error (used to be any non-"CE" value).
3. Any other value - disable data poison feature.

Note the macros describing the data poison-related fields of the ECC_CFG0
register need to be redefined in a way so they would be used to separately
switch the feature on/off and to select the type of the ECC error. As a
result the suggest solution turns into a proper ECC_CFG0 CSRs fields setup
based on the value written to the "inject_data_poison" DebugFS node.

Signed-off-by: Serge Semin <fancer.lancer@xxxxxxxxx>
---
drivers/edac/synopsys_edac.c | 29 ++++++++++++++++++++---------
1 file changed, 20 insertions(+), 9 deletions(-)

diff --git a/drivers/edac/synopsys_edac.c b/drivers/edac/synopsys_edac.c
index 90640b2f877a..3735c784fac9 100644
--- a/drivers/edac/synopsys_edac.c
+++ b/drivers/edac/synopsys_edac.c
@@ -123,6 +123,10 @@
#define ECC_CFG0_DIS_SCRUB BIT(4)
#define ECC_CFG0_MODE_MASK GENMASK(2, 0)

+/* ECC CFG1 register definitions */
+#define ECC_CFG1_POISON_BIT BIT(1)
+#define ECC_CFG1_POISON_EN BIT(0)
+
/* ECC status register definitions */
#define ECC_STAT_UE_MASK GENMASK(23, 16)
#define ECC_STAT_CE_MASK GENMASK(15, 8)
@@ -154,10 +158,6 @@
#define ECC_POISON1_BANK_MASK GENMASK(26, 24)
#define ECC_POISON1_ROW_MASK GENMASK(17, 0)

-/* DDRC ECC CE & UE poison mask */
-#define ECC_CEPOISON_MASK GENMASK(1, 0)
-#define ECC_UEPOISON_MASK BIT(0)
-
/* DDRC address mapping parameters */
#define DDR_ADDRMAP_NREGS 12

@@ -1787,10 +1787,14 @@ static ssize_t snps_inject_data_poison_read(struct file *filep, char __user *ubu
int pos;

regval = readl(priv->baseaddr + ECC_CFG1_OFST);
- errstr = FIELD_GET(ECC_CEPOISON_MASK, regval) == ECC_CEPOISON_MASK ?
- "Correctable Error" : "UnCorrectable Error";
+ if (!(regval & ECC_CFG1_POISON_EN))
+ errstr = "Off";
+ else if (regval & ECC_CFG1_POISON_BIT)
+ errstr = "CE";
+ else
+ errstr = "UE";

- pos = scnprintf(buf, sizeof(buf), "Data Poisoning: %s\n\r", errstr);
+ pos = scnprintf(buf, sizeof(buf), "%s\n", errstr);

return simple_read_from_buffer(ubuf, size, offp, buf, pos);
}
@@ -1801,6 +1805,7 @@ static ssize_t snps_inject_data_poison_write(struct file *filep, const char __us
struct mem_ctl_info *mci = filep->private_data;
struct snps_edac_priv *priv = mci->pvt_info;
char buf[SNPS_DBGFS_BUF_LEN];
+ u32 regval;
int rc;

rc = simple_write_to_buffer(buf, sizeof(buf), offp, ubuf, size);
@@ -1808,10 +1813,16 @@ static ssize_t snps_inject_data_poison_write(struct file *filep, const char __us
return rc;

writel(0, priv->baseaddr + DDR_SWCTL);
+
+ regval = readl(priv->baseaddr + ECC_CFG1_OFST);
if (strncmp(buf, "CE", 2) == 0)
- writel(ECC_CEPOISON_MASK, priv->baseaddr + ECC_CFG1_OFST);
+ regval |= ECC_CFG1_POISON_BIT | ECC_CFG1_POISON_EN;
+ else if (strncmp(buf, "UE", 2) == 0)
+ regval = (regval & ~ECC_CFG1_POISON_BIT) | ECC_CFG1_POISON_EN;
else
- writel(ECC_UEPOISON_MASK, priv->baseaddr + ECC_CFG1_OFST);
+ regval &= ~ECC_CFG1_POISON_EN;
+ writel(regval, priv->baseaddr + ECC_CFG1_OFST);
+
writel(1, priv->baseaddr + DDR_SWCTL);

return size;
--
2.41.0