RE: [PATCH v2 8/8] EDAC/aspeed: Add AST2700 support

From
Ryan Chen <ryan_chen@aspeedtech.com>
Date
2026-08-27 06:24:35
Also in
sashiko-reviews
Reply to
sashiko-bot@kernel.org, 2026-08-24
quoted hunk ↗ jump to hunk
Subject: Re: [PATCH v2 8/8] EDAC/aspeed: Add AST2700 support

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to
consider:
- [High] Returning IRQ_NONE for unhandled exclusive DRAMC interrupts relies
on the spurious IRQ detector, which permanently disables ECC reporting.
--
--- Patch [8]: [PATCH v2 8/8] EDAC/aspeed: Add AST2700 support --- commit
0d3da1d609b9b0868952fd656c60d988b1c32e40
Author: Ryan Chen [off-list ref]

EDAC/aspeed: Add AST2700 support

Add SDRAM ECC reporting for the Aspeed AST2700. Its DRAMC has a different
register layout, a split interrupt status/clear/mask scheme, DDR4/DDR5
memory and interrupt registers that are not key-protected.
quoted
diff --git a/drivers/edac/aspeed_edac.c b/drivers/edac/aspeed_edac.c
index e88d9d2646de7..88b7f5ff37273 100644
--- a/drivers/edac/aspeed_edac.c
+++ b/drivers/edac/aspeed_edac.c
[ ... ]
quoted
@@ -180,6 +209,54 @@ static irqreturn_t aspeed_mcr_isr(int irq, void
*arg)
[ ... ]
quoted
+static irqreturn_t ast2700_dramc_isr(int irq, void *arg) {
+	u32 int_sts, ecc_sts, fail_addr;
+	struct mem_ctl_info *mci = arg;
+	struct aspeed_edac *priv;
+	u8 rec_cnt, un_rec_cnt;
+	phys_addr_t addr;
+
+	priv = mci->pvt_info;
+
+	scoped_guard(raw_spinlock, &priv->lock) {
+		int_sts = readl(priv->regs + AST2700_INT_STS);
+		if (!(int_sts & AST2700_INT_ECC))
+			return IRQ_NONE;
[Severity: High]
If a non-ECC interrupt is left enabled by the bootloader in the mask register,
will returning IRQ_NONE here cause the kernel's spurious IRQ detector to
permanently disable the interrupt line?

Since the driver requests an exclusive, level-triggered interrupt, an unhandled
asserted line will cause an interrupt storm leading to a disabled ECC error
reporting mechanism.
The premise doesn't hold on this hardware. From u-boot, before Linux
starts:

  => md 12c0000c 1        # INT_MASK
  12c0000c: 0001ffff
  => mw 12c0000c ffffffff
  => md 12c0000c 1
  12c0000c: 0001ffff

INT_MASK is GENMASK(16, 0) - u-boot masks every implemented source -
and bits 31:17 are not implemented. So the read-modify-write leaves
0x0001ffcf: the two ECC bits are the only unmasked sources, nothing
else can raise the line, the handler cannot return IRQ_NONE in
practice, and the spurious detector cannot kick in. INT_STS reads
0x00004008, i.e. bits 3 and 14 asserted but masked.

Note IRQ_NONE is the safety net rather than the hazard: claiming every
interrupt unconditionally turns the same scenario into a livelock with
no diagnostic, which is why the v1 review asked for IRQ_NONE.

Taking full ownership of the mask register is still fair hardening,
since the driver does rely on the bootloader, but it isn't fixing a
present bug.

quoted
+
+		ecc_sts = readl(priv->regs + AST2700_ECC_STS);
+		fail_addr = readl(priv->regs + AST2700_ECC_FAIL_ADDR);
+
+		/* the interrupt registers are not key-protected; clear only ECC */
+		writel(int_sts & AST2700_INT_ECC, priv->regs + AST2700_INT_CLR);
+	}
[ ... ]
quoted
+static void ast2700_set_irq(struct aspeed_edac *priv, bool enable) {
+	u32 val;
+
+	guard(raw_spinlock_irqsave)(&priv->lock);
+
+	/* interrupts are enabled by clearing their mask bits */
+	val = readl(priv->regs + AST2700_INT_MASK);
+	if (enable)
+		val &= ~AST2700_INT_ECC;
+	else
+		val |= AST2700_INT_ECC;
+
+	writel(val, priv->regs + AST2700_INT_MASK);
[Severity: High]
Does this read-modify-write sequence leave non-ECC interrupts unmasked if
they were already enabled by the bootloader?

Taking full ownership of the mask register during initialization, perhaps by
masking all non-ECC sources explicitly instead of doing a read-modify-write,
might prevent the spurious interrupt issue mentioned above.
quoted
+}
--
Sashiko AI review ·
https://sashiko.dev/#/patchset/20260824-edac-v2-0-c8d8bb693586@aspeedte
ch.com?part=8
Thread (17 messages) 17 messages, 4 authors, last reply 2026-09-09
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help