lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [day] [month] [year] [list]
Message-ID: <20251009155752.773732-62-sashal@kernel.org>
Date: Thu,  9 Oct 2025 11:55:28 -0400
From: Sasha Levin <sashal@...nel.org>
To: patches@...ts.linux.dev,
	stable@...r.kernel.org
Cc: Ming Wang <wangming01@...ngson.cn>,
	Thomas Gleixner <tglx@...utronix.de>,
	Sasha Levin <sashal@...nel.org>,
	chenhuacai@...nel.org,
	jiaxun.yang@...goat.com,
	linux-mips@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: [PATCH AUTOSEL 6.17-6.1] irqchip/loongson-pch-lpc: Use legacy domain for PCH-LPC IRQ controller

From: Ming Wang <wangming01@...ngson.cn>

[ Upstream commit c33c43f71bda362b292a6e57ac41b64342dc87b3 ]

On certain Loongson platforms, drivers attempting to request a legacy
ISA IRQ directly via request_irq() (e.g., IRQ 4) may fail. The
virtual IRQ descriptor is not fully initialized and lacks a valid irqchip.

This issue does not affect ACPI-enumerated devices described in DSDT,
as their interrupts are properly mapped via the GSI translation path.
This indicates the LPC irqdomain itself is functional but is not correctly
handling direct VIRQ-to-HWIRQ mappings.

The root cause is the use of irq_domain_create_linear(). This API sets
up a domain for dynamic, on-demand mapping, typically triggered by a GSI
request. It does not pre-populate the mappings for the legacy VIRQ range
(0-15). Consequently, if no ACPI device claims a specific GSI
(e.g., GSI 4), the corresponding VIRQ (e.g., VIRQ 4) is never mapped to
the LPC domain. A direct call to request_irq(4, ...) then fails because
the kernel cannot resolve this VIRQ to a hardware interrupt managed by
the LPC controller.

The PCH-LPC interrupt controller is an i8259-compatible legacy device
that requires a deterministic, static 1-to-1 mapping for IRQs 0-15 to
support legacy drivers.

Fix this by replacing irq_domain_create_linear() with
irq_domain_create_legacy(). This API is specifically designed for such
controllers. It establishes the required static 1-to-1 VIRQ-to-HWIRQ
mapping for the entire legacy range (0-15) immediately upon domain
creation. This ensures that any VIRQ in this range is always resolvable,
making direct calls to request_irq() for legacy IRQs function correctly.

Signed-off-by: Ming Wang <wangming01@...ngson.cn>
Signed-off-by: Thomas Gleixner <tglx@...utronix.de>
Signed-off-by: Sasha Levin <sashal@...nel.org>
---

LLM Generated explanations, may be completely bogus:

YES: Switching the Loongson PCH-LPC domain from
`irq_domain_create_linear()` to `irq_domain_create_legacy()` removes a
long-standing bug that prevents legacy ISA interrupts from working when
drivers call `request_irq()` directly.

- The current code at `drivers/irqchip/irq-loongson-pch-lpc.c:203`
  creates a linear domain, which only instantiates mappings on demand.
  That leaves IRQs 0‑15 unmapped unless an ACPI GSI translation happens
  first, so `request_irq(4, …)` sees an uninitialized descriptor and
  fails exactly as described in the commit message.
- The fix replaces that call with `irq_domain_create_legacy(…,
  LPC_COUNT, 0, 0, …)` (same location) and adds a clarifying comment.
  Legacy domains eagerly map the whole 0‑15 range, invoking
  `pch_lpc_map()` for each slot so the chip/handler is set up before any
  driver requests the IRQ. This mirrors how other i8259-compatible
  controllers (e.g. `irq-i8259.c`) are wired.
- The regression was introduced when the controller first landed (Fixes:
  ee73f14ee9eb7, v6.0), so every stable kernel carrying Loongson support
  can hit it. Impact is high: legacy ISA drivers or firmware that still
  rely on numeric IRQs can never bind on affected systems.
- The change is tiny, localized to one driver, and does not alter the
  hardware programming model—only the domain helper is swapped. No new
  features, no ABI changes, and it parallels existing upstream practice,
  so regression risk is minimal.

Given the clear user-visible failure mode and the contained, well-
understood fix, this is an excellent candidate for stable backporting.

 drivers/irqchip/irq-loongson-pch-lpc.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/irqchip/irq-loongson-pch-lpc.c b/drivers/irqchip/irq-loongson-pch-lpc.c
index 2d4c3ec128b8f..912bf50a5c7ca 100644
--- a/drivers/irqchip/irq-loongson-pch-lpc.c
+++ b/drivers/irqchip/irq-loongson-pch-lpc.c
@@ -200,8 +200,13 @@ int __init pch_lpc_acpi_init(struct irq_domain *parent,
 		goto iounmap_base;
 	}
 
-	priv->lpc_domain = irq_domain_create_linear(irq_handle, LPC_COUNT,
-					&pch_lpc_domain_ops, priv);
+	/*
+	 * The LPC interrupt controller is a legacy i8259-compatible device,
+	 * which requires a static 1:1 mapping for IRQs 0-15.
+	 * Use irq_domain_create_legacy to establish this static mapping early.
+	 */
+	priv->lpc_domain = irq_domain_create_legacy(irq_handle, LPC_COUNT, 0, 0,
+						    &pch_lpc_domain_ops, priv);
 	if (!priv->lpc_domain) {
 		pr_err("Failed to create IRQ domain\n");
 		goto free_irq_handle;
-- 
2.51.0


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ