Commit 4785aa802853 ("cpuidle, ACPI: Evaluate LPI arch_flags for
broadcast timer") replaced the generic nonzero check for LPI
architectural context loss flags with arch_get_idle_state_flags().
RISC-V does not implement the helper, so it falls back to the stub
that returns 0. Consequently, CPUIDLE_FLAG_TIMER_STOP is not set when
an LPI state loses the hart timer context, preventing cpuidle from
using a broadcast timer for that state.

Implement the RISC-V helper and map the hart timer context loss flag
to CPUIDLE_FLAG_TIMER_STOP.

Fixes: 4785aa802853 ("cpuidle, ACPI: Evaluate LPI arch_flags for broadcast timer")
Cc: stable@vger.kernel.org
Acked-by: Sudeep Holla <sudeep.holla@kernel.org>
Reviewed-by: Yixun Lan <dlan@kernel.org>
Reviewed-by: Sunil V L <sunilvl@oss.qualcomm.com>
Reviewed-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Peixin Xie <peixin.xie@linux.spacemit.com>
---
Changes in v3:
- Use the PDF specification numbering, Chapter 3.1.2 and Table 4,
  following Sunil's clarification.
- Link to v2: https://patch.msgid.link/20260728-riscv-acpi-lpi-timer-v2-1-5a30e0021ec9@linux.spacemit.com

Changes in v2:
- Correct the FFH specification reference to Chapter 2.1.1.2,
  Table 3, as suggested by Yixun.
- Link to v1: https://patch.msgid.link/20260728-riscv-acpi-lpi-timer-v1-1-c722bce01e48@linux.spacemit.com
---
 arch/riscv/include/asm/acpi.h | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/arch/riscv/include/asm/acpi.h b/arch/riscv/include/asm/acpi.h
index 26ab37c171bc..f4db04f5bbe7 100644
--- a/arch/riscv/include/asm/acpi.h
+++ b/arch/riscv/include/asm/acpi.h
@@ -12,6 +12,8 @@
 #ifndef _ASM_ACPI_H
 #define _ASM_ACPI_H
 
+#include <linux/cpuidle.h>
+
 /* Basic configuration for ACPI */
 #ifdef CONFIG_ACPI
 
@@ -67,6 +69,23 @@ int acpi_get_riscv_isa(struct acpi_table_header *table,
 
 void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
 			     u32 *cboz_size, u32 *cbop_size);
+
+/*
+ * RISC-V Functional Fixed Hardware Specification Version v1.0.1,
+ * Chapter 3.1.2, Table 4: Arch. Context Lost Flags
+ */
+#define RISCV_LPI_HART_TIMER_CTXT_LOST		BIT(0)
+
+static inline unsigned int arch_get_idle_state_flags(u32 arch_flags)
+{
+	if (arch_flags & RISCV_LPI_HART_TIMER_CTXT_LOST)
+		return CPUIDLE_FLAG_TIMER_STOP;
+
+	return 0;
+}
+
+#define arch_get_idle_state_flags arch_get_idle_state_flags
+
 #else
 static inline void acpi_init_rintc_map(void) { }
 static inline struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)

---
base-commit: 1d6f0a217c662693d2307ee06851cb4bfcbf9529
change-id: 20260727-riscv-acpi-lpi-timer-6f1739a6c831

Best regards,
--  
Peixin Xie <peixin.xie@linux.spacemit.com>

On Mon, 3 Aug 2026, Peixin Xie wrote:

> Commit 4785aa802853 ("cpuidle, ACPI: Evaluate LPI arch_flags for
> broadcast timer") replaced the generic nonzero check for LPI
> architectural context loss flags with arch_get_idle_state_flags().
> RISC-V does not implement the helper, so it falls back to the stub
> that returns 0. Consequently, CPUIDLE_FLAG_TIMER_STOP is not set when
> an LPI state loses the hart timer context, preventing cpuidle from
> using a broadcast timer for that state.
> 
> Implement the RISC-V helper and map the hart timer context loss flag
> to CPUIDLE_FLAG_TIMER_STOP.
> 
> Fixes: 4785aa802853 ("cpuidle, ACPI: Evaluate LPI arch_flags for broadcast timer")
> Cc: stable@vger.kernel.org
> Acked-by: Sudeep Holla <sudeep.holla@kernel.org>
> Reviewed-by: Yixun Lan <dlan@kernel.org>
> Reviewed-by: Sunil V L <sunilvl@oss.qualcomm.com>
> Reviewed-by: Huisong Li <lihuisong@huawei.com>
> Signed-off-by: Peixin Xie <peixin.xie@linux.spacemit.com>

Thanks, queued for v7.3.


- Paul