[PATCH] riscv: andes: Rearrange Andes PLICSW to single-bit-per-hart strategy

[PATCH] riscv: andes: Rearrange Andes PLICSW to single-bit-per-hart strategy

[Randolph]

Source hart information is not necessary in IPI, so we could
use single-bit-per-hart strategy to rearrange PLICSW mapping.

Bit 0 of Interrupt Pending Bits is hardwired to 0.
Therefore, we use bit 1 to send IPI to hart 0,
bit 2 to hart 1, ..., and so on.

Signed-off-by: Randolph <randolph@andestech.com>
---
 arch/riscv/lib/andes_plicsw.c | 24 +++++++++++-------------
 1 file changed, 11 insertions(+), 13 deletions(-)

diff --git a/arch/riscv/lib/andes_plicsw.c b/arch/riscv/lib/andes_plicsw.c
index 7518408089..6fd49e873b 100644
--- a/arch/riscv/lib/andes_plicsw.c
+++ b/arch/riscv/lib/andes_plicsw.c
@@ -22,7 +22,7 @@
 #include <linux/err.h>
 
 /* pending register */
-#define PENDING_REG(base, hart)	((ulong)(base) + 0x1000 + ((hart) / 4) * 4)
+#define PENDING_REG(base)	((ulong)(base) + 0x1000)
 /* enable register */
 #define ENABLE_REG(base, hart)	((ulong)(base) + 0x2000 + (hart) * 0x80)
 /* claim register */
@@ -30,10 +30,11 @@
 /* priority register */
 #define PRIORITY_REG(base)	((ulong)(base) + PLICSW_PRIORITY_BASE)
 
-#define ENABLE_HART_IPI         (0x01010101)
-#define SEND_IPI_TO_HART(hart)  (0x1 << (hart))
+/* Bit 0 of PLIC-SW pending array is hardwired to zero, so we start from bit 1 */
+#define FIRST_AVAILABLE_BIT	0x2
+#define SEND_IPI_TO_HART(hart)	(FIRST_AVAILABLE_BIT << (hart))
 #define PLICSW_PRIORITY_BASE        0x4
-#define PLICSW_INTERRUPT_PER_HART   0x8
+#define PLICSW_INTERRUPT_PER_HART   0x1
 
 DECLARE_GLOBAL_DATA_PTR;
 
@@ -41,9 +42,8 @@ static int enable_ipi(int hart)
 {
 	unsigned int en;
 
-	en = ENABLE_HART_IPI << hart;
+	en = FIRST_AVAILABLE_BIT << hart;
 	writel(en, (void __iomem *)ENABLE_REG(gd->arch.plicsw, hart));
-	writel(en, (void __iomem *)ENABLE_REG(gd->arch.plicsw + 0x4, hart));
 
 	return 0;
 }
@@ -75,7 +75,7 @@ int riscv_init_ipi(void)
 	ret = uclass_find_first_device(UCLASS_CPU, &dev);
 	if (ret)
 		return ret;
-	else if (!dev)
+	if (!dev)
 		return -ENODEV;
 
 	ofnode_for_each_subnode(node, dev_ofnode(dev->parent)) {
@@ -105,10 +105,9 @@ int riscv_init_ipi(void)
 
 int riscv_send_ipi(int hart)
 {
-	unsigned int ipi = (SEND_IPI_TO_HART(hart) << (8 * gd->arch.boot_hart));
+	unsigned int ipi = SEND_IPI_TO_HART(hart);
 
-	writel(ipi, (void __iomem *)PENDING_REG(gd->arch.plicsw,
-				gd->arch.boot_hart));
+	writel(ipi, (void __iomem *)PENDING_REG(gd->arch.plicsw));
 
 	return 0;
 }
@@ -125,10 +124,9 @@ int riscv_clear_ipi(int hart)
 
 int riscv_get_ipi(int hart, int *pending)
 {
-	unsigned int ipi = (SEND_IPI_TO_HART(hart) << (8 * gd->arch.boot_hart));
+	unsigned int ipi = SEND_IPI_TO_HART(hart);
 
-	*pending = readl((void __iomem *)PENDING_REG(gd->arch.plicsw,
-						     gd->arch.boot_hart));
+	*pending = readl((void __iomem *)PENDING_REG(gd->arch.plicsw));
 	*pending = !!(*pending & ipi);
 
 	return 0;
-- 
2.34.1

Re: [PATCH] riscv: andes: Rearrange Andes PLICSW to single-bit-per-hart strategy

[Leo Liang]

On Thu, Oct 12, 2023 at 01:35:34PM +0800, Randolph wrote:
> Source hart information is not necessary in IPI, so we could
> use single-bit-per-hart strategy to rearrange PLICSW mapping.
> 
> Bit 0 of Interrupt Pending Bits is hardwired to 0.
> Therefore, we use bit 1 to send IPI to hart 0,
> bit 2 to hart 1, ..., and so on.
> 
> Signed-off-by: Randolph <randolph@andestech.com>
> ---
>  arch/riscv/lib/andes_plicsw.c | 24 +++++++++++-------------
>  1 file changed, 11 insertions(+), 13 deletions(-)

Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>