*** Add k230 reset management unit support ***
Subsystems (2)
cover *** Add k230 reset management unit support ***
From: Jack Wang <[email protected]>
This series adds support for reset management unit found on k230 soc and successfully wires it up to k230 machine;
The register design are implemented based on k230 trm and linux reset driver (drivers/reset/reset-k230.c)
detailed implementations:
1. write-enable mask for control registers(strobe in the upper 16 bits)
2. hardware auto-clearing for reset and flush request bits
3. latching of Done bits and w1c logic
4. qtest coverage for register read/write behaviors and hardware state modelin
There are some checkpatch style warnings (line length, comment formatting) that will be addressed in v2 based on reviewer feedback.
this is my first time for hardware-unit modeling so apologies for possible mistake. I am apprecitated for any feedback
Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU device
docs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 280 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 6 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +- include/hw/misc/k230_rmu.h | 76 ++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 125 ++++++++++++++++ tests/qtest/meson.build | 3 +- 11 files changed, 504 insertions(+), 4 deletions(-) create mode 100644 hw/misc/k230_rmu.c create mode 100644 include/hw/misc/k230_rmu.h create mode 100644 tests/qtest/k230-rmu-test.c
Signature · 1 line
2.53.0
RFC PATCH 1/2 hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model
From: Jack Wang <[email protected]>
Model the K230 Reset Management Unit (RMU / SYSCTL_RST) at 0x91101000. Its control registers gather the software-controllable reset lines of the SoC peripherals. The register semantics are modelled after the five reset types described by the Linux mainline driver drivers/reset/reset-k230.c:
- CPU0/CPU1: high-half per-bit write-enable strobe plus a done bit (bit12)
- FLUSH: the CPU register bit4, auto-cleared by hardware, no done bit
- HW_DONE: no write-enable; a done bit is latched when the reset fires
and acknowledged by software with write-1-to-clear
- SW_DONE: plain read/write storage, no write-enable and no done bit
QEMU does not reset the target peripherals, so the model collapses the hardware reset latency to zero: writing a reset request bit latches the paired done bit and auto-clears the request bit in the same access, which lets the kernel's readl_poll_timeout() loops succeed on the first read.
A qtest exercises the reset defaults, write-enable gating, HW_DONE latch and write-1-to-clear, repeated resets, flush auto-clear and SW_DONE storage behaviour.
--- hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 280 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 6 + include/hw/misc/k230_rmu.h | 76 ++++++++++ tests/qtest/k230-rmu-test.c | 125 ++++++++++++++++ tests/qtest/meson.build | 3 +- 7 files changed, 493 insertions(+), 1 deletion(-) create mode 100644 hw/misc/k230_rmu.c create mode 100644 include/hw/misc/k230_rmu.h create mode 100644 tests/qtest/k230-rmu-test.c
hw/misc/Kconfig +3 -0
@@ -251,6 +251,9 @@ config DJMEMC config IOSB bool +config K230_RMU + bool + config XLNX_VERSAL_TRNG bool
hw/misc/k230_rmu.c new file +280 -0
280 lines not on this page · raw message
hw/misc/meson.build +1 -0
@@ -111,6 +111,7 @@ system_ss.add(when: 'CONFIG_XLNX_VERSAL', if_true: files( system_ss.add(when: 'CONFIG_XLNX_VERSAL_TRNG', if_true: files( 'xlnx-versal-trng.c', )) +system_ss.add(when: 'CONFIG_K230_RMU', if_true: files('k230_rmu.c')) system_ss.add(when: 'CONFIG_STM32_RCC', if_true: files('stm32_rcc.c')) system_ss.add(when: 'CONFIG_STM32F2XX_SYSCFG', if_true: files('stm32f2xx_syscfg.c')) system_ss.add(when: 'CONFIG_STM32F4XX_SYSCFG', if_true: files('stm32f4xx_syscfg.c'))
hw/misc/trace-events +6 -0
@@ -417,3 +417,9 @@ iommu_testdev_dma_read(uint64_t gva, uint32_t len) "gva=0x%" PRIx64 " len=%u" iommu_testdev_dma_verify(uint32_t expected, uint32_t actual) "expected=0x%x actual=0x%x" iommu_testdev_dma_result(uint32_t result) "DMA completed result=0x%x" iommu_testdev_dma_armed(bool armed) "armed=%d" + +# k230_rmu.c +k230_rmu_read(uint64_t offset, uint32_t value) "K230 RMU read: [0x%" PRIx64 "] -> 0x%08" PRIx32 +k230_rmu_write(uint64_t offset, uint32_t value) "K230 RMU write: [0x%" PRIx64 "] <- 0x%08" PRIx32 +k230_rmu_flush(uint64_t offset) "K230 RMU flush auto-clear at offset 0x%" PRIx64 +k230_rmu_reset_device(void) "K230 RMU device reset"
include/hw/misc/k230_rmu.h new file +76 -0
76 lines not on this page · raw message
tests/qtest/k230-rmu-test.c new file +125 -0
125 lines not on this page · raw message
tests/qtest/meson.build +2 -1
@@ -294,7 +294,8 @@ qtests_riscv64 = ['riscv-csr-test'] + \ (config_all_devices.has_key('CONFIG_IOMMU_TESTDEV') and config_all_devices.has_key('CONFIG_RISCV_IOMMU') ? ['iommu-riscv-test'] : []) + \ - (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test'] : []) + (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test', + 'k230-rmu-test'] : []) qtests_hexagon = ['boot-serial-test']
Signature · 1 line
2.53.0
RFC PATCH 2/2 hw/riscv/k230: wire up the RMU device
In reply to jack wangFrom: Jack Wang <[email protected]>
Replace the "rmu" create_unimplemented_device() stub with a real K230RmuState instance mapped at the K230_DEV_RMU memmap entry, select K230_RMU from the K230 Kconfig, and document the device.
--- docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +++++++--- include/hw/riscv/k230.h | 2 ++ 4 files changed, 11 insertions(+), 3 deletions(-)
docs/system/riscv/k230.rst +1 -0
@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot options
hw/riscv/Kconfig +1 -0
@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMU
hw/riscv/k230.c +7 -3
@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );
@@ -206,6 +207,12 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,
@@ -268,9 +275,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);
include/hw/riscv/k230.h +2 -0
@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)
@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
Signature · 1 line
2.53.0
Hi all,
A polite ping for this RFC series.
Could anyone please take a look when you have a moment? Any feedback or guidance would be greatly appreciated.
Thanks,
Jack Wang
Quoted 45 lines
2026年7月9日 11:12,jack wang <[email protected]> 写道:
From: Jack Wang <[email protected]>
This series adds support for reset management unit found on k230 soc and successfully wires it up to k230 machine;
The register design are implemented based on k230 trm and linux reset driver (
drivers/reset/reset-k230.c)detailed implementations:
1. write-enable mask for control registers(strobe in the upper 16 bits)
2. hardware auto-clearing for reset and flush request bits
3. latching of Done bits and w1c logic
4. qtest coverage for register read/write behaviors and hardware state modelinThere are some checkpatch style warnings (line length, comment formatting) that will be addressed in v2 based on reviewer feedback.
this is my first time for hardware-unit modeling so apologies for possible mistake. I am apprecitated for any feedback
Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU devicedocs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 280 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 6 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +- include/hw/misc/k230_rmu.h | 76 ++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 125 ++++++++++++++++ tests/qtest/meson.build | 3 +- 11 files changed, 504 insertions(+), 4 deletions(-)create mode 100644
hw/misc/k230_rmu.c
create mode 100644include/hw/misc/k230_rmu.h
create mode 100644tests/qtest/k230-rmu-test.cSignature · 1 line
2.53.0
Re: [RFC PATCH 1/2] hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model
In reply to jack wangHi,
Quoted 13 lines · jack wang
From: Jack Wang <[email protected]>
Model the K230 Reset Management Unit (RMU /
SYSCTL_RST) at0x91101000. Its control registers gather the software-controllable reset lines of the SoC peripherals. The register semantics are modelled after the five reset types described by the Linux mainline driverdrivers/reset/reset-k230.c:- CPU0/CPU1: high-half per-bit write-enable strobe plus a done bit (bit12)
- FLUSH: the CPU register bit4, auto-cleared by hardware, no done bit
-HW_DONE: no write-enable; a done bit is latched when the reset fires
and acknowledged by software with write-1-to-clear
-SW_DONE: plain read/write storage, no write-enable and no done bit
Quoted 8 lines
QEMU does not reset the target peripherals, so the model collapses the hardware reset latency to zero: writing a reset request bit latches the paired done bit and auto-clears the request bit in the same access, which lets the kernel's
readl_poll_timeout()loops succeed on the first read.A qtest exercises the reset defaults, write-enable gating,
HW_DONElatch and write-1-to-clear, repeated resets, flush auto-clear andSW_DONEstorage behaviour.
The qtest fails with this patch:
29/352 qtest+qtest-riscv64 - qemu:qtest-riscv64/k230-rmu-test ERROR 0.33s killed by signal 6 SIGABRT
Error being thrown:
# starting QEMU: exec ./qemu-system-riscv64 -qtest unix:/tmp/qtest-3816689.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-3816689.qmp,id=char0 -object monitor-qmp,id=qmp0,chardev=char0 -display none -audio none -run-with exit-with-parent=on -machine k230 -accel qtest ok 1 /riscv64/k230/rmu/reset_values # starting QEMU: exec ./qemu-system-riscv64 -qtest unix:/tmp/qtest-3816689.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-3816689.qmp,id=char0 -object monitor-qmp,id=qmp0,chardev=char0 -display none -audio none -run-with exit-with-parent=on -machine k230 -accel qtest ** ERROR:../tests/qtest/k230-rmu-test.c:50:test_cpu_write_enable: assertion failed (rd(qts, K230_RMU_CPU0_CTRL) == K230_RMU_CPU_DONE): (0x00000000 == 0x00001000) not ok /riscv64/k230/rmu/cpu_we - ERROR:../tests/qtest/k230-rmu-test.c:50:test_cpu_write_enable: assertion failed (rd(qts, K230_RMU_CPU0_CTRL) == K230_RMU_CPU_DONE): (0x00000000 == 0x00001000) Bail out!
If we apply patch 2/2 the qtest succeeds. Which is good, but we aim for all tests being green for all individual patches to ease our lives when bisecting code.
You can either squash patch 2/2 into this one or move the new qtest to patch 2/2 where it'll pass.
The code LGTM aside from this bit:
Quoted 15 lines
Signed-off-by Jack Wang <[email protected]>--- hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 280 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 6 + include/hw/misc/k230_rmu.h | 76 ++++++++++ tests/qtest/k230-rmu-test.c | 125 ++++++++++++++++ tests/qtest/meson.build | 3 +- 7 files changed, 493 insertions(+), 1 deletion(-)create mode 100644
hw/misc/k230_rmu.c
create mode 100644include/hw/misc/k230_rmu.h
create mode 100644tests/qtest/k230-rmu-test.c
[...]
hw/misc/k230_rmu.c line 264
Quoted 6 lines
hw/misc/k230_rmu.c new file +6
-0+static void k230_rmu_class_init (ObjectClass *klass,const void *data) +{ +DeviceClass *dc =DEVICE_CLASS (klass); + + dc->realize = k230_rmu_realize; +device_class_set_legacy_reset (dc, k230_rmu_reset);
This is a deprecated API. If you look at include/hw/core/qdev.h:
---
* device_class_set_legacy_reset(): set the DeviceClass::reset method
* @dc: The device class
* @dev_reset: the reset function
*
* This function sets the DeviceClass::reset method. This is widely
* used in existing code, but new code should prefer to use the
* Resettable API as documented in docs/devel/reset.rst.
---
I suggest reading docs/devel/reset.rst to understand which reset phases might be applicable to this device. Some devices implement just the phases.hold cb, others implement enter and/or exit. Seeing k230_rmu_reset() I think you can get away with implementing it with just phases.hold but don't just take my word for it. Check out the docs and see what makes sense for this RMU.
Thanks,
Daniel
Quoted 269 lines
hw/misc/k230_rmu.c new file +16
-0+ dc->vmsd = &vmstate_k230_rmu; + dc->desc ="K230 Reset Management Unit" ; +} + +static const TypeInfo k230_rmu_info = { + .name =TYPE_K230_RMU , + .parent =TYPE_SYS_BUS_DEVICE , + .instance_size =sizeof (K230RmuState), + .class_init = k230_rmu_class_init, +}; + +static void k230_rmu_register_type (void ) +{ +type_register_static (&k230_rmu_info); +} +type_init (k230_rmu_register_type)hw/misc/meson.build +1
-0@@ -111,6 +111,7 @@ system_ss.add(when: 'CONFIG_XLNX_VERSAL', if_true: files( system_ss.add(when: 'CONFIG_XLNX_VERSAL_TRNG', if_true: files( 'xlnx-versal-trng.c', )) +system_ss.add(when: 'CONFIG_K230_RMU', if_true: files('k230_rmu.c')) system_ss.add(when: 'CONFIG_STM32_RCC', if_true: files('stm32_rcc.c')) system_ss.add(when: 'CONFIG_STM32F2XX_SYSCFG', if_true: files('stm32f2xx_syscfg.c')) system_ss.add(when: 'CONFIG_STM32F4XX_SYSCFG', if_true: files('stm32f4xx_syscfg.c'))hw/misc/trace-events +6
-0@@ -417,3 +417,9 @@ iommu_testdev_dma_read(uint64_t gva, uint32_t len) "gva=0x%" PRIx64 " len=%u" iommu_testdev_dma_verify(uint32_t expected, uint32_t actual) "expected=0x%x actual=0x%x" iommu_testdev_dma_result(uint32_t result) "DMA completed result=0x%x" iommu_testdev_dma_armed(bool armed) "armed=%d" + +# k230_rmu.c +k230_rmu_read(uint64_t offset, uint32_t value) "K230 RMU read: [0x%" PRIx64 "] -> 0x%08" PRIx32 +k230_rmu_write(uint64_t offset, uint32_t value) "K230 RMU write: [0x%" PRIx64 "] <- 0x%08" PRIx32 +k230_rmu_flush(uint64_t offset) "K230 RMU flush auto-clear at offset 0x%" PRIx64 +k230_rmu_reset_device(void) "K230 RMU device reset"include/hw/misc/k230_rmu.h new file +76
-0@@ -0,0 +1,76 @@ +/* +* K230 Reset Management Unit (RMU / SYSCTL_RST) +* +* K230 Technical Reference Manual V0.3.1 (2024-11-18): +* https://github.com/revyos/external-docs/blob/master/K230/en-us/K230_Technical_Reference_Manual_V0.3.1_20241118.pdf +* +* Register semantics cross-checked against the Linux mainline driver +* drivers/reset/reset-k230.c (compatible "canaan,k230-rst"). +* +* Copyright (c) 2026 Jack Wang <[email protected]> +* +* SPDX-License-Identifier: GPL-2.0-or-later +*/ + +#ifndef HW_MISC_K230_RMU_H +#define HW_MISC_K230_RMU_H + +#include "qemu/bitops.h" +#include "hw/core/sysbus.h" +#include "qom/object.h" + +#define TYPE_K230_RMU "riscv.k230.rmu" +OBJECT_DECLARE_SIMPLE_TYPE (K230RmuState,K230_RMU ) + +/* 1 KiB MMIO window, see K230_DEV_RMU in hw/riscv/k230.c. */ +#define K230_RMU_MMIO_SIZE 0x1000 +#define K230_RMU_NUM_REGS (K230_RMU_MMIO_SIZE / 4) + +/* Control register offsets used by drivers/reset/reset-k230.c. */ +#define K230_RMU_CPU0_CTRL 0x04 +#define K230_RMU_CPU1_CTRL 0x0C +#define K230_RMU_AI_CTRL 0x14 +#define K230_RMU_VPU_CTRL 0x1C +#define K230_RMU_PERI0_CTRL 0x20/* SW_DONE bank: TIMERx, WDTx, MAILBOX */ +#define K230_RMU_PERI1_CTRL 0x24/* SW_DONE bank: UARTx, I2Cx, GPIO */ +#define K230_RMU_HISYS_CTRL 0x2C +#define K230_RMU_SDIO_CTRL 0x34 +#define K230_RMU_USB_CTRL 0x3C +#define K230_RMU_SPI_CTRL 0x44 +#define K230_RMU_SEC_CTRL 0x4C +#define K230_RMU_DMA_CTRL 0x54 +#define K230_RMU_DECOMP_CTRL 0x5C +#define K230_RMU_SRAM_CTRL 0x64/* mixed: HW_DONE bits + SW_DONE bit1 */ +#define K230_RMU_NONAI2D_CTRL 0x6C +#define K230_RMU_MCTL_CTRL 0x74 +#define K230_RMU_ISP_CTRL 0x80/* mixed: HW_DONE bits + SW_DONE bits */ +#define K230_RMU_DPU_CTRL 0x88 +#define K230_RMU_DISP_CTRL 0x90 +#define K230_RMU_GPU_CTRL 0x98 +#define K230_RMU_AUDIO_CTRL 0xA4 +#define K230_RMU_SPI2AXI_CTRL 0xA8/* SW_DONE (active high) */ + +/* Bit layout shared by the CPU0/CPU1 control registers. */ +#define K230_RMU_CPU_RESET BIT(0)/* reset request, auto/soft cleared */ +#define K230_RMU_CPU_FLUSH BIT(4)/* L2 flush request, hw auto-clears */ +#define K230_RMU_CPU_DONE BIT(12)/* done bit, write-1-to-clear */ + +/* The high 16 bits are per-bit write-enable strobes (CPU0/CPU1 registers). */ +#define K230_RMU_WE_SHIFT 16 + +struct K230RmuState { +/*< private >*/ +SysBusDevice parent_obj; + +/*< public >*/ +MemoryRegion mmio; + +/* +* One 32-bit word per MMIO offset. Indexing by word offset keeps the +* VMState description trivial. Reset requests auto-clear, so what +* actually persists here is done bits and SW_DONE storage bits. +*/ +uint32_t regs[K230_RMU_NUM_REGS ]; +}; + +#endif /* HW_MISC_K230_RMU_H */ tests/qtest/k230-rmu-test.c new file +125
-0@@ -0,0 +1,125 @@ +/* +* QTest for the K230 Reset Management Unit. +* +* SPDX-License-Identifier: GPL-2.0-or-later +*/ +#include "qemu/osdep.h" +#include "qemu/bitops.h" +#include "libqtest.h" +#include "hw/misc/k230_rmu.h" + +#define K230_RMU_BASE 0x91101000ULL + +static inline uint32_t rd (QTestState *qts,uint64_t off) +{ +return qtest_readl (qts,K230_RMU_BASE + off); +} +static inline void wr (QTestState *qts,uint64_t off,uint32_t val) +{ +qtest_writel (qts,K230_RMU_BASE + off, val); +} + +/* Every known register reads back as 0 after reset. */ +static void test_reset_values (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); + + g_assert_cmphex(rd(qts, K230_RMU_CPU0_CTRL), ==, 0); + g_assert_cmphex(rd(qts, K230_RMU_AI_CTRL), ==, 0); + g_assert_cmphex(rd(qts, K230_RMU_HISYS_CTRL), ==, 0); + g_assert_cmphex(rd(qts, K230_RMU_PERI0_CTRL), ==, 0); + +qtest_quit (qts); +} + +/* +* CPU0: a low-half write without its strobe is ignored; with the strobe the +* reset request fires, latches done (bit12), and the request bit auto-clears. +*/ +static void test_cpu_write_enable (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); + +/* No strobe: the low-half write is dropped, reads back as 0. */ +wr (qts,K230_RMU_CPU0_CTRL ,K230_RMU_CPU_RESET ); + g_assert_cmphex(rd(qts, K230_RMU_CPU0_CTRL), ==, 0); + +/* With strobe (bit16): request bit0 takes effect, done (bit12) latches. */ +wr (qts,K230_RMU_CPU0_CTRL , +K230_RMU_CPU_RESET | (K230_RMU_CPU_RESET <<K230_RMU_WE_SHIFT )); + g_assert_cmphex(rd(qts, K230_RMU_CPU0_CTRL), ==, K230_RMU_CPU_DONE); + +qtest_quit (qts); +} + +/* +* HW_DONE: no write-enable needed. Writing HISYS request bit0 latches its +* done bit (bit4); writing 1 to the done bit clears it. +*/ +static void test_hw_done_and_w1c (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); +uint32_t v; + +wr (qts,K230_RMU_HISYS_CTRL ,BIT (0 ));/* no strobe */ + v =rd (qts,K230_RMU_HISYS_CTRL ); + g_assert_cmphex(v & BIT(4), ==, BIT(4)); /* done latched */ + g_assert_cmphex(v & BIT(0), ==, 0); /* request bit auto-cleared */ + +wr (qts,K230_RMU_HISYS_CTRL ,BIT (4 ));/* write-1-to-clear */ + g_assert_cmphex(rd(qts, K230_RMU_HISYS_CTRL) & BIT(4), ==, 0); + +qtest_quit (qts); +} + +/* Repeated resets: one HW_DONE line can be triggered again and again. */ +static void test_repeat_reset (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); + +for (int i =0 ; i <3 ; i++) { +wr (qts,K230_RMU_AI_CTRL ,BIT (0 ));/* request */ + g_assert_cmphex(rd(qts, K230_RMU_AI_CTRL) & BIT(31), ==, BIT(31)); +wr (qts,K230_RMU_AI_CTRL ,BIT (31 ));/* W1C done */ + g_assert_cmphex(rd(qts, K230_RMU_AI_CTRL) & BIT(31), ==, 0); + } + +qtest_quit (qts); +} + +/* FLUSH: the CPU0 bit4 flush request is auto-cleared, reads back as 0. */ +static void test_flush_auto_clear (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); + +wr (qts,K230_RMU_CPU0_CTRL , +K230_RMU_CPU_FLUSH | (K230_RMU_CPU_FLUSH <<K230_RMU_WE_SHIFT )); + g_assert_cmphex(rd(qts, K230_RMU_CPU0_CTRL) & K230_RMU_CPU_FLUSH, ==, 0); + +qtest_quit (qts); +} + +/* SW_DONE: the PERI0 low half is plain storage, no strobe and no done bit. */ +static void test_sw_done_storage (void ) +{ +QTestState *qts =qtest_init ("-machine k230" ); + +wr (qts,K230_RMU_PERI0_CTRL ,BIT (12 ) |BIT (13 ));/* WDT0 / WDT1 reset bits */ + g_assert_cmphex(rd(qts, K230_RMU_PERI0_CTRL), ==, (BIT(12) | BIT(13))); + +qtest_quit (qts); +} + +int main (int argc,char **argv) +{ +g_test_init (&argc, &argv,NULL ); + +qtest_add_func ("/k230/rmu/reset_values" , test_reset_values); +qtest_add_func ("/k230/rmu/cpu_we" , test_cpu_write_enable); +qtest_add_func ("/k230/rmu/hw_done_w1c" , test_hw_done_and_w1c); +qtest_add_func ("/k230/rmu/repeat_reset" , test_repeat_reset); +qtest_add_func ("/k230/rmu/flush" , test_flush_auto_clear); +qtest_add_func ("/k230/rmu/sw_done" , test_sw_done_storage); + +return g_test_run (); +}tests/qtest/meson.build +2
-1@@ -294,7 +294,8 @@ qtests_riscv64 = ['riscv-csr-test'] + \ (config_all_devices.has_key('CONFIG_IOMMU_TESTDEV') and config_all_devices.has_key('CONFIG_RISCV_IOMMU') ? ['iommu-riscv-test'] : []) + \ - (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test'] : []) + (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test', + 'k230-rmu-test'] : []) qtests_hexagon = ['boot-serial-test']
Re: [RFC PATCH 2/2] hw/riscv/k230: wire up the RMU device
In reply to jack wangHi,
Quoted 5 lines · jack wang
From: Jack Wang <[email protected]>
Replace the "rmu"
create_unimplemented_device()stub with a real K230RmuState instance mapped at theK230_DEV_RMUmemmap entry, selectK230_RMUfrom the K230Kconfig, and document the device.
So, as I said in patch 1, applying this patch will fix the qtest that you introduced there. I believe we want to either squash this patch in patch 1 or move the qtest creation to this patch. Both are fine to me.
Quoted 2 lines
Closes gevico/qemu-camp-2026-k230#11
Not sure what gevico is (seems like a bootcamp, according to Google at least) but the "Closes" tag usually contains a gitlab bug or any other link that is publicly available. It's better to remove it.
Thanks,
Daniel
Quoted 84 lines
Signed-off-by Jack Wang <[email protected]>--- docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +++++++--- include/hw/riscv/k230.h | 2 ++ 4 files changed, 11 insertions(+), 3 deletions(-)docs/system/riscv/k230.rst +1
-0@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot optionshw/riscv/Kconfig +1
-0@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMUhw/riscv/k230.c +7
-3@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );@@ -206,6 +207,12 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,@@ -268,9 +275,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);include/hw/riscv/k230.h +2
-0@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
Re: [RFC PATCH 1/2] hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model
In reply to jack wangHi Wang,
Thanks for working on the K230 RMU model.
I reviewed this series against the K230 TRM and the upstream Linux reset-k230 driver. The series builds and the supplied qtest passes, but I found several issues that should be addressed in v2.
1. The RMU currently acknowledges reset requests without actually resetting the target devices. Please connect at least the existing K230 watchdog devices and add a qtest which verifies that their state is reset through the RMU.
2. The register reset state does not match the TRM. Please initialize each modelled register with its documented reset value and update the reset-value test accordingly.
3. Some register access permissions and reset types are incorrect. In particular, some upper SW_DONE reset lines cannot be changed, reserved bits are writable, and CPU1 is treated as self-clearing even though it requires separate assert and deassert operations.
4. Patch 1 adds k230-rmu-test while the K230 machine still uses the unimplemented RMU stub. K230_RMU is enabled and wired only by patch 2, so patch 1 cannot pass its own test when checked out independently. Please reorder the changes, move the test to patch 2, or squash the two patches.
5. The new source, header and qtest files should be added to the K230 section in MAINTAINERS. Please address the checkpatch errors and warnings before sending v2 as well.
6. Please consider modelling the reset timing registers. At minimum, their reset values and writable fields should be represented. Real-time delays may remain simplified if the zero-latency behavior is documented and preserves guest-visible ordering.
7. Please add coverage for the corrected reset values, CPU1 transitions, upper SW_DONE fields, reserved bits, and reset propagation to the watchdogs.
Regards,
Junze Cao
Re: [RFC PATCH 2/2] hw/riscv/k230: wire up the RMU device
In reply to Daniel Henrique BarbozaOn Thu, Jul 16, 2026 at 02:56:30PM +0800, Daniel Henrique Barboza wrote:
Hi Daniel,
Quoted 14 lines
Hi,
Quoted 5 lines · jack wang
From: Jack Wang <[email protected]>
Replace the "rmu"
create_unimplemented_device()stub with a real K230RmuState instance mapped at theK230_DEV_RMUmemmap entry, selectK230_RMUfrom the K230Kconfig, and document the device.So, as I said in patch 1, applying this patch will fix the qtest that you introduced there. I believe we want to either squash this patch in patch 1 or move the qtest creation to this patch. Both are fine to me.
Quoted 1 line
Quoted 6 lines
Quoted 1 line
Closes gevico/qemu-camp-2026-k230#11Not sure what gevico is (seems like a bootcamp, according to Google at least) but the "Closes" tag usually contains a gitlab bug or any other link that is publicly available. It's better to remove it.
I believe this link refers to:
https://github.com/gevico/qemu-camp-2026-k230/issues/11
However, according to the QEMU development collaboration workflow, we don't need to add it.
Gevico, here is the English transliteration for the Chinese "格维". This is a major open-source community in China focused on technical software, which I currently organize and own.
Within our community, we run a QEMU technical training camp primarily centered on the RISC-V architecture and exploratory work regarding GPGPU microarchitecture.
One of our key projects involves hardware modeling for the K230, where we are implementing a digital satellite-borne computer. The recent contributions seen upstream were primarily patches sent by participants of this project. We are currently in the process of contributing our results from this stage to the QEMU upstream.
If you are interested in this, you can visit the link below.
https://qemu.gevico.online/tutorial/2026/ch3/qemu-k230/
Thanks,
Chao
Quoted 81 lines
Thanks,
DanielQuoted 75 lines
Signed-off-by Jack Wang <[email protected]>--- docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +++++++--- include/hw/riscv/k230.h | 2 ++ 4 files changed, 11 insertions(+), 3 deletions(-)docs/system/riscv/k230.rst +1
-0@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot optionshw/riscv/Kconfig +1
-0@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMUhw/riscv/k230.c +7
-3@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );@@ -206,6 +207,12 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,@@ -268,9 +275,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);include/hw/riscv/k230.h +2
-0@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
Re: [RFC PATCH 2/2] hw/riscv/k230: wire up the RMU device
Quoted 37 lines · Chao Liu
On Thu, Jul 16, 2026 at 02:56:30PM +0800, Daniel Henrique Barboza wrote:
Hi Daniel,
Quoted 14 lines
Hi,
Quoted 5 lines · jack wang
From: Jack Wang <[email protected]>
Replace the "rmu"
create_unimplemented_device()stub with a real K230RmuState instance mapped at theK230_DEV_RMUmemmap entry, selectK230_RMUfrom the K230Kconfig, and document the device.So, as I said in patch 1, applying this patch will fix the qtest that you introduced there. I believe we want to either squash this patch in patch 1 or move the qtest creation to this patch. Both are fine to me.
Quoted 1 line
Quoted 6 lines
Quoted 1 line
Closes gevico/qemu-camp-2026-k230#11Not sure what gevico is (seems like a bootcamp, according to Google at least) but the "Closes" tag usually contains a gitlab bug or any other link that is publicly available. It's better to remove it.
I believe this link refers to:
https://github.com/gevico/qemu-camp-2026-k230/issues/11However, according to the QEMU development collaboration workflow, we don't need to add it.
Gevico, here is the English transliteration for the Chinese "格维". This is a major open-source community in China focused on technical software, which I currently organize and own.
Within our community, we run a QEMU technical training camp primarily centered on the RISC-V architecture and exploratory work regarding GPGPU microarchitecture.
Quoted 5 lines
One of our key projects involves hardware modeling for the K230, where we are implementing a digital satellite-borne computer. The recent contributions seen upstream were primarily patches sent by participants of this project. We are currently in the process of contributing our results from this stage to the QEMU upstream.
Thanks for the info. This seems very interesting indeed, and please keep the QEMU contributions flowing :D
Cheers,
Daniel
Quoted 87 lines
If you are interested in this, you can visit the link below.
https://qemu.gevico.online/tutorial/2026/ch3/qemu-k230/Thanks,
ChaoQuoted 81 lines
Thanks,
DanielQuoted 75 lines
Signed-off-by Jack Wang <[email protected]>--- docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +++++++--- include/hw/riscv/k230.h | 2 ++ 4 files changed, 11 insertions(+), 3 deletions(-)docs/system/riscv/k230.rst +1
-0@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot optionshw/riscv/Kconfig +1
-0@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMUhw/riscv/k230.c +7
-3@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );@@ -206,6 +207,12 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,@@ -268,9 +275,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);include/hw/riscv/k230.h +2
-0@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
Re: [RFC PATCH 2/2] hw/riscv/k230: wire up the RMU device
Quoted 40 lines · Daniel Henrique Barboza
Quoted 37 lines · Chao Liu
On Thu, Jul 16, 2026 at 02:56:30PM +0800, Daniel Henrique Barboza wrote:
Hi Daniel,
Quoted 14 lines
Hi,
Quoted 5 lines · jack wang
From: Jack Wang <[email protected]>
Replace the "rmu"
create_unimplemented_device()stub with a real K230RmuState instance mapped at theK230_DEV_RMUmemmap entry, selectK230_RMUfrom the K230Kconfig, and document the device.So, as I said in patch 1, applying this patch will fix the qtest that you introduced there. I believe we want to either squash this patch in patch 1 or move the qtest creation to this patch. Both are fine to me.
Quoted 1 line
Quoted 6 lines
Quoted 1 line
Closes gevico/qemu-camp-2026-k230#11Not sure what gevico is (seems like a bootcamp, according to Google at least) but the "Closes" tag usually contains a gitlab bug or any other link that is publicly available. It's better to remove it.
I believe this link refers to:
https://github.com/gevico/qemu-camp-2026-k230/issues/11However, according to the QEMU development collaboration workflow, we don't need to add it.
Gevico, here is the English transliteration for the Chinese "格维". This is a major open-source community in China focused on technical software, which I currently organize and own.
Within our community, we run a QEMU technical training camp primarily centered on the RISC-V architecture and exploratory work regarding GPGPU microarchitecture.
Quoted 9 lines
Quoted 5 lines
One of our key projects involves hardware modeling for the K230, where we are implementing a digital satellite-borne computer. The recent contributions seen upstream were primarily patches sent by participants of this project. We are currently in the process of contributing our results from this stage to the QEMU upstream.
Thanks for the info. This seems very interesting indeed, and please keep the QEMU contributions flowing :D
Yes, sir. :)
Thanks,
Chao
Quoted 91 lines
Cheers,
DanielQuoted 87 lines
If you are interested in this, you can visit the link below.
https://qemu.gevico.online/tutorial/2026/ch3/qemu-k230/Thanks,
ChaoQuoted 81 lines
Thanks,
DanielQuoted 75 lines
Signed-off-by Jack Wang <[email protected]>--- docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 10 +++++++--- include/hw/riscv/k230.h | 2 ++ 4 files changed, 11 insertions(+), 3 deletions(-)docs/system/riscv/k230.rst +1
-0@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot optionshw/riscv/Kconfig +1
-0@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMUhw/riscv/k230.c +7
-3@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );@@ -206,6 +207,12 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,@@ -268,9 +275,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);include/hw/riscv/k230.h +2
-0@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
v2 cover hw/riscv/k230: add the K230 Reset Management Unit
In reply to jack wangReferences: <[email protected]>
This v2 patch reply on review of
This series models the Kendryte K230 Reset Management Unit (RMU), a bank of reset-control registers at 0x91101000, and wires it into the K230 SoC. It replaces the create_unimplemented_device("rmu") stub so guest reset drivers (drivers/reset/reset-k230.c) work against a real model.
Patch 1 adds the standalone device model; patch 2 wires it into the SoC and adds the qtest. The qtest is deliberately placed in patch 2 accorinding to suggestion from Daniel: it drives "-machine k230", which only exists once the device is wired up.
Register semantics, reset values and reserved-bit masks are taken from the K230 TRM chapter 2.1 "Reset" (V0.3.1) and cross-checked against the mainline Linux driver.
Changes since v1:
- Switched to the Resettable API (phases.hold instead of the legacy device_class_set_legacy_reset()), per Daniel's review.
- Model real reset propagation: the RMU now cold-resets the two linked watchdogs via "wdt0"/"wdt1" QOM links when their PERI0 reset bits are written (Cao review).
- CPU1 reset request is now a two-step assert/deassert (not self-clearing), matching the hardware (Cao review ).
- Filled in the documented per-register reset values and reserved-bit (writable) masks from the TRM; the *_rst_done status bits reset to 0, so they follow the per-bit "Reset" column rather than the summarised total.
- Added the reset-time-control (*_RST_TIM) registers as plain storage with their documented reset values.
- Fixed the storage mask so the upper half of non-write-enable registers is writable (surfaced by the new TIM registers).
- New qtests: reset values, TIM storage, reserved-bit read-only, CPU1 two-step, and real watchdog reset propagation.
- Moved the qtest from patch 1 to patch 2 for bisectability; added the MAINTAINERS entries.
Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU device
MAINTAINERS | 3 + docs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 425 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 7 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 15 +- include/hw/misc/k230_rmu.h | 126 +++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 225 +++++++++++++++++++ tests/qtest/meson.build | 3 +- 12 files changed, 808 insertions(+), 4 deletions(-) create mode 100644 hw/misc/k230_rmu.c create mode 100644 include/hw/misc/k230_rmu.h create mode 100644 tests/qtest/k230-rmu-test.c
Signature · 1 line
2.53.0
PATCH v2 1/2 hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model
The K230 Reset Management Unit (RMU) is a bank of reset-control registers at 0x91101000. Each register gathers the software-controllable reset lines of a group of peripherals. Register semantics are modelled after the reset types described in the Linux mainline driver drivers/reset/reset-k230.c and cross-checked against the K230 TRM chapter 2.1 "Reset":
- CPU0 : high-half write-enable strobe and a done bit (bit12); the
reset request is self-clearing.
- CPU1 : same layout as CPU0 but the reset request is NOT self-
clearing: software asserts it and later deasserts it as two separate operations.
- FLUSH : bit4 of the CPU registers; hardware auto-clears it, no done.
- HW_DONE : no write-enable; a done bit is latched when the reset fires
and cleared by software (write-1-to-clear).
- SW_DONE : plain read/write storage, no write-enable and no done bit.
- *_RST_TIM : reset-time-control registers; plain read/write storage that comes up with its documented reset value.
For most groups QEMU has no real peripheral to reset, so the model collapses the hardware reset latency to zero: writing a reset request bit latches the matching done bit in the same access and (for self-clearing lines) auto- clears the request bit, which lets the kernel's readl_poll_timeout() loops succeed on the first read. The reset values and reserved-bit masks of every modelled register are taken from the K230 TRM chapter 2.1.5.
An optional pair of "wdt0"/"wdt1" QOM links lets the machine connect the two watchdogs; asserting their PERI0 reset bits then performs a real device_cold_reset() on the linked WDT devices.
This commit adds:
- Device model in hw/misc/k230_rmu.c with per-register pairing tables
- Header with register offsets and bit definitions
- Kconfig and meson.build integration
- Trace events for read/write/reset operations
--- MAINTAINERS | 2 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 425 +++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 7 + include/hw/misc/k230_rmu.h | 126 +++++++++++ 6 files changed, 564 insertions(+) create mode 100644 hw/misc/k230_rmu.c create mode 100644 include/hw/misc/k230_rmu.h
MAINTAINERS +2 -0
@@ -1808,8 +1808,10 @@ M: Chao Liu <[email protected]> L: [email protected] S: Maintained F: docs/system/riscv/k230.rst +F: hw/misc/k230_rmu.c F: hw/riscv/k230.c F: hw/watchdog/k230_wdt.c +F: include/hw/misc/k230_rmu.h F: include/hw/riscv/k230.h F: include/hw/watchdog/k230_wdt.h F: tests/qtest/k230-wdt-test.c
hw/misc/Kconfig +3 -0
@@ -251,6 +251,9 @@ config DJMEMC config IOSB bool +config K230_RMU + bool + config XLNX_VERSAL_TRNG bool
hw/misc/k230_rmu.c new file +425 -0
425 lines not on this page · raw message
hw/misc/meson.build +1 -0
@@ -111,6 +111,7 @@ system_ss.add(when: 'CONFIG_XLNX_VERSAL', if_true: files( system_ss.add(when: 'CONFIG_XLNX_VERSAL_TRNG', if_true: files( 'xlnx-versal-trng.c', )) +system_ss.add(when: 'CONFIG_K230_RMU', if_true: files('k230_rmu.c')) system_ss.add(when: 'CONFIG_STM32_RCC', if_true: files('stm32_rcc.c')) system_ss.add(when: 'CONFIG_STM32F2XX_SYSCFG', if_true: files('stm32f2xx_syscfg.c')) system_ss.add(when: 'CONFIG_STM32F4XX_SYSCFG', if_true: files('stm32f4xx_syscfg.c'))
hw/misc/trace-events +7 -0
@@ -417,3 +417,10 @@ iommu_testdev_dma_read(uint64_t gva, uint32_t len) "gva=0x%" PRIx64 " len=%u" iommu_testdev_dma_verify(uint32_t expected, uint32_t actual) "expected=0x%x actual=0x%x" iommu_testdev_dma_result(uint32_t result) "DMA completed result=0x%x" iommu_testdev_dma_armed(bool armed) "armed=%d" + +# k230_rmu.c +k230_rmu_read(uint64_t offset, uint32_t value) "K230 RMU read: [0x%" PRIx64 "] -> 0x%08" PRIx32 +k230_rmu_write(uint64_t offset, uint32_t value) "K230 RMU write: [0x%" PRIx64 "] <- 0x%08" PRIx32 +k230_rmu_flush(uint64_t offset) "K230 RMU flush auto-clear at offset 0x%" PRIx64 +k230_rmu_reset_device(void) "K230 RMU device reset" +k230_rmu_target_reset(uint64_t offset, unsigned target) "K230 RMU cold-reset target at offset 0x%" PRIx64 " target %u"
include/hw/misc/k230_rmu.h new file +126 -0
126 lines not on this page · raw message
Signature · 1 line
2.53.0
PATCH v2 2/2 hw/riscv/k230: wire up the RMU device
In reply to Jack WangThe previous commit added the K230 RMU device model but left it unconnected. This commit integrates it into the K230 SoC:
- Instantiate K230RmuState in K230SoCState structure
- Realize and memory-map the device at 0x91101000 in k230_soc_realize()
- Connect the two watchdogs via the RMU "wdt0"/"wdt1" QOM links so that a software reset request in PERI0 cold-resets the real WDT devices
- Remove the create_unimplemented_device("rmu") stub
- Add CONFIG_K230_RMU=y selection in hw/riscv/Kconfig
- Update documentation to list RMU in supported devices
- Add the qtest and its MAINTAINERS entry
The qtest lives here rather than with the device model so the series stays
bisectable: it drives "-machine k230", which only exists once the device is wired up. It covers all reset types, the documented reset values, the reserved-bit masks, the CPU1 two-step assert/deassert sequence and the real watchdog reset propagation.
--- MAINTAINERS | 1 + docs/system/riscv/k230.rst | 1 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 15 ++- include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 225 ++++++++++++++++++++++++++++++++++++ tests/qtest/meson.build | 3 +- 7 files changed, 244 insertions(+), 4 deletions(-) create mode 100644 tests/qtest/k230-rmu-test.c
MAINTAINERS +1 -0
@@ -1814,6 +1814,7 @@ F: hw/watchdog/k230_wdt.c F: include/hw/misc/k230_rmu.h F: include/hw/riscv/k230.h F: include/hw/watchdog/k230_wdt.h +F: tests/qtest/k230-rmu-test.c F: tests/qtest/k230-wdt-test.c RX Machines
docs/system/riscv/k230.rst +1 -0
@@ -19,6 +19,7 @@ The ``k230`` machine supports the following devices: * Core Local Interruptor (CLINT) * Platform-Level Interrupt Controller (PLIC) * 2 K230 Watchdog Timer +* K230 Reset Management Unit (RMU) * 5 UART Boot options
hw/riscv/Kconfig +1 -0
@@ -149,3 +149,4 @@ config K230 select SERIAL_MM select UNIMP select K230_WDT + select K230_RMU
hw/riscv/k230.c +12 -3
@@ -110,6 +110,7 @@ static void k230_soc_init(Object *obj)object_initialize_child (obj,"c908-cpu" , cpu0,TYPE_RISCV_HART_ARRAY );object_initialize_child (obj,"k230-wdt0" , &s->wdt[0 ],TYPE_K230_WDT );object_initialize_child (obj,"k230-wdt1" , &s->wdt[1 ],TYPE_K230_WDT ); +object_initialize_child (obj,"k230-rmu" , &s->rmu,TYPE_K230_RMU );qdev_prop_set_uint32 (DEVICE (cpu0),"hartid-base" ,0 );qdev_prop_set_string (DEVICE (cpu0),"cpu-type" ,TYPE_RISCV_CPU_THEAD_C908 );
@@ -206,6 +207,17 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)sysbus_connect_irq (SYS_BUS_DEVICE (&s->wdt[1 ]),0 ,qdev_get_gpio_in (DEVICE (s->c908_plic),K230_WDT1_IRQ )); +/* RMU (reset management unit) */ +/* Link the watchdogs so the RMU can reset them (before realize). */ +object_property_set_link (OBJECT (&s->rmu),"wdt0" , +OBJECT (&s->wdt[0 ]), &error_abort); +object_property_set_link (OBJECT (&s->rmu),"wdt1" , +OBJECT (&s->wdt[1 ]), &error_abort); +if (!sysbus_realize (SYS_BUS_DEVICE (&s->rmu), errp)) { +return ; + } +sysbus_mmio_map (SYS_BUS_DEVICE (&s->rmu),0 , memmap[K230_DEV_RMU ].base); +/* unimplemented devices */ create_unimplemented_device("kpu.l2-cache", memmap[K230_DEV_KPU_L2_CACHE].base,
@@ -268,9 +280,6 @@ static void k230_soc_realize(DeviceState *dev, Error **errp)create_unimplemented_device ("cmu" , memmap[K230_DEV_CMU ].base, memmap[K230_DEV_CMU ].size); -create_unimplemented_device ("rmu" , memmap[K230_DEV_RMU ].base, - memmap[K230_DEV_RMU ].size); -create_unimplemented_device ("boot" , memmap[K230_DEV_BOOT ].base, memmap[K230_DEV_BOOT ].size);
include/hw/riscv/k230.h +2 -0
@@ -18,6 +18,7 @@#include "hw/core/boards.h" #include "hw/riscv/riscv_hart.h" #include "hw/watchdog/k230_wdt.h" +#include "hw/misc/k230_rmu.h" #define C908_CPU_HARTID (0)
@@ -33,6 +34,7 @@ typedef struct K230SoCState {RISCVHartArrayState c908_cpu;/* Small core */ K230WdtState wdt[2 ]; +K230RmuState rmu;MemoryRegion sram;MemoryRegion bootrom;
tests/qtest/k230-rmu-test.c new file +225 -0
225 lines not on this page · raw message
tests/qtest/meson.build +2 -1
@@ -294,7 +294,8 @@ qtests_riscv64 = ['riscv-csr-test'] + \ (config_all_devices.has_key('CONFIG_IOMMU_TESTDEV') and config_all_devices.has_key('CONFIG_RISCV_IOMMU') ? ['iommu-riscv-test'] : []) + \ - (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test'] : []) + (config_all_devices.has_key('CONFIG_K230') ? ['k230-wdt-test', + 'k230-rmu-test'] : []) qtests_hexagon = ['boot-serial-test']
Signature · 1 line
2.53.0
Re: [PATCH v2 0/2] hw/riscv/k230: add the K230 Reset Management Unit
In reply to Jack WangHi all,
A polite ping for this RFC series.
I just fix problems generated by chao liu and Daniel.
Could anyone please take a look when you have a moment? Any feedback or guidance would be greatly appreciated.
Thanks,
Jack Wang
Quoted 63 lines
2026年7月20日 02:02,Jack Wang <[email protected]> 写道:
References: <[email protected]>
This v2 patch reply on review of
This series models the Kendryte K230 Reset Management Unit (RMU), a bank of reset-control registers at0x91101000, and wires it into the K230 SoC. It replaces thecreate_unimplemented_device("rmu")stub so guest reset drivers (drivers/reset/reset-k230.c) work against a real model.Patch 1 adds the standalone device model; patch 2 wires it into the SoC and adds the qtest. The qtest is deliberately placed in patch 2 accorinding to suggestion from Daniel: it drives "-machine k230", which only exists once the device is wired up.
Register semantics, reset values and reserved-bit masks are taken from the K230 TRM chapter 2.1 "Reset" (V0.3.1) and cross-checked against the mainline Linux driver.
Changes since v1:
- Switched to the Resettable API (phases.hold instead of the legacydevice_class_set_legacy_reset()), per Daniel's review.
- Model real reset propagation: the RMU now cold-resets the two linked watchdogs via "wdt0"/"wdt1" QOM links when their PERI0 reset bits are written (Cao review).
- CPU1 reset request is now a two-step assert/deassert (not self-clearing), matching the hardware (Cao review ).
- Filled in the documented per-register reset values and reserved-bit (writable) masks from the TRM; the*_rst_donestatus bits reset to 0, so they follow the per-bit "Reset" column rather than the summarised total.
- Added the reset-time-control (*_RST_TIM) registers as plain storage with their documented reset values.
- Fixed the storage mask so the upper half of non-write-enable registers is writable (surfaced by the new TIM registers).
- New qtests: reset values, TIM storage, reserved-bit read-only, CPU1 two-step, and real watchdog reset propagation.
- Moved the qtest from patch 1 to patch 2 for bisectability; added theMAINTAINERSentries.Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU deviceMAINTAINERS | 3 + docs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 425 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 7 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 15 +- include/hw/misc/k230_rmu.h | 126 +++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 225 +++++++++++++++++++ tests/qtest/meson.build | 3 +- 12 files changed, 808 insertions(+), 4 deletions(-)create mode 100644
hw/misc/k230_rmu.c
create mode 100644include/hw/misc/k230_rmu.h
create mode 100644tests/qtest/k230-rmu-test.cSignature · 1 line
2.53.0
Re: [PATCH v2 0/2] hw/riscv/k230: add the K230 Reset Management Unit
In reply to Jack WangHi all,
A polite ping for this RFC series.
I fix problems raised by chao liu and Daniel. I apologize for previously missing the qemu-riscv Mailing list, it has been added to this ping mail. Any feedback or guidance would be greatly appreciated.
Thanks for your time,
Jack Wang
Quoted 64 lines
References: <[email protected]>
This v2 patch reply on review of
This series models the Kendryte K230 Reset Management Unit (RMU), a bank of reset-control registers at0x91101000, and wires it into the K230 SoC. It replaces thecreate_unimplemented_device("rmu")stub so guest reset drivers (drivers/reset/reset-k230.c) work against a real model.Patch 1 adds the standalone device model; patch 2 wires it into the SoC and adds the qtest. The qtest is deliberately placed in patch 2 accorinding to suggestion from Daniel: it drives "-machine k230", which only exists once the device is wired up.
Register semantics, reset values and reserved-bit masks are taken from the K230 TRM chapter 2.1 "Reset" (V0.3.1) and cross-checked against the mainline Linux driver.
Changes since v1:
- Switched to the Resettable API (phases.hold instead of the legacydevice_class_set_legacy_reset()), per Daniel's review.
- Model real reset propagation: the RMU now cold-resets the two linked watchdogs via "wdt0"/"wdt1" QOM links when their PERI0 reset bits are written (Cao review).
- CPU1 reset request is now a two-step assert/deassert (not self-clearing), matching the hardware (Cao review ).
- Filled in the documented per-register reset values and reserved-bit (writable) masks from the TRM; the*_rst_donestatus bits reset to 0, so they follow the per-bit "Reset" column rather than the summarised total.
- Added the reset-time-control (*_RST_TIM) registers as plain storage with their documented reset values.
- Fixed the storage mask so the upper half of non-write-enable registers is writable (surfaced by the new TIM registers).
- New qtests: reset values, TIM storage, reserved-bit read-only, CPU1 two-step, and real watchdog reset propagation.
- Moved the qtest from patch 1 to patch 2 for bisectability; added theMAINTAINERSentries.Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU deviceMAINTAINERS | 3 + docs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 425 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 7 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 15 +- include/hw/misc/k230_rmu.h | 126 +++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 225 +++++++++++++++++++ tests/qtest/meson.build | 3 +- 12 files changed, 808 insertions(+), 4 deletions(-)create mode 100644
hw/misc/k230_rmu.c
create mode 100644include/hw/misc/k230_rmu.h
create mode 100644tests/qtest/k230-rmu-test.cSignature · 1 line
2.53.0
Re: [PATCH v2 0/2] hw/riscv/k230: add the K230 Reset Management Unit
In reply to Jack WangPing for this patch
Thanks for your time
JW..
Quoted 63 lines
2026年7月20日 02:02,Jack Wang <[email protected]> 写道:
References: <[email protected]>
This v2 patch reply on review of
This series models the Kendryte K230 Reset Management Unit (RMU), a bank of reset-control registers at0x91101000, and wires it into the K230 SoC. It replaces thecreate_unimplemented_device("rmu")stub so guest reset drivers (drivers/reset/reset-k230.c) work against a real model.Patch 1 adds the standalone device model; patch 2 wires it into the SoC and adds the qtest. The qtest is deliberately placed in patch 2 accorinding to suggestion from Daniel: it drives "-machine k230", which only exists once the device is wired up.
Register semantics, reset values and reserved-bit masks are taken from the K230 TRM chapter 2.1 "Reset" (V0.3.1) and cross-checked against the mainline Linux driver.
Changes since v1:
- Switched to the Resettable API (phases.hold instead of the legacydevice_class_set_legacy_reset()), per Daniel's review.
- Model real reset propagation: the RMU now cold-resets the two linked watchdogs via "wdt0"/"wdt1" QOM links when their PERI0 reset bits are written (Cao review).
- CPU1 reset request is now a two-step assert/deassert (not self-clearing), matching the hardware (Cao review ).
- Filled in the documented per-register reset values and reserved-bit (writable) masks from the TRM; the*_rst_donestatus bits reset to 0, so they follow the per-bit "Reset" column rather than the summarised total.
- Added the reset-time-control (*_RST_TIM) registers as plain storage with their documented reset values.
- Fixed the storage mask so the upper half of non-write-enable registers is writable (surfaced by the new TIM registers).
- New qtests: reset values, TIM storage, reserved-bit read-only, CPU1 two-step, and real watchdog reset propagation.
- Moved the qtest from patch 1 to patch 2 for bisectability; added theMAINTAINERSentries.Jack Wang (2):
hw/misc/k230_rmu: add Kendryte K230 Reset Management Unit model hw/riscv/k230: wire up the RMU deviceMAINTAINERS | 3 + docs/system/riscv/k230.rst | 1 + hw/misc/Kconfig | 3 + hw/misc/k230_rmu.c | 425 ++++++++++++++++++++++++++++++++++++ hw/misc/meson.build | 1 + hw/misc/trace-events | 7 + hw/riscv/Kconfig | 1 + hw/riscv/k230.c | 15 +- include/hw/misc/k230_rmu.h | 126 +++++++++++ include/hw/riscv/k230.h | 2 + tests/qtest/k230-rmu-test.c | 225 +++++++++++++++++++ tests/qtest/meson.build | 3 +- 12 files changed, 808 insertions(+), 4 deletions(-)create mode 100644
hw/misc/k230_rmu.c
create mode 100644include/hw/misc/k230_rmu.h
create mode 100644tests/qtest/k230-rmu-test.cSignature · 1 line
2.53.0