From: Bin Meng <bin.meng@processmission.com>
To: QEMU <qemu-devel@nongnu.org>
Cc: Conor Dooley <conor.dooley@microchip.com>,
Alistair Francis <alistair.francis@wdc.com>,
Chao Liu <chao.liu@processmission.com>,
Conor Dooley <conor@kernel.org>,
Daniel Henrique Barboza <daniel.barboza@oss.qualcomm.com>,
Liu Zhiwei <zhiwei_liu@linux.alibaba.com>,
Palmer Dabbelt <palmer@dabbelt.com>,
Paolo Bonzini <pbonzini@redhat.com>,
Sebastian Huber <sebastian.huber@embedded-brains.de>,
Weiwei Li <liwei1518@gmail.com>,
qemu-riscv@nongnu.org
Subject: [PATCH v4 05/27] hw/riscv: pfsoc: Map the SiFive L2CC memory windows
Date: Wed, 14 Oct 2026 09:22:45 +0800 [thread overview]
Message-ID: <20261014012323.322790-6-bin.meng@processmission.com> (raw)
In-Reply-To: <20261014012323.322790-1-bin.meng@processmission.com>
L2-LIM consists of cache SRAM ways controlled by L2CC WayEnable, and
L2 Zero is the controller's scratchpad allocation interface. Creating
their RAM stand-ins in the SoC and passing a LIM link splits ownership
of the register state and the memory views it controls.
Instantiate the shared SiFive L2CC with 15 masters and the existing
2 MiB modeled Zero aperture. Let the controller own the backing and
both windows, and map its register, LIM, and Zero regions from the SoC.
Remove the external RAM allocations and the l2-lim link.
The device restores the visible LIM range after reset or migration.
L2 Zero keeps its fixed aperture as a RAM approximation of scratchpad
access rather than tracking the remaining LIM capacity.
Signed-off-by: Bin Meng <bin.meng@processmission.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
---
(no changes since v3)
Changes in v3:
- Configure and instantiate the shared SiFive L2CC
- Map the register, LIM, and Zero windows exposed by the controller
- Remove the external RAM allocations and l2-lim link
- Keep the PolarFire master count and modeled Zero aperture unchanged
include/hw/riscv/microchip_pfsoc.h | 2 ++
hw/riscv/microchip_pfsoc.c | 45 ++++++++++--------------------
hw/riscv/Kconfig | 1 +
3 files changed, 17 insertions(+), 31 deletions(-)
diff --git a/include/hw/riscv/microchip_pfsoc.h b/include/hw/riscv/microchip_pfsoc.h
index 612fd2b69c..18f5003850 100644
--- a/include/hw/riscv/microchip_pfsoc.h
+++ b/include/hw/riscv/microchip_pfsoc.h
@@ -27,6 +27,7 @@
#include "hw/cpu/cluster.h"
#include "hw/dma/sifive_pdma.h"
#include "hw/misc/mchp_pfsoc_dmc.h"
+#include "hw/misc/sifive_l2cc.h"
#include "hw/misc/mchp_pfsoc_ioscb.h"
#include "hw/misc/mchp_pfsoc_sysreg.h"
#include "hw/net/cadence_gem.h"
@@ -43,6 +44,7 @@ typedef struct MicrochipPFSoCState {
RISCVHartArrayState e_cpus;
RISCVHartArrayState u_cpus;
DeviceState *plic;
+ SiFiveL2CCState l2cc;
MchpPfSoCDdrSgmiiPhyState ddr_sgmii_phy;
MchpPfSoCDdrCfgState ddr_cfg;
MchpPfSoCIoscbState ioscb;
diff --git a/hw/riscv/microchip_pfsoc.c b/hw/riscv/microchip_pfsoc.c
index 8106dbb3ad..d0d8c8f84c 100644
--- a/hw/riscv/microchip_pfsoc.c
+++ b/hw/riscv/microchip_pfsoc.c
@@ -174,6 +174,12 @@ static void microchip_pfsoc_soc_instance_init(Object *obj)
object_initialize_child(obj, "dma-controller", &s->dma,
TYPE_SIFIVE_PDMA);
+ object_initialize_child(obj, "l2-cache-controller", &s->l2cc,
+ TYPE_SIFIVE_L2CC);
+ qdev_prop_set_uint32(DEVICE(&s->l2cc), "num-masters", 15);
+ qdev_prop_set_uint64(DEVICE(&s->l2cc), "zero-size",
+ microchip_pfsoc_memmap[MICROCHIP_PFSOC_L2ZERO].size);
+
object_initialize_child(obj, "sysreg", &s->sysreg,
TYPE_MCHP_PFSOC_SYSREG);
@@ -200,8 +206,6 @@ static void microchip_pfsoc_soc_realize(DeviceState *dev, Error **errp)
MemoryRegion *system_memory = get_system_memory();
MemoryRegion *rsvd0_mem = g_new(MemoryRegion, 1);
MemoryRegion *e51_dtim_mem = g_new(MemoryRegion, 1);
- MemoryRegion *l2lim_mem = g_new(MemoryRegion, 1);
- MemoryRegion *l2zero_mem = g_new(MemoryRegion, 1);
MemoryRegion *envm_data = g_new(MemoryRegion, 1);
MemoryRegion *qspi_xip_mem = g_new(MemoryRegion, 1);
char *plic_hart_config;
@@ -260,35 +264,14 @@ static void microchip_pfsoc_soc_realize(DeviceState *dev, Error **errp)
RISCV_ACLINT_DEFAULT_MTIMECMP, RISCV_ACLINT_DEFAULT_MTIME,
iks->clint_timebase_freq, false);
- /* L2 cache controller */
- create_unimplemented_device("microchip.pfsoc.l2cc",
- memmap[MICROCHIP_PFSOC_L2CC].base, memmap[MICROCHIP_PFSOC_L2CC].size);
-
- /*
- * Add L2-LIM at reset size.
- * This should be reduced in size as the L2 Cache Controller WayEnable
- * register is incremented. Unfortunately I don't see a nice (or any) way
- * to handle reducing or blocking out the L2 LIM while still allowing it
- * be re returned to all enabled after a reset. For the time being, just
- * leave it enabled all the time. This won't break anything, but will be
- * too generous to misbehaving guests.
- */
- memory_region_init_ram(l2lim_mem, NULL, "microchip.pfsoc.l2lim",
- memmap[MICROCHIP_PFSOC_L2LIM].size, &error_fatal);
- memory_region_add_subregion(system_memory,
- memmap[MICROCHIP_PFSOC_L2LIM].base,
- l2lim_mem);
-
- /*
- * HSS decompresses into the L2 zero-device window and executes there.
- * Model it as RAM because QEMU does not model the backing L2 cache.
- */
- memory_region_init_ram(l2zero_mem, NULL, "microchip.pfsoc.l2zero",
- memmap[MICROCHIP_PFSOC_L2ZERO].size,
- &error_fatal);
- memory_region_add_subregion(system_memory,
- memmap[MICROCHIP_PFSOC_L2ZERO].base,
- l2zero_mem);
+ /* The L2CC owns the SRAM backing and both software-visible windows */
+ sysbus_realize(SYS_BUS_DEVICE(&s->l2cc), errp);
+ sysbus_mmio_map(SYS_BUS_DEVICE(&s->l2cc), SIFIVE_L2CC_MMIO_REGS,
+ memmap[MICROCHIP_PFSOC_L2CC].base);
+ sysbus_mmio_map(SYS_BUS_DEVICE(&s->l2cc), SIFIVE_L2CC_MMIO_LIM,
+ memmap[MICROCHIP_PFSOC_L2LIM].base);
+ sysbus_mmio_map(SYS_BUS_DEVICE(&s->l2cc), SIFIVE_L2CC_MMIO_ZERO,
+ memmap[MICROCHIP_PFSOC_L2ZERO].base);
/* create PLIC hart topology configuration string */
plic_hart_config = riscv_plic_hart_config_string(ms->smp.cpus);
diff --git a/hw/riscv/Kconfig b/hw/riscv/Kconfig
index d06ac26648..46a499f70f 100644
--- a/hw/riscv/Kconfig
+++ b/hw/riscv/Kconfig
@@ -21,6 +21,7 @@ config MICROCHIP_PFSOC
select MCHP_PFSOC_MMUART
select MCHP_PFSOC_SYSREG
select RISCV_ACLINT
+ select SIFIVE_L2CC
select SIFIVE_PDMA
select SIFIVE_PLIC
select UNIMP
--
2.53.0
next prev parent reply other threads:[~2026-10-14 1:24 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-14 1:22 [PATCH v4 00/27] hw/riscv: Restore Microchip PolarFire SoC Icicle Kit firmware boot Bin Meng
2026-10-14 1:22 ` [PATCH v4 01/27] hw/riscv: pfsoc: Correct the L2LIM maximum mapped size Bin Meng
2026-10-14 1:22 ` [PATCH v4 02/27] hw/riscv: pfsoc: Map the L2 zero device window Bin Meng
2026-10-14 1:22 ` [PATCH v4 03/27] hw/misc: pfsoc: Support DDR training with newer HSS Bin Meng
2026-10-14 1:22 ` [PATCH v4 04/27] hw/misc: Add SiFive L2 cache controller model Bin Meng
2026-10-14 1:22 ` Bin Meng [this message]
2026-10-14 1:22 ` [PATCH v4 06/27] hw/riscv: sifive_u: Correct the L2LIM maximum mapped size Bin Meng
2026-10-14 1:22 ` [PATCH v4 07/27] hw/riscv: sifive_u: Use the shared SiFive L2 cache controller Bin Meng
2026-10-14 1:22 ` [PATCH v4 08/27] tests/qtest: Add SiFive L2 cache controller coverage Bin Meng
2026-10-14 1:22 ` [PATCH v4 09/27] hw/sd: sdhci: Run ADMA independently of MMIO Bin Meng
2026-10-14 1:22 ` [PATCH v4 10/27] tests/qtest: add Cadence SDHCI ADMA pacing/MMIO reproducer Bin Meng
2026-10-14 1:22 ` [PATCH v4 11/27] hw/sd: sd: Keep high-capacity memory blocks at 512 bytes Bin Meng
2026-10-14 1:22 ` [PATCH v4 12/27] hw/sd: cadence: Add 64-bit system bus property Bin Meng
2026-10-14 1:22 ` [PATCH v4 13/27] hw/riscv: pfsoc: Enable 64-bit SDHCI system bus support Bin Meng
2026-10-14 1:22 ` [PATCH v4 14/27] hw/rtc: Add PolarFire SoC RTC model Bin Meng
2026-10-14 1:22 ` [PATCH v4 15/27] hw/riscv: pfsoc: Add PolarFire SoC RTC to Icicle Kit Bin Meng
2026-10-14 1:22 ` [PATCH v4 16/27] tests/qtest: Add PolarFire SoC RTC coverage Bin Meng
2026-10-14 1:22 ` [PATCH v4 17/27] hw/misc: pfsoc: Model PolarFire SoC serial number service Bin Meng
2026-10-14 1:22 ` [PATCH v4 18/27] hw/misc: pfsoc: Honor service notification requests Bin Meng
2026-10-14 1:22 ` [PATCH v4 19/27] hw/misc: pfsoc: Let IOSCB own service interrupt state Bin Meng
2026-10-14 1:23 ` [PATCH v4 20/27] hw/riscv: pfsoc: Correct PolarFire SoC DDR aliases Bin Meng
2026-10-14 1:23 ` [PATCH v4 21/27] hw/riscv: pfsoc: Fix Icicle Kit RAM size at 2 GiB Bin Meng
2026-10-14 1:23 ` [PATCH v4 22/27] hw/riscv: pfsoc: Default Icicle Kit to five harts Bin Meng
2026-10-14 1:23 ` [PATCH v4 23/27] docs/system/riscv: Document Icicle Kit HSS boot Bin Meng
2026-10-14 1:23 ` [PATCH v4 24/27] docs/system/riscv: pfsoc: Document CLINT topology for direct Linux boot Bin Meng
2026-10-14 1:23 ` [PATCH v4 25/27] tests/functional/riscv64: Add Icicle Kit firmware boot test Bin Meng
2026-10-14 1:23 ` [PATCH v4 26/27] MAINTAINERS: Add PolarFire SoC Icicle Kit maintainer Bin Meng
2026-10-14 1:23 ` [PATCH v4 27/27] MAINTAINERS: Move Conor Dooley to PolarFire SoC reviewer Bin Meng
2026-10-14 4:16 ` Philippe Mathieu-Daudé
2026-10-14 8:06 ` Daniel P. Berrangé
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20261014012323.322790-6-bin.meng@processmission.com \
--to=bin.meng@processmission.com \
--cc=alistair.francis@wdc.com \
--cc=chao.liu@processmission.com \
--cc=conor.dooley@microchip.com \
--cc=conor@kernel.org \
--cc=daniel.barboza@oss.qualcomm.com \
--cc=liwei1518@gmail.com \
--cc=palmer@dabbelt.com \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=qemu-riscv@nongnu.org \
--cc=sebastian.huber@embedded-brains.de \
--cc=zhiwei_liu@linux.alibaba.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.