From: Matthew Brost <matthew.brost@intel.com>
To: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: "Anoop, Vijay" <anoop.c.vijay@intel.com>,
<intel-xe@lists.freedesktop.org>, <badal.nilawar@intel.com>,
<rodrigo.vivi@intel.com>, <aravind.iddamsetty@intel.com>,
<riana.tauro@intel.com>, <ravi.kishore.koppuravuri@intel.com>,
<anshuman.gupta@intel.com>, <matthew.d.roper@intel.com>,
<michael.j.ruhl@intel.com>, <paul.e.luse@intel.com>,
<mohamed.mansoor.v@intel.com>, <kam.nasim@intel.com>
Subject: Re: [PATCH v2 1/1] drm/xe/sysctrl: Add system controller component for Xe3p dGPU platforms
Date: Tue, 6 Jan 2026 17:24:48 -0800 [thread overview]
Message-ID: <aV214ECqe28iHfz1@lstrano-desk.jf.intel.com> (raw)
In-Reply-To: <aV1WekOuQFlqdEne@soc-5CG1426VCC.clients.intel.com>
On Tue, Jan 06, 2026 at 10:37:46AM -0800, Umesh Nerlige Ramappa wrote:
> Hi Anoop,
>
> Thanks for incorporating most of the comments here. I have a few more in
> this revision.
>
> On Fri, Jan 02, 2026 at 08:54:50AM -0800, Anoop, Vijay wrote:
> > From: Anoop Vijay <anoop.c.vijay@intel.com>
> >
> > Add a new system controller (sysctrl) component for Intel Xe3p dGPU
> > platforms.
> >
> > This component provides the foundational infrastructure for communication
> > with the System Controller firmware using MKHI protocol over a mailbox
> > interface.
> >
> > Key features introduced:
> > - Detection and initialization of System Controller interface on Xe3p
> > dGPU platforms
> > - Mailbox communication with System Controller firmware
> > - Fragmented message transfer for large command payloads
>
> Let's break down the patches as Matt suggested.
>
> >
> > This implementation establishes the base for future System Controller
> > feature enablement and firmware command handling.
> >
> > Signed-off-by: Anoop Vijay <anoop.c.vijay@intel.com>
> > ---
> > drivers/gpu/drm/xe/Makefile | 2 +
> > drivers/gpu/drm/xe/regs/xe_sysctrl_regs.h | 44 +++
> > drivers/gpu/drm/xe/xe_device.c | 5 +
> > drivers/gpu/drm/xe/xe_device_types.h | 6 +
> > drivers/gpu/drm/xe/xe_pci.c | 2 +
> > drivers/gpu/drm/xe/xe_pci_types.h | 1 +
> > drivers/gpu/drm/xe/xe_sysctrl.c | 62 ++++
> > drivers/gpu/drm/xe/xe_sysctrl.h | 18 +
> > drivers/gpu/drm/xe/xe_sysctrl_mailbox.c | 409 ++++++++++++++++++++++
> > drivers/gpu/drm/xe/xe_sysctrl_mailbox.h | 77 ++++
> > drivers/gpu/drm/xe/xe_sysctrl_types.h | 23 ++
> > 11 files changed, 649 insertions(+)
> > create mode 100644 drivers/gpu/drm/xe/regs/xe_sysctrl_regs.h
> > create mode 100644 drivers/gpu/drm/xe/xe_sysctrl.c
> > create mode 100644 drivers/gpu/drm/xe/xe_sysctrl.h
> > create mode 100644 drivers/gpu/drm/xe/xe_sysctrl_mailbox.c
> > create mode 100644 drivers/gpu/drm/xe/xe_sysctrl_mailbox.h
> > create mode 100644 drivers/gpu/drm/xe/xe_sysctrl_types.h
> >
> > diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
> > index 2b20c79d7ec9..947fbcac65d5 100644
> > --- a/drivers/gpu/drm/xe/Makefile
> > +++ b/drivers/gpu/drm/xe/Makefile
> > @@ -121,6 +121,8 @@ xe-y += xe_bb.o \
> > xe_step.o \
> > xe_survivability_mode.o \
> > xe_sync.o \
> > + xe_sysctrl.o \
> > + xe_sysctrl_mailbox.o \
> > xe_tile.o \
> > xe_tile_sysfs.o \
> > xe_tlb_inval.o \
> > diff --git a/drivers/gpu/drm/xe/regs/xe_sysctrl_regs.h b/drivers/gpu/drm/xe/regs/xe_sysctrl_regs.h
> > new file mode 100644
> > index 000000000000..6627a9c32c4f
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/regs/xe_sysctrl_regs.h
> > @@ -0,0 +1,44 @@
> > +/* SPDX-License-Identifier: MIT */
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#ifndef _XE_SYSCTRL_REGS_H_
> > +#define _XE_SYSCTRL_REGS_H_
> > +
> > +#include "xe_regs.h"
> > +
> > +#define SYSCTRL_BASE_OFFSET 0xDB000
> > +#define SYSCTRL_BASE (SOC_BASE + SYSCTRL_BASE_OFFSET)
> > +#define SYSCTRL_MAILBOX_INDEX 0x03
> > +#define SC_BAR_LENGTH 0x1000
> > +
> > +#define SC_MB_CTRL XE_REG(SYSCTRL_BASE + 0x10)
> > +#define SC_MB_CTRL_RUN_BUSY REG_BIT(31)
> > +#define SC_MB_CTRL_IRQ REG_BIT(30)
> > +#define SC_MB_CTRL_RUN_BUSY_OUT REG_BIT(29)
> > +#define SC_MB_CTRL_PARAM3_MASK REG_GENMASK(28, 24)
> > +#define SC_MB_CTRL_PARAM2_MASK REG_GENMASK(23, 16)
> > +#define SC_MB_CTRL_PARAM1_MASK REG_GENMASK(15, 8)
> > +#define SC_MB_CTRL_COMMAND_MASK REG_GENMASK(7, 0)
> > +
> > +#define SC_MB_DATA0 XE_REG(SYSCTRL_BASE + 0x14)
> > +#define SC_MB_DATA1 XE_REG(SYSCTRL_BASE + 0x18)
> > +#define SC_MB_DATA2 XE_REG(SYSCTRL_BASE + 0x1C)
> > +#define SC_MB_DATA3 XE_REG(SYSCTRL_BASE + 0x20)
> > +
> > +#define MKHI_FRAME_PHASE REG_BIT(24)
> > +#define MKHI_FRAME_CURRENT_MASK REG_GENMASK(21, 16)
> > +#define MKHI_FRAME_TOTAL_MASK REG_GENMASK(13, 8)
> > +#define MKHI_FRAME_COMMAND_MASK REG_GENMASK(7, 0)
> > +
> > +#define SC_MB_FRAME_SIZE 16
> > +#define SC_MB_MAX_FRAMES 64
> > +#define SC_MB_MAX_MESSAGE_SIZE (SC_MB_FRAME_SIZE * SC_MB_MAX_FRAMES)
> > +#define SC_MKHI_COMMAND 5
> > +
> > +#define SC_MB_DEFAULT_TIMEOUT_MS 500
> > +#define SC_MB_RETRY_TIMEOUT_MS 20
> > +#define SC_MB_POLL_INTERVAL_US 100
>
> 1)
> I missed this the last time. Can we change SC to SYSCTRL here for
> everything? I see that for the mailbox, you are using SC_MB, but at the end
> of the day SC still means SYSCTRL and consistently using the same name helps
> readability.
>
> 2)
> The MKHI_FRAME definitions can move to the xe_sysctrl_mailbox.c since that's
> the only code accessing it. See more comments below.
>
>
> > +
> > +#endif /* _XE_SYSCTRL_REGS_H_ */
> > diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> > index e101d290b2a6..805d48dd954d 100644
> > --- a/drivers/gpu/drm/xe/xe_device.c
> > +++ b/drivers/gpu/drm/xe/xe_device.c
> > @@ -66,6 +66,7 @@
> > #include "xe_survivability_mode.h"
> > #include "xe_sriov.h"
> > #include "xe_svm.h"
> > +#include "xe_sysctrl.h"
> > #include "xe_tile.h"
> > #include "xe_ttm_stolen_mgr.h"
> > #include "xe_ttm_sys_mgr.h"
> > @@ -1032,6 +1033,10 @@ int xe_device_probe(struct xe_device *xe)
> > if (err)
> > goto err_unregister_display;
> >
> > + err = xe_sysctrl_init(xe);
> > + if (err)
> > + goto err_unregister_display;
> > +
> > err = xe_device_sysfs_init(xe);
> > if (err)
> > goto err_unregister_display;
> > diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> > index a85be9ba175e..6295b2c35d4a 100644
> > --- a/drivers/gpu/drm/xe/xe_device_types.h
> > +++ b/drivers/gpu/drm/xe/xe_device_types.h
> > @@ -29,6 +29,7 @@
> > #include "xe_sriov_vf_ccs_types.h"
> > #include "xe_step_types.h"
> > #include "xe_survivability_mode_types.h"
> > +#include "xe_sysctrl_types.h"
> > #include "xe_tile_sriov_vf_types.h"
> > #include "xe_validation.h"
> >
> > @@ -340,6 +341,8 @@ struct xe_device {
> > u8 has_soc_remapper_telem:1;
> > /** @info.has_sriov: Supports SR-IOV */
> > u8 has_sriov:1;
> > + /** @info.has_sysctrl: Supports System Controller */
> > + u8 has_sysctrl:1;
> > /** @info.has_usm: Device has unified shared memory support */
> > u8 has_usm:1;
> > /** @info.has_64bit_timestamp: Device supports 64-bit timestamps */
> > @@ -606,6 +609,9 @@ struct xe_device {
> > /** @heci_gsc: graphics security controller */
> > struct xe_heci_gsc heci_gsc;
> >
> > + /** @sc: System Controller */
> > + struct xe_sysctrl sc;
> > +
> > /** @nvm: discrete graphics non-volatile memory */
> > struct intel_dg_nvm_dev *nvm;
> >
> > diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
> > index 91e0553a8163..b6dc3030b673 100644
> > --- a/drivers/gpu/drm/xe/xe_pci.c
> > +++ b/drivers/gpu/drm/xe/xe_pci.c
> > @@ -426,6 +426,7 @@ static const struct xe_device_desc cri_desc = {
> > .has_soc_remapper_sysctrl = true,
> > .has_soc_remapper_telem = true,
> > .has_sriov = true,
> > + .has_sysctrl = true,
> > .max_gt_per_tile = 2,
> > .require_force_probe = true,
> > .va_bits = 57,
> > @@ -701,6 +702,7 @@ static int xe_info_init_early(struct xe_device *xe,
> > xe->info.has_soc_remapper_telem = desc->has_soc_remapper_telem;
> > xe->info.has_sriov = xe_configfs_primary_gt_allowed(to_pci_dev(xe->drm.dev)) &&
> > desc->has_sriov;
> > + xe->info.has_sysctrl = desc->has_sysctrl;
> > xe->info.has_mem_copy_instr = desc->has_mem_copy_instr;
> > xe->info.skip_guc_pc = desc->skip_guc_pc;
> > xe->info.skip_mtcfg = desc->skip_mtcfg;
> > diff --git a/drivers/gpu/drm/xe/xe_pci_types.h b/drivers/gpu/drm/xe/xe_pci_types.h
> > index 5f20f56571d1..53e44a32883d 100644
> > --- a/drivers/gpu/drm/xe/xe_pci_types.h
> > +++ b/drivers/gpu/drm/xe/xe_pci_types.h
> > @@ -56,6 +56,7 @@ struct xe_device_desc {
> > u8 has_soc_remapper_sysctrl:1;
> > u8 has_soc_remapper_telem:1;
> > u8 has_sriov:1;
> > + u8 has_sysctrl:1;
> > u8 needs_scratch:1;
> > u8 skip_guc_pc:1;
> > u8 skip_mtcfg:1;
> > diff --git a/drivers/gpu/drm/xe/xe_sysctrl.c b/drivers/gpu/drm/xe/xe_sysctrl.c
> > new file mode 100644
> > index 000000000000..e0e7b0ecf2bf
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/xe_sysctrl.c
> > @@ -0,0 +1,62 @@
> > +// SPDX-License-Identifier: MIT
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#include <drm/drm_managed.h>
> > +#include <linux/device.h>
> > +#include <linux/mutex.h>
> > +
> > +#include "regs/xe_sysctrl_regs.h"
> > +#include "xe_device.h"
> > +#include "xe_printk.h"
> > +#include "xe_soc_remapper.h"
> > +#include "xe_sysctrl.h"
> > +#include "xe_sysctrl_mailbox.h"
> > +#include "xe_sysctrl_types.h"
> > +
> > +static void xe_sysctrl_fini(void *arg)
> > +{
> > + struct xe_sysctrl *sc = arg;
> > + struct xe_device *xe = sc_to_xe(sc);
>
> Instead of using sc_to_xe here, just pass the xe as arg.
> > +
> > + if (!xe->soc_remapper.set_sysctrl_region)
> > + return;
> > +
> > + xe->soc_remapper.set_sysctrl_region(xe, 0);
> > +}
> > +
> > +/**
> > + * xe_sysctrl_init - Initialize SC subsystem
> > + * @xe: xe device instance
> > + *
> > + * Entry point for SC initialization, called from xe_device_probe().
> > + * This function checks platform support and initializes the system controller.
> > + *
> > + * Return: 0 on success, error code on failure
> > + */
> > +int xe_sysctrl_init(struct xe_device *xe)
> > +{
> > + struct xe_sysctrl *sc = &xe->sc;
> > + int ret;
> > +
> > + if (!xe->info.has_sysctrl)
> > + return 0;
> > +
> > + ret = devm_add_action_or_reset(xe->drm.dev, xe_sysctrl_fini, sc);
> > + if (ret)
> > + return ret;
> > +
> > + if (!xe->soc_remapper.set_sysctrl_region)
> > + return -ENODEV;
> > +
> > + xe->soc_remapper.set_sysctrl_region(xe, SYSCTRL_MAILBOX_INDEX);
> > +
> > + ret = drmm_mutex_init(&xe->drm, &sc->cmd_lock);
> > + if (ret)
> > + return ret;
> > +
> > + xe_sysctrl_mailbox_init(sc);
> > +
> > + return 0;
> > +}
> > diff --git a/drivers/gpu/drm/xe/xe_sysctrl.h b/drivers/gpu/drm/xe/xe_sysctrl.h
> > new file mode 100644
> > index 000000000000..fe90d6577d54
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/xe_sysctrl.h
> > @@ -0,0 +1,18 @@
> > +/* SPDX-License-Identifier: MIT */
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#ifndef _XE_SYSCTRL_H_
> > +#define _XE_SYSCTRL_H_
> > +
> > +struct xe_device;
> > +
> > +static inline struct xe_device *sc_to_xe(struct xe_sysctrl *sc)
> > +{
> > + return container_of(sc, struct xe_device, sc);
> > +}
>
> I would move this to the xe_sysctrl_mailbox.c. Also compile fails here due
> to missing struct xe_sysctrl declaration.
>
> > +
> > +int xe_sysctrl_init(struct xe_device *xe);
> > +
> > +#endif /* _XE_SYSCTRL_H_ */
> > diff --git a/drivers/gpu/drm/xe/xe_sysctrl_mailbox.c b/drivers/gpu/drm/xe/xe_sysctrl_mailbox.c
> > new file mode 100644
> > index 000000000000..940ea535da2e
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/xe_sysctrl_mailbox.c
> > @@ -0,0 +1,409 @@
> > +// SPDX-License-Identifier: MIT
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#include <linux/bitfield.h>
> > +#include <linux/errno.h>
> > +#include <linux/minmax.h>
> > +#include <linux/mutex.h>
> > +#include <linux/slab.h>
> > +#include <linux/string.h>
> > +#include <linux/types.h>
> > +
> > +#include "regs/xe_sysctrl_regs.h"
> > +#include "xe_device.h"
> > +#include "xe_mmio.h"
> > +#include "xe_pm.h"
> > +#include "xe_printk.h"
> > +#include "xe_sysctrl.h"
> > +#include "xe_sysctrl_mailbox.h"
> > +#include "xe_sysctrl_types.h"
> > +
> > +static bool xe_sysctrl_mailbox_wait_bit_clear(struct xe_sysctrl *sc, u32 bit_mask,
> > + unsigned int timeout_ms)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + int ret;
> > +
> > + ret = xe_mmio_wait32_not(mmio, SC_MB_CTRL, bit_mask, bit_mask,
> > + timeout_ms * 1000, NULL, false);
> > +
> > + return ret == 0;
> > +}
> > +
> > +static bool xe_sysctrl_mailbox_wait_bit_set(struct xe_sysctrl *sc, u32 bit_mask,
> > + unsigned int timeout_ms)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + int ret;
> > +
> > + ret = xe_mmio_wait32(mmio, SC_MB_CTRL, bit_mask, bit_mask,
> > + timeout_ms * 1000, NULL, false);
> > +
> > + return ret == 0;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_write_frame(struct xe_sysctrl *sc, const void *frame,
> > + size_t len)
> > +{
> > + static const struct xe_reg regs[] = {
> > + SC_MB_DATA0, SC_MB_DATA1, SC_MB_DATA2, SC_MB_DATA3
> > + };
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + u32 val[SC_MB_FRAME_SIZE / sizeof(u32)] = {0};
> > + u32 dw = DIV_ROUND_UP(len, sizeof(u32));
> > + u32 i;
> > +
> > + memcpy(val, frame, len);
> > +
> > + for (i = 0; i < dw; i++)
> > + xe_mmio_write32(mmio, regs[i], val[i]);
> > +
> > + return 0;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_read_frame(struct xe_sysctrl *sc, void *frame,
> > + size_t len)
> > +{
> > + static const struct xe_reg regs[] = {
> > + SC_MB_DATA0, SC_MB_DATA1, SC_MB_DATA2, SC_MB_DATA3
> > + };
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + u32 val[SC_MB_FRAME_SIZE / sizeof(u32)] = {0};
> > + u32 dw = DIV_ROUND_UP(len, sizeof(u32));
> > + u32 i;
> > +
> > + for (i = 0; i < dw; i++)
> > + val[i] = xe_mmio_read32(mmio, regs[i]);
> > +
> > + memcpy(frame, val, len);
> > +
> > + return 0;
> > +}
> > +
> > +static void xe_sysctrl_mailbox_clear_response(struct xe_sysctrl *sc)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > +
> > + xe_mmio_rmw32(mmio, SC_MB_CTRL, SC_MB_CTRL_RUN_BUSY_OUT, 0);
> > +}
> > +
> > +static int xe_sysctrl_mailbox_prepare_command(struct xe_sysctrl *sc,
> > + u8 group_id, u8 command,
> > + const void *data_in, size_t data_in_len,
> > + u8 **mbox_cmd, size_t *cmd_size)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_sysctrl_mailbox_mkhi_msg_hdr *mkhi_hdr;
> > + size_t size;
> > + u8 *buffer;
> > +
> > + size = sizeof(*mkhi_hdr) + data_in_len;
> > + if (size > SC_MB_MAX_MESSAGE_SIZE) {
> > + xe_err(xe, "sysctrl: Message too large: %zu bytes\n", size);
> > + return -EINVAL;
> > + }
> > +
> > + buffer = kmalloc(size, GFP_KERNEL);
> > + if (!buffer)
> > + return -ENOMEM;
> > +
> > + mkhi_hdr = (struct xe_sysctrl_mailbox_mkhi_msg_hdr *)buffer;
> > + mkhi_hdr->data = cpu_to_le32(FIELD_PREP(MKHI_HDR_GROUP_ID_MASK, group_id) |
> > + FIELD_PREP(MKHI_HDR_COMMAND_MASK, command & 0x7F) |
> > + FIELD_PREP(MKHI_HDR_IS_RESPONSE, 0) |
> > + FIELD_PREP(MKHI_HDR_RESERVED_MASK, 0) |
> > + FIELD_PREP(MKHI_HDR_RESULT_MASK, 0));
> > +
> > + if (data_in && data_in_len)
> > + memcpy(buffer + sizeof(*mkhi_hdr), data_in, data_in_len);
> > +
> > + *mbox_cmd = buffer;
> > + *cmd_size = size;
> > +
> > + return 0;
> > +}
>
> In all the functions above, you do not use the sc pointer at all, so why
> not just pass xe to the functions? If you do that, then in most functions
> below you will not use sc pointer and you can do away with the sc_to_xe
> conversion altogether and pass xe everywhere. Note that you only need the sc
> object in
>
> xe_sysctrl_init()
> xe_sysctrl_mailbox_send_frames()
> xe_sysctrl_mailbox_init()
>
> > +
> > +static int xe_sysctrl_mailbox_send_frames(struct xe_sysctrl *sc, const u8 *mbox_cmd,
> > + size_t cmd_size, unsigned int timeout_ms)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + u32 ctrl_reg, total_frames, frame;
> > + size_t bytes_sent, frame_size;
> > +
> > + total_frames = DIV_ROUND_UP(cmd_size, SC_MB_FRAME_SIZE);
> > +
> > + if (!xe_sysctrl_mailbox_wait_bit_clear(sc, SC_MB_CTRL_RUN_BUSY, timeout_ms)) {
> > + xe_err(xe, "sysctrl: Mailbox busy\n");
> > + return -EBUSY;
> > + }
> > +
> > + sc->phase_bit ^= 1;
> > + bytes_sent = 0;
> > +
> > + for (frame = 0; frame < total_frames; frame++) {
> > + frame_size = min(cmd_size - bytes_sent, (size_t)SC_MB_FRAME_SIZE);
> > +
> > + if (xe_sysctrl_mailbox_write_frame(sc, mbox_cmd + bytes_sent, frame_size)) {
> > + xe_err(xe, "sysctrl: Failed to write frame %u\n", frame);
> > + sc->phase_bit ^= 1;
> > + return -EIO;
> > + }
> > +
> > + ctrl_reg = SC_MB_CTRL_RUN_BUSY |
> > + FIELD_PREP(MKHI_FRAME_CURRENT_MASK, frame) |
> > + FIELD_PREP(MKHI_FRAME_TOTAL_MASK, total_frames - 1) |
> > + FIELD_PREP(MKHI_FRAME_COMMAND_MASK, SC_MKHI_COMMAND) |
> > + (sc->phase_bit ? MKHI_FRAME_PHASE : 0);
> > +
> > + xe_mmio_write32(mmio, SC_MB_CTRL, ctrl_reg);
> > +
> > + if (!xe_sysctrl_mailbox_wait_bit_clear(sc, SC_MB_CTRL_RUN_BUSY, timeout_ms)) {
> > + xe_err(xe, "sysctrl: Frame %u acknowledgment timeout\n", frame);
> > + return -ETIMEDOUT;
> > + }
> > +
> > + bytes_sent += frame_size;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_process_first_frame(struct xe_sysctrl *sc,
> > + const struct xe_sysctrl_mailbox_mkhi_msg_hdr *req,
> > + void *out,
> > + size_t frame_size,
> > + size_t *payload_bytes)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + u32 frame_data[4];
> > + struct xe_sysctrl_mailbox_mkhi_msg_hdr *resp_hdr;
> > + size_t hdr_size = sizeof(*resp_hdr);
> > + size_t payload_size;
> > + int ret;
>
> I would rearrange the variables in decreasing line length. Not a coding
> guideline, but is generally advised to do so.
>
> struct xe_sysctrl_mailbox_mkhi_msg_hdr *resp_hdr;
> struct xe_device *xe = sc_to_xe(sc);
> size_t hdr_size = sizeof(*resp_hdr);
> size_t payload_size;
> u32 frame_data[4];
> int ret;
>
> Also, please avoid hardcoded 4 in frame_data dimension. It could be
> frame_data[SC_MB_FRAME_SIZE / sizeof(u32)]
>
> > +
> > + ret = xe_sysctrl_mailbox_read_frame(sc, frame_data, frame_size);
> > + if (ret)
> > + return ret;
> > +
> > + resp_hdr = (struct xe_sysctrl_mailbox_mkhi_msg_hdr *)frame_data;
> > +
> > + if (!XE_SYSCTRL_MKHI_HDR_IS_RESPONSE(resp_hdr) ||
> > + XE_SYSCTRL_MKHI_HDR_GROUP_ID(resp_hdr) != XE_SYSCTRL_MKHI_HDR_GROUP_ID(req) ||
> > + XE_SYSCTRL_MKHI_HDR_COMMAND(resp_hdr) != XE_SYSCTRL_MKHI_HDR_COMMAND(req)) {
> > + xe_err(xe, "SC: Response header mismatch\n");
> > + return -EPROTO;
> > + }
> > +
> > + if (XE_SYSCTRL_MKHI_HDR_RESULT(resp_hdr) != 0) {
> > + xe_err(xe, "SC: Firmware error: 0x%02lx\n",
> > + XE_SYSCTRL_MKHI_HDR_RESULT(resp_hdr));
> > + return -EIO;
> > + }
> > +
> > + payload_size = frame_size - hdr_size;
> > + if (payload_size > 0)
> > + memcpy(out, (u8 *)frame_data + hdr_size, payload_size);
> > +
> > + *payload_bytes = payload_size;
> > +
> > + xe_sysctrl_mailbox_clear_response(sc);
> > +
> > + return 0;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_process_frame(struct xe_sysctrl *sc,
> > + void *out, size_t frame_size,
> > + unsigned int timeout_ms)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + int ret;
> > +
> > + if (!xe_sysctrl_mailbox_wait_bit_set(sc, SC_MB_CTRL_RUN_BUSY_OUT, timeout_ms)) {
> > + xe_err(xe, "sysctrl: Response frame timeout\n");
> > + return -ETIMEDOUT;
> > + }
> > +
> > + ret = xe_sysctrl_mailbox_read_frame(sc, out, frame_size);
> > + if (ret)
> > + return ret;
> > +
> > + xe_sysctrl_mailbox_clear_response(sc);
> > +
> > + return 0;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_receive_frames(struct xe_sysctrl *sc,
> > + const struct xe_sysctrl_mailbox_mkhi_msg_hdr *req,
> > + void *data_out, size_t data_out_len,
> > + size_t *rdata_len, unsigned int timeout_ms)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + struct xe_sysctrl_mailbox_mkhi_msg_hdr *mkhi_hdr;
> > + u32 ctrl_reg, total_frames, frame;
> > + size_t hdr_size = sizeof(*mkhi_hdr);
> > + u8 *out = data_out;
> > + size_t received = 0;
> > + size_t frame_size;
> > + int ret = 0;
> > +
> > + if (!xe_sysctrl_mailbox_wait_bit_set(sc, SC_MB_CTRL_RUN_BUSY_OUT, timeout_ms)) {
> > + xe_err(xe, "sysctrl: Response frame 0 timeout\n");
> > + return -ETIMEDOUT;
> > + }
> > +
> > + ctrl_reg = xe_mmio_read32(mmio, SC_MB_CTRL);
> > + total_frames = FIELD_GET(MKHI_FRAME_TOTAL_MASK, ctrl_reg) + 1;
> > +
> > + if (total_frames == 1)
> > + frame_size = min(hdr_size + data_out_len, (size_t)SC_MB_FRAME_SIZE);
> > + else
> > + frame_size = SC_MB_FRAME_SIZE;
> > +
> > + ret = xe_sysctrl_mailbox_process_first_frame(sc, req, out, frame_size, &received);
> > + if (ret)
> > + return ret;
> > +
> > + out += received;
> > +
> > + for (frame = 1; frame < total_frames; frame++) {
> > + size_t remaining = data_out_len - received;
> > +
> > + frame_size = min_t(size_t, remaining, SC_MB_FRAME_SIZE);
> > +
> > + ret = xe_sysctrl_mailbox_process_frame(sc, out, frame_size, timeout_ms);
> > + if (ret)
> > + break;
> > +
> > + received += frame_size;
> > + out += frame_size;
> > + }
>
> I fell there should be a simpler way to implement this function. I will get
> back if I can think of anything.
>
> > +
> > + *rdata_len = received;
> > +
> > + return ret;
> > +}
> > +
> > +static int xe_sysctrl_mailbox_send_command(struct xe_sysctrl *sc,
> > + const u8 *mbox_cmd, size_t cmd_size,
> > + void *data_out, size_t data_out_len,
> > + size_t *rdata_len, unsigned int timeout_ms)
> > +{
> > + const struct xe_sysctrl_mailbox_mkhi_msg_hdr *mkhi_hdr;
> > + size_t received;
> > + int ret;
> > +
> > + ret = xe_sysctrl_mailbox_send_frames(sc, mbox_cmd, cmd_size, timeout_ms);
> > + if (ret)
> > + return ret;
> > +
> > + if (!data_out || !rdata_len)
> > + return 0;
> > +
> > + mkhi_hdr = (const struct xe_sysctrl_mailbox_mkhi_msg_hdr *)mbox_cmd;
> > +
> > + ret = xe_sysctrl_mailbox_receive_frames(sc, mkhi_hdr, data_out, data_out_len,
> > + &received, timeout_ms);
> > + if (ret)
> > + return ret;
> > +
> > + *rdata_len = received;
> > +
> > + return 0;
> > +}
> > +
> > +/**
> > + * xe_sysctrl_send_command - Send command to System Controller via mailbox
> > + * @handle: XE device handle
> > + * @cmd_buffer: Pointer to xe_sysctrl_mailbox_command structure
> > + * @rdata_len: Pointer to store actual response data size (can be NULL)
> > + *
> > + * Send a command to the System Controller using MKHI protocol. Handles
> > + * command preparation, fragmentation, transmission, and response reception.
> > + *
> > + * Return: 0 on success, negative error code on failure
> > + */
> > +int xe_sysctrl_send_command(void *handle, void *cmd_buffer, size_t *rdata_len)
> > +{
> > + struct xe_device *xe = handle;
> > + struct xe_sysctrl *sc = &xe->sc;
> > + struct xe_sysctrl_mailbox_command *cmd = cmd_buffer;
> > + u8 *mbox_cmd = NULL;
> > + size_t cmd_size = 0;
> > + u8 group_id, command_code;
> > + int ret = 0;
> > +
> > + if (!xe) {
> > + pr_err("sysctrl: Invalid device handle\n");
> > + return -EINVAL;
> > + }
> > +
> > + if (!cmd) {
> > + xe_err(xe, "sysctrl: Invalid command buffer\n");
> > + return -EINVAL;
> > + }
> > +
> > + if (!xe->info.has_sysctrl)
> > + return -ENODEV;
> > +
> > + group_id = XE_SYSCTRL_APP_HDR_GROUP_ID(&cmd->header);
> > + command_code = XE_SYSCTRL_APP_HDR_COMMAND(&cmd->header);
> > +
> > + if (!cmd->data_in && cmd->data_in_len) {
> > + xe_err(xe, "sysctrl: Invalid input parameters\n");
> > + return -EINVAL;
> > + }
> > +
> > + if (!cmd->data_out && cmd->data_out_len) {
> > + xe_err(xe, "sysctrl: Invalid output parameters\n");
> > + return -EINVAL;
> > + }
> > +
> > + might_sleep();
> > +
> > + ret = xe_sysctrl_mailbox_prepare_command(sc, group_id, command_code,
> > + cmd->data_in, cmd->data_in_len,
> > + &mbox_cmd, &cmd_size);
> > + if (ret) {
> > + xe_err(xe, "sysctrl: Failed to prepare command: %d\n", ret);
> > + return ret;
> > + }
> > +
> > + guard(xe_pm_runtime)(xe);
> > +
> > + guard(mutex)(&sc->cmd_lock);
> > +
> > + ret = xe_sysctrl_mailbox_send_command(sc, mbox_cmd, cmd_size,
> > + cmd->data_out, cmd->data_out_len, rdata_len,
> > + SC_MB_DEFAULT_TIMEOUT_MS);
> > + if (ret)
> > + xe_err(xe, "sysctrl: Mailbox command failed: %d\n", ret);
> > +
> > + kfree(mbox_cmd);
> > +
> > + return ret;
> > +}
> > +
> > +/**
> > + * xe_sysctrl_mailbox_init - Initialize the System Controller mailbox state
> > + * @sc: System controller structure
> > + */
> > +void xe_sysctrl_mailbox_init(struct xe_sysctrl *sc)
> > +{
> > + struct xe_device *xe = sc_to_xe(sc);
> > + struct xe_mmio *mmio = xe_root_tile_mmio(xe);
> > + u32 ctrl_reg;
> > +
> > + ctrl_reg = xe_mmio_read32(mmio, SC_MB_CTRL);
> > + sc->phase_bit = (ctrl_reg & MKHI_FRAME_PHASE) ? 1 : 0;
> > +
> > + xe_mmio_rmw32(mmio, SC_MB_CTRL, MKHI_FRAME_PHASE, 0);
> > +}
> > diff --git a/drivers/gpu/drm/xe/xe_sysctrl_mailbox.h b/drivers/gpu/drm/xe/xe_sysctrl_mailbox.h
> > new file mode 100644
> > index 000000000000..3e472418ebd0
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/xe_sysctrl_mailbox.h
> > @@ -0,0 +1,77 @@
> > +/* SPDX-License-Identifier: MIT */
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#ifndef __XE_SYSCTRL_MAILBOX_H__
> > +#define __XE_SYSCTRL_MAILBOX_H__
> > +
> > +#include <linux/bitfield.h>
> > +#include <linux/types.h>
> > +
> > +struct xe_sysctrl;
> > +
> > +#define MKHI_HDR_GROUP_ID_MASK GENMASK(7, 0)
> > +#define MKHI_HDR_COMMAND_MASK GENMASK(14, 8)
> > +#define MKHI_HDR_IS_RESPONSE BIT(15)
> > +#define MKHI_HDR_RESERVED_MASK GENMASK(23, 16)
> > +#define MKHI_HDR_RESULT_MASK GENMASK(31, 24)
>
> Same comment as (2) above. If the MKHI defines are solely used by
> xe_sysctrl_mailbox.c, then you can just move them there. I assume that the
> app is only using the APP_HDR and is not concerned with the underlying
> protocol/transport.
>
> > +
> > +struct xe_sysctrl_mailbox_mkhi_msg_hdr {
> > + __le32 data;
> > +} __packed;
> > +
> > +#define APP_HDR_GROUP_ID_MASK GENMASK(7, 0)
> > +#define APP_HDR_COMMAND_MASK GENMASK(15, 8)
> > +#define APP_HDR_VERSION_MASK GENMASK(23, 16)
> > +#define APP_HDR_RESERVED_MASK GENMASK(31, 24)
> > +
> > +struct xe_sysctrl_mailbox_app_msg_hdr {
> > + __le32 data;
> > +} __packed;
> > +
> > +#define XE_SYSCTRL_APP_HDR_GROUP_ID(hdr) \
> > + FIELD_GET(APP_HDR_GROUP_ID_MASK, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_APP_HDR_COMMAND(hdr) \
> > + FIELD_GET(APP_HDR_COMMAND_MASK, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_APP_HDR_VERSION(hdr) \
> > + FIELD_GET(APP_HDR_VERSION_MASK, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_MKHI_HDR_GROUP_ID(hdr) \
> > + FIELD_GET(MKHI_HDR_GROUP_ID_MASK, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_MKHI_HDR_COMMAND(hdr) \
> > + FIELD_GET(MKHI_HDR_COMMAND_MASK, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_MKHI_HDR_IS_RESPONSE(hdr) \
> > + FIELD_GET(MKHI_HDR_IS_RESPONSE, le32_to_cpu((hdr)->data))
> > +
> > +#define XE_SYSCTRL_MKHI_HDR_RESULT(hdr) \
> > + FIELD_GET(MKHI_HDR_RESULT_MASK, le32_to_cpu((hdr)->data))
>
> Same here. MKHI defines can move to the C file. Either that or the APP api
> can be moved to a separate header - xe_sysctrl_mailbox_api.h.
>
> Regards,
> Umesh
>
> > +
> > +/**
> > + * struct xe_sysctrl_mailbox_command - System Controller mailbox command structure
> > + */
> > +struct xe_sysctrl_mailbox_command {
> > + /** @header: Application message header containing command information */
> > + struct xe_sysctrl_mailbox_app_msg_hdr header;
> > +
> > + /** @data_in: Pointer to input payload data (can be NULL if no input data) */
> > + void *data_in;
> > +
> > + /** @data_in_len: Size of input payload in bytes (0 if no input data) */
> > + size_t data_in_len;
> > +
> > + /** @data_out: Pointer to output buffer for response data (can be NULL if no response) */
> > + void *data_out;
> > +
> > + /** @data_out_len: Size of output buffer in bytes (0 if no response expected) */
> > + size_t data_out_len;
> > +};
> > +
> > +void xe_sysctrl_mailbox_init(struct xe_sysctrl *sc);
> > +int xe_sysctrl_send_command(void *handle, void *cmd_buffer, size_t *rdata_len);
The above function looks to be unused too.
Also why 'void *handle' rather than xe_device or xe_sysctrl?
Matt
> > +
> > +#endif /* __XE_SYSCTRL_MAILBOX_H__ */
> > diff --git a/drivers/gpu/drm/xe/xe_sysctrl_types.h b/drivers/gpu/drm/xe/xe_sysctrl_types.h
> > new file mode 100644
> > index 000000000000..88a34967688b
> > --- /dev/null
> > +++ b/drivers/gpu/drm/xe/xe_sysctrl_types.h
> > @@ -0,0 +1,23 @@
> > +/* SPDX-License-Identifier: MIT */
> > +/*
> > + * Copyright © 2026 Intel Corporation
> > + */
> > +
> > +#ifndef _XE_SYSCTRL_TYPES_H_
> > +#define _XE_SYSCTRL_TYPES_H_
> > +
> > +#include <linux/mutex.h>
> > +#include <linux/types.h>
> > +
> > +/**
> > + * struct xe_sysctrl - System Controller driver context
> > + */
> > +struct xe_sysctrl {
> > + /** @cmd_lock: Mutex protecting mailbox command operations */
> > + struct mutex cmd_lock;
> > +
> > + /** @phase_bit: MKHI message boundary phase toggle bit */
> > + u32 phase_bit;
> > +};
> > +
> > +#endif /* _XE_SYSCTRL_TYPES_H_ */
> > --
> > 2.43.0
> >
next prev parent reply other threads:[~2026-01-07 1:24 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-01-02 16:54 [PATCH v2 0/1] drm/xe/sysctrl: Add system controller component for Xe3p dGPU platforms Anoop, Vijay
2026-01-02 16:54 ` [PATCH v2 1/1] " Anoop, Vijay
2026-01-02 23:15 ` Matthew Brost
2026-01-05 6:29 ` Riana Tauro
2026-01-06 18:37 ` Umesh Nerlige Ramappa
2026-01-07 1:24 ` Matthew Brost [this message]
2026-01-02 17:01 ` ✗ CI.checkpatch: warning for drm/xe/sysctrl: Add system controller component for Xe3p dGPU platforms (rev2) Patchwork
2026-01-02 17:02 ` ✓ CI.KUnit: success " Patchwork
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=aV214ECqe28iHfz1@lstrano-desk.jf.intel.com \
--to=matthew.brost@intel.com \
--cc=anoop.c.vijay@intel.com \
--cc=anshuman.gupta@intel.com \
--cc=aravind.iddamsetty@intel.com \
--cc=badal.nilawar@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=kam.nasim@intel.com \
--cc=matthew.d.roper@intel.com \
--cc=michael.j.ruhl@intel.com \
--cc=mohamed.mansoor.v@intel.com \
--cc=paul.e.luse@intel.com \
--cc=ravi.kishore.koppuravuri@intel.com \
--cc=riana.tauro@intel.com \
--cc=rodrigo.vivi@intel.com \
--cc=umesh.nerlige.ramappa@intel.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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox