[PATCH v10 07/17] media: mali-c55: Add Mali-C55 ISP driver
Laurent Pinchart
laurent.pinchart at ideasonboard.com
Mon Jun 30 01:35:19 PDT 2025
On Mon, Jun 30, 2025 at 10:37:47AM +0300, Sakari Ailus wrote:
> On Sun, Jun 29, 2025 at 09:35:47PM +0300, Laurent Pinchart wrote:
> > On Sat, Jun 28, 2025 at 11:06:54PM +0300, Sakari Ailus wrote:
> > > On 6/24/25 13:21, Daniel Scally wrote:
> >
> > [snip]
> >
> > > > diff --git a/drivers/media/platform/arm/mali-c55/mali-c55-isp.c b/drivers/media/platform/arm/mali-c55/mali-c55-isp.c
> > > > new file mode 100644
> > > > index 0000000000000000000000000000000000000000..20d4d16c75fbf0d5519ecadb5ed1d080bdae05de
> > > > --- /dev/null
> > > > +++ b/drivers/media/platform/arm/mali-c55/mali-c55-isp.c
> > > > @@ -0,0 +1,656 @@
> > > > +// SPDX-License-Identifier: GPL-2.0
> > > > +/*
> > > > + * ARM Mali-C55 ISP Driver - Image signal processor
> > > > + *
> > > > + * Copyright (C) 2024 Ideas on Board Oy
> > >
> > > It's 2025 already.
> > >
> > > > + */
> > > > +
> > > > +#include <linux/delay.h>
> > > > +#include <linux/iopoll.h>
> > > > +#include <linux/property.h>
> > > > +#include <linux/string.h>
> > > > +
> > > > +#include <linux/media/arm/mali-c55-config.h>
> > >
> > > If this is a UAPI header, please include uapi in the path, too.
> > >
> > > Earlier such headers have been under include/uapi/linux, I don't object
> > > putting new ones elsewhere in principle though. Just check with Hans and
> > > Laurent, too... I don't have an opinion yet really.
> >
> > With each new media header we add to include/uapi/linux/, I wish
> > stronger and stronger that we had created include/uapi/linux/media/. We
> > don't have to do it now, my regret will just grow stronger :-)
>
> I'm fine with using include/uapi/linux/media/.
>
> > > > +/* NOT const because the default needs to be filled in at runtime */
> > > > +static struct v4l2_ctrl_config mali_c55_isp_v4l2_custom_ctrls[] = {
> > > > + {
> > > > + .ops = &mali_c55_isp_ctrl_ops,
> > > > + .id = V4L2_CID_MALI_C55_CAPABILITIES,
> > > > + .name = "Mali-C55 ISP Capabilities",
> > > > + .type = V4L2_CTRL_TYPE_BITMASK,
> > > > + .min = 0,
> > > > + .max = MALI_C55_GPS_PONG_FITTED |
> > > > + MALI_C55_GPS_WDR_FITTED |
> > > > + MALI_C55_GPS_COMPRESSION_FITTED |
> > > > + MALI_C55_GPS_TEMPER_FITTED |
> > > > + MALI_C55_GPS_SINTER_LITE_FITTED |
> > > > + MALI_C55_GPS_SINTER_FITTED |
> > > > + MALI_C55_GPS_IRIDIX_LTM_FITTED |
> > > > + MALI_C55_GPS_IRIDIX_GTM_FITTED |
> > > > + MALI_C55_GPS_CNR_FITTED |
> > > > + MALI_C55_GPS_FRSCALER_FITTED |
> > > > + MALI_C55_GPS_DS_PIPE_FITTED,
> > > > + .def = 0,
> > > > + },
> > > > +};
> > > > +
> > > > +static int mali_c55_isp_init_controls(struct mali_c55 *mali_c55)
> > > > +{
> > > > + struct v4l2_ctrl_handler *handler = &mali_c55->isp.handler;
> > > > + struct v4l2_ctrl *capabilities;
> > > > + int ret;
> > > > +
> > > > + ret = v4l2_ctrl_handler_init(handler, 1);
> > > > + if (ret)
> > > > + return ret;
> > > > +
> > > > + mali_c55_isp_v4l2_custom_ctrls[0].def = mali_c55->capabilities;
> > >
> > > The capabilities here are still specific to a device, not global, in
> > > principle at least. Can you move it here, as a local variable?
> > >
> > > > +
> > > > + capabilities = v4l2_ctrl_new_custom(handler,
> > > > + &mali_c55_isp_v4l2_custom_ctrls[0],
> > > > + NULL);
> > > > + if (capabilities)
> > > > + capabilities->flags |= V4L2_CTRL_FLAG_READ_ONLY;
> > > > +
> > > > + if (handler->error) {
> > > > + dev_err(mali_c55->dev, "failed to register capabilities control\n");
> > > > + v4l2_ctrl_handler_free(handler);
> > > > + return handler->error;
> > >
> > > v4l2_ctrl_handler_free() will return the error soon, presumably sooner
> > > than the above code makes it to upstream. Before that, this pattern
> > > won't work as v4l2_ctrl_handler_free() also resets the handler's error
> > > field. :-)
> > >
> > > > diff --git a/drivers/media/platform/arm/mali-c55/mali-c55-registers.h b/drivers/media/platform/arm/mali-c55/mali-c55-registers.h
> > > > new file mode 100644
> > > > index 0000000000000000000000000000000000000000..36a81be0191a15da91809dd2da5d279716f6d725
> > > > --- /dev/null
> > > > +++ b/drivers/media/platform/arm/mali-c55/mali-c55-registers.h
> > > > @@ -0,0 +1,318 @@
> > > > +/* SPDX-License-Identifier: GPL-2.0 */
> > > > +/*
> > > > + * ARM Mali-C55 ISP Driver - Register definitions
> > > > + *
> > > > + * Copyright (C) 2024 Ideas on Board Oy
> > > > + */
> > > > +
> > > > +#ifndef _MALI_C55_REGISTERS_H
> > > > +#define _MALI_C55_REGISTERS_H
> > > > +
> > > > +#include <linux/bits.h>
> > > > +
> > > > +/* ISP Common 0x00000 - 0x000ff */
> > > > +
> > > > +#define MALI_C55_REG_API 0x00000
> > > > +#define MALI_C55_REG_PRODUCT 0x00004
> > > > +#define MALI_C55_REG_VERSION 0x00008
> > > > +#define MALI_C55_REG_REVISION 0x0000c
> > > > +#define MALI_C55_REG_PULSE_MODE 0x0003c
> > > > +#define MALI_C55_REG_INPUT_MODE_REQUEST 0x0009c
> > > > +#define MALI_C55_INPUT_SAFE_STOP 0x00
> > > > +#define MALI_C55_INPUT_SAFE_START 0x01
> > > > +#define MALI_C55_REG_MODE_STATUS 0x000a0
> > > > +#define MALI_C55_REG_INTERRUPT_MASK_VECTOR 0x00030
> > > > +#define MALI_C55_INTERRUPT_MASK_ALL GENMASK(31, 0)
> > > > +
> > > > +#define MALI_C55_REG_GLOBAL_MONITOR 0x00050
> > > > +
> > > > +#define MALI_C55_REG_GEN_VIDEO 0x00080
> > > > +#define MALI_C55_REG_GEN_VIDEO_ON_MASK BIT(0)
> > > > +#define MALI_C55_REG_GEN_VIDEO_MULTI_MASK BIT(1)
> > > > +#define MALI_C55_REG_GEN_PREFETCH_MASK GENMASK(31, 16)
> > > > +
> > > > +#define MALI_C55_REG_MCU_CONFIG 0x00020
> > > > +#define MALI_C55_REG_MCU_CONFIG_OVERRIDE_MASK BIT(0)
> > > > +#define MALI_C55_REG_MCU_CONFIG_WRITE_MASK BIT(1)
> > > > +#define MALI_C55_MCU_CONFIG_WRITE(x) ((x) << 1)
> > >
> > > Is x unsigned?
> >
> > Does it matter ? The reason why the BIT() macro uses (UL(1) << (nr))
> > instead of (1 << (nr)) is (if I'm not mistaken) to avoid incorrect
> > handling of bit 31. As long as x doesn't take negative values and
> > doesn't extend to bit 31, it should be fine.
>
> For that reason exactly. If you're unsure, maybe at least cast it as
> unsigned?
The MCU_CONFIG_WRITE field is on bit long, so there's no practical
issue. I also don't expect the driver to pass a negative value. We could
still add an unsigned cast if there's a general rule that all register
fields must be cast to unsigned. This isn't done in the vast majority of
drivers though.
> > > > +#define MALI_C55_REG_MCU_CONFIG_WRITE_PING BIT(1)
> > > > +#define MALI_C55_REG_MCU_CONFIG_WRITE_PONG 0x00
> > > > +#define MALI_C55_REG_MULTI_CONTEXT_MODE_MASK BIT(8)
> > > > +#define MALI_C55_REG_PING_PONG_READ 0x00024
> > > > +#define MALI_C55_REG_PING_PONG_READ_MASK BIT(2)
> > > > +#define MALI_C55_INTERRUPT_BIT(x) BIT(x)
> > > > +
> > > > +#define MALI_C55_REG_GLOBAL_PARAMETER_STATUS 0x00068
> > > > +#define MALI_C55_GPS_PONG_FITTED BIT(0)
> > > > +#define MALI_C55_GPS_WDR_FITTED BIT(1)
> > > > +#define MALI_C55_GPS_COMPRESSION_FITTED BIT(2)
> > > > +#define MALI_C55_GPS_TEMPER_FITTED BIT(3)
> > > > +#define MALI_C55_GPS_SINTER_LITE_FITTED BIT(4)
> > > > +#define MALI_C55_GPS_SINTER_FITTED BIT(5)
> > > > +#define MALI_C55_GPS_IRIDIX_LTM_FITTED BIT(6)
> > > > +#define MALI_C55_GPS_IRIDIX_GTM_FITTED BIT(7)
> > > > +#define MALI_C55_GPS_CNR_FITTED BIT(8)
> > > > +#define MALI_C55_GPS_FRSCALER_FITTED BIT(9)
> > > > +#define MALI_C55_GPS_DS_PIPE_FITTED BIT(10)
> > > > +
> > > > +#define MALI_C55_REG_BLANKING 0x00084
> > > > +#define MALI_C55_REG_HBLANK_MASK GENMASK(15, 0)
> > > > +#define MALI_C55_REG_VBLANK_MASK GENMASK(31, 16)
> > > > +#define MALI_C55_VBLANK(x) ((x) << 16)
> > >
> > > Same question for the bit shifts left elsewhere in the header.
> > >
> > > > +
> > > > +#define MALI_C55_REG_HC_START 0x00088
> > > > +#define MALI_C55_HC_START(h) (((h) & 0xffff) << 16)
> > > > +#define MALI_C55_REG_HC_SIZE 0x0008c
> > > > +#define MALI_C55_HC_SIZE(h) ((h) & 0xffff)
> > > > +#define MALI_C55_REG_VC_START_SIZE 0x00094
> > > > +#define MALI_C55_VC_START(v) ((v) & 0xffff)
> > > > +#define MALI_C55_VC_SIZE(v) (((v) & 0xffff) << 16)
> > > > +
> > > > +/* Ping/Pong Configuration Space */
> > > > +#define MALI_C55_REG_BASE_ADDR 0x18e88
> > > > +#define MALI_C55_REG_BYPASS_0 0x18eac
> > > > +#define MALI_C55_REG_BYPASS_0_VIDEO_TEST BIT(0)
> > > > +#define MALI_C55_REG_BYPASS_0_INPUT_FMT BIT(1)
> > > > +#define MALI_C55_REG_BYPASS_0_DECOMPANDER BIT(2)
> > > > +#define MALI_C55_REG_BYPASS_0_SENSOR_OFFSET_WDR BIT(3)
> > > > +#define MALI_C55_REG_BYPASS_0_GAIN_WDR BIT(4)
> > > > +#define MALI_C55_REG_BYPASS_0_FRAME_STITCH BIT(5)
> > > > +#define MALI_C55_REG_BYPASS_1 0x18eb0
> > > > +#define MALI_C55_REG_BYPASS_1_DIGI_GAIN BIT(0)
> > > > +#define MALI_C55_REG_BYPASS_1_FE_SENSOR_OFFS BIT(1)
> > > > +#define MALI_C55_REG_BYPASS_1_FE_SQRT BIT(2)
> > > > +#define MALI_C55_REG_BYPASS_1_RAW_FE BIT(3)
> > > > +#define MALI_C55_REG_BYPASS_2 0x18eb8
> > > > +#define MALI_C55_REG_BYPASS_2_SINTER BIT(0)
> > > > +#define MALI_C55_REG_BYPASS_2_TEMPER BIT(1)
> > > > +#define MALI_C55_REG_BYPASS_3 0x18ebc
> > > > +#define MALI_C55_REG_BYPASS_3_SQUARE_BE BIT(0)
> > > > +#define MALI_C55_REG_BYPASS_3_SENSOR_OFFSET_PRE_SH BIT(1)
> > > > +#define MALI_C55_REG_BYPASS_3_MESH_SHADING BIT(3)
> > > > +#define MALI_C55_REG_BYPASS_3_WHITE_BALANCE BIT(4)
> > > > +#define MALI_C55_REG_BYPASS_3_IRIDIX BIT(5)
> > > > +#define MALI_C55_REG_BYPASS_3_IRIDIX_GAIN BIT(6)
> > > > +#define MALI_C55_REG_BYPASS_4 0x18ec0
> > > > +#define MALI_C55_REG_BYPASS_4_DEMOSAIC_RGB BIT(1)
> > > > +#define MALI_C55_REG_BYPASS_4_PF_CORRECTION BIT(3)
> > > > +#define MALI_C55_REG_BYPASS_4_CCM BIT(4)
> > > > +#define MALI_C55_REG_BYPASS_4_CNR BIT(5)
> > > > +#define MALI_C55_REG_FR_BYPASS 0x18ec4
> > > > +#define MALI_C55_REG_DS_BYPASS 0x18ec8
> > > > +#define MALI_C55_BYPASS_CROP BIT(0)
> > > > +#define MALI_C55_BYPASS_SCALER BIT(1)
> > > > +#define MALI_C55_BYPASS_GAMMA_RGB BIT(2)
> > > > +#define MALI_C55_BYPASS_SHARPEN BIT(3)
> > > > +#define MALI_C55_BYPASS_CS_CONV BIT(4)
> > > > +#define MALI_C55_REG_ISP_RAW_BYPASS 0x18ecc
> > > > +#define MALI_C55_ISP_RAW_BYPASS_BYPASS_MASK BIT(0)
> > > > +#define MALI_C55_ISP_RAW_BYPASS_FR_BYPASS_MASK GENMASK(9, 8)
> > > > +#define MALI_C55_ISP_RAW_BYPASS_RAW_FR_BYPASS (2 << 8)
> > > > +#define MALI_C55_ISP_RAW_BYPASS_RGB_FR_BYPASS (1 << 8)
> > >
> > > BIT() or make these unsigned.
> >
> > It's a 2 bits field, BIT() isn't appropriate.
>
> That leaves us with the other alternative, doesn't it?
Absolutely, although unsigned is unnecessary in this case as we're not
approaching bit 31.
--
Regards,
Laurent Pinchart
More information about the linux-arm-kernel
mailing list