Message-ID: <508d18e5-3479-4cce-9b61-a40b91491f1f@suse.de>
Date: Tue, 4 Feb 2025 09:55:42 +0100
From: Thomas Zimmermann <tzimmermann@...e.de>
To: Gwenael Georgeault <ggeorgea@...rox.com>,
 David Airlie <airlied@...hat.com>, Jocelyn Falempe <jfalempe@...hat.com>,
 dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] drm/mgag200: Added support for the new device G200eH5

Hi


Am 03.02.25 um 14:07 schrieb Gwenael Georgeault:

Thanks for sending this patch and welcome to the kernel community

> - Added the new device ID
> - Added new pll algorithm
>
> Co-authored-by: Mamadou Insa Diop <mdiop@...rox.com>
> ---
>  drivers/gpu/drm/mgag200/Makefile          |   1 +
>  drivers/gpu/drm/mgag200/mgag200_drv.c     |   4 +
>  drivers/gpu/drm/mgag200/mgag200_drv.h     |   7 +-
>  drivers/gpu/drm/mgag200/mgag200_g200eh5.c | 212 ++++++++++++++++++++++

Great, all new code located in a single file. That's how it is intended.

The code looks correct, but I have plenty of comments on the style.


>  4 files changed, 222 insertions(+), 2 deletions(-)
>  create mode 100644 drivers/gpu/drm/mgag200/mgag200_g200eh5.c
>
> diff --git a/drivers/gpu/drm/mgag200/Makefile 
> b/drivers/gpu/drm/mgag200/Makefile
> index 5a02203fad12..94f063c8722a 100644
> --- a/drivers/gpu/drm/mgag200/Makefile
> +++ b/drivers/gpu/drm/mgag200/Makefile
> @@ -6,6 +6,7 @@ mgag200-y := \
>      mgag200_g200.o \
>      mgag200_g200eh.o \
>      mgag200_g200eh3.o \
> +    mgag200_g200eh5.o \
>      mgag200_g200er.o \
>      mgag200_g200ev.o \
>      mgag200_g200ew3.o \
> diff --git a/drivers/gpu/drm/mgag200/mgag200_drv.c 
> b/drivers/gpu/drm/mgag200/mgag200_drv.c
> index 069fdd2dc8f6..1c257f5b5136 100644
> --- a/drivers/gpu/drm/mgag200/mgag200_drv.c
> +++ b/drivers/gpu/drm/mgag200/mgag200_drv.c
> @@ -214,6 +214,7 @@ static const struct pci_device_id 
> mgag200_pciidlist[] = {
>      { PCI_VENDOR_ID_MATROX, 0x534, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 
> G200_ER },
>      { PCI_VENDOR_ID_MATROX, 0x536, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 
> G200_EW3 },
>      { PCI_VENDOR_ID_MATROX, 0x538, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 
> G200_EH3 },
> +    { PCI_VENDOR_ID_MATROX, 0x53A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 
> G200_EH5 },
>      {0,}
>  };
>
> @@ -256,6 +257,9 @@ mgag200_pci_probe(struct pci_dev *pdev, const 
> struct pci_device_id *ent)
>      case G200_EH3:
>          mdev = mgag200_g200eh3_device_create(pdev, &mgag200_driver);
>          break;
> +    case G200_EH5:
> +        mdev = mgag200_g200eh5_device_create(pdev, &mgag200_driver);
> +        break;
>      case G200_ER:
>          mdev = mgag200_g200er_device_create(pdev, &mgag200_driver);
>          break;
> diff --git a/drivers/gpu/drm/mgag200/mgag200_drv.h 
> b/drivers/gpu/drm/mgag200/mgag200_drv.h
> index 0608fc63e588..065ba09d109b 100644
> --- a/drivers/gpu/drm/mgag200/mgag200_drv.h
> +++ b/drivers/gpu/drm/mgag200/mgag200_drv.h
> @@ -196,6 +196,7 @@ enum mga_type {
>      G200_EV,
>      G200_EH,
>      G200_EH3,
> +    G200_EH5,
>      G200_ER,
>      G200_EW3,
>  };
> @@ -333,11 +334,13 @@ void mgag200_g200eh_pixpllc_atomic_update(struct 
> drm_crtc *crtc, struct drm_atom
>  struct mga_device *mgag200_g200eh_device_create(struct pci_dev *pdev,
>                          const struct drm_driver *drv);
>  struct mga_device *mgag200_g200eh3_device_create(struct pci_dev *pdev,
> -                         const struct drm_driver *drv);
> +                        const struct drm_driver *drv);
> +struct mga_device *mgag200_g200eh5_device_create(struct pci_dev *pdev,
> +                        const struct drm_driver *drv);
>  struct mga_device *mgag200_g200er_device_create(struct pci_dev *pdev,
>                          const struct drm_driver *drv);
>  struct mga_device *mgag200_g200ew3_device_create(struct pci_dev *pdev,
> -                         const struct drm_driver *drv);
> +                        const struct drm_driver *drv);
>
>  /*
>   * mgag200_mode.c
> diff --git a/drivers/gpu/drm/mgag200/mgag200_g200eh5.c 
> b/drivers/gpu/drm/mgag200/mgag200_g200eh5.c
> new file mode 100644
> index 000000000000..5e39504785d8
> --- /dev/null
> +++ b/drivers/gpu/drm/mgag200/mgag200_g200eh5.c

The coding style in this file is off. Jocelyn already pointed out quite 
a few issues. Please run checkpatch.pl on the patch file before 
submitting and fix the errors and warnings. It's in the scripts/ directory.

./scripts/checkpatch.pl --strict <filename>

> @@ -0,0 +1,212 @@
> +// SPDX-License-Identifier: GPL-2.0-only

Empty line here.

> +#include <linux/pci.h>
> +
> +#include <drm/drm_atomic.h>
> +#include <drm/drm_atomic_helper.h>
> +#include <drm/drm_drv.h>
> +#include <drm/drm_gem_atomic_helper.h>
> +#include <drm/drm_probe_helper.h>
> +
> +#include "mgag200_drv.h"
> +
> +/*
> + * PIXPLLC
> + */
> +
> +static int mgag200_g200eh5_pixpllc_atomic_check(struct drm_crtc *crtc,
> +                    struct drm_atomic_state *new_state)
> +{
> +

No empty line here.

> + static u64 ulVCOMax     = 10000000000ULL;   // units in Hz (10 GHz)
> +    static u64 ulVCOMin     = 2500000000LL;     // units in Hz (2.5 GHz)
> +    static u64 ulPLLFreqRef = 25000000ULL;      // units in Hz (25 MHz)

The camel case and hungarian notation is not used within the kernel. The 
code that uses this has likely been copied from somewhere else. 
Fixed-size types (e.g., u64) should be avoided IIRC; unless they are 
necessary. Using 'static' is ok if you also make it 'const', so that 
there are real constant. For these numbers, you could also look at 
<linux/units.h> for SI multipliers.

For these constants, I'd write something like

static const unsigned long long VCO_MAX = 10 * GIGA // Hz
static const unsigned long long VCO_MAX = 2500 * MEGA // Hz
static const unsigned long long PLL_FREQ_REF = 25  * MEGA // Hz


> +
> +    struct drm_crtc_state *new_crtc_state = 
> drm_atomic_get_new_crtc_state(new_state, crtc);

This file appears to be using 4 spaces per level of indention. Indention 
within the kernel is 1 tab. Each tab is equivalent to 8 space.

> + struct mgag200_crtc_state *new_mgag200_crtc_state = 
> to_mgag200_crtc_state(new_crtc_state);
> +    long   clock = new_crtc_state->mode.clock;

Multiple spaces after 'long'. There are similar cases below.

> + struct mgag200_pll_values *pixpllc = &new_mgag200_crtc_state->pixpllc;
> +
> +    u64 ulFDelta     = 0xFFFFFFFFFFFFFFFFULL;
> +
> +    u16 ulMultMax    = (u16)(ulVCOMax / ulPLLFreqRef);    // 400 (0x190)
> +    u16 ulMultMin    = (u16)(ulVCOMin / ulPLLFreqRef);    // 100 (0x64)
> +
> +    u64 ulFTmpDelta;
> +    u64 ulComputedFo;
> +
> +    u16 ulTestM;
> +    u8  ulTestDivA;
> +    u8  ulTestDivB;
> +    u64 ulFoHz;
> +    int iDone = 0;
> +
> +    u8 ucM = 0;
> +    u8 ucN = 0;
> +    u8 ucP = 0;
> +
> +    ulFoHz = (u64)clock * 1000ULL;

For such conversions, you should use HZ_PER_KHZ, also found in 
<linux/units.h>. Makes it clear what it does.

> +
> +
> +    for (ulTestM = ulMultMin; ulTestM <= ulMultMax; ulTestM++) { // 
> This gives 100 <= M <= 400
> +        for (ulTestDivA = 8; ulTestDivA > 0; ulTestDivA--) { // This 
> gives 1 <= A <= 8
> +            for (ulTestDivB = 1; ulTestDivB <= ulTestDivA; 
> ulTestDivB++) {
> +                // This gives 1 <= B <= A
> +                ulComputedFo = (ulPLLFreqRef * ulTestM) /
> +                    (4 * ulTestDivA * ulTestDivB);
> +
> +                if (ulComputedFo > ulFoHz)
> +                    ulFTmpDelta = ulComputedFo - ulFoHz;
> +                else
> +                    ulFTmpDelta = ulFoHz - ulComputedFo;
> +
> +                if (ulFTmpDelta < ulFDelta) {
> +                    ulFDelta = ulFTmpDelta;
> +                    ucM = (u8)(0xFF & ulTestM);
> +                    ucN = (u8)(
> +                    (0x7 & (ulTestDivA - 1))
> +                    | (0x70 & (0x7 & (ulTestDivB - 1)) << 4)
> +                    );
> +                    ucP = (u8)(1 & (ulTestM >> 8));
> +                }
> +                if (ulFDelta == 0) {
> +                    iDone = 1;
> +                    break;
> +                }
> +            } // End of DivB if (iDone)
> +            if (iDone)
> +                break;
> +        } // End of DivA Loop
> +
> +        if (iDone)
> +            break;
> +    } // End of M Loop

As with all other models, frequency calculation is not easily 
understandable. I haven't found a way to make these nested loops 
readable, so it's OK to do this here as well.

But you should remove these '// End of' comments. This is something that 
the formating should make obvious.

> +
> +    pixpllc->m = ucM + 1;
> +    pixpllc->n = ucN + 1;
> +    pixpllc->p = ucP + 1;
> +    pixpllc->s = 0;
> +
> +    return 0;
> +    }
> +
> +
> +
> +/*
> + * Mode-setting pipeline
> + */
> +
> +static const struct drm_plane_helper_funcs 
> mgag200_g200eh5_primary_plane_helper_funcs = {
> +    MGAG200_PRIMARY_PLANE_HELPER_FUNCS,
> +};
> +
> +static const struct drm_plane_funcs 
> mgag200_g200eh5_primary_plane_funcs = {
> +    MGAG200_PRIMARY_PLANE_FUNCS,
> +};
> +
> +static const struct drm_crtc_helper_funcs 
> mgag200_g200eh5_crtc_helper_funcs = {
> +    MGAG200_CRTC_HELPER_FUNCS,
> +};
> +
> +static const struct drm_crtc_funcs mgag200_g200eh5_crtc_funcs = {
> +    MGAG200_CRTC_FUNCS,
> +};
> +
> +static int mgag200_g200eh5_pipeline_init(struct mga_device *mdev)
> +{
> +    struct drm_device *dev = &mdev->base;
> +    struct drm_plane *primary_plane = &mdev->primary_plane;
> +    struct drm_crtc *crtc = &mdev->crtc;
> +    int ret;
> +
> +    ret = drm_universal_plane_init(dev, primary_plane, 0,
> +        &mgag200_g200eh5_primary_plane_funcs,
> +        mgag200_primary_plane_formats,
> +        mgag200_primary_plane_formats_size,
> +        mgag200_primary_plane_fmtmods,
> +        DRM_PLANE_TYPE_PRIMARY, NULL);

This should be indented as shown at

https://elixir.bootlin.com/linux/v6.13.1/source/Documentation/process/coding-style.rst#L497

The file will tell you how to format the code.

Best regards
Thomas

> + if (ret) {
> +        drm_err(dev, "drm_universal_plane_init() failed: %d\n", ret);
> +        return ret;
> +    }
> +    drm_plane_helper_add(primary_plane, 
> &mgag200_g200eh5_primary_plane_helper_funcs);
> +    drm_plane_enable_fb_damage_clips(primary_plane);
> +
> +    ret = drm_crtc_init_with_planes(dev, crtc, primary_plane, NULL,
> +        &mgag200_g200eh5_crtc_funcs, NULL);
> +    if (ret) {
> +        drm_err(dev, "drm_crtc_init_with_planes() failed: %d\n", ret);
> +        return ret;
> +    }
> +
> +    drm_crtc_helper_add(crtc, &mgag200_g200eh5_crtc_helper_funcs);
> +
> +    /* FIXME: legacy gamma tables, but atomic gamma doesn't work 
> without */
> +    drm_mode_crtc_set_gamma_size(crtc, MGAG200_LUT_SIZE);
> +    drm_crtc_enable_color_mgmt(crtc, 0, false, MGAG200_LUT_SIZE);
> +    ret = mgag200_vga_bmc_output_init(mdev);
> +
> +    if (ret)
> +        return ret;
> +
> +    return 0;
> +}
> +
> +/*
> + * DRM device
> + */
> +
> +static const struct mgag200_device_info mgag200_g200eh5_device_info =
> +    MGAG200_DEVICE_INFO_INIT(2048, 2048, 0, false, 1, 0, false);
> +
> +static const struct mgag200_device_funcs mgag200_g200eh5_device_funcs 
> = {
> +    .pixpllc_atomic_check = mgag200_g200eh5_pixpllc_atomic_check,
> +    .pixpllc_atomic_update = mgag200_g200eh_pixpllc_atomic_update, // 
> same as G200EH
> +};
> +
> +struct mga_device *mgag200_g200eh5_device_create(struct pci_dev *pdev,
> +    const struct drm_driver *drv)
> +{
> +
> +    struct mga_device *mdev;
> +    struct drm_device *dev;
> +    resource_size_t vram_available;
> +    int ret;
> +
> +    mdev = devm_drm_dev_alloc(&pdev->dev, drv, struct mga_device, base);
> +
> +    if (IS_ERR(mdev))
> +        return mdev;
> +    dev = &mdev->base;
> +
> +    pci_set_drvdata(pdev, dev);
> +
> +    ret = mgag200_init_pci_options(pdev, 0x00000120, 0x0000b000);
> +    if (ret)
> +        return ERR_PTR(ret);
> +
> +    ret = mgag200_device_preinit(mdev);
> +    if (ret)
> +        return ERR_PTR(ret);
> +
> +    ret = mgag200_device_init(mdev, &mgag200_g200eh5_device_info,
> +        &mgag200_g200eh5_device_funcs);
> +
> +    if (ret)
> +        return ERR_PTR(ret);
> +
> +    mgag200_g200eh_init_registers(mdev); // same as G200EH
> +    vram_available = mgag200_device_probe_vram(mdev);
> +
> +    ret = mgag200_mode_config_init(mdev, vram_available);
> +    if (ret)
> +        return ERR_PTR(ret);
> +
> +    ret = mgag200_g200eh5_pipeline_init(mdev);
> +    if (ret)
> +        return ERR_PTR(ret);
> +
> +    drm_mode_config_reset(dev);
> +    drm_kms_helper_poll_init(dev);
> +
> +    return mdev;
> +}

-- 
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstrasse 146, 90461 Nuernberg, Germany
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
HRB 36809 (AG Nuernberg)

Powered by blists - more mailing lists