Re: [RFC PATCH v2 1/9] mailbox: Add Amlogic Meson Message-Handling-Unit

From: Jassi Brar
Date: Tue Jun 28 2016 - 11:06:25 EST

On Tue, Jun 28, 2016 at 7:30 PM, Neil Armstrong <narmstrong@xxxxxxxxxxxx> wrote:
> On 06/25/2016 07:50 PM, Jassi Brar wrote:
>> -#define INTR_STAT_OFS 0x0
>> -#define INTR_SET_OFS 0x8
>> -#define INTR_CLR_OFS 0x10
>> -
>> -#define MHU_LP_OFFSET 0x0
>> -#define MHU_HP_OFFSET 0x20
>> -#define MHU_SEC_OFFSET 0x200
>> -#define TX_REG_OFFSET 0x100
>> +#define INTR_SET_OFS 0x0
>> +#define INTR_STAT_OFS 0x4
>> +#define INTR_CLR_OFS 0x8
>>
>> -#define MHU_CHANS 3
>> +#define MHU_LP_OFFSET 0x10
>> +#define MHU_HP_OFFSET 0x1c
>> +
>> +#define TX_REG_OFFSET 0x24
>> +
>> +#define MHU_CHANS 2
>>
>> ^^^^^^^^ this is precisely the difference if we ignore cosmetic
>> differences. So the IP is essentially the same.
>
> Well, no. The overall design is similar. but clearly it's a different IP.
>
If your this patch works on your platform, then the diff from arm_mhu
tells the controller is the same but only with re-ordered registers.
And you already call it 'MHU' :)


>> [snip]
>>
>>> -------------------------------------------------------------------------------------------
>>> From: Neil Armstrong <narmstrong@xxxxxxxxxxxx>
>>> Subject: [PATCH] mailbox: arm_mhu: Add support for Amlogic Meson MHU
>>>
>> Is there some version of MHU specified anywhere in manuals? It seems
>> weird Amlogic took the IP and only rearranged the registers. Maybe the
>> order is specific to non-Amba version of the IP? Lets call it that.
>
> I think Amlogic took an early Juno platform release and re-implemented the
> MHU using the same concept but following their design rules.
>
>>
>>> To achieve this integration, add support for generic probe from amba
>>> or platform.
>>> Move all register offsets to a data structure passed in either amba id or
>>> platform dt id match table.
>>>
>>> Signed-off-by: Neil Armstrong <narmstrong@xxxxxxxxxxxx>
>>> ---
>>> drivers/mailbox/arm_mhu.c | 217 ++++++++++++++++++++++++++++++++++++++--------
>>> 1 file changed, 181 insertions(+), 36 deletions(-)
>>>
>>> diff --git a/drivers/mailbox/arm_mhu.c b/drivers/mailbox/arm_mhu.c
>>> index 99befa7..d7fb843 100644
>>> --- a/drivers/mailbox/arm_mhu.c
>>> +++ b/drivers/mailbox/arm_mhu.c
>>> @@ -22,45 +22,68 @@
>>> #include <linux/io.h>
>>> #include <linux/module.h>
>>> #include <linux/amba/bus.h>
>>> +#include <linux/of_platform.h>
>>> +#include <linux/platform_device.h>
>>> #include <linux/mailbox_controller.h>
>>>
>>> -#define INTR_STAT_OFS 0x0
>>> -#define INTR_SET_OFS 0x8
>>> -#define INTR_CLR_OFS 0x10
>>> +#define MHU_INTR_STAT_OFS 0x0
>>> +#define MHU_INTR_SET_OFS 0x8
>>> +#define MHU_INTR_CLR_OFS 0x10
>>>
>>> #define MHU_LP_OFFSET 0x0
>>> #define MHU_HP_OFFSET 0x20
>>> #define MHU_SEC_OFFSET 0x200
>>> -#define TX_REG_OFFSET 0x100
>>> +#define MHU_TX_REG_OFFSET 0x100
>>>
>>> -#define MHU_CHANS 3
>>> +#define MESON_INTR_SET_OFS 0x0
>>> +#define MESON_INTR_STAT_OFS 0x4
>>> +#define MESON_INTR_CLR_OFS 0x8
>>> +
>>> +#define MESON_MHU_LP_OFFSET 0x10
>>> +#define MESON_MHU_HP_OFFSET 0x1c
>>> +#define MESON_MHU_TX_OFFSET 0x24
>>> +
>>> +#define MAX_MHU_CHANS 3
>>>
>> MHU_CHANS always 3 doesn't hurt. Lets keep it unchanged.
>>
>>> struct mhu_link {
>>> unsigned irq;
>>> - void __iomem *tx_reg;
>>> - void __iomem *rx_reg;
>>> + void __iomem *tx_stat_reg;
>>> + void __iomem *tx_set_reg;
>>> + void __iomem *tx_clr_reg;
>>> + void __iomem *rx_stat_reg;
>>> + void __iomem *rx_set_reg;
>>> + void __iomem *rx_clr_reg;
>>> };
>>
>> Yeah, this is OK.
>>
>>
>>>
>>> struct arm_mhu {
>>> void __iomem *base;
>>> - struct mhu_link mlink[MHU_CHANS];
>>> - struct mbox_chan chan[MHU_CHANS];
>>> + struct mhu_link mlink[MAX_MHU_CHANS];
>>> + struct mbox_chan chan[MAX_MHU_CHANS];
>>> struct mbox_controller mbox;
>>> };
>> just leave it MHU_CHANS
>>
>>>
>>> +struct arm_mhu_data {
>>> + unsigned int channels;
>>> + int rx_offsets[MAX_MHU_CHANS];
>>> + int tx_offsets[MAX_MHU_CHANS];
>>> + unsigned int intr_stat_offs;
>>> + unsigned int intr_set_offs;
>>> + unsigned int intr_clr_offs;
>>> +};
>> This is unnecessary. Please remove it and code will be simpler -
>> assign rx/tx_regs directly in probe.
>
> I won't assume the platform driver is only for Amlogic, it does not
> make sense.
>
It is unlikely other platforms will come with more random register
arrangements of MHU.


>>
>>> +
>>> static irqreturn_t mhu_rx_interrupt(int irq, void *p)
>>> {
>>> struct mbox_chan *chan = p;
>>> struct mhu_link *mlink = chan->con_priv;
>>> u32 val;
>>>
>>> - val = readl_relaxed(mlink->rx_reg + INTR_STAT_OFS);
>>> + val = readl_relaxed(mlink->rx_stat_reg);
>>> if (!val)
>>> return IRQ_NONE;
>>>
>>> mbox_chan_received_data(chan, (void *)&val);
>>>
>>> - writel_relaxed(val, mlink->rx_reg + INTR_CLR_OFS);
>>> + writel_relaxed(val, mlink->rx_clr_reg);
>>>
>>> return IRQ_HANDLED;
>>> }
>>> @@ -68,7 +91,7 @@ static irqreturn_t mhu_rx_interrupt(int irq, void *p)
>>> static bool mhu_last_tx_done(struct mbox_chan *chan)
>>> {
>>> struct mhu_link *mlink = chan->con_priv;
>>> - u32 val = readl_relaxed(mlink->tx_reg + INTR_STAT_OFS);
>>> + u32 val = readl_relaxed(mlink->tx_stat_reg);
>>>
>>> return (val == 0);
>>> }
>>> @@ -78,7 +101,7 @@ static int mhu_send_data(struct mbox_chan *chan, void *data)
>>> struct mhu_link *mlink = chan->con_priv;
>>> u32 *arg = data;
>>>
>>> - writel_relaxed(*arg, mlink->tx_reg + INTR_SET_OFS);
>>> + writel_relaxed(*arg, mlink->tx_set_reg);
>>>
>>> return 0;
>>> }
>>> @@ -89,8 +112,8 @@ static int mhu_startup(struct mbox_chan *chan)
>>> u32 val;
>>> int ret;
>>>
>>> - val = readl_relaxed(mlink->tx_reg + INTR_STAT_OFS);
>>> - writel_relaxed(val, mlink->tx_reg + INTR_CLR_OFS);
>>> + val = readl_relaxed(mlink->tx_stat_reg);
>>> + writel_relaxed(val, mlink->tx_clr_reg);
>>>
>>> ret = request_irq(mlink->irq, mhu_rx_interrupt,
>>> IRQF_SHARED, "mhu_link", chan);
>>> @@ -117,52 +140,155 @@ static const struct mbox_chan_ops mhu_ops = {
>>> .last_tx_done = mhu_last_tx_done,
>>> };
>>>
>>> -static int mhu_probe(struct amba_device *adev, const struct amba_id *id)
>>> +static struct arm_mhu_data arm_mhu_amba_data = {
>>> + .channels = 3,
>>> + .rx_offsets = {MHU_LP_OFFSET, MHU_HP_OFFSET, MHU_SEC_OFFSET},
>>> + .tx_offsets = {MHU_LP_OFFSET + MHU_TX_REG_OFFSET,
>>> + MHU_HP_OFFSET + MHU_TX_REG_OFFSET,
>>> + MHU_SEC_OFFSET + MHU_TX_REG_OFFSET},
>>> + .intr_stat_offs = MHU_INTR_STAT_OFS,
>>> + .intr_set_offs = MHU_INTR_SET_OFS,
>>> + .intr_clr_offs = MHU_INTR_CLR_OFS,
>>> +};
>>> +
>>> +static const struct arm_mhu_data meson_mhu_data = {
>>> + .channels = 2,
>>> + .rx_offsets = {MESON_MHU_LP_OFFSET, MESON_MHU_HP_OFFSET},
>>> + .tx_offsets = {MESON_MHU_LP_OFFSET + MESON_MHU_TX_OFFSET,
>>> + MESON_MHU_HP_OFFSET + MESON_MHU_TX_OFFSET},
>>> + .intr_stat_offs = MESON_INTR_STAT_OFS,
>>> + .intr_set_offs = MESON_INTR_SET_OFS,
>>> + .intr_clr_offs = MESON_INTR_CLR_OFS,
>>> +};
>>> +
>> These could be directly set in amba vs platform probes ?
>
> It does not make sense to assume the platform driver is only for
> amlogic gxbb, it could match other vendors aswell.
>
Perhaps you didn't get my suggestion.

> The amba could force a single struct, but it's smarter to use the
> same mechanism and associate the struct to an ID.
>
>> Thanks.
>>
>
> My main question is : do you really want to transform this simple driver into
> a dirty multi-bus generic mailbox driver ?
> The meson_mhu is only 199 lines and this patch adds 181 lines to the arm_mhu driver.
>
> I'll personally push to have two separate drivers here.
>
It is a shame if we need to copy a driver only due to changed register
offsets. Let me give it a shot and see how worse off we would be.

Thanks.