Re: [PATCH v6 2/2] soc: mediatek: add mt6779 devapc driver

[Neal Liu]

Hi Chun-Kuang,

On Tue, 2020-08-18 at 14:20 +0800, Neal Liu wrote:
> Hi Chun-Kuang,
> 
> On Tue, 2020-08-18 at 10:44 +0800, Neal Liu wrote:
> > Hi Chun-Kuang,
> > 
> > On Mon, 2020-08-17 at 23:13 +0800, Chun-Kuang Hu wrote:
> > > Hi, Neal:
> > > 
> > > Neal Liu <neal.liu@xxxxxxxxxxxx> 於 2020年8月17日 週一 下午12:02寫道：
> > > >
> > > > Hi Chun-Kuang,
> > > >
> > > > On Sat, 2020-08-15 at 11:03 +0800, Chun-Kuang Hu wrote:
> > > > > Hi, Neal:
> > > > >
> > > > > Neal Liu <neal.liu@xxxxxxxxxxxx> 於 2020年8月13日 週四 上午11:33寫道：
> > > > > >
> > > > > > MediaTek bus fabric provides TrustZone security support and data
> > > > > > protection to prevent slaves from being accessed by unexpected
> > > > > > masters.
> > > > > > The security violation is logged and sent to the processor for
> > > > > > further analysis or countermeasures.
> > > > > >
> > > > > > Any occurrence of security violation would raise an interrupt, and
> > > > > > it will be handled by mtk-devapc driver. The violation
> > > > > > information is printed in order to find the murderer.
> > > > > >
> > > > > > Signed-off-by: Neal Liu <neal.liu@xxxxxxxxxxxx>
> > > > > > ---
> > > > >
> > > > > [snip]
> > > > >
> > > > > > +/*
> > > > > > + * devapc_violation_irq - the devapc Interrupt Service Routine (ISR) will dump
> > > > > > + *                        violation information including which master violates
> > > > > > + *                        access slave.
> > > > > > + */
> > > > > > +static irqreturn_t devapc_violation_irq(int irq_number,
> > > > > > +                                       struct mtk_devapc_context *ctx)
> > > > > > +{
> > > > > > +       /*
> > > > > > +        * Mask slave's irq before clearing vio status.
> > > > > > +        * Must do it to avoid nested interrupt and prevent
> > > > > > +        * unexpected behavior.
> > > > > > +        */
> > > > > > +       mask_module_irq(ctx, true);
> > > > >
> > > > > I still don't understand why nested interrupt happen. If two CPU
> > > > > process different devapc interrupt at the same time, mask interrupt
> > > > > could not prevent these two CPU to sync vio dbg at the same time. As I
> > > > > know, in ARM CPU, only CPU0 process irq handler, and all devapc
> > > > > interrupt has the same priority, so why nested interrupt happen? Could
> > > > > you explain more detail about how nested interrupt happen?
> > > >
> > > > If there is another violation happened before previous violation is
> > > > fully handled, nested interrupt would happen.
> > > >
> > > > Let's me take an example:
> > > > vio A happen
> > > > enter A ISR
> > > > ...             vio B happen
> > > > finish A ISR    enter B ISR
> > > >                 ...
> > > >                 finish B ISR
> > > >
> > > > We mask all module's irq to avoid nested interrupt.
> > > 
> > > This is not 'nested' interrupt. After A ISR is finished, B ISR happen.
> > > So A ISR and B ISR are consecutive interrupt, not nested interrupt.
> > > To compare mask irq and no mask irq, Let's consider this situation:
> > > 
> > > 1. 1000 consecutive violation happen, the time period between two
> > > violation is 0.01 ms, so the total time is 10ms. (In 10ms, 1000
> > > violation happen)
> > > 2. One ISR handle time is 1 ms, so in one ISR handler, 100 violation happen.
> > > 
> > > For mask irq solution, 10 ISR handler is trigger. For no mask irq
> > > solution, 11 ISR handler is trigger.
> > > I think these two solution have similar result, and no mask irq
> > > solution print more information (If these 1000 violation is trigger by
> > > 20 different driver, no mask solution may show one more driver than
> > > mask solution)
> > > So I think it's not necessary to mask irq in irq handler.
> > > 
> > 
> > No, my example is B ISR is entered before A ISR finished.
> > Why this is not nested?
> > vio A happen
> > enter A ISR
> > ...             vio B happen
> > ...		enter B ISR
> > finish A ISR
> >                 ...
> > 		...
> >                 finish B ISR
> > 
> 
> I have some misunderstanding about how ARM CPU & GIC works. I'll confirm
> it and get back to you. Please ignore previous mail thread.
> Thanks !

Yes, you are right. There is only 1 CPU (CPU0) will process irq handler
in ARM CPU. Nested interrupt will never happen.
The reason why I have misunderstanding is that Mediatek has some
customization about irq handling for ARM CPUs. But it would not applied
in upstream kernel.

Let remove mask/unmask module irq during ISR. The diff would be:

diff --git a/drivers/soc/mediatek/mtk-devapc.c
b/drivers/soc/mediatek/mtk-devapc.c
index 5189b3f4d63f..0ba61d742e0e 100644
--- a/drivers/soc/mediatek/mtk-devapc.c
+++ b/drivers/soc/mediatek/mtk-devapc.c
@@ -183,24 +183,10 @@ static void devapc_extract_vio_dbg(struct
mtk_devapc_context *ctx)
 static irqreturn_t devapc_violation_irq(int irq_number,
                                        struct mtk_devapc_context *ctx)
 {
-       /*
-        * Mask slave's irq before clearing vio status.
-        * Must do it to avoid nested interrupt and prevent
-        * unexpected behavior.
-        */
-       mask_module_irq(ctx, true);
-
        while (devapc_sync_vio_dbg(ctx))
                devapc_extract_vio_dbg(ctx);

-       /*
-        * Ensure that violation info are written
-        * before further operations
-        */
-       smp_mb();
-
        clear_vio_status(ctx);
-       mask_module_irq(ctx, false);

        return IRQ_HANDLED;
 }

Is that okay for you?
Thanks !

> 
> > > >
> > > > >
> > > > > > +
> > > > > > +       while (devapc_sync_vio_dbg(ctx))
> > > > > > +               devapc_extract_vio_dbg(ctx);
> > > > > > +
> > > > > > +       /*
> > > > > > +        * Ensure that violation info are written
> > > > > > +        * before further operations
> > > > > > +        */
> > > > > > +       smp_mb();
> > > > > > +
> > > > > > +       clear_vio_status(ctx);
> > > > > > +       mask_module_irq(ctx, false);
> > > > > > +
> > > > > > +       return IRQ_HANDLED;
> > > > > > +}
> > > > > > +
> > > > >
> > > > > [snip]
> > > > >
> > > > > > +
> > > > > > +static int mtk_devapc_remove(struct platform_device *pdev)
> > > > > > +{
> > > > > > +       struct mtk_devapc_context *ctx = platform_get_drvdata(pdev);
> > > > > > +
> > > > > > +       stop_devapc(ctx);
> > > > > > +
> > > > > > +       if (ctx->infra_clk)
> > > > >
> > > > > This always true.
> > > >
> > > > Does it mean that remove function would be called only if probe function
> > > > is returned successfully?
> > > 
> > > Yes.
> > > 
> > > > Is there any chance this function would be called directly?
> > > 
> > > No.
> > > 
> > > Regards,
> > > Chun-Kuang.
> > > 
> > > >
> > > > >
> > > > > Regards,
> > > > > Chun-Kuang.
> > > > >
> > > > > > +               clk_disable_unprepare(ctx->infra_clk);
> > > > > > +
> > > > > > +       return 0;
> > > > > > +}
> > > > > > +
> > > > > > +static struct platform_driver mtk_devapc_driver = {
> > > > > > +       .probe = mtk_devapc_probe,
> > > > > > +       .remove = mtk_devapc_remove,
> > > > > > +       .driver = {
> > > > > > +               .name = KBUILD_MODNAME,
> > > > > > +               .of_match_table = mtk_devapc_dt_match,
> > > > > > +       },
> > > > > > +};
> > > > > > +
> > > > > > +module_platform_driver(mtk_devapc_driver);
> > > > > > +
> > > > > > +MODULE_DESCRIPTION("Mediatek Device APC Driver");
> > > > > > +MODULE_AUTHOR("Neal Liu <neal.liu@xxxxxxxxxxxx>");
> > > > > > +MODULE_LICENSE("GPL");
> > > > > > --
> > > > > > 1.7.9.5
> > > > > > _______________________________________________
> > > > > > Linux-mediatek mailing list
> > > > > > Linux-mediatek@xxxxxxxxxxxxxxxxxxx
> > > > > > http://lists.infradead.org/mailman/listinfo/linux-mediatek
> > > >
> > 
> > 
> 
>