Message-ID: <CAAOTY_82nN8U34-LW8RTqTLEH2MwTg+B5KMHjha3=sbOXPdWSw@mail.gmail.com>
Date:   Wed, 29 Jul 2020 10:22:06 +0800
From:   Chun-Kuang Hu <chunkuang.hu@...nel.org>
To:     Neal Liu <neal.liu@...iatek.com>
Cc:     Chun-Kuang Hu <chunkuang.hu@...nel.org>,
        Rob Herring <robh+dt@...nel.org>,
        Matthias Brugger <matthias.bgg@...il.com>,
        "devicetree@...r.kernel.org" <devicetree@...r.kernel.org>,
        wsd_upstream <wsd_upstream@...iatek.com>,
        lkml <linux-kernel@...r.kernel.org>,
        "moderated list:ARM/Mediatek SoC support" 
        <linux-mediatek@...ts.infradead.org>,
        Linux ARM <linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH v3 2/2] soc: mediatek: add mtk-devapc driver

Neal Liu <neal.liu@...iatek.com> 於 2020年7月29日 週三 上午10:10寫道：
>
> Hi Chun-Kuang,
>
> On Tue, 2020-07-28 at 23:35 +0800, Chun-Kuang Hu wrote:
> > Hi, Neal:
> >
> > Neal Liu <neal.liu@...iatek.com> 於 2020年7月28日 週二 上午11:52寫道：
> > >
> > > Hi Chun-Kuang,
> > >
> > > On Mon, 2020-07-27 at 22:47 +0800, Chun-Kuang Hu wrote:
> > > > Hi, Neal:
> > > >
> > > > Neal Liu <neal.liu@...iatek.com> 於 2020年7月27日 週一 上午11:06寫道：
> > > > >
> > > > > Hi Chun-Kuang,
> > > > >
> > > > > On Fri, 2020-07-24 at 23:55 +0800, Chun-Kuang Hu wrote:
> > > > > > Hi, Neal:
> > > > > >
> > > > > > Neal Liu <neal.liu@...iatek.com> 於 2020年7月24日 週五 下午2:55寫道：
> > > > > > >
> > > > > > > Hi Chun-Kuang,
> > > > > > >
> > > > > > > On Fri, 2020-07-24 at 00:32 +0800, Chun-Kuang Hu wrote:
> > > > > > > > Hi, Neal:
> > > > > > > >
> > > > > > > > Neal Liu <neal.liu@...iatek.com> 於 2020年7月23日 週四 下午2:11寫道：
> > > > > > > > >
> > > > > > > > > Hi Chun-Kuang,
> > > > > > > > >
> > > > > > > > > On Wed, 2020-07-22 at 22:25 +0800, Chun-Kuang Hu wrote:
> > > > > > > > > > Hi, Neal:
> > > > > > > > > >
> > > > > > > > > > Neal Liu <neal.liu@...iatek.com> 於 2020年7月22日 週三 上午11:49寫道：
> > > > > > > > > > >
> > > > > > > > > > > Hi Chun-Kuang,
> > > > > > > > > > >
> > > > > > > > > > > On Wed, 2020-07-22 at 07:21 +0800, Chun-Kuang Hu wrote:
> > > > > > > > > > > > Hi, Neal:
> > > > > > > > > > > >
> > > > > > > > > > > > Neal Liu <neal.liu@...iatek.com> 於 2020年7月21日 週二 下午12:00寫道：
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +/*
> > > > > > > > > > > > > + * mtk_devapc_dump_vio_dbg - get the violation index and dump the full violation
> > > > > > > > > > > > > + *                           debug information.
> > > > > > > > > > > > > + */
> > > > > > > > > > > > > +static bool mtk_devapc_dump_vio_dbg(struct mtk_devapc_context *ctx, u32 vio_idx)
> > > > > > > > > > > > > +{
> > > > > > > > > > > > > +       u32 shift_bit;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       if (check_vio_mask(ctx, vio_idx))
> > > > > > > > > > > > > +               return false;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       if (!check_vio_status(ctx, vio_idx))
> > > > > > > > > > > > > +               return false;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       shift_bit = get_shift_group(ctx, vio_idx);
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       if (sync_vio_dbg(ctx, shift_bit))
> > > > > > > > > > > > > +               return false;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       devapc_extract_vio_dbg(ctx);
> > > > > > > > > > > >
> > > > > > > > > > > > I think get_shift_group(), sync_vio_dbg(), and
> > > > > > > > > > > > devapc_extract_vio_dbg() should be moved out of vio_idx for-loop (the
> > > > > > > > > > > > loop in devapc_violation_irq()) because these three function is not
> > > > > > > > > > > > related to vio_idx.
> > > > > > > > > > > > Another question: when multiple vio_idx violation occur, vio_addr is
> > > > > > > > > > > > related to which one vio_idx? The latest happened one?
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Actually, it's related to vio_idx. But we don't use it directly on these
> > > > > > > > > > > function. I think below snip code might be better way to understand it.
> > > > > > > > > > >
> > > > > > > > > > > for (...)
> > > > > > > > > > > {
> > > > > > > > > > >         check_vio_mask()
> > > > > > > > > > >         check_vio_status()
> > > > > > > > > > >
> > > > > > > > > > >         // if get vio_idx, mask it temporarily
> > > > > > > > > > >         mask_module_irq(true)
> > > > > > > > > > >         clear_vio_status()
> > > > > > > > > > >
> > > > > > > > > > >         // dump violation info
> > > > > > > > > > >         get_shift_group()
> > > > > > > > > > >         sync_vio_dbg()
> > > > > > > > > > >         devapc_extract_vio_dbg()
> > > > > > > > > > >
> > > > > > > > > > >         // unmask
> > > > > > > > > > >         mask_module_irq(false)
> > > > > > > > > > > }
> > > > > > > > > >
> > > > > > > > > > This snip code does not explain any thing. I could rewrite this code as:
> > > > > > > > > >
> > > > > > > > > > for (...)
> > > > > > > > > > {
> > > > > > > > > >     check_vio_mask()
> > > > > > > > > >     check_vio_status()
> > > > > > > > > >
> > > > > > > > > >     // if get vio_idx, mask it temporarily
> > > > > > > > > >     mask_module_irq(true)
> > > > > > > > > >     clear_vio_status()
> > > > > > > > > >     // unmask
> > > > > > > > > >     mask_module_irq(false)
> > > > > > > > > > }
> > > > > > > > > >
> > > > > > > > > > // dump violation info
> > > > > > > > > > get_shift_group()
> > > > > > > > > > sync_vio_dbg()
> > > > > > > > > > devapc_extract_vio_dbg()
> > > > > > > > > >
> > > > > > > > > > And my version is identical with your version, isn't it?
> > > > > > > > >
> > > > > > > > > Sorry, I did not explain it clearly. Let's me try again.
> > > > > > > > > The reason why I put "dump violation info" between mask & unmask context
> > > > > > > > > is because it has to stop interrupt first before dump violation info,
> > > > > > > > > and then unmask it to prepare next violation.
> > > > > > > > > These sequence guarantee that if multiple violation is triggered, we
> > > > > > > > > still have information to debug.
> > > > > > > > > If the code sequence in your version and multiple violation is
> > > > > > > > > triggered, there might be no any information but keeps entering ISR.
> > > > > > > > > Finally, system might be abnormal and watchdog timeout.
> > > > > > > > > In this case, we still don't have any information to debug.
> > > > > > > >
> > > > > > > > I still don't understand why no information to debug. For example when
> > > > > > > > vio_idx 5, 10, 15 has violation,
> > > > > > > > You would mask vio_idx 5 to get information, but vio_idx 10, 15 does
> > > > > > > > not mask yet.
> > > > > > > > In your words, when vio_idx 10, 15 not mask, you would not get any
> > > > > > > > debug information when you process vio_idx 5.
> > > > > > > >
> > > > > > > > In my version, I would clear all status, why keeps entering ISR?
> > > > > > >
> > > > > > > Think about this case, if someone tries to dump "AAA" module's register.
> > > > > > > It would keep read reg base, base+0x4, base+0x8, ...
> > > > > > > All these registers are in the same slave, which would be same vio_idx.
> > > > > > > (Take vio_idx 5 as example)
> > > > > > > In this case, vio_idx 5 will keep triggering interrupt. If you did not
> > > > > > > do "dump violation info" between mask & unmask, you cannot get any
> > > > > > > violation info until the last interrupt being handled.
> > > > > > > Normally, system will crash before last interrupt coming.
> > > > > >
> > > > > > You have said that first vio_addr would be kept until it's 'handled'.
> > > > > > So the first vio_addr reg_base would be kept even though other
> > > > > > violation happen. And I could handle (clear status and dump info) it
> > > > > > then vio_addr would next violation's address. I'm confused with your
> > > > > > statement. If AAA is dumping register of vio_idx 5, BBB is dumping
> > > > > > register of vio_idx 10, CCC is dumping register of vio_idx 15, I think
> > > > > > you should mask all vio_idx not only one. So the code would be
> > > > > >
> > > > > > for all vio_idx {
> > > > > >     mask_module_irq(true)
> > > > > > }
> > > > > >
> > > > > > devapc_extract_vio_dbg()
> > > > > >
> > > > > > for all vio_idx {
> > > > > >     clear_vio_status()
> > > > > >     mask_module_irq(false)
> > > > > > }
> > > > > >
> > > > >
> > > > > I'm also consider this solution and I think it's much better to
> > > > > understand hardware behavior.
> > > > >
> > > > > devapc_dump_vio_dbg()
> > > > > {
> > > > >         while(1) {
> > > > >                 // might have multiple shift_bit raised
> > > > >                 shift_bit = get_shift_group()
> > > > >                 if (shift_bit >= 0 && shift bit <= 31)
> > > > >                         sync_vio_dbg(shift_bit)
> > > > >                         extract_vio_dbg()
> > > >
> > > > According to your statement, when multiple violation occur, only the
> > > > first one is kept, others are dropped. I think we just need to dump
> > > > debug info once.
> > > >
> > > > Because only one violation information would be kept, why not only one
> > > > group (equal to no group)?
> > > >
> > > > Regards,
> > > > Chun-Kuang.
> > >
> > > Let's me give you an example of devapc design.
> > > vio_idx: 0, 1, 2 -> group 0 (shift_bit: 0)
> > > vio_idx: 3, 4, 5 -> group 1 (shift_bit: 1)
> > > ...
> > >
> > > Each group violation will keep one violation (the first one). If vio_idx
> > > 0 is triggered first, vio_idx 1 is triggered next, then group 0 will
> > > just keep vio_idx 0 violation info.
> > > If vio_idx 2 is triggered first, vio_idx 3 is triggered next, then group
> > > 0 will keep vio_idx 2 violation info, group 1 will keep vio_idx 3's.
> > >
> > > We have to scan all groups and dump everything we have.
> > > Thanks !
> > >
> >
> > Could we let all vio_idx be group 0 so that we could just sync one
> > group? It's bad to spend too much time in irq handler.
> > When we set pd_vio_shift_sel_reg, it seems we could set multiple group
> > together, couldn't it?
> >
> > Regards,
> > Chun-Kuang.
> >
>
> No, Which group vio_idx belongs to is determined by hardware. Software
> cannot change its group.
> There is very low possibility that multiple groups has violation at the
> same time, so it would not spend much time to handle it.
> It also cannot shift multiple groups at the same time since there is
> only one vio_info(rw, vio_addr, master_id, ...) exist at a time.
> devapc_extract_vio_dbg() function is doing this step.
>

So this flow is OK for me. Would you please add comment for this
information so that we could understand how hardware work.

Regards,
Chun-Kuang.

> Thanks !
>
> > > >
> > > > >                 else
> > > > >                         break
> > > > >         }
> > > > > }
> > > > >
> > > > > devapc_violation_irq()
> > > > > {
> > > > >         for all vio_idx {
> > > > >                 mask_module_irq(true)
> > > > >         }
> > > > >
> > > > >         devapc_dump_vio_dbg()
> > > > >
> > > > >         for all vio_idx {
> > > > >                 clear_vio_status()
> > > > >                 mask_module_irq(false)
> > > > >         }
> > > > > }
> > > > >
> > > > > Is it more clear for this control flow?
> > > > > Thanks !
> > > > >
> > > > > > >
> > > > > > > >
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > About your question, vio_addr would be the first one.
> > > > > > > > > >
> > > > > > > > > > So other vio_addr would be dropped? Or hardware would keep all
> > > > > > > > > > vio_addr and you have some way to get all vio_addr?
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > > In this case, hardware will drop other violation info and keep the first
> > > > > > > > > one until it been handled.
> > > > > > > >
> > > > > > > > Does 'handled' mean status is cleared?
> > > > > > >
> > > > > > > "handled" means clear status and dump violation info.
> > > > > > >
> > > > > > > >
> > > > > > > > Regards,
> > > > > > > > Chun-Kuang.
> > > > > > > >
> > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       return true;
> > > > > > > > > > > > > +}
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +/*
> > > > > > > > > > > > > + * 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)
> > > > > > > > > > > > > +{
> > > > > > > > > > > > > +       u32 vio_idx;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       for (vio_idx = 0; vio_idx < ctx->vio_idx_num; vio_idx++) {
> > > > > > > > > > > > > +               if (!mtk_devapc_dump_vio_dbg(ctx, vio_idx))
> > > > > > > > > > > > > +                       continue;
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +               /* Ensure that violation info are written before
> > > > > > > > > > > > > +                * further operations
> > > > > > > > > > > > > +                */
> > > > > > > > > > > > > +               smp_mb();
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +               /*
> > > > > > > > > > > > > +                * Mask slave's irq before clearing vio status.
> > > > > > > > > > > > > +                * Must do it to avoid nested interrupt and prevent
> > > > > > > > > > > > > +                * unexpected behavior.
> > > > > > > > > > > > > +                */
> > > > > > > > > > > > > +               mask_module_irq(ctx, vio_idx, true);
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +               clear_vio_status(ctx, vio_idx);
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +               mask_module_irq(ctx, vio_idx, false);
> > > > > > > > > > > > > +       }
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +       return IRQ_HANDLED;
> > > > > > > > > > > > > +}
> > > > > > > > > > > > > +
> > > > > > > > > > > > > +/*
> > > > > > >
> > > > >
> > >
>

Powered by blists - more mailing lists