From: Chun-Kuang Hu <chunkuang.hu@kernel.org>
To: Neal Liu <neal.liu@mediatek.com>
Cc: Chun-Kuang Hu <chunkuang.hu@kernel.org>,
Rob Herring <robh+dt@kernel.org>,
Matthias Brugger <matthias.bgg@gmail.com>,
devicetree@vger.kernel.org,
wsd_upstream <wsd_upstream@mediatek.com>,
lkml <linux-kernel@vger.kernel.org>,
"moderated list:ARM/Mediatek SoC support"
<linux-mediatek@lists.infradead.org>,
Linux ARM <linux-arm-kernel@lists.infradead.org>
Subject: Re: [PATCH v5 2/2] soc: mediatek: add mt6779 devapc driver
Date: Tue, 11 Aug 2020 07:14:43 +0800 [thread overview]
Message-ID: <CAAOTY__dYy2yN5MHNJP5=_v48oVWPu2GRZW=EuKT=-U1Pfu7wQ@mail.gmail.com> (raw)
In-Reply-To: <1597030994.7823.5.camel@mtkswgap22>
Hi, Neal:
Neal Liu <neal.liu@mediatek.com> 於 2020年8月10日 週一 上午11:43寫道:
>
> Hi Chun-Kuang,
>
> On Fri, 2020-08-07 at 23:52 +0800, Chun-Kuang Hu wrote:
> > Hi, Neal:
> >
> > Neal Liu <neal.liu@mediatek.com> 於 2020年8月7日 週五 上午10:34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@mediatek.com>
> > > ---
> >
> > [snip]
> >
> > > +
> > > +#define PHY_DEVAPC_TIMEOUT 0x10000
> > > +
> > > +/*
> > > + * devapc_sync_vio_dbg - do "shift" mechansim" to get full violation information.
> > > + * shift mechanism is depends on devapc hardware design.
> > > + * Mediatek devapc set multiple slaves as a group.
> > > + * When violation is triggered, violation info is kept
> > > + * inside devapc hardware.
> > > + * Driver should do shift mechansim to sync full violation
> > > + * info to VIO_DBGs registers.
> > > + *
> > > + */
> > > +static int devapc_sync_vio_dbg(struct mtk_devapc_context *ctx)
> > > +{
> > > + void __iomem *pd_vio_shift_sta_reg;
> > > + void __iomem *pd_vio_shift_sel_reg;
> > > + void __iomem *pd_vio_shift_con_reg;
> > > + int min_shift_group;
> > > + int ret;
> > > + u32 val;
> > > +
> > > + pd_vio_shift_sta_reg = ctx->infra_base +
> > > + ctx->data->vio_shift_sta_offset;
> > > + pd_vio_shift_sel_reg = ctx->infra_base +
> > > + ctx->data->vio_shift_sel_offset;
> > > + pd_vio_shift_con_reg = ctx->infra_base +
> > > + ctx->data->vio_shift_con_offset;
> > > +
> > > + /* Find the minimum shift group which has violation */
> > > + val = readl(pd_vio_shift_sta_reg);
> > > + if (!val)
> > > + return false;
> > > +
> > > + min_shift_group = __ffs(val);
> > > +
> > > + /* Assign the group to sync */
> > > + writel(0x1 << min_shift_group, pd_vio_shift_sel_reg);
> > > +
> > > + /* Start syncing */
> > > + writel(0x1, pd_vio_shift_con_reg);
> > > +
> > > + ret = readl_poll_timeout(pd_vio_shift_con_reg, val, val == 0x3, 0,
> > > + PHY_DEVAPC_TIMEOUT);
> > > + if (ret) {
> > > + dev_err(ctx->dev, "%s: Shift violation info failed\n", __func__);
> > > + return false;
> > > + }
> > > +
> > > + /* Stop syncing */
> > > + writel(0x0, pd_vio_shift_con_reg);
> > > + writel(0x0, pd_vio_shift_sel_reg);
> >
> > This is redundant because you set this register before start syncing.
>
> No, we don't set this reg before start syncing.
>
I'm talking about pd_vio_shift_sel_reg, and I find this before start syncing:
/* Assign the group to sync */
writel(0x1 << min_shift_group, pd_vio_shift_sel_reg);
/* Start syncing */
writel(0x1, pd_vio_shift_con_reg);
> >
> > > + writel(0x1 << min_shift_group, pd_vio_shift_sta_reg);
> >
> > You read this register to find minimum shift group, but you write it
> > back into this register, so this function would get the same minimum
> > shift group in next time, isn't it?
>
> No. The operation means write clear. We won't get the same minimum shift
> group after clear this bit.
>
Add comment for this because this is not trivial.
> >
> > > +
> > > + return true;
> > > +}
> > > +
> > > +/*
> > > + * devapc_extract_vio_dbg - extract full violation information after doing
> > > + * shift mechanism.
> > > + */
> > > +static void devapc_extract_vio_dbg(struct mtk_devapc_context *ctx)
> > > +{
> > > + struct mtk_devapc_vio_dbgs *vio_dbgs;
> >
> > struct mtk_devapc_vio_dbgs vio_dbgs;
> >
> > Use stack instead of allocating from heap.
>
> Why it cannot use heap if the memory is handled correctly?
>
You could use heap but allocating memory from heap would cost much
time. In the worst case, it would trigger buddy system to break a page
for slub. Using stack cost almost no time, but it has some limitation.
Stack memory is small and it should be used for local variable, and
vio_dbgs match this limitation, so stack is better than heap.
> >
> > > + void __iomem *vio_dbg0_reg;
> > > + void __iomem *vio_dbg1_reg;
> > > +
> > > + vio_dbgs = devm_kzalloc(ctx->dev, sizeof(struct mtk_devapc_vio_dbgs),
> > > + GFP_KERNEL);
> > > + if (!vio_dbgs)
> > > + return;
> > > +
> > > + vio_dbg0_reg = ctx->infra_base + ctx->data->vio_dbg0_offset;
> > > + vio_dbg1_reg = ctx->infra_base + ctx->data->vio_dbg1_offset;
> > > +
> > > + vio_dbgs->vio_dbg0 = readl(vio_dbg0_reg);
> > > + vio_dbgs->vio_dbg1 = readl(vio_dbg1_reg);
> > > +
> > > + /* Print violation information */
> > > + if (vio_dbgs->dbg0_bits.vio_w)
> > > + dev_info(ctx->dev, "Write Violation\n");
> > > + else if (vio_dbgs->dbg0_bits.vio_r)
> > > + dev_info(ctx->dev, "Read Violation\n");
> > > +
> > > + dev_info(ctx->dev, "Bus ID:0x%x, Dom ID:0x%x, Vio Addr:0x%x\n",
> > > + vio_dbgs->dbg0_bits.mstid, vio_dbgs->dbg0_bits.dmnid,
> > > + vio_dbgs->vio_dbg1);
> > > +}
> > > +
> >
> > [snip]
> >
> > > +
> > > +static int mtk_devapc_probe(struct platform_device *pdev)
> > > +{
> > > + struct device_node *node = pdev->dev.of_node;
> > > + struct mtk_devapc_context *ctx;
> > > + u32 devapc_irq;
> > > + int ret;
> > > +
> > > + if (IS_ERR(node))
> > > + return -ENODEV;
> > > +
> > > + ctx = devm_kzalloc(&pdev->dev, sizeof(*ctx), GFP_KERNEL);
> > > + if (!ctx)
> > > + return -ENOMEM;
> > > +
> > > + ctx->data = of_device_get_match_data(&pdev->dev);
> > > + ctx->dev = &pdev->dev;
> > > +
> > > + ctx->infra_base = of_iomap(node, 0);
> > > + if (!ctx->infra_base)
> > > + return -EINVAL;
> > > +
> > > + devapc_irq = irq_of_parse_and_map(node, 0);
> > > + if (!devapc_irq)
> > > + return -EINVAL;
> > > +
> > > + ctx->infra_clk = devm_clk_get(&pdev->dev, "devapc-infra-clock");
> > > + if (IS_ERR(ctx->infra_clk))
> > > + return -EINVAL;
> > > +
> > > + if (clk_prepare_enable(ctx->infra_clk))
> > > + return -EINVAL;
> > > +
> > > + ret = devm_request_irq(&pdev->dev, devapc_irq,
> > > + (irq_handler_t)devapc_violation_irq,
> > > + IRQF_TRIGGER_NONE, "devapc", ctx);
> > > + if (ret) {
> > > + clk_disable_unprepare(ctx->infra_clk);
> > > + return ret;
> > > + }
> > > +
> > > + platform_set_drvdata(pdev, ctx);
> > > +
> > > + start_devapc(ctx);
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +static int mtk_devapc_remove(struct platform_device *pdev)
> > > +{
> > > + struct mtk_devapc_context *ctx = platform_get_drvdata(pdev);
> > > +
> >
> > stop_devapc(ctx);
>
> We don't have to do any further operations to stop devapc hw.
>
After this driver is removed, I think we should restore hardware to
the status before probing. Before probe(), devapc hardware is stopped
(pd_apc_con_reg is a default value and all vio irq is masked), so it
should be the same status after remove(). This concept is the same as
what you do for infra_clk.
Regards,
Chun-Kuang.
> >
> > Regards,
> > Chun-Kuang.
> >
> > > + if (ctx->infra_clk)
> > > + clk_disable_unprepare(ctx->infra_clk);
> > > +
> > > + return 0;
> > > +}
> > > +
>
next prev parent reply other threads:[~2020-08-10 23:15 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-08-07 2:22 [PATCH v5] Add MediaTek MT6779 devapc driver Neal Liu
2020-08-07 2:22 ` [PATCH v5 1/2] dt-bindings: devapc: add bindings for mtk-devapc Neal Liu
2020-08-07 2:22 ` [PATCH v5 2/2] soc: mediatek: add mt6779 devapc driver Neal Liu
2020-08-07 15:52 ` Chun-Kuang Hu
2020-08-10 3:43 ` Neal Liu
2020-08-10 23:14 ` Chun-Kuang Hu [this message]
2020-08-11 2:42 ` Neal Liu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAAOTY__dYy2yN5MHNJP5=_v48oVWPu2GRZW=EuKT=-U1Pfu7wQ@mail.gmail.com' \
--to=chunkuang.hu@kernel.org \
--cc=devicetree@vger.kernel.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mediatek@lists.infradead.org \
--cc=matthias.bgg@gmail.com \
--cc=neal.liu@mediatek.com \
--cc=robh+dt@kernel.org \
--cc=wsd_upstream@mediatek.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).