From: Robin Murphy <robin.murphy@arm.com>
To: Yong Wu <yong.wu@mediatek.com>, Tomasz Figa <tfiga@google.com>
Cc: Rob Herring <robh+dt@kernel.org>, Joerg Roedel <joro@8bytes.org>,
Matthias Brugger <matthias.bgg@gmail.com>,
Will Deacon <Will.Deacon@arm.com>,
Daniel Kurtz <djkurtz@google.com>,
Lucas Stach <l.stach@pengutronix.de>,
Mark Rutland <Mark.Rutland@arm.com>,
Catalin Marinas <Catalin.Marinas@arm.com>,
"linux-mediatek@lists.infradead.org"
<linux-mediatek@lists.infradead.org>,
Sasha Hauer <kernel@pengutronix.de>,
"srv_heupstream@mediatek.com" <srv_heupstream@mediatek.com>,
"devicetree@vger.kernel.org" <devicetree@vger.kernel.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"linux-arm-kernel@lists.infradead.org"
<linux-arm-kernel@lists.infradead.org>,
"iommu@lists.linux-foundation.org"
<iommu@lists.linux-foundation.org>
Subject: Re: [PATCH 2/5] iommu/mediatek: Add mt8173 IOMMU driver
Date: Fri, 20 Mar 2015 19:14:59 +0000 [thread overview]
Message-ID: <550C71B3.1080708@arm.com> (raw)
In-Reply-To: <1426677749.22581.38.camel@mhfsdcap03>
On 18/03/15 11:22, Yong Wu wrote:
> Hi Tomasz,
> Thanks very much for your review. please help check below.
> The others I will fix in the next version.
>
> Hi Robin,
> There are some place I would like you can have a look and give me
> some suggestion.
>
> On Wed, 2015-03-11 at 19:53 +0900, Tomasz Figa wrote:
>> Hi,
>>
>> Please find next part of my comments inline.
>>
>> On Fri, Mar 6, 2015 at 7:48 PM, <yong.wu@mediatek.com> wrote:
>>
>> [snip]
>>
>>> +/*
>>> + * pimudev is a global var for dma_alloc_coherent.
>>> + * It is not accepatable, we will delete it if "domain_alloc" is enabled
>>
>> It looks like we indeed need to use dma_alloc_coherent() and we don't
>> have a good way to pass the device pointer to domain_init callback.
>>
>> If you don't expect SoCs in the nearest future to have multiple M4U
>> blocks, then I guess this global variable could stay, after changing
>> the comment into an explanation why it's correct. Also it should be
>> moved to the top of the file, below #include directives, as this is
>> where usually global variables are located.
> @Robin,
> We have merged this patch[0] in order to delete the global var, But
> it seems that your patch of "arm64:IOMMU" isn't based on it right row.
> it will build fail.
Yeah, I've not yet managed to try pulling in that series (much as I
approve of it), partly as I know doing so is going to lean towards a
not-insignificant rework and I'd rather avoid picking up more unmerged
dependencies to block getting _something_ in for arm64 (which we can
then improve).
>
> [0]:http://lists.linuxfoundation.org/pipermail/iommu/2015-January/011939.html
>
>>> + */
>>> +static struct device *pimudev;
>>> +
> [snip]
>>> +
>>> +static int mtk_iommu_attach_device(struct iommu_domain *domain,
>>> + struct device *dev)
>>> +{
>>> + unsigned long flags;
>>> + struct mtk_iommu_domain *priv = domain->priv;
>>> + struct mtk_iommu_info *piommu = priv->piommuinfo;
>>> + struct of_phandle_args out_args = {0};
>>> + struct device *imudev;
>>> + unsigned int i = 0;
>>> +
>>> + if (!piommu)
>>
>> Could you explain when this can happen?
> If we call arch_setup_dma_ops to create a iommu domain,
> it will enter iommu_dma_attach_device, then enter here. At that time, we
> don't add the private data to this "struct iommu_domain *".
> @Robin, Could this be improved?
Calling arch_setup_dma_ops() from the driver looks plain wrong,
especially given that you apparently attach the IOMMU to itself - if you
want your own domain you should use iommu_dma_create_domain(). I admit
that still leaves you having to dance around a bit in order to tear down
the automatic domains for now, but hopefully we'll get the core code
sorted out sooner rather than later.
>>
>>> + goto imudev;
>>
>> return 0;
>>
>>> + else
>>
>> No else needed.
>>
>>> + imudev = piommu->dev;
>>> +
>>> + spin_lock_irqsave(&priv->portlock, flags);
>>
>> What is protected by this spinlock?
> We will write a register of the local arbiter while config port. If
> some modules are in the same local arbiter, it may be overwrite. so I
> add it here.
>>
>>> +
>>> + while (!of_parse_phandle_with_args(dev->of_node, "iommus",
>>> + "#iommu-cells", i, &out_args)) {
>>> + if (1 == out_args.args_count) {
>>
>> Can we be sure that this is actually referring to our IOMMU?
>>
>> Maybe this should be rewritten to
>>
>> if (out_args.np != imudev->of_node)
>> continue;
>> if (out_args.args_count != 1) {
>> dev_err(imudev, "invalid #iommu-cells property for IOMMU %s\n",
>>
>> }
>>
>>> + unsigned int portid = out_args.args[0];
>>> +
>>> + dev_dbg(dev, "iommu add port:%d\n", portid);
>>
>> imudev should be used here instead of dev.
>>
>>> +
>>> + mtk_iommu_config_port(piommu, portid);
>>> +
>>> + if (i == 0)
>>> + dev->archdata.dma_ops =
>>> + piommu->dev->archdata.dma_ops;
>>
>> Shouldn't this be set automatically by IOMMU or DMA mapping core?
> @Robin,
> In the original "arm_iommu_attach_device" of arm/mm, it will call
> set_dma_ops to add iommu_ops for each iommu device.
> But iommu_dma_attach_device don't help this, so I have to add it here.
> Could this be improved?
If you implemented a simple of_xlate callback so that the core code
handles the dma_ops as intended, I think the simplest cheat would be to
check the client device's domain, either on attachment or when they
start mapping/unmapping, and move them to your own domain if necessary.
I'm putting together a v3 of the DMA mapping series, so I'll have a look
to see if I can squeeze in a way to make that a bit less painful until
we solve it properly.
Robin.
>>
>>> + }
>>> + i++;
>>> + }
>>> +
>>> + spin_unlock_irqrestore(&priv->portlock, flags);
>>> +
>>> +imudev:
>>> + return 0;
>>> +}
>>> +
>>> +static void mtk_iommu_detach_device(struct iommu_domain *domain,
>>> + struct device *dev)
>>> +{
>>
>> No hardware (de)configuration or clean-up necessary?
> I will add it. Actually we design like this:If a device have attached to
> iommu domain, it won't detach from it.
>>
>>> +}
>>> +
> [snip]
>>
>>> +
>>> + piommu->protect_va = devm_kmalloc(piommu->dev, MTK_PROTECT_PA_ALIGN*2,
>>
>> style: Operators like * should have space on both sides.
>>
>>> + GFP_KERNEL);
>>
>> Shouldn't dma_alloc_coherent() be used for this?
> We don't care the data in it. I think they are the same. Could you
> help tell me why dma_alloc_coherent may be better.
>>
>>> + if (!piommu->protect_va)
>>> + goto protect_err;
>>
>> Please return -ENOMEM here directly, as there is nothing to clean up
>> in this case.
>>
> [snip]
>>
>>> + dev_err(piommu->dev, "IRQ request %d failed\n",
>>> + piommu->irq);
>>> + goto hw_err;
>>> + }
>>> +
>>> + iommu_set_fault_handler(domain, mtk_iommu_fault_handler, piommu);
>>
>> I don't see any other drivers doing this. Isn't this for upper layers,
>> so that they can set their own generic fault handlers?
> I think that this function is related with the iommu domain, we
> have only one multimedia iommu domain. so I add it after the iommu
> domain are created.
>>
>>> +
>>> + dev_set_drvdata(piommu->dev, piommu);
>>
>> This should be set before allowing the interrupt to fire. In other
>> words, the driver should be fully set up at the time of enabling the
>> IRQ.
>>
>>> +
>>> + return 0;
>>
>> style: Missing blank line.
>>
>>> +hw_err:
>>> + arch_teardown_dma_ops(piommu->dev);
>>> +pte_err:
>>> + kmem_cache_destroy(piommu->m4u_pte_kmem);
>>> +protect_err:
>>> + dev_err(piommu->dev, "probe error\n");
>>
>> Please replace this with specific messages for all errors (in case the
>> called function doesn't already print one like kmalloc and friends).
>>
>>> + return 0;
>>
>> Returning 0, which means success, doesn't look like a good idea for
>> signalling a failure. Please return the correct error code as received
>> from function that errors out if possible.
>>
>> End of part 3.
>>
>> Best regards,
>> Tomasz
>
>
>
next prev parent reply other threads:[~2015-03-20 19:15 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-03-06 10:48 [RFC PATCH 0/5] MT8173 IOMMU support yong.wu
2015-03-06 10:48 ` [PATCH 1/5] soc: mediatek: Add SMI driver yong.wu
2015-03-06 11:30 ` Paul Bolle
2015-03-09 11:57 ` Yong Wu
2015-03-09 17:59 ` Paul Bolle
2015-03-09 21:54 ` Arnd Bergmann
2015-03-10 6:17 ` Yingjoe Chen
2015-03-09 3:26 ` Yingjoe Chen
2015-03-09 21:56 ` Arnd Bergmann
2015-03-10 6:27 ` Yingjoe Chen
2015-03-10 9:05 ` Arnd Bergmann
2015-03-10 9:24 ` Lucas Stach
2015-03-09 11:03 ` Sascha Hauer
2015-03-06 10:48 ` [PATCH 2/5] iommu/mediatek: Add mt8173 IOMMU driver yong.wu
2015-03-06 10:58 ` Will Deacon
2015-03-09 12:11 ` Yong Wu
2015-03-17 15:14 ` Will Deacon
2015-03-06 17:15 ` Mitchel Humpherys
2015-03-09 12:16 ` Yong Wu
2015-03-09 16:57 ` Mitchel Humpherys
2015-03-08 4:12 ` Tomasz Figa
2015-03-12 14:16 ` Yong Wu
2015-03-09 8:24 ` Daniel Kurtz
2015-03-09 11:11 ` Tomasz Figa
2015-03-09 14:46 ` Yingjoe Chen
2015-03-09 17:00 ` Tomasz Figa
2015-03-10 3:41 ` Yingjoe Chen
2015-03-10 4:06 ` Tomasz Figa
2015-03-11 10:53 ` Tomasz Figa
2015-03-18 11:22 ` Yong Wu
2015-03-20 19:14 ` Robin Murphy [this message]
2015-04-14 6:50 ` Yong Wu
2015-03-27 9:41 ` Tomasz Figa
2015-04-14 6:31 ` Yong Wu
2015-04-15 2:20 ` Tomasz Figa
2015-04-15 7:06 ` Yong Wu
2015-04-15 7:41 ` Tomasz Figa
2015-04-29 6:23 ` Yong Wu
2015-03-06 10:48 ` [PATCH 3/5] dt-bindings: mediatek: Add smi dts binding yong.wu
2015-03-06 11:13 ` Mark Rutland
2015-03-09 12:55 ` Yong Wu
2015-04-14 9:07 ` Yong Wu
2015-04-14 10:06 ` Mark Rutland
2015-04-14 13:49 ` Yong Wu
2015-04-14 13:55 ` Yong Wu
2015-04-14 13:56 ` Mark Rutland
2015-03-06 14:48 ` Sergei Shtylyov
2015-03-09 12:32 ` Yong Wu
2015-03-06 10:48 ` [PATCH 4/5] dt-bindings: iommu: Add binding for mediatek IOMMU yong.wu
2015-03-06 11:21 ` Mark Rutland
2015-03-09 11:30 ` Yong Wu
2015-03-06 10:48 ` [PATCH 5/5] dts: mt8173: Add iommu/smi nodes for mt8173 yong.wu
2015-03-07 15:20 ` Daniel Kurtz
2015-03-09 12:18 ` Yong Wu
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=550C71B3.1080708@arm.com \
--to=robin.murphy@arm.com \
--cc=Catalin.Marinas@arm.com \
--cc=Mark.Rutland@arm.com \
--cc=Will.Deacon@arm.com \
--cc=devicetree@vger.kernel.org \
--cc=djkurtz@google.com \
--cc=iommu@lists.linux-foundation.org \
--cc=joro@8bytes.org \
--cc=kernel@pengutronix.de \
--cc=l.stach@pengutronix.de \
--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=robh+dt@kernel.org \
--cc=srv_heupstream@mediatek.com \
--cc=tfiga@google.com \
--cc=yong.wu@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).