linux-kernel - Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20120125.094020.983282777619146490.hdoyu@nvidia.com>
Date:	Wed, 25 Jan 2012 08:40:20 +0100
From:	Hiroshi Doyu <hdoyu@...dia.com>
To:	"joro@...tes.org" <joro@...tes.org>
CC:	"iommu@...ts.linux-foundation.org" <iommu@...ts.linux-foundation.org>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	"linux-tegra@...r.kernel.org" <linux-tegra@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linaro-mm-sig-bounces@...ts.linaro.org" 
	<linaro-mm-sig-bounces@...ts.linaro.org>
Subject: Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

Hi Joerg,

From: Joerg Roedel <joro@...tes.org>
Subject: Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver
Date: Mon, 23 Jan 2012 16:00:48 +0100
Message-ID: <20120123150048.GB6269@...tes.org>

> Hi,
> 
> first, some questions about the GART on your platform.
> 
> 1) How many GARTs are usually implemented? One GART per device or is
>    there a single GART covering all devices, or a mix of both?

There's a single GART in the system which takes care of all devices.

> 2) Are devices allowed to DMA outside of the remappable range or will
>    this fail?

Fail.

> 
> Besides that I think IOMMU-API is not yet fully ready for GART-like
> drivers like this one. But I will merge it anyway to get things moving.
> But please answer or fix my objections first.
> 
> On Thu, Jan 05, 2012 at 09:11:48AM +0200, Hiroshi DOYU wrote:
> > +static void gart_iommu_domain_destroy(struct iommu_domain *domain)
> > +{
> > +	struct gart_device *gart = domain->priv;
> > +
> > +	spin_lock(&gart->client_lock);
> > +	if (!list_empty(&gart->client)) {
> > +		struct gart_client *c;
> > +
> > +		list_for_each_entry(c, &gart->client, list)
> > +			dev_err(gart->dev,
> > +				"%s is still attached\n", dev_name(c->dev));
> > +	}
> > +	spin_unlock(&gart->client_lock);
> 
> gart needs a NULL check. When you create a domain and immediatly destroy
> it without ever attaching a device this code will dereference a NULL
> pointer.

Fixed.

> As a general improvement I suggest that you introduce a gart_domain
> structure and store it in domain->priv (assigned in domain_init) instead
> of using the hardware descriptor.
> 
> > +static int gart_iommu_attach_dev(struct iommu_domain *domain,
> > +				 struct device *dev)
> > +{
> > +	struct gart_device *gart;
> > +	struct gart_client *client, *c;
> > +	int err = 0;
> > +
> > +	gart = dev_get_drvdata(dev->parent);
> > +	if (!gart)
> > +		return -EINVAL;
> > +	domain->priv = gart;
> 
> What happens when devices behind different GARTs are assigned to the
> same domain? domain->priv can only hold a pointer to one hardware GART.
> This can be solved by a 'struct gart_domain' holding a linked list to
> all gart_devices in this domain.

As discussed in SMMU thread, I'll revisit the above multiple iommu
device support later.

Attached the update patch.

View attachment "0001-ARM-IOMMU-Tegra20-Add-iommu_ops-for-GART-driver.patch" of type "text/x-patch" (14203 bytes)