[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200306100955.GB50020@myrica>
Date: Fri, 6 Mar 2020 11:09:55 +0100
From: Jean-Philippe Brucker <jean-philippe@...aro.org>
To: Hanjun Guo <guohanjun@...wei.com>
Cc: Joerg Roedel <joro@...tes.org>, iommu@...ts.linux-foundation.org,
linux-kernel@...r.kernel.org, linux-arm-msm@...r.kernel.org,
linux-mediatek@...ts.infradead.org,
virtualization@...ts.linux-foundation.org,
Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
Sudeep Holla <sudeep.holla@....com>,
Rob Clark <robdclark@...il.com>, Sean Paul <sean@...rly.run>,
Will Deacon <will@...nel.org>,
Robin Murphy <robin.murphy@....com>,
Matthias Brugger <matthias.bgg@...il.com>,
Thierry Reding <thierry.reding@...il.com>,
Andy Gross <agross@...nel.org>,
Bjorn Andersson <bjorn.andersson@...aro.org>,
Linuxarm <linuxarm@...wei.com>
Subject: Re: [PATCH 00/14] iommu: Move iommu_fwspec out of 'struct device'
On Fri, Mar 06, 2020 at 04:39:37PM +0800, Hanjun Guo wrote:
> Hi Joerg,
>
> On 2020/2/28 23:08, Joerg Roedel wrote:
> > Hi,
> >
> > here is a patch-set to rename iommu_param to dev_iommu and
> > establish it as a struct for generic per-device iommu-data.
> > Also move the iommu_fwspec pointer from struct device into
> > dev_iommu to have less iommu-related pointers in struct
> > device.
> >
> > The bigger part of this patch-set moves the iommu_priv
> > pointer from struct iommu_fwspec to dev_iommu, making is
> > usable for iommu-drivers which do not use fwspecs.
> >
> > The changes for that were mostly straightforward, except for
> > the arm-smmu (_not_ arm-smmu-v3) and the qcom iommu driver.
> > Unfortunatly I don't have the hardware for those, so any
> > testing of these drivers is greatly appreciated.
>
> I tested this patch set on Kunpeng 920 ARM64 server which
> using smmu-v3 with ACPI booting, but triggered a NULL
> pointer dereference and panic at boot:
I think that's because patch 01/14 move the fwspec access too early. In
err = pci_for_each_dma_alias(to_pci_dev(dev),
iort_pci_iommu_init, &info);
if (!err && iort_pci_rc_supports_ats(node))
dev->iommu_fwspec->flags |= IOMMU_FWSPEC_PCI_RC_ATS;
the iommu_fwspec is only valid if iort_pci_iommu_init() initialized it
successfully, if err == 0. The following might fix it:
diff --git a/drivers/acpi/arm64/iort.c b/drivers/acpi/arm64/iort.c
index 0e981d7f3c7d..7d04424189df 100644
--- a/drivers/acpi/arm64/iort.c
+++ b/drivers/acpi/arm64/iort.c
@@ -1015,7 +1015,7 @@ const struct iommu_ops *iort_iommu_configure(struct device *dev)
return ops;
if (dev_is_pci(dev)) {
- struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
+ struct iommu_fwspec *fwspec;
struct pci_bus *bus = to_pci_dev(dev)->bus;
struct iort_pci_alias_info info = { .dev = dev };
@@ -1028,7 +1028,8 @@ const struct iommu_ops *iort_iommu_configure(struct device *dev)
err = pci_for_each_dma_alias(to_pci_dev(dev),
iort_pci_iommu_init, &info);
- if (!err && iort_pci_rc_supports_ats(node))
+ fwspec = dev_iommu_fwspec_get(dev);
+ if (fwspec && iort_pci_rc_supports_ats(node))
fwspec->flags |= IOMMU_FWSPEC_PCI_RC_ATS;
} else {
int i = 0;
Note that this use of iommu_fwspec will be removed by the ATS cleanup
series [1], but this change should work as a temporary fix.
Thanks,
Jean
[1] https://lore.kernel.org/linux-iommu/20200213165049.508908-10-jean-philippe@linaro.org/
Powered by blists - more mailing lists