[PATCH v3] iommu/mediatek: Move attach_device after iommu-group is ready for M4Uv1
Yong Wu
yong.wu at mediatek.com
Fri Jan 26 00:34:02 PST 2018
On Thu, 2018-01-25 at 12:02 +0000, Robin Murphy wrote:
> On 25/01/18 11:14, Yong Wu wrote:
> > In the commit 05f80300dc8b, the iommu framework has supposed all the
> > iommu drivers have their owner iommu-group, it get rid of the FIXME
> > workarounds while the group is NULL. But the flow of Mediatek M4U gen1
> > looks a bit trick that it will hang at this case:
> >
> > ==========================================
> > Unable to handle kernel NULL pointer dereference at virtual address 00000030
> > pgd = c0004000
> > [00000030] *pgd=00000000
> > PC is at mutex_lock+0x28/0x54
> > LR is at iommu_attach_device+0xa4/0xd4
> > pc : [<c07632e8>] lr : [<c04736fc>] psr: 60000013
> > sp : df0edbb8 ip : df0edbc8 fp : df0edbc4
> > r10: c114da14 r9 : df2a3e40 r8 : 00000003
> > r7 : df27a210 r6 : df2a90c4 r5 : 00000030 r4 : 00000000
> > r3 : df0f8000 r2 : fffff000 r1 : df29c610 r0 : 00000030
> > Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
> > xxx
> > (mutex_lock) from [<c04736fc>] (iommu_attach_device+0xa4/0xd4)
> > (iommu_attach_device) from [<c011b9dc>] (__arm_iommu_attach_device+0x28/0x90)
> > (__arm_iommu_attach_device) from [<c011ba60>] (arm_iommu_attach_device+0x1c/0x30)
> > (arm_iommu_attach_device) from [<c04759ac>] (mtk_iommu_add_device+0xfc/0x214)
> > (mtk_iommu_add_device) from [<c0472aa4>] (add_iommu_group+0x3c/0x68)
> > (add_iommu_group) from [<c047d044>] (bus_for_each_dev+0x78/0xac)
> > (bus_for_each_dev) from [<c04734a4>] (bus_set_iommu+0xb0/0xec)
> > (bus_set_iommu) from [<c0476310>] (mtk_iommu_probe+0x328/0x368)
> > (mtk_iommu_probe) from [<c048189c>] (platform_drv_probe+0x5c/0xc0)
> > (platform_drv_probe) from [<c047f510>] (driver_probe_device+0x2f4/0x4d8)
> > (driver_probe_device) from [<c047f800>] (__driver_attach+0x10c/0x128)
> > (__driver_attach) from [<c047d044>] (bus_for_each_dev+0x78/0xac)
> > (bus_for_each_dev) from [<c047ec78>] (driver_attach+0x2c/0x30)
> > (driver_attach) from [<c047e640>] (bus_add_driver+0x1e0/0x278)
> > (bus_add_driver) from [<c048052c>] (driver_register+0x88/0x108)
> > (driver_register) from [<c04817ec>] (__platform_driver_register+0x50/0x58)
> > (__platform_driver_register) from [<c0b31380>] (m4u_init+0x24/0x28)
> > (m4u_init) from [<c0101c38>] (do_one_initcall+0xf0/0x17c)
> > =========================
> >
> > The root cause is that the device's iommu-group is NULL while
> > arm_iommu_attach_device is called. This patch prepare a new iommu-group
> > for the iommu consumer devices to fix this issue.
> >
> > CC: Robin Murphy <robin.murphy at arm.com>
> > CC: Honghui Zhang <honghui.zhang at mediatek.com>
> > Fixes: 05f80300dc8b ('iommu: Finish making iommu_group support mandatory')
> > Reported-by: Ryder Lee <ryder.lee at mediatek.com>
> > Signed-off-by: Yong Wu <yong.wu at mediatek.com>
> > ---
> > changes notes:
> > v3: don't use the global variable and allocate a new iommu group before
> > arm_iommu_attach_device following Robin's suggestion.
> >
> > v2: http://lists.infradead.org/pipermail/linux-mediatek/2018-January/011810.html
> > Add mtk_domain_v1=NULL in domain_free for symmetry.
> >
> > v1: https://patchwork.kernel.org/patch/10176255/
> > ---
> > drivers/iommu/mtk_iommu_v1.c | 49 ++++++++++++++++++++++----------------------
> > 1 file changed, 25 insertions(+), 24 deletions(-)
> >
> > diff --git a/drivers/iommu/mtk_iommu_v1.c b/drivers/iommu/mtk_iommu_v1.c
> > index 542930c..aca76d2 100644
> > --- a/drivers/iommu/mtk_iommu_v1.c
> > +++ b/drivers/iommu/mtk_iommu_v1.c
> > @@ -418,20 +418,12 @@ static int mtk_iommu_create_mapping(struct device *dev,
> > m4udev->archdata.iommu = mtk_mapping;
> > }
> >
> > - ret = arm_iommu_attach_device(dev, mtk_mapping);
> > - if (ret)
> > - goto err_release_mapping;
> > -
> > return 0;
> > -
> > -err_release_mapping:
> > - arm_iommu_release_mapping(mtk_mapping);
> > - m4udev->archdata.iommu = NULL;
> > - return ret;
> > }
> >
> > static int mtk_iommu_add_device(struct device *dev)
> > {
> > + struct dma_iommu_mapping *mtk_mapping;
> > struct of_phandle_args iommu_spec;
> > struct of_phandle_iterator it;
> > struct mtk_iommu_data *data;
> > @@ -452,9 +444,30 @@ static int mtk_iommu_add_device(struct device *dev)
> > if (!dev->iommu_fwspec || dev->iommu_fwspec->ops != &mtk_iommu_ops)
> > return -ENODEV; /* Not a iommu client device */
> >
> > + /*
> > + * This is a short-term bodge because the ARM DMA code doesn't
> > + * understand multi-device groups, but we have to call into it
> > + * successfully (and not just rely on a normal IOMMU API attach
> > + * here) in order to set the correct DMA API ops on @dev.
> > + */
> > + group = iommu_group_alloc();
> > + if (IS_ERR(group))
> > + return PTR_ERR(group);
> > +
> > + err = iommu_group_add_device(group, dev);
> > + iommu_group_put(group);
> > + if (err)
> > + return err;
> > +
> > data = dev->iommu_fwspec->iommu_priv;
> > - iommu_device_link(&data->iommu, dev);
> > + mtk_mapping = data->dev->archdata.iommu;
> > + err = arm_iommu_attach_device(dev, mtk_mapping);
> > + if (err) {
> > + iommu_group_remove_device(dev);
> > + return err;
> > + }
> >
> > + iommu_device_link(&data->iommu, dev);
> > group = iommu_group_get_for_dev(dev);
>
> This call now does nothing, so you may as well remove it (and the
> subsequent iommu_group_put)...
>
> > if (IS_ERR(group))
> > return PTR_ERR(group);
> > @@ -479,20 +492,8 @@ static void mtk_iommu_remove_device(struct device *dev)
> >
> > static struct iommu_group *mtk_iommu_device_group(struct device *dev)
> > {
> > - struct mtk_iommu_data *data = dev->iommu_fwspec->iommu_priv;
> > -
> > - if (!data)
> > - return ERR_PTR(-ENODEV);
> > -
> > - /* All the client devices are in the same m4u iommu-group */
> > - if (!data->m4u_group) {
> > - data->m4u_group = iommu_group_alloc();
> > - if (IS_ERR(data->m4u_group))
> > - dev_err(dev, "Failed to allocate M4U IOMMU group\n");
> > - } else {
> > - iommu_group_ref_get(data->m4u_group);
> > - }
> > - return data->m4u_group;
> > + /* The iommu-group has always been allocated in add_device. */
> > + return NULL;
>
> ... and get rid of this and the mtk_iommu_ops.device_group callback
> entirely.
OK. Thanks.
>
> If you're moving the arm_iommu_attach_device() call anyway though,
> another clean option would be put it after iommu_group_get_for_dev() and
> switch .device_group to generic_device_group - I *think* that should be
> directly equivalent to the open-coded iommu_group_alloc/add_device.
In the v1 and v2, I followed this method. then I got a problem in [1]
which always create a new domain, that is not our expectation.Thus, I
have to use a global variable to restore our special M4U domain.
>
> I admit I was somewhat hoping we could confine the ARM-specific stuff to
> mtk_iommu_create_mapping() and let all the "correct" IOMMU API code stay
> in the right places even if it would strictly be unused. On reflection
> though, that's clearly a bit silly, since we can always easily bring it
> back via a revert in future.
>
I remember a patch[2], It looks try to synchronize the flow of ARM/DMA
and ARM64/DMA. Maybe this will help?.
[1]
http://elixir.free-electrons.com/linux/latest/source/drivers/iommu/iommu.c#L1027
[2] https://patchwork.kernel.org/patch/8357801/
> Robin.
>
> > }
> >
> > static int mtk_iommu_hw_init(const struct mtk_iommu_data *data)
> >
More information about the linux-arm-kernel
mailing list