[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210519191450.GA3469078@robh.at.kernel.org>
Date: Wed, 19 May 2021 14:14:50 -0500
From: Rob Herring <robh@...nel.org>
To: Wang Xingang <wangxingang5@...wei.com>
Cc: will@...nel.org, joro@...tes.org, frowand.list@...il.com,
helgaas@...nel.org, gregkh@...uxfoundation.org,
iommu@...ts.linux-foundation.org, linux-kernel@...r.kernel.org,
linux-pci@...r.kernel.org, devicetree@...r.kernel.org,
xieyingtai@...wei.com
Subject: Re: [PATCH v2] iommu/of: Fix pci_request_acs() before enumerating
PCI devices
On Mon, May 17, 2021 at 01:17:05PM +0000, Wang Xingang wrote:
> From: Xingang Wang <wangxingang5@...wei.com>
>
> When booting with devicetree, the pci_request_acs() is called after the
> enumeration and initialization of PCI devices, thus the ACS is not
> enabled. This patch add check for IOMMU in of_core_init(), and call
> pci_request_acs() when iommu is detected, making sure that the ACS will
> be enabled.
>
> Fixes: 6bf6c24720d33 ("iommu/of: Request ACS from the PCI core when
> configuring IOMMU linkage")
> Signed-off-by: Xingang Wang <wangxingang5@...wei.com>
> ---
> drivers/iommu/of_iommu.c | 1 -
> drivers/of/base.c | 9 ++++++++-
> 2 files changed, 8 insertions(+), 2 deletions(-)
>
> Change log:
> v1->v2:
> - remove pci_request_acs() in of_iommu_configure
> - check and call pci_request_acs() in of_core_init()
>
> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> index a9d2df001149..54a14da242cc 100644
> --- a/drivers/iommu/of_iommu.c
> +++ b/drivers/iommu/of_iommu.c
> @@ -205,7 +205,6 @@ const struct iommu_ops *of_iommu_configure(struct device *dev,
> .np = master_np,
> };
>
> - pci_request_acs();
> err = pci_for_each_dma_alias(to_pci_dev(dev),
> of_pci_iommu_init, &info);
> } else {
> diff --git a/drivers/of/base.c b/drivers/of/base.c
> index 48e941f99558..95cd8f0e5435 100644
> --- a/drivers/of/base.c
> +++ b/drivers/of/base.c
> @@ -24,6 +24,7 @@
> #include <linux/of.h>
> #include <linux/of_device.h>
> #include <linux/of_graph.h>
> +#include <linux/pci.h>
> #include <linux/spinlock.h>
> #include <linux/slab.h>
> #include <linux/string.h>
> @@ -166,7 +167,7 @@ void __of_phandle_cache_inv_entry(phandle handle)
> void __init of_core_init(void)
> {
> struct device_node *np;
> -
> + bool of_iommu_detect = false;
>
> /* Create the kset, and register existing nodes */
> mutex_lock(&of_mutex);
> @@ -180,6 +181,12 @@ void __init of_core_init(void)
> __of_attach_node_sysfs(np);
> if (np->phandle && !phandle_cache[of_phandle_cache_hash(np->phandle)])
> phandle_cache[of_phandle_cache_hash(np->phandle)] = np;
> +
> + /* Detect IOMMU and make sure ACS will be enabled */
> + if (!of_iommu_detect && of_get_property(np, "iommu-map", NULL)) {
> + of_iommu_detect = true;
> + pci_request_acs();
> + }
Private DT internal init code doesn't seem like the right place for
this. If this needs to be ordered WRT PCI device enumeration, then
somewhere in the PCI host bridge or bus init code seems like the right
place to me.
Also, shouldn't this be conditional on 'iommu-map' being in the host
bridge or a parent or ??? rather than just any iommu-map anywhere in the
DT.
> }
> mutex_unlock(&of_mutex);
>
> --
> 2.19.1
>
Powered by blists - more mailing lists