[PATCH V3 0/2] PCI: Resolve Loongson's LS7A PCI problems

[PATCH V3 0/2] PCI: Resolve Loongson's LS7A PCI problems

[Huacai Chen]

This patchset attempt to resolves Loongson's LS7A PCI problems: the
first patch remove pci_disable_device() in pcie_port_device_remove() to
avoid poweroff/reboot failure; the second patch improves the mrrs quirk
for LS7A chipset; 

V1 -> V2:
1, Update commit messages and comments.

V2 -> V3:
1, Simply remove pci_disable_device() in pcie_port_device_remove() to
   solve poweroff/reboot failure.
2, Update commit messages and comments.

Huacai Chen, Tiezhu Yang and Jianmin Lv(2):
 PCI: Omit pci_disable_device() in pcie_port_device_remove();
 PCI: loongson: Improve the MRRS quirk for LS7A.

Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Signed-off-by: Jianmin Lv <lvjianmin@loongson.cn> 
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
---
 drivers/pci/controller/pci-loongson.c | 44 ++++++++++++-----------------------
 drivers/pci/pci.c                     |  6 +++++
 drivers/pci/pcie/portdrv.c            |  1 -
 include/linux/pci.h                   |  1 +
 4 files changed, 22 insertions(+), 30 deletions(-)
--
2.27.0

[PATCH V3 1/2] PCI: Omit pci_disable_device() in pcie_port_device_remove()

[Huacai Chen]

This patch has a long story.

After cc27b735ad3a7557 ("PCI/portdrv: Turn off PCIe services during
shutdown") we observe poweroff/reboot failures on systems with LS7A
chipset.

We found that if we remove "pci_command &= ~PCI_COMMAND_MASTER" in
do_pci_disable_device(), it can work well. The hardware engineer says
that the root cause is that CPU is still accessing PCIe devices while
poweroff/reboot, and if we disable the Bus Master Bit at this time, the
PCIe controller doesn't forward requests to downstream devices, and also
does not send TIMEOUT to CPU, which causes CPU wait forever (hardware
deadlock).

To be clear, the sequence is like this:

  - CPU issues MMIO read to device below Root Port

  - LS7A Root Port fails to forward transaction to secondary bus
    because of LS7A Bus Master defect

  - CPU hangs waiting for response to MMIO read

Then how is userspace able to use a device after the device is removed?

To give more details, let's take the graphics driver (e.g. amdgpu) as
an example. The userspace programs call printf() to display "shutting
down xxx service" during shutdown/reboot, or the kernel calls printk()
to display something during shutdown/reboot. These can happen at any
time, even after we call pcie_port_device_remove() to disable the pcie
port on the graphic card.

The call stack is: printk() --> call_console_drivers() --> con->write()
--> vt_console_print() --> fbcon_putcs()

This scenario happens because userspace programs (or the kernel itself)
don't know whether a device is 'usable', they just use it, at any time.

This hardware behavior is a PCIe protocol violation (Bus Master should
not be involved in CPU MMIO transactions), and it will be fixed in new
revisions of hardware (add timeout mechanism for CPU read request,
whether or not Bus Master bit is cleared).

On some x86 platforms, radeon/amdgpu devices can cause similar problems
[1][2].

Once before I add a quirk to solve the LS7A problem but looks ugly.
After long time discussions, Bjorn Helgaas suggest simply remove the
pci_disable_device() in pcie_port_device_remove() and this patch do it
exactly.

[1] https://bugs.freedesktop.org/show_bug.cgi?id=97980
[2] https://bugs.freedesktop.org/show_bug.cgi?id=98638

Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
---
 drivers/pci/pcie/portdrv.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/pci/pcie/portdrv.c b/drivers/pci/pcie/portdrv.c
index 2cc2e60bcb39..9fe1fbca6476 100644
--- a/drivers/pci/pcie/portdrv.c
+++ b/drivers/pci/pcie/portdrv.c
@@ -501,7 +501,6 @@ static void pcie_port_device_remove(struct pci_dev *dev)
 {
 	device_for_each_child(&dev->dev, NULL, remove_iter);
 	pci_free_irq_vectors(dev);
-	pci_disable_device(dev);
 }
 
 /**
-- 
2.39.0

[PATCH V3 2/2] PCI: loongson: Improve the MRRS quirk for LS7A

[Huacai Chen]

In new revision of LS7A, some PCIe ports support larger value than 256,
but their maximum supported MRRS values are not detectable. Moreover,
the current loongson_mrrs_quirk() cannot avoid devices increasing its
MRRS after pci_enable_device(), and some devices (e.g. Realtek 8169)
will actually set a big value in its driver. So the only possible way
is configure MRRS of all devices in BIOS, and add a pci host bridge bit
flag (i.e., no_inc_mrrs) to stop the increasing MRRS operations.

However, according to PCIe Spec, it is legal for an OS to program any
value for MRRS, and it is also legal for an endpoint to generate a Read
Request with any size up to its MRRS. As the hardware engineers say, the
root cause here is LS7A doesn't break up large read requests. In detail,
LS7A PCIe port reports CA (Completer Abort) if it receives a Memory Read
request with a size that's "too big" ("too big" means larger than the
PCIe ports can handle, which means 256 for some ports and 4096 for the
others, and of course this is a problem in the LS7A's hardware design).

Link: Link: https://bugzilla.kernel.org/show_bug.cgi?id=216884
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
---
 drivers/pci/controller/pci-loongson.c | 44 +++++++++------------------
 drivers/pci/pci.c                     |  6 ++++
 include/linux/pci.h                   |  1 +
 3 files changed, 22 insertions(+), 29 deletions(-)

diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
index 05c50408f13b..759ec211c17b 100644
--- a/drivers/pci/controller/pci-loongson.c
+++ b/drivers/pci/controller/pci-loongson.c
@@ -75,37 +75,23 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
 			DEV_LS7A_LPC, system_bus_quirk);
 
-static void loongson_mrrs_quirk(struct pci_dev *dev)
+static void loongson_mrrs_quirk(struct pci_dev *pdev)
 {
-	struct pci_bus *bus = dev->bus;
-	struct pci_dev *bridge;
-	static const struct pci_device_id bridge_devids[] = {
-		{ PCI_VDEVICE(LOONGSON, DEV_PCIE_PORT_0) },
-		{ PCI_VDEVICE(LOONGSON, DEV_PCIE_PORT_1) },
-		{ PCI_VDEVICE(LOONGSON, DEV_PCIE_PORT_2) },
-		{ 0, },
-	};
-
-	/* look for the matching bridge */
-	while (!pci_is_root_bus(bus)) {
-		bridge = bus->self;
-		bus = bus->parent;
-		/*
-		 * Some Loongson PCIe ports have a h/w limitation of
-		 * 256 bytes maximum read request size. They can't handle
-		 * anything larger than this. So force this limit on
-		 * any devices attached under these ports.
-		 */
-		if (pci_match_id(bridge_devids, bridge)) {
-			if (pcie_get_readrq(dev) > 256) {
-				pci_info(dev, "limiting MRRS to 256\n");
-				pcie_set_readrq(dev, 256);
-			}
-			break;
-		}
-	}
+	/*
+	 * Some Loongson PCIe ports have h/w limitations of maximum read
+	 * request size. They can't handle anything larger than this. So
+	 * force this limit on any devices attached under these ports.
+	 */
+	struct pci_host_bridge *bridge = pci_find_host_bridge(pdev->bus);
+
+	bridge->no_inc_mrrs = 1;
 }
-DECLARE_PCI_FIXUP_ENABLE(PCI_ANY_ID, PCI_ANY_ID, loongson_mrrs_quirk);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
+			DEV_PCIE_PORT_0, loongson_mrrs_quirk);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
+			DEV_PCIE_PORT_1, loongson_mrrs_quirk);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
+			DEV_PCIE_PORT_2, loongson_mrrs_quirk);
 
 static void loongson_pci_pin_quirk(struct pci_dev *pdev)
 {
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index fba95486caaf..ae88210a12c7 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -6033,6 +6033,7 @@ int pcie_set_readrq(struct pci_dev *dev, int rq)
 {
 	u16 v;
 	int ret;
+	struct pci_host_bridge *bridge = pci_find_host_bridge(dev->bus);
 
 	if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
 		return -EINVAL;
@@ -6051,6 +6052,11 @@ int pcie_set_readrq(struct pci_dev *dev, int rq)
 
 	v = (ffs(rq) - 8) << 12;
 
+	if (bridge->no_inc_mrrs) {
+		if (rq > pcie_get_readrq(dev))
+			return -EINVAL;
+	}
+
 	ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
 						  PCI_EXP_DEVCTL_READRQ, v);
 
diff --git a/include/linux/pci.h b/include/linux/pci.h
index adffd65e84b4..3df2049ec4a8 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -572,6 +572,7 @@ struct pci_host_bridge {
 	void		*release_data;
 	unsigned int	ignore_reset_delay:1;	/* For entire hierarchy */
 	unsigned int	no_ext_tags:1;		/* No Extended Tags */
+	unsigned int	no_inc_mrrs:1;		/* No Increase MRRS */
 	unsigned int	native_aer:1;		/* OS may use PCIe AER */
 	unsigned int	native_pcie_hotplug:1;	/* OS may use PCIe hotplug */
 	unsigned int	native_shpc_hotplug:1;	/* OS may use SHPC hotplug */
-- 
2.39.0

Re: [PATCH V3 1/2] PCI: Omit pci_disable_device() in pcie_port_device_remove()

[Bjorn Helgaas]

On Wed, Feb 01, 2023 at 10:38:02AM +0800, Huacai Chen wrote:
> This patch has a long story.

Understatement of the year :)

> @@ -501,7 +501,6 @@ static void pcie_port_device_remove(struct pci_dev *dev)
>  {
>  	device_for_each_child(&dev->dev, NULL, remove_iter);
>  	pci_free_irq_vectors(dev);
> -	pci_disable_device(dev);

Interesting.  What I had in mind was keeping the original
pcie_port_device_remove() unchanged, adding the new
pcie_portdrv_shutdown(), and omitting pci_disable_device() just from 
pcie_portdrv_shutdown().

I haven't thought about the implications of omitting
pci_disable_device() from the .remove() method
(pcie_port_device_remove()).

Just pointing this out quickly before going to bed in case you have a
chance to think about what's the best strategy :)

Bjorn

Re: [PATCH V3 1/2] PCI: Omit pci_disable_device() in pcie_port_device_remove()

[Huacai Chen]

On Wed, Feb 1, 2023 at 11:42 AM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> On Wed, Feb 01, 2023 at 10:38:02AM +0800, Huacai Chen wrote:
> > This patch has a long story.
>
> Understatement of the year :)
>
> > @@ -501,7 +501,6 @@ static void pcie_port_device_remove(struct pci_dev *dev)
> >  {
> >       device_for_each_child(&dev->dev, NULL, remove_iter);
> >       pci_free_irq_vectors(dev);
> > -     pci_disable_device(dev);
>
> Interesting.  What I had in mind was keeping the original
> pcie_port_device_remove() unchanged, adding the new
> pcie_portdrv_shutdown(), and omitting pci_disable_device() just from
> pcie_portdrv_shutdown().
>
> I haven't thought about the implications of omitting
> pci_disable_device() from the .remove() method
> (pcie_port_device_remove()).
>
> Just pointing this out quickly before going to bed in case you have a
> chance to think about what's the best strategy :)
Sorry that I misunderstood, V4 will be coming quickly.

Huacai
>
> Bjorn