The PCI rcar-gen2 does not call any clk_prepare_enable().
This lead to an access failure when the driver tries to access
the IP (at least on a RZ/N1D platform).
Prepare and enable clocks using the bulk version of
clk_prepare_enable() in order to prepare and enable all clocks
attached to this device.
Signed-off-by: Herve Codina <[email protected]>
---
drivers/pci/controller/pci-rcar-gen2.c | 28 ++++++++++++++++++++++++--
1 file changed, 26 insertions(+), 2 deletions(-)
diff --git a/drivers/pci/controller/pci-rcar-gen2.c b/drivers/pci/controller/pci-rcar-gen2.c
index 35804ea394fd..528bc3780e01 100644
--- a/drivers/pci/controller/pci-rcar-gen2.c
+++ b/drivers/pci/controller/pci-rcar-gen2.c
@@ -8,6 +8,7 @@
* Author: Valentine Barshak <[email protected]>
*/
+#include <linux/clk.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/interrupt.h>
@@ -99,6 +100,8 @@ struct rcar_pci {
struct resource mem_res;
struct resource *cfg_res;
int irq;
+ struct clk_bulk_data *clocks;
+ int nclocks;
};
/* PCI configuration space operations */
@@ -282,6 +285,7 @@ static int rcar_pci_probe(struct platform_device *pdev)
struct rcar_pci *priv;
struct pci_host_bridge *bridge;
void __iomem *reg;
+ int ret;
bridge = devm_pci_alloc_host_bridge(dev, sizeof(*priv));
if (!bridge)
@@ -305,13 +309,25 @@ static int rcar_pci_probe(struct platform_device *pdev)
priv->mem_res = *mem_res;
priv->cfg_res = cfg_res;
+ ret = devm_clk_bulk_get_all(dev, &priv->clocks);
+ if (ret < 0) {
+ dev_err(dev, "failed to get clocks %d\n", ret);
+ return ret;
+ }
+ priv->nclocks = ret;
+
+ ret = clk_bulk_prepare_enable(priv->nclocks, priv->clocks);
+ if (ret)
+ return ret;
+
priv->irq = platform_get_irq(pdev, 0);
priv->reg = reg;
priv->dev = dev;
if (priv->irq < 0) {
dev_err(dev, "no valid irq found\n");
- return priv->irq;
+ ret = priv->irq;
+ goto disable_clocks;
}
bridge->ops = &rcar_pci_ops;
@@ -320,7 +336,15 @@ static int rcar_pci_probe(struct platform_device *pdev)
rcar_pci_setup(priv);
- return pci_host_probe(bridge);
+ ret = pci_host_probe(bridge);
+ if (ret < 0)
+ goto disable_clocks;
+
+ return 0;
+
+disable_clocks:
+ clk_bulk_disable_unprepare(priv->nclocks, priv->clocks);
+ return ret;
}
static const struct of_device_id rcar_pci_of_match[] = {
--
2.35.1
Hi Hervé,
Thanks for your patch!
On Thu, Apr 14, 2022 at 9:40 AM Herve Codina <[email protected]> wrote:
> The PCI rcar-gen2 does not call any clk_prepare_enable().
Correct, this driver manages the clocks indirectly through Runtime PM.
> This lead to an access failure when the driver tries to access
> the IP (at least on a RZ/N1D platform).
I expect adding
power-domans = <&sysctrl>;
to the pci_usb node makes this patch redundant.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Hi Geert,
On Thu, 14 Apr 2022 10:45:54 +0200
Geert Uytterhoeven <[email protected]> wrote:
> Hi Hervé,
>
> Thanks for your patch!
>
> On Thu, Apr 14, 2022 at 9:40 AM Herve Codina <[email protected]> wrote:
> > The PCI rcar-gen2 does not call any clk_prepare_enable().
>
> Correct, this driver manages the clocks indirectly through Runtime PM.
>
> > This lead to an access failure when the driver tries to access
> > the IP (at least on a RZ/N1D platform).
>
> I expect adding
>
> power-domans = <&sysctrl>;
>
> to the pci_usb node makes this patch redundant.
Seems not enough.
I tried what you suggest :
- Added 'power-domains = <&systrl>;' to the pci_usb node
- Added missing '#power-domain-cells = <0>;' to sysctrl node
- Reverted my patch.
The system crashed at boot:
--- 8< ---
...
[ 0.705309] loop: module loaded
[ 0.709597] ehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver
[ 0.716804] ehci-pci: EHCI PCI platform driver
[ 0.721716] ohci_hcd: USB 1.1 'Open' Host Controller (OHCI) Driver
[ 0.728522] ohci-pci: OHCI PCI platform driver
[ 0.733581] usbcore: registered new interface driver usb-storage
[ 0.740458] ThumbEE CPU extension supported.
[ 0.745093] Registering SWP/SWPB emulation handler
[ 0.797518] rzn1-pinctrl 40067000.pinctrl: probed
[ 0.803311] pci-rcar-gen2 40030000.pci: host bridge /soc/pci@40030000 ranges:
[ 0.811255] pci-rcar-gen2 40030000.pci: MEM 0x0040020000..0x004002ffff -> 0x0040020000
[ 0.820373] pci-rcar-gen2 40030000.pci: IB MEM 0x0080000000..0x00bfffffff -> 0x0080000000
[ 0.829609] 8<--- cut here ---
[ 0.832958] Unhandled fault: external abort on non-linefetch (0x1008) at 0x90b5f848
[ 0.841259] [90b5f848] *pgd=82149811, *pte=40030653, *ppte=40030453
[ 0.848093] Internal error: : 1008 [#1] SMP ARM
[ 0.853024] Modules linked in:
[ 0.856398] CPU: 0 PID: 31 Comm: kworker/u4:1 Not tainted 5.18.0-rc2-00009-g803ee9fd9fa5-dirty #5
[ 0.865998] Hardware name: Generic DT based system
[ 0.871176] Workqueue: events_unbound deferred_probe_work_func
[ 0.877539] PC is at rcar_pci_probe+0x15c/0x2f8
[ 0.882454] LR is at _raw_spin_unlock_irqrestore+0x24/0x2c
[ 0.888434] pc : [<803ea428>] lr : [<804dc9b0>] psr: 60000013
[ 0.895193] sp : 90aa5e40 ip : 8217c4e0 fp : 00000000
[ 0.900857] r10: 80e7bd30 r9 : 80000000 r8 : 40000000
[ 0.906532] r7 : 80000000 r6 : 8217c410 r5 : 821d3400 r4 : 90b5f000
[ 0.913580] r3 : 00000009 r2 : 5c120fb6 r1 : 60000013 r0 : 00000000
[ 0.920646] Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
[ 0.928365] Control: 10c5387d Table: 8000406a DAC: 00000051
...
--- 8< ---
I also added a trace printk in r9a06g032-clocks.c and
r9a06g032_attach_dev() was never called.
Did I miss to set something ?
Regards,
Hervé
Hi Hervé,
On Thu, Apr 14, 2022 at 1:29 PM Herve Codina <[email protected]> wrote:
> On Thu, 14 Apr 2022 10:45:54 +0200
> Geert Uytterhoeven <[email protected]> wrote:
> > On Thu, Apr 14, 2022 at 9:40 AM Herve Codina <[email protected]> wrote:
> > > The PCI rcar-gen2 does not call any clk_prepare_enable().
> >
> > Correct, this driver manages the clocks indirectly through Runtime PM.
> >
> > > This lead to an access failure when the driver tries to access
> > > the IP (at least on a RZ/N1D platform).
> >
> > I expect adding
> >
> > power-domans = <&sysctrl>;
> >
> > to the pci_usb node makes this patch redundant.
>
> Seems not enough.
> I tried what you suggest :
> - Added 'power-domains = <&systrl>;' to the pci_usb node
> - Added missing '#power-domain-cells = <0>;' to sysctrl node
> - Reverted my patch.
>
> The system crashed at boot:
> [ 0.832958] Unhandled fault: external abort on non-linefetch (0x1008) at 0x90b5f848
That's indeed a typical symptom of accessing a module's registers
while the module's clock is disabled.
> I also added a trace printk in r9a06g032-clocks.c and
> r9a06g032_attach_dev() was never called.
>
> Did I miss to set something ?
Do you have CONFIG_PM and CONFIG_PM_GENERIC_DOMAINS
enabled?
Apparently ARCH_RZN1 does not select these options yet.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Hi Geert,
On Thu, 14 Apr 2022 13:48:22 +0200
Geert Uytterhoeven <[email protected]> wrote:
> Hi Hervé,
>
> On Thu, Apr 14, 2022 at 1:29 PM Herve Codina <[email protected]> wrote:
> > On Thu, 14 Apr 2022 10:45:54 +0200
> > Geert Uytterhoeven <[email protected]> wrote:
> > > On Thu, Apr 14, 2022 at 9:40 AM Herve Codina <[email protected]> wrote:
> > > > The PCI rcar-gen2 does not call any clk_prepare_enable().
> > >
> > > Correct, this driver manages the clocks indirectly through Runtime PM.
> > >
> > > > This lead to an access failure when the driver tries to access
> > > > the IP (at least on a RZ/N1D platform).
> > >
> > > I expect adding
> > >
> > > power-domans = <&sysctrl>;
> > >
> > > to the pci_usb node makes this patch redundant.
> >
> > Seems not enough.
> > I tried what you suggest :
> > - Added 'power-domains = <&systrl>;' to the pci_usb node
> > - Added missing '#power-domain-cells = <0>;' to sysctrl node
> > - Reverted my patch.
> >
> > The system crashed at boot:
>
> > [ 0.832958] Unhandled fault: external abort on non-linefetch (0x1008) at 0x90b5f848
>
> That's indeed a typical symptom of accessing a module's registers
> while the module's clock is disabled.
>
> > I also added a trace printk in r9a06g032-clocks.c and
> > r9a06g032_attach_dev() was never called.
> >
> > Did I miss to set something ?
>
> Do you have CONFIG_PM and CONFIG_PM_GENERIC_DOMAINS
> enabled?
> Apparently ARCH_RZN1 does not select these options yet.
>
Thanks a lot for pointing this.
I added select CONFIG_PM and CONFIG_PM_GENERIC_DOMAINS
in ARCH_RZN1 and it works.
I will remove my patch calling clk_bulk_prepare_enable() and
add some new patches to enable power domains in the v3 series.
Regards,
Hervé