Skip to contentSkip to conversation

PCI: rcar: Add missing COMMON_CLK dependency

PATCH In review [email protected] 6 participants 13 messages Sep 7, 2021 – Sep 30, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 PCI: rcar: Add missing COMMON_CLK dependency

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

---
 drivers/pci/controller/Kconfig | 2 ++
 1 file changed, 2 insertions(+)
drivers/pci/controller/Kconfig +2 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
 config PCIE_RCAR_HOST
 	bool "Renesas R-Car PCIe host controller"
 	depends on ARCH_RENESAS || COMPILE_TEST
+	depends on COMMON_CLK
 	depends on PCI_MSI_IRQ_DOMAIN
 	help
 	  Say Y here if you want PCIe controller support on R-Car SoCs in host
@@ -74,6 +75,7 @@ config PCIE_RCAR_HOST
 config PCIE_RCAR_EP
 	bool "Renesas R-Car PCIe endpoint controller"
 	depends on ARCH_RENESAS || COMPILE_TEST
+	depends on COMMON_CLK
 	depends on PCI_ENDPOINT
 	help
 	  Say Y here if you want PCIe controller support on R-Car SoCs in
Signature · 1 line
2.33.0
#1 lore raw
Quoted 17 lines · On Tue, Sep 07, 2021 at 04:45:12PM +0200, [email protected] wrote:

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>
Quoted 3 lines

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

Hi Stephen,

can you please have a look into this please to see if there is a better way of fixing this breakage ? We should aim for one of the upcoming rcX, thank you very much.

Lorenzo

Quoted 27 lines
---
 drivers/pci/controller/Kconfig | 2 ++
 1 file changed, 2 insertions(+)
drivers/pci/controller/Kconfig +2 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
 config PCIE_RCAR_HOST
 	bool "Renesas R-Car PCIe host controller"
 	depends on ARCH_RENESAS || COMPILE_TEST
+	depends on COMMON_CLK
 	depends on PCI_MSI_IRQ_DOMAIN
 	help
 	  Say Y here if you want PCIe controller support on R-Car SoCs in host
@@ -74,6 +75,7 @@ config PCIE_RCAR_HOST
 config PCIE_RCAR_EP
 	bool "Renesas R-Car PCIe endpoint controller"
 	depends on ARCH_RENESAS || COMPILE_TEST
+	depends on COMMON_CLK
 	depends on PCI_ENDPOINT
 	help
 	  Say Y here if you want PCIe controller support on R-Car SoCs in
Signature · 1 line
2.33.0
#2 lore raw
In reply to [email protected]

Hi Marek,

Thanks for your patch!

Quoted 5 lines · On Tue, Sep 7, 2021 at 4:45 PM <[email protected]> wrote:

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'

This is a link failure for the host driver...

Quoted 24 lines

This should be OK, since all platforms shipping this controller also need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>

---

(quoted diff) +1 -1
+CC Stephen, please double-check whether this is the right approach or
    whether there is some better option
---
 drivers/pci/controller/Kconfig | 2 ++
 1 file changed, 2 insertions(+)
drivers/pci/controller/Kconfig +0 -0
index 326f7d13024f..ee6f5e525d3a 100644

drivers/pci/controller/Kconfig config PCI_RCAR_GEN2 · line 69

Quoted 4 lines
drivers/pci/controller/Kconfig +1 -0
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
 config PCIE_RCAR_HOST
        bool "Renesas R-Car PCIe host controller"
        depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK

This part is OK.

drivers/pci/controller/Kconfig config PCIE_RCAR_HOST · line 78

Quoted 8 lines
drivers/pci/controller/Kconfig +1 -0
        depends on PCI_MSI_IRQ_DOMAIN
        help
          Say Y here if you want PCIe controller support on R-Car SoCs in host
@@ -74,6 +75,7 @@ config PCIE_RCAR_HOST
 config PCIE_RCAR_EP
        bool "Renesas R-Car PCIe endpoint controller"
        depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK

... so why did you add a dependency to the endpoint driver, too?

drivers/pci/controller/Kconfig config PCIE_RCAR_HOST · line 81

Quoted 3 lines
drivers/pci/controller/Kconfig +0 -0
        depends on PCI_ENDPOINT
        help
          Say Y here if you want PCIe controller support on R-Car SoCs in

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
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
#3 lore raw

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 11 lines

drivers/pci/controller/Kconfig config PCI_RCAR_GEN2 · line 69

Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK

This part is OK.

This part is also identical in the patch from Arnd, so you can just pick that one as a fix and be done with it:

[PATCH] PCI: rcar: add COMMON_CLK dependency
https://patchwork.kernel.org/project/linux-pci/patch/[email protected]/

#4 lore raw
Quoted 16 lines · Marek Vasut

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 11 lines

drivers/pci/controller/Kconfig config PCI_RCAR_GEN2 · line 69

Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK

This part is OK.

Quoted 5 lines

This part is also identical in the patch from Arnd, so you can just pick that one as a fix and be done with it:

[PATCH] PCI: rcar: add COMMON_CLK dependency
https://patchwork.kernel.org/project/linux-pci/patch/[email protected]/

It is not strictly identical (Arnd's patch only touches the COMPILE_TEST option).

Bjorn, shall we pick Arnd's patch up then ? We should be fixing this in one of the upcoming -rcs since we introduced it in the last merge window.

Thanks,
Lorenzo

#5 lore raw
Quoted 23 lines · Lorenzo Pieralisi
Quoted 16 lines · Marek Vasut

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 11 lines

drivers/pci/controller/Kconfig config PCI_RCAR_GEN2 · line 69

Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK

This part is OK.

Quoted 5 lines

This part is also identical in the patch from Arnd, so you can just pick that one as a fix and be done with it:

[PATCH] PCI: rcar: add COMMON_CLK dependency
https://patchwork.kernel.org/project/linux-pci/patch/[email protected]/

Quoted 6 lines

It is not strictly identical (Arnd's patch only touches the COMPILE_TEST option).

Bjorn, shall we pick Arnd's patch up then ? We should be fixing this in one of the upcoming -rcs since we introduced it in the last merge window.

IIUC, a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook") appeared in v5.15-rc1 and added a use of __clk_is_enabled(), which is only available when COMMON_CLK=y.

PCIE_RCAR_HOST depends on ARCH_RENESAS || COMPILE_TEST. ARCH_RENESAS is an ARM64 platform, so when COMPILE_TEST is not set, I think we get COMMON_CLK=y via this:

config ARM64
select COMMON_CLK

But when ARCH_RENESAS is not set and COMPILE_TEST=y, there's nothing that enforces the dependency on COMMON_CLK. Personally I like the first hunk of Marek's patch at [1] because the dependency on COMMON_CLK is explicit:

  config PCIE_RCAR_HOST
    depends on ARCH_RENESAS || COMPILE_TEST
    depends on COMMON_CLK

Whereas Arnd's patch [2] implicitly depends on the fact that the platform selects COMMON_CLK:

  config PCIE_RCAR_HOST
    depends on ARCH_RENESAS || (COMMON_CLK && COMPILE_TEST)

But either is fine and I agree we should fix it soonish.

Bjorn

[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/[email protected]/

#6 lore raw
Quoted 32 lines · Bjorn Helgaas
Quoted 16 lines · Lorenzo Pieralisi
Quoted 14 lines · Marek Vasut

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 10 lines
Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK
Quoted 3 lines

Bjorn, shall we pick Arnd's patch up then ? We should be fixing this in one of the upcoming -rcs since we introduced it in the last merge window.

IIUC, a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook") appeared in v5.15-rc1 and added a use of __clk_is_enabled(), which is only available when COMMON_CLK=y.

PCIE_RCAR_HOST depends on ARCH_RENESAS || COMPILE_TEST. ARCH_RENESAS is an ARM64 platform, so when COMPILE_TEST is not set, I think we get COMMON_CLK=y via this:

config ARM64
select COMMON_CLK

Quoted 8 lines

But when ARCH_RENESAS is not set and COMPILE_TEST=y, there's nothing that enforces the dependency on COMMON_CLK. Personally I like the first hunk of Marek's patch at [1] because the dependency on COMMON_CLK is explicit:

  config PCIE_RCAR_HOST
    depends on ARCH_RENESAS || COMPILE_TEST
    depends on COMMON_CLK

Agreed, Marek's version is clearer than mine, please use that.

Arnd

#7 lore raw
Quoted 38 lines · Arnd Bergmann
Quoted 37 lines · Bjorn Helgaas
Quoted 16 lines · Lorenzo Pieralisi
Quoted 14 lines · Marek Vasut

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 10 lines
Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK
Quoted 3 lines

Bjorn, shall we pick Arnd's patch up then ? We should be fixing this in one of the upcoming -rcs since we introduced it in the last merge window.

IIUC, a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook") appeared in v5.15-rc1 and added a use of __clk_is_enabled(), which is only available when COMMON_CLK=y.

PCIE_RCAR_HOST depends on ARCH_RENESAS || COMPILE_TEST. ARCH_RENESAS is an ARM64 platform, so when COMPILE_TEST is not set, I think we get COMMON_CLK=y via this:

config ARM64
select COMMON_CLK

But when ARCH_RENESAS is not set and COMPILE_TEST=y, there's nothing that enforces the dependency on COMMON_CLK. Personally I like the first hunk of Marek's patch at [1] because the dependency on COMMON_CLK is explicit:

Quoted 5 lines
Quoted 3 lines
  config PCIE_RCAR_HOST
    depends on ARCH_RENESAS || COMPILE_TEST
    depends on COMMON_CLK

Agreed, Marek's version is clearer than mine, please use that.

But PCIE_RCAR_EP does not need the dependency.

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
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
#8 lore raw
Quoted 39 lines · Geert Uytterhoeven
Quoted 38 lines · Arnd Bergmann
Quoted 37 lines · Bjorn Helgaas
Quoted 16 lines · Lorenzo Pieralisi
Quoted 14 lines · Marek Vasut

On 9/21/21 6:08 PM, Geert Uytterhoeven wrote:

[...]

Quoted 10 lines
Quoted 9 lines
drivers/pci/controller/Kconfig +1 -0
index 326f7d13024f..ee6f5e525d3a 100644
@@ -66,6 +66,7 @@ config PCI_RCAR_GEN2
  config PCIE_RCAR_HOST
         bool "Renesas R-Car PCIe host controller"
         depends on ARCH_RENESAS || COMPILE_TEST
+       depends on COMMON_CLK
Quoted 3 lines

Bjorn, shall we pick Arnd's patch up then ? We should be fixing this in one of the upcoming -rcs since we introduced it in the last merge window.

IIUC, a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook") appeared in v5.15-rc1 and added a use of __clk_is_enabled(), which is only available when COMMON_CLK=y.

PCIE_RCAR_HOST depends on ARCH_RENESAS || COMPILE_TEST. ARCH_RENESAS is an ARM64 platform, so when COMPILE_TEST is not set, I think we get COMMON_CLK=y via this:

config ARM64
select COMMON_CLK

But when ARCH_RENESAS is not set and COMPILE_TEST=y, there's nothing that enforces the dependency on COMMON_CLK. Personally I like the first hunk of Marek's patch at [1] because the dependency on COMMON_CLK is explicit:

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
  config PCIE_RCAR_HOST
    depends on ARCH_RENESAS || COMPILE_TEST
    depends on COMMON_CLK

Agreed, Marek's version is clearer than mine, please use that.

But PCIE_RCAR_EP does not need the dependency.

Right, that's why I said the *first* hunk of Marek's patch. I would apply that and skip the second one.

maintainer
#9 lore raw
In reply to [email protected]

+linux-clk as I don't regularly read my inbox :/

Quoting [email protected] (2021-09-07 07:45:12)

Quoted 17 lines

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>
Quoted 3 lines

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

Stop using __clk_is_enabled()? I don't quite understand what's going on in the code but __clk_is_enabled() should really go away. I thought we were close to doing that but now I see a handful of calls have come up. The API should be replaced by clk_hw_is_enabled() and then removed. We move it to clk_hw API so that only clk providers can look at it.

Sigh!

Anyway, fixing the dependency is "ok" but really the long term fix would be to not use a "is this clk enabled" sort of API. If I'm reading the code correctly, this is some sort of fault handler that's trying to avoid hanging the bus while handling the fault so it tries to make sure the clk is enabled first? Is it a problem if the clk is not actually enabled here? Would runtime PM enable state of the device work just as well?

#10 lore raw

Hi Stephen,

Quoted 24 lines · Stephen Boyd

+linux-clk as I don't regularly read my inbox :/

Quoting [email protected] (2021-09-07 07:45:12)

Quoted 17 lines

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>
Quoted 3 lines

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

Quoted 5 lines

Stop using __clk_is_enabled()? I don't quite understand what's going on in the code but __clk_is_enabled() should really go away. I thought we were close to doing that but now I see a handful of calls have come up. The API should be replaced by clk_hw_is_enabled() and then removed. We move it to clk_hw API so that only clk providers can look at it.

But this is not a clk provider...

Quoted 1 line

Sigh!

;-)

Quoted 7 lines

Anyway, fixing the dependency is "ok" but really the long term fix would be to not use a "is this clk enabled" sort of API. If I'm reading the code correctly, this is some sort of fault handler that's trying to avoid hanging the bus while handling the fault so it tries to make sure the clk is enabled first? Is it a problem if the clk is not actually enabled here? Would runtime PM enable state of the device work just as well?

Thanks, checking Runtime PM state is a good suggestion. Doing so would require caching a pointer to the PCIe struct device instead of the struct clk.
However, pcie_bus_clk is not the module clock, which is managed by Runtime PM, but the PCIe bus clock, which is managed explicitly by the driver.
However, I believe that we are checking the wrong clock, as register access needs the module clock to be enabled, not the PCIe bus clock? As the driver just calls pm_runtime_get_sync() and clk_prepare_enable() in .probe(), and never touches Runtime PM status or the PCIe bus clock during the further lifetime of the driver (it cannot be unloaded), both the module clock and the PCIe bus clock should always[*] be enabled when the static copy of the remapped PCIe controller address is valid. [*] Modulo system-wide power transitions like s2ram. Does
pm_runtime_suspended() reflect that state, too?

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
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
maintainer
#11 lore raw

Quoting Geert Uytterhoeven (2021-09-30 01:01:24)

Quoted 33 lines

Hi Stephen,

Quoted 24 lines · Stephen Boyd

+linux-clk as I don't regularly read my inbox :/

Quoting [email protected] (2021-09-07 07:45:12)

Quoted 17 lines

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>
Quoted 3 lines

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

Quoted 5 lines

Stop using __clk_is_enabled()? I don't quite understand what's going on in the code but __clk_is_enabled() should really go away. I thought we were close to doing that but now I see a handful of calls have come up. The API should be replaced by clk_hw_is_enabled() and then removed. We move it to clk_hw API so that only clk providers can look at it.

Quoted 5 lines

But this is not a clk provider...

Quoted 1 line

Sigh!

;-)

Exactly!

Quoted 9 lines
Quoted 8 lines

Anyway, fixing the dependency is "ok" but really the long term fix would be to not use a "is this clk enabled" sort of API. If I'm reading the code correctly, this is some sort of fault handler that's trying to avoid hanging the bus while handling the fault so it tries to make sure the clk is enabled first? Is it a problem if the clk is not actually enabled here? Would runtime PM enable state of the device work just as well?

Quoted 16 lines

Thanks, checking Runtime PM state is a good suggestion. Doing so would require caching a pointer to the PCIe struct device instead of the struct clk.
However, pcie_bus_clk is not the module clock, which is managed by Runtime PM, but the PCIe bus clock, which is managed explicitly by the driver.
However, I believe that we are checking the wrong clock, as register access needs the module clock to be enabled, not the PCIe bus clock? As the driver just calls pm_runtime_get_sync() and clk_prepare_enable() in .probe(), and never touches Runtime PM status or the PCIe bus clock during the further lifetime of the driver (it cannot be unloaded), both the module clock and the PCIe bus clock should always[*] be enabled when the static copy of the remapped PCIe controller address is valid. [*] Modulo system-wide power transitions like s2ram. Does
pm_runtime_suspended() reflect that state, too?

Great! If that's all correct then simply removing the call to __clk_is_enabled() should work. Can we do that?

#12 lore raw

Hi Stephen,

Quoted 66 lines · Stephen Boyd

Quoting Geert Uytterhoeven (2021-09-30 01:01:24)

Quoted 31 lines · Stephen Boyd
Quoted 24 lines

+linux-clk as I don't regularly read my inbox :/

Quoting [email protected] (2021-09-07 07:45:12)

Quoted 17 lines

From: Marek Vasut <[email protected]>

Add COMMON_CLK dependency, otherwise the following build error occurs:
arm-linux-gnueabi-ld: drivers/pci/controller/pcie-rcar-host.o: in function `rcar_pcie_aarch32_abort_handler':
pcie-rcar-host.c:(.text+0xdd0): undefined reference to `__clk_is_enabled'
This should be OK, since all platforms shipping this controller also
need COMMON_CLK enabled for their clock driver.

Fixes a115b1bd3af0 ("PCI: rcar: Add L1 link state fix into data abort hook")
Cc Bjorn Helgaas <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Lorenzo Pieralisi <[email protected]>
Cc Stephen Boyd <[email protected]>
Cc Wolfram Sang <[email protected]>
Cc Yoshihiro Shimoda <[email protected]>
Quoted 3 lines

---
+CC Stephen, please double-check whether this is the right approach or
whether there is some better option

Quoted 5 lines

Stop using __clk_is_enabled()? I don't quite understand what's going on in the code but __clk_is_enabled() should really go away. I thought we were close to doing that but now I see a handful of calls have come up. The API should be replaced by clk_hw_is_enabled() and then removed. We move it to clk_hw API so that only clk providers can look at it.

Quoted 5 lines

But this is not a clk provider...

Quoted 1 line

Sigh!

;-)

Exactly!

Quoted 9 lines
Quoted 8 lines

Anyway, fixing the dependency is "ok" but really the long term fix would be to not use a "is this clk enabled" sort of API. If I'm reading the code correctly, this is some sort of fault handler that's trying to avoid hanging the bus while handling the fault so it tries to make sure the clk is enabled first? Is it a problem if the clk is not actually enabled here? Would runtime PM enable state of the device work just as well?

Quoted 16 lines

Thanks, checking Runtime PM state is a good suggestion. Doing so would require caching a pointer to the PCIe struct device instead of the struct clk.
However, pcie_bus_clk is not the module clock, which is managed by Runtime PM, but the PCIe bus clock, which is managed explicitly by the driver.
However, I believe that we are checking the wrong clock, as register access needs the module clock to be enabled, not the PCIe bus clock? As the driver just calls pm_runtime_get_sync() and clk_prepare_enable() in .probe(), and never touches Runtime PM status or the PCIe bus clock during the further lifetime of the driver (it cannot be unloaded), both the module clock and the PCIe bus clock should always[*] be enabled when the static copy of the remapped PCIe controller address is valid. [*] Modulo system-wide power transitions like s2ram. Does
pm_runtime_suspended() reflect that state, too?

Quoted 2 lines

Great! If that's all correct then simply removing the call to __clk_is_enabled() should work. Can we do that?

We first have to double-check that pm_runtime_suspended() reflects the state, as the reason behind the fault handler is to fix lock-ups during system-wide power transitions.

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
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