x86/PCI: Get rid of custom x86 model comparison
PATCH 1/2 x86/PCI: Get rid of custom x86 model comparison
Switch the platform code to use x86_id_table and accompanying API instead of custom comparison against x86 CPU model.
--- arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
arch/x86/pci/intel_mid_pci.c +15 -2
@@ -28,10 +28,12 @@#include <linux/io.h> #include <linux/smp.h> +#include <asm/cpu_device_id.h> #include <asm/segment.h> #include <asm/pci_x86.h> #include <asm/hw_irq.h> #include <asm/io_apic.h> +#include <asm/intel-family.h> #include <asm/intel-mid.h> #define PCIE_CAP_OFFSET 0x100
@@ -211,9 +213,16 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where, where, size, value); } +static const struct x86_cpu_id intel_mid_cpu_ids[] = { +X86_MATCH_INTEL_FAM6_MODEL (ATOM_SILVERMONT_MID ,NULL ), + {} +}; +static int intel_mid_pci_irq_enable (struct pci_dev *dev) { +const struct x86_cpu_id *id;struct irq_alloc_info info; +u16 model =0 ;int polarity;int ret;u8 gsi;
@@ -227,8 +236,12 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)return ret; } -switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID : polarity =IOAPIC_POL_HIGH ;/* Special treatment for IRQ0 */
Signature · 1 line
2.27.0
PATCH 2/2 x86/PCI: Describe @reg for type1_access_ok()
Describe missed parameter in documentation of type1_access_ok(). Otherwise we get the following warning:
CHECK arch/x86/pci/intel_mid_pci.c
CC arch/x86/pci/intel_mid_pci.o
arch/x86/pci/intel_mid_pci.c:152: warning: Function parameter or member 'reg' not described in 'type1_access_ok'
--- arch/x86/pci/intel_mid_pci.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/pci/intel_mid_pci.c +1 -0
@@ -141,6 +141,7 @@ static int pci_device_update_fixed(struct pci_bus *bus, unsigned int devfn,* type1_access_ok - check whether to use type 1 * @bus: bus number * @devfn: device & function in question +* @reg: configuration register offset * * If the bus is on a Lincroft chip and it exists, or is not on a Lincroft at * all, the we can go ahead with any reads & writes. If it's on a Lincroft,
Signature · 1 line
2.27.0
Re: [PATCH v1 2/2] x86/PCI: Describe @reg for type1_access_ok()
Quoted 2 lines · Andy Shevchenko
Describe missed parameter in documentation of
type1_access_ok(). Otherwise we get the following warning:
Would you mind including the "make" invocation that runs this checking? I assume it's something like "make C=2 arch/x86/pci/intel_mid_pci.o"?
I'm not in the habit of running these checks, but maybe seeing how easy it is will help me and others get in the habit.
Quoted 24 lines
CHECK
arch/x86/pci/intel_mid_pci.c
CCarch/x86/pci/intel_mid_pci.o
arch/x86/pci/intel_mid_pci.c:152: warning: Function parameter or member 'reg' not described in 'type1_access_ok'Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/pci/intel_mid_pci.c | 1 + 1 file changed, 1 insertion(+)arch/x86/pci/intel_mid_pci.c +1
-0@@ -141,6 +141,7 @@ static int pci_device_update_fixed(struct pci_bus *bus, unsigned int devfn,* type1_access_ok - check whether to use type 1 * @bus: bus number * @devfn: device & function in question +* @reg: configuration register offset * * If the bus is on a Lincroft chip and it exists, or is not on a Lincroft at * all, the we can go ahead with any reads & writes. If it's on a Lincroft, Signature · 1 line
2.27.0
Re: [PATCH v1 2/2] x86/PCI: Describe @reg for type1_access_ok()
Quoted 7 lines · Bjorn Helgaas
Quoted 2 lines · Andy Shevchenko
Describe missed parameter in documentation of
type1_access_ok(). Otherwise we get the following warning:Would you mind including the "make" invocation that runs this checking? I assume it's something like "make C=2
arch/x86/pci/intel_mid_pci.o"?
No, it is not sparse, it's a kernel doc validation.
I guess make W=1 does it, but I can repeat my command line publicly again :-)
make W=1 C=1 CF=-D__CHECK_ENDIAN__ -j64
Quoted 27 lines
I'm not in the habit of running these checks, but maybe seeing how easy it is will help me and others get in the habit.
Quoted 24 lines
CHECK
arch/x86/pci/intel_mid_pci.c
CCarch/x86/pci/intel_mid_pci.o
arch/x86/pci/intel_mid_pci.c:152: warning: Function parameter or member 'reg' not described in 'type1_access_ok'Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/pci/intel_mid_pci.c | 1 + 1 file changed, 1 insertion(+)arch/x86/pci/intel_mid_pci.c +1
-0@@ -141,6 +141,7 @@ static int pci_device_update_fixed(struct pci_bus *bus, unsigned int devfn,* type1_access_ok - check whether to use type 1 * @bus: bus number * @devfn: device & function in question +* @reg: configuration register offset * * If the bus is on a Lincroft chip and it exists, or is not on a Lincroft at * all, the we can go ahead with any reads & writes. If it's on a Lincroft, Signature · 1 line
2.27.0
Signature · 2 lines
With Best Regards, Andy Shevchenko
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 245
Quoted 9 lines · Andy Shevchenko
arch/x86/pci/intel_mid_pci.c +6
-2-switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID : polarity =IOAPIC_POL_HIGH ;
The diff makes it _look_ like there's a behavior change, swapping silvermont and tangier. It appears from intel_mid_arch_setup() that INTEL_FAM6_ATOM_SILVERMONT_MID and tangier are related:
#define INTEL_FAM6_ATOM_SILVERMONT_MID 0x4A /* Merriefield */
...
case 0x3C:
case 0x4A:
__intel_mid_cpu_chip = INTEL_MID_CPU_CHIP_TANGIER;
x86_platform.legacy.rtc = 1;
break;But that's probably worth a note in the changelog. If you added a patch
to intel_mid_arch_setup() to this series to use the intel-family.h
#defines, this would be much more self explanatory.
This also all makes me wonder if intel_mid_identify_cpu() is really even necessary.
Does this fix any kinds of problems? It's a diffstat-challenged cleanup if not:
arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
That's not usually how cleanups look. :)
Quoted 44 lines · Andy Shevchenko
Switch the platform code to use
x86_id_tableand accompanying API instead of custom comparison against x86 CPU model.Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)arch/x86/pci/intel_mid_pci.c +9
-0@@ -28,10 +28,12 @@#include <linux/io.h> #include <linux/smp.h> +#include <asm/cpu_device_id.h> #include <asm/segment.h> #include <asm/pci_x86.h> #include <asm/hw_irq.h> #include <asm/io_apic.h> +#include <asm/intel-family.h> #include <asm/intel-mid.h> #define PCIE_CAP_OFFSET 0x100@@ -211,9 +213,16 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where, where, size, value); } +static const struct x86_cpu_id intel_mid_cpu_ids[] = { +X86_MATCH_INTEL_FAM6_MODEL (ATOM_SILVERMONT_MID ,NULL ), + {} +}; +static int intel_mid_pci_irq_enable (struct pci_dev *dev) { +const struct x86_cpu_id *id;struct irq_alloc_info info; +u16 model =0 ;int polarity;int ret;u8 gsi;@@ -227,8 +236,12 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)return ret;
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 244
Quoted 10 lines
arch/x86/pci/intel_mid_pci.c +6
-2} -switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID :
Is there a magic decoder ring somewhere that connects INTEL_MID_CPU_CHIP_TANGIER and INTEL_FAM6_ATOM_SILVERMONT_MID?
I don't know how to verify that the new code is equivalent to the old.
Or maybe the new code is *better* than the old, in which case the subject/commit log should mention that it's fixing or improving something.
Also, there are a number of other places that check for "intel_mid_identify_cpu() == INTEL_MID_CPU_CHIP_TANGIER":
mrfld_pinctrl_init
register_mrfld_power_btn
mrfld_legacy_rtc_init
mrfld_sd_init
spidev_platform_data
register_mid_wdt
sfi_parse_devs
atomisp_css_input_set_mode
Maybe they should all be changed together? Or maybe this needs an explanation about why some places need intel_mid_identify_cpu() and others need x86_match_cpu()?
Quoted 6 lines
arch/x86/pci/intel_mid_pci.c +0
-0polarity =IOAPIC_POL_HIGH ;/* Special treatment for IRQ0 */ Signature · 1 line
2.27.0
Re: [PATCH v1 2/2] x86/PCI: Describe @reg for type1_access_ok()
In reply to Andy ShevchenkoQuoted 5 lines · Andy Shevchenko
Quoted 4 lines · Bjorn Helgaas
Quoted 2 lines · Andy Shevchenko
Describe missed parameter in documentation of
type1_access_ok(). Otherwise we get the following warning:
Quoted 7 lines
Quoted 3 lines
Would you mind including the "make" invocation that runs this checking? I assume it's something like "make C=2
arch/x86/pci/intel_mid_pci.o"?No, it is not sparse, it's a kernel doc validation.
I guessmake W=1does it, but I can repeat my command line publicly again :-)make W=1 C=1 CF=-D__CHECK_ENDIAN__ -j64
Thanks. "make W=1 arch/x86/pci/" is enough. I would just put this in the commit log, e.g.,
Otherwise "make W=1 arch/x86/pci/" produces the following warning:
CHECK arch/x86/pci/intel_mid_pci.c
CC arch/x86/pci/intel_mid_pci.o
arch/x86/pci/intel_mid_pci.c:152: warning: Function parameter or member 'reg' not described in 'type1_access_ok'
Re: [PATCH v1 2/2] x86/PCI: Describe @reg for type1_access_ok()
Quoted 14 lines · Bjorn Helgaas
Quoted 5 lines · Andy Shevchenko
Quoted 4 lines · Bjorn Helgaas
Quoted 2 lines · Andy Shevchenko
Describe missed parameter in documentation of
type1_access_ok(). Otherwise we get the following warning:Quoted 7 lines
Quoted 3 lines
Would you mind including the "make" invocation that runs this checking? I assume it's something like "make C=2
arch/x86/pci/intel_mid_pci.o"?No, it is not sparse, it's a kernel doc validation.
I guessmake W=1does it, but I can repeat my command line publicly again :-)make W=1 C=1 CF=-D__CHECK_ENDIAN__ -j64
Quoted 8 lines
Thanks. "make W=1
arch/x86/pci/" is enough. I would just put this in the commit log, e.g.,Otherwise "make W=1
arch/x86/pci/" produces the following warning:CHECK
arch/x86/pci/intel_mid_pci.c
CCarch/x86/pci/intel_mid_pci.o
arch/x86/pci/intel_mid_pci.c:152: warning: Function parameter or member 'reg' not described in 'type1_access_ok'
Will do.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 15 lines · Dave Hansen
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 245
Quoted 9 lines · Andy Shevchenko
arch/x86/pci/intel_mid_pci.c +6
-2-switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID : polarity =IOAPIC_POL_HIGH ;The diff makes it _look_ like there's a behavior change, swapping silvermont and tangier. It appears from
intel_mid_arch_setup()thatINTEL_FAM6_ATOM_SILVERMONT_MIDand tangier are related:
Quoted 8 lines
#define
INTEL_FAM6_ATOM_SILVERMONT_MID0x4A/* Merriefield */...
case 0x3C: case 0x4A: __intel_mid_cpu_chip = INTEL_MID_CPU_CHIP_TANGIER; x86_platform.legacy.rtc = 1; break;
Yes. The original code is too old and was submitted without relying on existing kernel helpers.
Quoted 3 lines
But that's probably worth a note in the changelog. If you added a patch
tointel_mid_arch_setup()to this series to use theintel-family.h
#defines, this would be much more self explanatory.
No, I'm not going to do that at all. The idea behind is to get rid of this ugly customisation.
Quoted 2 lines
This also all makes me wonder if
intel_mid_identify_cpu()is really even necessary.
Exactly!
Quoted 7 lines
Does this fix any kinds of problems? It's a diffstat-challenged cleanup if not:
arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)That's not usually how cleanups look. :)
I understand. This will help to do a real clean up in the future.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 47 lines · Bjorn Helgaas
Quoted 46 lines · Andy Shevchenko
Switch the platform code to use
x86_id_tableand accompanying API instead of custom comparison against x86 CPU model.Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)arch/x86/pci/intel_mid_pci.c +9
-0@@ -28,10 +28,12 @@#include <linux/io.h> #include <linux/smp.h> +#include <asm/cpu_device_id.h> #include <asm/segment.h> #include <asm/pci_x86.h> #include <asm/hw_irq.h> #include <asm/io_apic.h> +#include <asm/intel-family.h> #include <asm/intel-mid.h> #define PCIE_CAP_OFFSET 0x100@@ -211,9 +213,16 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where, where, size, value); } +static const struct x86_cpu_id intel_mid_cpu_ids[] = { +X86_MATCH_INTEL_FAM6_MODEL (ATOM_SILVERMONT_MID ,NULL ), + {} +}; +static int intel_mid_pci_irq_enable (struct pci_dev *dev) { +const struct x86_cpu_id *id;struct irq_alloc_info info; +u16 model =0 ;int polarity;int ret;u8 gsi;@@ -227,8 +236,12 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)return ret; }
Quoted 11 lines
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 244
Quoted 8 lines
arch/x86/pci/intel_mid_pci.c +6
-2-switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID :Is there a magic decoder ring somewhere that connects
INTEL_MID_CPU_CHIP_TANGIERandINTEL_FAM6_ATOM_SILVERMONT_MID?
Yes. And the idea is to get rid of it.
Quoted 17 lines
I don't know how to verify that the new code is equivalent to the old.
Or maybe the new code is *better* than the old, in which case the subject/commit log should mention that it's fixing or improving something.
Also, there are a number of other places that check for "
intel_mid_identify_cpu()==INTEL_MID_CPU_CHIP_TANGIER":
mrfld_pinctrl_init
register_mrfld_power_btn
mrfld_legacy_rtc_init
mrfld_sd_init
spidev_platform_data
register_mid_wdt
sfi_parse_devs
Quoted 1 line
atomisp_css_input_set_mode
This has been pending in Mauro's tree.
Quoted 3 lines
Maybe they should all be changed together? Or maybe this needs an explanation about why some places need
intel_mid_identify_cpu()and others needx86_match_cpu()?
No. The rest is subject to huge clean up (complete removal) in the future. I don't want to waste time on something which I will remove for sure.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 60 lines · Andy Shevchenko
Quoted 47 lines · Bjorn Helgaas
Quoted 46 lines · Andy Shevchenko
Switch the platform code to use
x86_id_tableand accompanying API instead of custom comparison against x86 CPU model.Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/pci/intel_mid_pci.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)arch/x86/pci/intel_mid_pci.c +9
-0@@ -28,10 +28,12 @@#include <linux/io.h> #include <linux/smp.h> +#include <asm/cpu_device_id.h> #include <asm/segment.h> #include <asm/pci_x86.h> #include <asm/hw_irq.h> #include <asm/io_apic.h> +#include <asm/intel-family.h> #include <asm/intel-mid.h> #define PCIE_CAP_OFFSET 0x100@@ -211,9 +213,16 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where, where, size, value); } +static const struct x86_cpu_id intel_mid_cpu_ids[] = { +X86_MATCH_INTEL_FAM6_MODEL (ATOM_SILVERMONT_MID ,NULL ), + {} +}; +static int intel_mid_pci_irq_enable (struct pci_dev *dev) { +const struct x86_cpu_id *id;struct irq_alloc_info info; +u16 model =0 ;int polarity;int ret;u8 gsi;@@ -227,8 +236,12 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)return ret; }Quoted 11 lines
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 244
Quoted 8 lines
arch/x86/pci/intel_mid_pci.c +6
-2-switch (intel_mid_identify_cpu ()) { -case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID :Is there a magic decoder ring somewhere that connects
INTEL_MID_CPU_CHIP_TANGIERandINTEL_FAM6_ATOM_SILVERMONT_MID?
Quoted 1 line
Yes. And the idea is to get rid of it.
OK. You don't want to even include a mention of it in the commit log to help people connect the dots and verify that this change is correct?
Quoted 19 lines
Quoted 17 lines
I don't know how to verify that the new code is equivalent to the old.
Or maybe the new code is *better* than the old, in which case the subject/commit log should mention that it's fixing or improving something.
Also, there are a number of other places that check for "
intel_mid_identify_cpu()==INTEL_MID_CPU_CHIP_TANGIER":
mrfld_pinctrl_init
register_mrfld_power_btn
mrfld_legacy_rtc_init
mrfld_sd_init
spidev_platform_data
register_mid_wdt
sfi_parse_devsQuoted 1 line
atomisp_css_input_set_mode
Quoted 8 lines
This has been pending in Mauro's tree.
Quoted 3 lines
Maybe they should all be changed together? Or maybe this needs an explanation about why some places need
intel_mid_identify_cpu()and others needx86_match_cpu()?No. The rest is subject to huge clean up (complete removal) in the future. I don't want to waste time on something which I will remove for sure.
OK, I'll leave this to the x86 guys. If I were merging it I would want a little more explanation, but this isn't PCI-related at all and I'm sure they understand this better than I do.
Bjorn
Quoted 5 lines · Bjorn Helgaas
Quoted 4 lines · Andy Shevchenko
Quoted 3 lines · Bjorn Helgaas
Quoted 2 lines · Andy Shevchenko
Switch the platform code to use
x86_id_tableand accompanying API instead of custom comparison against x86 CPU model.
...
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
arch/x86/pci/intel_mid_pci.c static int intel_mid_pci_irq_enable(struct pci_dev *dev) · line 244
Quoted 7 lines
arch/x86/pci/intel_mid_pci.c +6
-1-case INTEL_MID_CPU_CHIP_TANGIER : + id =x86_match_cpu (intel_mid_cpu_ids); +if (id) + model = id->model; + +switch (model) { +case INTEL_FAM6_ATOM_SILVERMONT_MID :Is there a magic decoder ring somewhere that connects
INTEL_MID_CPU_CHIP_TANGIERandINTEL_FAM6_ATOM_SILVERMONT_MID?
Quoted 5 lines
Quoted 1 line
Yes. And the idea is to get rid of it.
OK. You don't want to even include a mention of it in the commit log to help people connect the dots and verify that this change is correct?
I think I missed it. I will update changelog for v2.
Signature · 2 lines
With Best Regards, Andy Shevchenko