x86: acpi: Refactor XSDT handling in acpi_add_table()
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
There is no need to have an assignment to NULL for XSDT pointer. Therefore, no need to assign it when rsdt_address is not set. Because of above changes we may decrease indentation level as well.
While here, drop unnecessary parentheses.
--- arch/x86/lib/acpi_table.c | 37 +++++++++++++++++++------------------ 1 file changed, 19 insertions(+), 18 deletions(-)
arch/x86/lib/acpi_table.c +19 -18
@@ -109,14 +109,11 @@ static void acpi_add_table(struct acpi_rsdp *rsdp, void *table) {int i, entries_num;struct acpi_rsdt *rsdt; -struct acpi_xsdt *xsdt =NULL ; +struct acpi_xsdt *xsdt;/* The RSDT is mandatory while the XSDT is not */ rsdt = (struct acpi_rsdt *)rsdp->rsdt_address; -if (rsdp->xsdt_address) - xsdt = (struct acpi_xsdt *)((u32 )rsdp->xsdt_address); -/* This should always be MAX_ACPI_TABLES */ entries_num =ARRAY_SIZE (rsdt->entry);
@@ -135,30 +132,34 @@ static void acpi_add_table(struct acpi_rsdp *rsdp, void *table)/* Fix RSDT length or the kernel will assume invalid entries */ rsdt->header.length =sizeof (struct acpi_table_header ) + - (sizeof (u32) * (i +1 )); +sizeof (u32) * (i +1 );/* Re-calculate checksum */ rsdt->header.checksum =0 ; rsdt->header.checksum =table_compute_checksum ((u8 *)rsdt, rsdt->header.length); +/* The RSDT is mandatory while the XSDT is not */ +if (!rsdp->xsdt_address) +return ; +/* * And now the same thing for the XSDT. We use the same index as for * now we want the XSDT and RSDT to always be in sync in U-Boot */ -if (xsdt) { -/* Add table to the XSDT */ - xsdt->entry[i] = (u64 )(u32 )table; - -/* Fix XSDT length */ - xsdt->header.length =sizeof (struct acpi_table_header ) + - (sizeof (u64) * (i +1 )); - -/* Re-calculate checksum */ - xsdt->header.checksum =0 ; - xsdt->header.checksum =table_compute_checksum ((u8 *)xsdt, - xsdt->header.length); - } + xsdt = (struct acpi_xsdt *)((u32 )rsdp->xsdt_address); + +/* Add table to the XSDT */ + xsdt->entry[i] = (u64 )(u32 )table; + +/* Fix XSDT length */ + xsdt->header.length =sizeof (struct acpi_table_header ) + +sizeof (u64) * (i +1 ); + +/* Re-calculate checksum */ + xsdt->header.checksum =0 ; + xsdt->header.checksum =table_compute_checksum ((u8 *)xsdt, + xsdt->header.length); }static void acpi_create_facs (struct acpi_facs *facs)
Signature · 1 line
2.25.0
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
Quoted 11 lines · Andy Shevchenko
There is no need to have an assignment to NULL for XSDT pointer. Therefore, no need to assign it when
rsdt_addressis not set. Because of above changes we may decrease indentation level as well.While here, drop unnecessary parentheses.
Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/lib/acpi_table.c | 37 +++++++++++++++++++------------------ 1 file changed, 19 insertions(+), 18 deletions(-)
Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
Regards,
Simon
Quoted 73 lines
arch/x86/lib/acpi_table.c +19
-19@@ -109,14 +109,11 @@ static void acpi_add_table(struct acpi_rsdp *rsdp, void *table) {int i, entries_num;struct acpi_rsdt *rsdt; -struct acpi_xsdt *xsdt =NULL ; +struct acpi_xsdt *xsdt;/* The RSDT is mandatory while the XSDT is not */ rsdt = (struct acpi_rsdt *)rsdp->rsdt_address; -if (rsdp->xsdt_address) - xsdt = (struct acpi_xsdt *)((u32 )rsdp->xsdt_address); -/* This should always be MAX_ACPI_TABLES */ entries_num =ARRAY_SIZE (rsdt->entry);@@ -135,30 +132,34 @@ static void acpi_add_table(struct acpi_rsdp *rsdp, void *table)/* Fix RSDT length or the kernel will assume invalid entries */ rsdt->header.length =sizeof (struct acpi_table_header ) + - (sizeof (u32) * (i +1 )); +sizeof (u32) * (i +1 );/* Re-calculate checksum */ rsdt->header.checksum =0 ; rsdt->header.checksum =table_compute_checksum ((u8 *)rsdt, rsdt->header.length); +/* The RSDT is mandatory while the XSDT is not */ +if (!rsdp->xsdt_address) +return ; +/* * And now the same thing for the XSDT. We use the same index as for * now we want the XSDT and RSDT to always be in sync in U-Boot */ -if (xsdt) { -/* Add table to the XSDT */ - xsdt->entry[i] = (u64 )(u32 )table; - -/* Fix XSDT length */ - xsdt->header.length =sizeof (struct acpi_table_header ) + - (sizeof (u64) * (i +1 )); - -/* Re-calculate checksum */ - xsdt->header.checksum =0 ; - xsdt->header.checksum =table_compute_checksum ((u8 *)xsdt, - xsdt->header.length); - } + xsdt = (struct acpi_xsdt *)((u32 )rsdp->xsdt_address); + +/* Add table to the XSDT */ + xsdt->entry[i] = (u64 )(u32 )table; + +/* Fix XSDT length */ + xsdt->header.length =sizeof (struct acpi_table_header ) + +sizeof (u64) * (i +1 ); + +/* Re-calculate checksum */ + xsdt->header.checksum =0 ; + xsdt->header.checksum =table_compute_checksum ((u8 *)xsdt, + xsdt->header.length); }static void acpi_create_facs (struct acpi_facs *facs) --2.25.0
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 8 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
Quoted 5 lines · Andy Shevchenko
Quoted 2 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:
Quoted 8 lines
Quoted 5 lines
Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Quoted 5 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.
How do we add the required linux,name ACPI property into the ACPI tables for a device?
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci {
p2sb at d,0 {
n {
gpio-n {So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
Quoted 4 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!) probably have a better idea on what is needed. The core DM support and tests are there.
Regards,
Simon
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 9 lines · Simon Glass
Quoted 8 lines · Andy Shevchenko
Quoted 7 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Quoted 11 lines
Quoted 4 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
Quoted 7 lines
pci { p2sb at d,0 { n { gpio-n {So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,
Quoted 6 lines
Quoted 3 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!)
Haven't you seen the series to introduce ACPI for ARM in U-Boot recently?
Quoted 2 lines
probably have a better idea on what is needed. The core DM support and tests are there.
I think with a such big rework it's not big deal to simple move it outside of arch/x86 to the lib/acpi or so.
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
Quoted 13 lines · Andy Shevchenko
Quoted 11 lines · Simon Glass
Quoted 10 lines · Andy Shevchenko
Quoted 8 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
dm: core: Add basic ACPI support
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
static const struct acpi_device_id bxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};
So we have to put INT3452 in the ACPI table.
Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {
Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,
Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
Quoted 9 lines
Quoted 7 lines
Quoted 4 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!)
Haven't you seen the series to introduce ACPI for ARM in U-Boot recently?
Yes, and that author is awaiting us getting this series in so that he can build on it.
Quoted 6 lines
Quoted 3 lines
probably have a better idea on what is needed. The core DM support and tests are there.
I think with a such big rework it's not big deal to simple move it outside of
arch/x86to the lib/acpi or so.
My intention was to put generic ACPI things in there though, not x86-specific stuff. I assume that ARM would have its own stuff in arch.arm
Bin, what do you think?
Regards,
Simon
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 107 lines · Simon Glass
Hi Andy,
Quoted 13 lines · Andy Shevchenko
Quoted 11 lines · Simon Glass
Quoted 10 lines · Andy Shevchenko
Quoted 8 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
dm: core: Add basic ACPI support
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
Quoted 9 lines
Quoted 7 lines
Quoted 4 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!)
Haven't you seen the series to introduce ACPI for ARM in U-Boot recently?
Yes, and that author is awaiting us getting this series in so that he can build on it.
Quoted 4 lines
Quoted 3 lines
probably have a better idea on what is needed. The core DM support and tests are there.
Quoted 7 lines
Quoted 2 lines
I think with a such big rework it's not big deal to simple move it outside of
arch/x86to the lib/acpi or so.My intention was to put generic ACPI things in there though, not x86-specific stuff. I assume that ARM would have its own stuff in arch.arm
That's what my assumption too.
Quoted 1 line
Bin, what do you think?
I will take some time to review the remaining patches though.
Regards,
Bin
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
In reply to Simon GlassQuoted 19 lines · Simon Glass
Quoted 18 lines · Andy Shevchenko
Quoted 16 lines · Simon Glass
Quoted 4 lines · Andy Shevchenko
Quoted 2 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Quoted 8 lines
Quoted 5 lines
Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Quoted 7 lines
Quoted 2 lines
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
dm: core: Add basic ACPI support
Okay, I will try to find a time to look at it first.
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
Quoted 7 lines
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Wait, this is not a *name*, this is ACPI _HID. ACPI _HID, of course, should be somewhere in board code.
I was thinking myself about some U-Boot framework that actually takes ACPI _HID from the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI _HID in each DTS.
But to the current topic, you put *instance* (not even _HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.
Quoted 39 lines
Quoted 16 lines
Quoted 16 lines
Quoted 4 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!)
Haven't you seen the series to introduce ACPI for ARM in U-Boot recently?
Yes, and that author is awaiting us getting this series in so that he can build on it.
Good that we aware.
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
probably have a better idea on what is needed. The core DM support and tests are there.
I think with a such big rework it's not big deal to simple move it outside of
arch/x86to the lib/acpi or so.My intention was to put generic ACPI things in there though, not x86-specific stuff. I assume that ARM would have its own stuff in arch.arm
Yes, that what I'm talking about. So we are on the same page here.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Antwort: Re: [PATCH v1] x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
-----"U-Boot" <[email protected]> schrieb: -----
Quoted 29 lines
[snip]
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
Quoted 7 lines
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Wait, this is not a *name*, this is ACPI
_HID. ACPI_HID, of course, should be somewhere in board code.
Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.
There is a related discussion in another thread, here is the link:
https://lists.denx.de/pipermail/u-boot/2020-February/398856.html
I have brought that up, but I'm no expert in this area, so any feedback would be welcome.
That discussion is not only about inferring _HID, but also about the idea of inferring other ACPI properties from device tree (the example discussed is the HID offset).
regards, Wolfgang
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
On Tue, Mar 3, 2020 at 12:00 PM Wolfgang Wallner
<[email protected]> wrote:
...
Quoted 10 lines
Quoted 9 lines
Wait, this is not a *name*, this is ACPI
_HID. ACPI_HID, of course, should be somewhere in board code.I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.
Quoted 2 lines
There is a related discussion in another thread, here is the link:
https://lists.denx.de/pipermail/u-boot/2020-February/398856.html
Thank you for pointing this out.
I totally agree with you. I really do not like the idea of polluting DT with ACPI bits.
Quoted 6 lines
I have brought that up, but I'm no expert in this area, so any feedback would be welcome.
That discussion is not only about inferring
_HID, but also about the idea of inferring other ACPI properties from device tree (the example discussed is the HID offset).
Yes, that's what I'm implying above as well. I'm on the same page with you!
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
In reply to Andy ShevchenkoHi Andy,
Quoted 57 lines · Andy Shevchenko
Quoted 19 lines · Simon Glass
Quoted 18 lines · Andy Shevchenko
Quoted 16 lines · Simon Glass
Quoted 4 lines · Andy Shevchenko
Quoted 2 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Quoted 8 lines
Quoted 5 lines
Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Quoted 7 lines
Quoted 2 lines
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
dm: core: Add basic ACPI support
Okay, I will try to find a time to look at it first.
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
Quoted 7 lines
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Wait, this is not a *name*, this is ACPI
_HID. ACPI_HID, of course, should be somewhere in board code.
Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.
Why are you so opposed to using device tree for this? The GPIO and pinctrl drivers are intended to be generic....what a pain to add all this stuff into the tables in the driver!
When other platforms use APL we can move some .dts nodes over into a intel-apl.dtsi file (or similar) to deal with any duplication. Of course we don't want duplication.
Re the thread that Wolfgang references, I'm going to have a close look at that and hopefully simplify things. We still need quite a bit more patches to be reviewed before it is worth sending again, I think.
Quoted 4 lines
But to the current topic, you put *instance* (not even
_HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.
OK, so are you saying the property name (linux-name) should change? We have acpi,name elsewhere but I don't think that is the _HID.
Or are you saying that the "INT3452:" should be factored out and it should know the 00/01/0203 by its position in the device tree?
Quoted 43 lines
Quoted 43 lines
Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
Also, I'm not sure how your rework helps ARM (or any other architecture) people with their approach to ACPI enabling (most of the files are under x86).
I kept x86-specific tables in the x86 directories. Of course I might be wrong about this. But then, people who use ACPI on ARM (ick!)
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Haven't you seen the series to introduce ACPI for ARM in U-Boot recently?
Yes, and that author is awaiting us getting this series in so that he can build on it.
Good that we aware.
Yes.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
probably have a better idea on what is needed. The core DM support and tests are there.
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
I think with a such big rework it's not big deal to simple move it outside of
arch/x86to the lib/acpi or so.My intention was to put generic ACPI things in there though, not x86-specific stuff. I assume that ARM would have its own stuff in arch.arm
Yes, that what I'm talking about. So we are on the same page here.
OK good.
Regards,
Simpon
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 24 lines · Simon Glass
Quoted 23 lines · Andy Shevchenko
Quoted 22 lines · Simon Glass
Quoted 20 lines · Andy Shevchenko
Quoted 9 lines · Simon Glass
Quoted 8 lines · Andy Shevchenko
Quoted 7 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
u-boot-dm/coral-working
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Quoted 6 lines
Quoted 6 lines
Quoted 4 lines
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
dm: core: Add basic ACPI support
Okay, I will try to find a time to look at it first.
I started looking at them from the above mentioned patch.
1/ Can we do include/acpi/ folder for ACPI related headers?
2/ How this is supposed to be compiled?
+ table_compute_checksum((xsdt, xsdt->header.length);
...means this series should go thru bisectability tests (something like aiaiai https://lwn.net/Articles/488992/ script provides)
3/ This one looks not 64-bit compatible.
+ printf("RSDP %08lx %06x (v%02d %.6s)\n", (ulong)map_to_sysmem(rsdp),
+ rsdp->length, rsdp->revision, rsdp->oem_id);
...means that types for printing and all explicit casting should be revisited.
Till this one "acpi: Add some tables required by the generation code" looks okay (in terms of approach).
This one "acpi: Add generation code for devices" requires quite a good review. So, I would recommend to split the series (and this patch in particular) to smaller chunks. So does this "acpi: Add functions to generate ACPI code". They are unreviewable.
Perhaps first pile some generalization that ARM people may start their work...
Quoted 33 lines
Quoted 26 lines
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
Quoted 7 lines
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Wait, this is not a *name*, this is ACPI
_HID. ACPI_HID, of course, should be somewhere in board code.Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.
Quoted 3 lines
Why are you so opposed to using device tree for this? The GPIO and pinctrl drivers are intended to be generic....what a pain to add all this stuff into the tables in the driver!
So, this is a trade off between C code and DTS. I'm okay to use DTS for the stuff that belongs to it. But then, if we enable DTS for ACPI tables generation, we have to provide a mean to do it without driver involvement.
How to generate the table for the device U-Boot has no driver for?
Quoted 7 lines
When other platforms use APL we can move some .dts nodes over into a
intel-apl.dtsifile (or similar) to deal with any duplication. Of course we don't want duplication.Re the thread that Wolfgang references, I'm going to have a close look at that and hopefully simplify things. We still need quite a bit more patches to be reviewed before it is worth sending again, I think.
Yes, please. My main point here is to avoid data duplication because it's simpler to pollute DTS with it.
Quoted 9 lines
Quoted 3 lines
But to the current topic, you put *instance* (not even
_HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.OK, so are you saying the property name (linux-name) should change? We have acpi,name elsewhere but I don't think that is the
_HID.Or are you saying that the "INT3452:" should be factored out and it should know the 00/01/0203 by its position in the device tree?
It shouldn't be anywhere in the U-Boot, it's complete OS business. What you have in U-Boot is ACPI _HID (_UID, etc.), and device path (e.g. \\_SB_.GPO0), no-one should rely on OS (Linux, Windows, etc) internals. We have already an issue with GPIO pin numbers on Chromebook with Intel Cherryview SoC.
This
+ linux-name = "INT3452:00";
is wrong in both sides -- left, as a property name, and right, as an *instance* in some OS we must not rely on ever.
The question is why do you need it?
Quoted 28 lines
Quoted 28 lines
Quoted 28 lines
Quoted 16 lines
Quoted 16 lines
Quoted 4 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
I'm not sure I understood how the mentioned source file related to P2SB case. In that file PCIe functions and USB ports are described.
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
Quoted 29 lines · Andy Shevchenko
Quoted 27 lines · Simon Glass
Quoted 26 lines · Andy Shevchenko
Quoted 25 lines · Simon Glass
Quoted 12 lines · Andy Shevchenko
Quoted 11 lines · Simon Glass
Quoted 10 lines · Andy Shevchenko
Quoted 8 lines · Simon Glass
On Thu, 27 Feb 2020 at 06:00, Andy Shevchenko
<[email protected]> wrote:Could you take a look at the ACPI series?
It was sent out about a month ago and has a refactor to this function.
u-boot-dm/coral-working
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
There are tons of changes. Care to point what changes are more important (generic to all x86)?
I'm not quite sure about that...but x86 patches have an x86: tag, so perhaps that helps?
Okay, some like 50 of them or even more? I really don't want to spend time on the board related patches like "x86: apl:".
Well that's why I add the tags, so you can see what they relate to. This is probably a good one to review:
Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
dm: core: Add basic ACPI support
Okay, I will try to find a time to look at it first.
I started looking at them from the above mentioned patch.
1/ Can we do
include/acpi/folder for ACPI related headers?
OK
Quoted 4 lines
2/ How this is supposed to be compiled?
+ table_compute_checksum((xsdt,xsdt->header.length);
...means this series should go thru bisectability tests (something like aiaiai https://lwn.net/Articles/488992/ script provides)
I'm not sure what you mean by that?
Quoted 4 lines
3/ This one looks not 64-bit compatible.
+ printf("RSDP %08lx %06x (v%02d %.6s)\n", (ulong)map_to_sysmem(rsdp),
+rsdp->length,rsdp->revision,rsdp->oem_id);
...means that types for printing and all explicit casting should be revisited.
I don't want to print 64-bit addresses if I can help it. I don't even want to use them as they are a pain to look at! If we start using 64-bit U-Boot I still think we will put the ACPI tables below 2GB.
Quoted 9 lines
Till this one "acpi: Add some tables required by the generation code" looks okay (in terms of approach).
This one "acpi: Add generation code for devices" requires quite a good review. So, I would recommend to split the series (and this patch in particular) to smaller chunks. So does this "acpi: Add functions to generate ACPI code". They are unreviewable.
OK I can split that first part off as a series and then split the next patches.
Quoted 40 lines
Perhaps first pile some generalization that ARM people may start their work...
Quoted 33 lines
Quoted 26 lines
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
P.S. Briefly looking at the last ~30 patches I can say that the idea looks good, implementation needs more work. For example, there is 'linux,name' property. Shouldn't be referred at all. Linux names and other type of enumerations is utterly opaque to the outside world.How do we add the required linux,name ACPI property into the ACPI tables for a device?
There must not be Linux device names or anything Linux related (like hardcoded GPIO numbers) in the ACPI table.
Apparently the Intel GPIO driver requires that name. See for example here:
https://elixir.bootlin.com/linux/latest/source/drivers/pinctrl/intel/pinctrl-broxton.c#L999
Quoted 7 lines
static const struct
acpi_device_idbxt_pinctrl_acpi_match[] = { { "INT3452", (kernel_ulong_t)apl_pinctrl_soc_data }, { "INT34D1", (kernel_ulong_t)bxt_pinctrl_soc_data }, { }
};So we have to put INT3452 in the ACPI table.
Wait, this is not a *name*, this is ACPI
_HID. ACPI_HID, of course, should be somewhere in board code.Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.Quoted 3 lines
Why are you so opposed to using device tree for this? The GPIO and pinctrl drivers are intended to be generic....what a pain to add all this stuff into the tables in the driver!
Quoted 5 lines
So, this is a trade off between C code and DTS. I'm okay to use DTS for the stuff that belongs to it. But then, if we enable DTS for ACPI tables generation, we have to provide a mean to do it without driver involvement.
How to generate the table for the device U-Boot has no driver for?
We add a driver. The driver might only generate ACPI tables, but it is still a driver.
Quoted 11 lines
Quoted 8 lines
When other platforms use APL we can move some .dts nodes over into a
intel-apl.dtsifile (or similar) to deal with any duplication. Of course we don't want duplication.Re the thread that Wolfgang references, I'm going to have a close look at that and hopefully simplify things. We still need quite a bit more patches to be reviewed before it is worth sending again, I think.
Yes, please. My main point here is to avoid data duplication because it's simpler to pollute DTS with it.
OK I have done that in v3.
Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
But to the current topic, you put *instance* (not even
_HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.OK, so are you saying the property name (linux-name) should change? We have acpi,name elsewhere but I don't think that is the
_HID.
Quoted 8 lines
Quoted 2 lines
Or are you saying that the "INT3452:" should be factored out and it should know the 00/01/0203 by its position in the device tree?
It shouldn't be anywhere in the U-Boot, it's complete OS business. What you have in U-Boot is ACPI
_HID(_UID, etc.), and device path (e.g. \\_SB_.GPO0), no-one should rely on OS (Linux, Windows, etc) internals. We have already an issue with GPIO pin numbers on Chromebook with Intel Cherryview SoC.
Right but how can ACPI code refer to the GPIO if it cannot reference it?
Quoted 9 lines
This
+ linux-name = "INT3452:00";
is wrong in both sides -- left, as a property name, and right, as an *instance* in some OS we must not rely on ever.
The question is why do you need it?
To generate ACPI code which references that GPIO.
See chromeos_acpi_gpio_generate().
Can you suggest a better property name? Maybe acpi,linux-name? But it isn't really an ACPI name.
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
I'm not sure I understood how the mentioned source file related to P2SB case. In that file PCIe functions and USB ports are described.
It covers all devices that need an ACPI name. I would like this name to be kept with the rest of the device data, unless if is the same for all devices of a certain type, in which case I suppose we could use a function (e.g. the root node).
Regards,
Simon
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 4 lines · Simon Glass
Quoted 3 lines · Andy Shevchenko
Quoted 2 lines · Simon Glass
Quoted 1 line · Andy Shevchenko
On Tue, Mar 3, 2020 at 1:36 AM Simon Glass <[email protected]> wrote:
...
Quoted 6 lines
Quoted 4 lines
2/ How this is supposed to be compiled?
+ table_compute_checksum((xsdt,xsdt->header.length);
...means this series should go thru bisectability tests (something like aiaiai https://lwn.net/Articles/488992/ script provides)I'm not sure what you mean by that?
Isn't above may not be compiled without error?
aiaiai is a script which allows to check bisectability (compile-time) errors like above.
Quoted 8 lines
Quoted 4 lines
3/ This one looks not 64-bit compatible.
+ printf("RSDP %08lx %06x (v%02d %.6s)\n", (ulong)map_to_sysmem(rsdp),
+rsdp->length,rsdp->revision,rsdp->oem_id);
...means that types for printing and all explicit casting should be revisited.I don't want to print 64-bit addresses if I can help it. I don't even want to use them as they are a pain to look at! If we start using 64-bit U-Boot I still think we will put the ACPI tables below 2GB.
So, it should be documented then.
...
Quoted 9 lines
Quoted 7 lines
Till this one "acpi: Add some tables required by the generation code" looks okay (in terms of approach).
This one "acpi: Add generation code for devices" requires quite a good review. So, I would recommend to split the series (and this patch in particular) to smaller chunks. So does this "acpi: Add functions to generate ACPI code". They are unreviewable.
OK I can split that first part off as a series and then split the next patches.
Thank you.
...
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.Why are you so opposed to using device tree for this? The GPIO and pinctrl drivers are intended to be generic....what a pain to add all this stuff into the tables in the driver!
Quoted 8 lines
Quoted 5 lines
So, this is a trade off between C code and DTS. I'm okay to use DTS for the stuff that belongs to it. But then, if we enable DTS for ACPI tables generation, we have to provide a mean to do it without driver involvement.
How to generate the table for the device U-Boot has no driver for?
We add a driver. The driver might only generate ACPI tables, but it is still a driver.
Hmm... This is interesting concept. So, each time we would like to get only tables we will need to create a stub driver.
...
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
Quoted 3 lines
But to the current topic, you put *instance* (not even
_HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.OK, so are you saying the property name (linux-name) should change? We have acpi,name elsewhere but I don't think that is the
_HID.Or are you saying that the "INT3452:" should be factored out and it should know the 00/01/0203 by its position in the device tree?
Quoted 7 lines
Quoted 5 lines
It shouldn't be anywhere in the U-Boot, it's complete OS business. What you have in U-Boot is ACPI
_HID(_UID, etc.), and device path (e.g. \\_SB_.GPO0), no-one should rely on OS (Linux, Windows, etc) internals. We have already an issue with GPIO pin numbers on Chromebook with Intel Cherryview SoC.Right but how can ACPI code refer to the GPIO if it cannot reference it?
What do you mean? Any example where you need *instance* over *_HID*?
Quoted 4 lines
Quoted 4 lines
This
+ linux-name = "INT3452:00";
Quoted 8 lines
Quoted 4 lines
is wrong in both sides -- left, as a property name, and right, as an *instance* in some OS we must not rely on ever.
The question is why do you need it?
To generate ACPI code which references that GPIO.
See
chromeos_acpi_gpio_generate().
I can't see right now. Do you have web browser thru source code? GitLab instance doesn't show recent x86 repository.
Quoted 2 lines
Can you suggest a better property name? Maybe acpi,linux-name? But it isn't really an ACPI name.
ACPI _HID. No instance, please. If you are using instance outside of Linux kernel code it points to wrong design and I have strong opinion not to support this kind of design.
Quoted 34 lines
Quoted 34 lines
Quoted 33 lines
Quoted 33 lines
Quoted 33 lines
Quoted 16 lines
Quoted 16 lines
Quoted 4 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
Quoted 7 lines
Quoted 2 lines
I'm not sure I understood how the mentioned source file related to P2SB case. In that file PCIe functions and USB ports are described.
It covers all devices that need an ACPI name. I would like this name to be kept with the rest of the device data, unless if is the same for all devices of a certain type, in which case I suppose we could use a function (e.g. the root node).
I didn't get what you mean under ACPI name here. ACPI _HID? Yes, somewhere we need to keep ACPI _HID for the devices. I agree (DTS or board code is not so important -- choose best one in your opinion).
Signature · 2 lines
With Best Regards, Andy Shevchenko
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Hi Andy,
Quoted 14 lines · Andy Shevchenko
Quoted 4 lines · Simon Glass
Quoted 3 lines · Andy Shevchenko
Quoted 2 lines · Simon Glass
Quoted 1 line · Andy Shevchenko
On Tue, Mar 3, 2020 at 1:36 AM Simon Glass <[email protected]> wrote:
...
Quoted 5 lines
Quoted 4 lines
2/ How this is supposed to be compiled?
+ table_compute_checksum((xsdt,xsdt->header.length);
...means this series should go thru bisectability tests (something like aiaiai https://lwn.net/Articles/488992/ script provides)
Quoted 6 lines
Quoted 1 line
I'm not sure what you mean by that?
Isn't above may not be compiled without error?
aiaiai is a script which allows to check bisectability (compile-time) errors like above.
OK I see. In U-Boot we use buildman for this. It builds entire branches and I do that before sending patches.
Quoted 10 lines
Quoted 8 lines
Quoted 4 lines
3/ This one looks not 64-bit compatible.
+ printf("RSDP %08lx %06x (v%02d %.6s)\n", (ulong)map_to_sysmem(rsdp),
+rsdp->length,rsdp->revision,rsdp->oem_id);
...means that types for printing and all explicit casting should be revisited.I don't want to print 64-bit addresses if I can help it. I don't even want to use them as they are a pain to look at! If we start using 64-bit U-Boot I still think we will put the ACPI tables below 2GB.
So, it should be documented then.
OK.
Quoted 32 lines
...
Quoted 9 lines
Quoted 7 lines
Till this one "acpi: Add some tables required by the generation code" looks okay (in terms of approach).
This one "acpi: Add generation code for devices" requires quite a good review. So, I would recommend to split the series (and this patch in particular) to smaller chunks. So does this "acpi: Add functions to generate ACPI code". They are unreviewable.
OK I can split that first part off as a series and then split the next patches.
Thank you.
...
Quoted 15 lines
Quoted 15 lines
Quoted 10 lines
Quoted 6 lines
I was thinking myself about some U-Boot framework that actually takes ACPI
_HIDfrom the driver. So, when you define in U-Boot device tree a compatible string (for U-Boot use), in the driver it will have in the class structure the callback / field / stubstructure to use when ACPI generate tables is enabled. It will drop duplication of compatible with ACPI_HIDin each DTS.Why are you so opposed to using device tree for this? The GPIO and pinctrl drivers are intended to be generic....what a pain to add all this stuff into the tables in the driver!
So, this is a trade off between C code and DTS. I'm okay to use DTS for the stuff that belongs to it. But then, if we enable DTS for ACPI tables generation, we have to provide a mean to do it without driver involvement.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
How to generate the table for the device U-Boot has no driver for?
We add a driver. The driver might only generate ACPI tables, but it is still a driver.
Hmm... This is interesting concept. So, each time we would like to get only tables we will need to create a stub driver.
Yes, if it is not actually used in U-Boot.
Quoted 19 lines
...
Quoted 16 lines
Quoted 15 lines
Quoted 9 lines
Quoted 3 lines
But to the current topic, you put *instance* (not even
_HID) to the DT with property called "linux,name". It's inappropriate. NAK for that for sure.OK, so are you saying the property name (linux-name) should change? We have acpi,name elsewhere but I don't think that is the
_HID.Or are you saying that the "INT3452:" should be factored out and it should know the 00/01/0203 by its position in the device tree?
It shouldn't be anywhere in the U-Boot, it's complete OS business. What you have in U-Boot is ACPI
_HID(_UID, etc.), and device path (e.g. \\_SB_.GPO0), no-one should rely on OS (Linux, Windows, etc) internals. We have already an issue with GPIO pin numbers on Chromebook with Intel Cherryview SoC.
Quoted 3 lines
Quoted 1 line
Right but how can ACPI code refer to the GPIO if it cannot reference it?
What do you mean? Any example where you need *instance* over *_HID*?
I suppose I just don't understand this very well. Does ACPI code access things via a _HID?
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
This
+ linux-name = "INT3452:00";
is wrong in both sides -- left, as a property name, and right, as an *instance* in some OS we must not rely on ever.
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
The question is why do you need it?
To generate ACPI code which references that GPIO.
See
chromeos_acpi_gpio_generate().I can't see right now. Do you have web browser thru source code? GitLab instance doesn't show recent x86 repository.
You can see u-boot-dm/coral-working for the source code.
See here for what I am talking about:
Quoted 7 lines
Quoted 3 lines
Can you suggest a better property name? Maybe acpi,linux-name? But it isn't really an ACPI name.
ACPI
_HID. No instance, please. If you are using instance outside of Linux kernel code it points to wrong design and I have strong opinion not to support this kind of design.
OK let me know what you think of the above. If provides a way to access these GPIOs.
Quoted 38 lines
Quoted 38 lines
Quoted 37 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 17 lines
Quoted 17 lines
Quoted 5 lines
On top of that, I think we rather need to have a conversion layer than putting some names inside DT, like \_SB_.GPO0 should be generated automatically from DT node. That said, I don't like DT being polluted with non-DT stuff.
Well DT is the configuration mechanism for U-Boot.
\_SB_.GPO0 is a special case since it actually refers to pinctrl (ACPI seems to make no distinction between pinctrl and GPIO) and this node is inside p2sb:
pci { p2sb at d,0 { n { gpio-n {Quoted 7 lines
Quoted 2 lines
So the automatically generated path would have p2sb in it. The same work-around is in coreboot.
It's not a coreboot, we may do better, right?
So, generation can strip p2sb (special case) from all p2sb devices. However, I'm not sure I understand how p2sb is involved in GPIO enumeration,Well the only other way to create a path is to work up to the root and build it node by node. I wonder if we could make p2sb be transparent? I tried that but hit a problem.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Coreboot has these really awful (IMO) functions that are repeated for every SoC:
https://github.com/coreboot/coreboot/blob/master/src/soc/amd/stoneyridge/chip.c
so I want to avoid that.
I'm not sure I understood how the mentioned source file related to P2SB case. In that file PCIe functions and USB ports are described.
Quoted 8 lines
Quoted 4 lines
It covers all devices that need an ACPI name. I would like this name to be kept with the rest of the device data, unless if is the same for all devices of a certain type, in which case I suppose we could use a function (e.g. the root node).
I didn't get what you mean under ACPI name here. ACPI
_HID? Yes, somewhere we need to keep ACPI_HIDfor the devices. I agree (DTS or board code is not so important -- choose best one in your opinion).
OK, DTS.
Regards,
Simon
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
In reply to Andy ShevchenkoQuoted 5 lines · Andy Shevchenko
There is no need to have an assignment to NULL for XSDT pointer. Therefore, no need to assign it when
rsdt_addressis not set. Because of above changes we may decrease indentation level as well.
Quoted 7 lines
While here, drop unnecessary parentheses.
Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/lib/acpi_table.c | 37 +++++++++++++++++++------------------ 1 file changed, 19 insertions(+), 18 deletions(-)
This patch looks good to me and I believe we can put this in v2020.04.
Regards,
Bin
PATCH 1/1 x86: acpi: Refactor XSDT handling in acpi_add_table()
Quoted 16 lines · Bin Meng
Quoted 5 lines · Andy Shevchenko
There is no need to have an assignment to NULL for XSDT pointer. Therefore, no need to assign it when
rsdt_addressis not set. Because of above changes we may decrease indentation level as well.Quoted 7 lines
While here, drop unnecessary parentheses.
Signed-off-by Andy Shevchenko <[email protected]>--- arch/x86/lib/acpi_table.c | 37 +++++++++++++++++++------------------ 1 file changed, 19 insertions(+), 18 deletions(-)
Quoted 4 lines
This patch looks good to me and I believe we can put this in v2020.04.
Reviewed-by Bin Meng <[email protected]>
applied to u-boot-x86, thanks!