ACPI/IORT: Workaround for IORT ID count "minus one" issue
Subsystems (2)
Hanjun Guo
maintainerPATCH 1/1 ACPI/IORT: Workaround for IORT ID count "minus one" issue
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function iort_id_map() did something wrong from the start, which bails out if:
the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS reference
PCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS reference
Two mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID 0x1100 will map to ITS 0xC4 not 0xD4 if we update '>=' to '>'.
So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platforms
---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.
drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)
drivers/acpi/arm64/iort.c +52 -3
@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {
@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus +* one, but the IORT code ingored the "minus one", and some firmware +* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware. +*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ; + }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + } *rid_out = map->output_base + (rid_in - map->input_base);return 0 ;
@@ -1631,5 +1679,6 @@ void __init acpi_iort_init(void)return ; } +iort_check_id_count_workaround (iort_table);iort_init_platform_devices (); }
Signature · 1 line
1.7.12.4
+
Quoted 36 lines · Guohanjun (Hanjun Guo)
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.
Quoted 6 lines
So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2],
Hi Hanjun,
only
Quoted 2 lines
HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table,
Are you asserting that other platforms are ok on the basis that NumIds = large power of 2 - 1, e.g. 0xffff? Is this strictly proper?
if we break other platforms, we can add that later.
Quoted 1 line
I think that it would be better to audit others now as well as best as reasonably possible. There is somewhat limited coverage in [2].
Thanks,
John
Quoted 95 lines
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platformsCc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)drivers/acpi/arm64/iort.c +52
-3@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus +* one, but the IORT code ingored the "minus one", and some firmware +* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware. +*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ; + }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + } *rid_out = map->output_base + (rid_in - map->input_base);return 0 ;@@ -1631,5 +1679,6 @@ void __init acpi_iort_init(void)return ; } +iort_check_id_count_workaround (iort_table);iort_init_platform_devices (); }
Hanjun Guo
maintainerQuoted 46 lines · John Garry
+
Quoted 36 lines · Guohanjun (Hanjun Guo)
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base:0x1000
ID Count:0x100
Output base:0x1000
Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base:0x1100
ID Count:0x100
Output base:0x2000
Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.Quoted 6 lines
So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2],
Quoted 7 lines
Hi Hanjun,
only
Quoted 2 lines
HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table,
Are you asserting that other platforms are ok on the basis that NumIds = large power of 2 - 1, e.g.
0xffff? Is this strictly proper?
No, some platforms with no opensource ACPI tables, are not covered.
Quoted 5 lines
if we break other platforms, we can add that later.
Quoted 1 line
I think that it would be better to audit others now as well as best as reasonably possible. There is somewhat limited coverage in [2].
I will Cc people form Mavell, Ampere, and Ard who is know Socionext very well, that's the best I can do.
Thanks
Hanjun
Quoted 119 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platformsCc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)drivers/acpi/arm64/iort.c +41
-3@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus +* one, but the IORT code ingored the "minus one", and some firmware +* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware.
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 367
Quoted 10 lines
drivers/acpi/arm64/iort.c +10
-0+*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ; + }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + }
This seems needlessly repetitive and convoluted... how about refactoring to something like:
map_max = map->input_base + map->id_count; if (apply_id_count_workaround) map_max--;
?
Robin.
Quoted 10 lines
drivers/acpi/arm64/iort.c +1
-0*rid_out = map->output_base + (rid_in - map->input_base);return 0 ;@@ -1631,5 +1679,6 @@ void __init acpi_iort_init(void)return ; } +iort_check_id_count_workaround (iort_table);iort_init_platform_devices (); }
Hanjun Guo
maintainerQuoted 131 lines · Robin Murphy
Quoted 119 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base:0x1000
ID Count:0x100
Output base:0x1000
Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base:0x1100
ID Count:0x100
Output base:0x2000
Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platformsCc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.
drivers/acpi/arm64/iort.c| 55 ++++++++++++++++++++++++++++++++++++++++++++---1 file changed, 52 insertions(+), 3 deletions(-)drivers/acpi/arm64/iort.c +35
-0@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node, return status; } +struct iort_workaround_oem_info { + char oem_id[ACPI_OEM_ID_SIZE + 1]; + char oem_table_id[ACPI_OEM_TABLE_ID_SIZE + 1]; + u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id = "HISI ", + .oem_table_id = "HIP07 ", + .oem_revision = 0, + }, { + .oem_id = "HISI ", + .oem_table_id = "HIP08 ", + .oem_revision = 0, + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ + int i; + + for (i = 0; i < ARRAY_SIZE(wa_info); i++) { + if (!memcmp(wa_info[i].oem_id, tbl->oem_id, ACPI_OEM_ID_SIZE) && + !memcmp(wa_info[i].oem_table_id, tbl->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; + pr_warn(FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n"); + break; + } + } +} + static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, u32 *rid_out) {@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, return -ENXIO; }- if (
rid_in<map->input_base||
- (rid_in>=map->input_base+map->id_count))
- return-ENXIO;
+ /*
+ * IORT spec says Number of IDs = The number of IDs in the range minus + * one, but the IORT code ingored the "minus one", and some firmware + * did that too, so apply a workaround here to keep compatible with + * both new and old versions of the firmware.
+ */
+ if (apply_id_count_workaround) {Quoted 8 lines
+ if (
rid_in<map->input_base||
+ (rid_in>=map->input_base+map->id_count)) + return-ENXIO;
+ } else {
+ if (rid_in<map->input_base||
+ (rid_in>map->input_base+map->id_count)) + return-ENXIO;
+ }This seems needlessly repetitive and convoluted... how about refactoring to something like:
Quoted 3 lines
map_max =
map->input_base+map->id_count;
if (apply_id_count_workaround)
map_max--;
Much better, thanks! I will update my patch.
Thanks
Hanjun
Quoted 3 lines · Robin Murphy
Quoted 2 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the
Why is it confusing ? Because we botched the kernel code :) ?
Quoted 5 lines
Quoted 5 lines
IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
Quoted 8 lines
Quoted 8 lines
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
Add a Link: tag, when I read a commit log I want to have a reference to the patches relevant to the commit in question (which in turn will help understand what Pankaj suggested).
Quoted 25 lines
Quoted 25 lines
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
Quoted 6 lines
Quoted 6 lines
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
Add a Link: tag to a message-ID
Quoted 1 line
Quoted 1 line
It is useless in a commit log - this is a moving target.
I can rewrite this commit log if you think it is faster.
Quoted 58 lines
Quoted 58 lines
Cc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)drivers/acpi/arm64/iort.c +36
-0@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {
Quoted 8 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 353
Quoted 8 lines
drivers/acpi/arm64/iort.c +2
-3@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus
Section, page, table number please, "IORT spec says" is too vague.
Quoted 1 line
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 355
Quoted 1 line
drivers/acpi/arm64/iort.c +1
-0+* one, but the IORT code ingored the "minus one", and some firmware
s/ingored/ignored/
Quoted 2 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 357
Quoted 2 lines
drivers/acpi/arm64/iort.c +2
-0+* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware.
It is not "new" vs "old" it is spec compliant vs non-spec compliant.
Quoted 5 lines
Quoted 5 lines
drivers/acpi/arm64/iort.c +5
-0+*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ;
Quoted 8 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 367
Quoted 5 lines
drivers/acpi/arm64/iort.c +5
-0+ }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + }This seems needlessly repetitive and convoluted... how about refactoring to something like:
+1
Quoted 4 lines
map_max = map->input_base + map->id_count; if (apply_id_count_workaround) map_max--;
You can even turn it into an inline function (ie iort_get_map_max()) with the comment above in it so that the quirk is isolated instead of having it in the middle of iort_id_map().
I am fine either way. We need test coverage since I feel this may break a number of systems (ie I don't think it should be merged as a fix).
Lorenzo
Hanjun Guo
maintainerQuoted 6 lines · Lorenzo Pieralisi
Quoted 3 lines · Robin Murphy
Quoted 2 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the
Why is it confusing ? Because we botched the kernel code :) ?
I think 'minus one' is not bringing any benefit :)
Quoted 51 lines
Quoted 6 lines
Quoted 6 lines
IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
Quoted 8 lines
Quoted 8 lines
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
Add a Link: tag, when I read a commit log I want to have a reference to the patches relevant to the commit in question (which in turn will help understand what Pankaj suggested).
Quoted 25 lines
Quoted 25 lines
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
Quoted 6 lines
Quoted 6 lines
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
Quoted 7 lines
Add a Link: tag to a message-ID
Quoted 1 line
Quoted 1 line
It is useless in a commit log - this is a moving target.
I can rewrite this commit log if you think it is faster.
That will be very helpful, please do so, thanks!
Quoted 70 lines
Quoted 59 lines
Quoted 59 lines
Cc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>---
RFC->v1:
- Print warning when matched the workaround info, suggested by Pankaj.drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)drivers/acpi/arm64/iort.c +36
-0@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {Quoted 8 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 353
Quoted 8 lines
drivers/acpi/arm64/iort.c +2
-3@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus Section, page, table number please, "IORT spec says" is too vague.
Quoted 8 lines
Quoted 1 line
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 355
Quoted 1 line
drivers/acpi/arm64/iort.c +1
-0+* one, but the IORT code ingored the "minus one", and some firmware s/ingored/ignored/
Quoted 2 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 357
Quoted 2 lines
drivers/acpi/arm64/iort.c +2
-0+* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware. It is not "new" vs "old" it is spec compliant vs non-spec compliant.
Agreed.
Quoted 18 lines
Quoted 14 lines
drivers/acpi/arm64/iort.c static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in, · line 367
Quoted 11 lines
drivers/acpi/arm64/iort.c +10
-0+*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ; + }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + }This seems needlessly repetitive and convoluted... how about refactoring to something like:
+1
Quoted 1 line
Quoted 7 lines
Quoted 3 lines
map_max = map->input_base + map->id_count; if (apply_id_count_workaround) map_max--;You can even turn it into an inline function (ie
iort_get_map_max()) with the comment above in it so that the quirk is isolated instead of having it in the middle ofiort_id_map().
I vote for this one, it's self-contained.
Quoted 4 lines
I am fine either way. We need test coverage since I feel this may break a number of systems (ie I don't think it should be merged as a fix).
Will you resend this patch with commit log and the updated code? or let me do that? Both are ok to me, let's get it tested for longer time if we merge it ASAP.
Thanks
Hanjun
On Fri, Jan 10, 2020 at 02:22:22PM +0800, Hanjun Guo wrote:
[...]
Quoted 7 lines
Quoted 3 lines
I am fine either way. We need test coverage since I feel this may break a number of systems (ie I don't think it should be merged as a fix).
Will you resend this patch with commit log and the updated code? or let me do that? Both are ok to me, let's get it tested for longer time if we merge it ASAP.
I will write the commit log in reply to the original patch, please update the code and repost the whole thing (it is better you repost it to avoid confusing Catalin and Will).
This technically is not a fix; we can try to get it into v5.6 but I am a bit nervous since it can break other platforms, we have to fix it though so better sooner than later.
Thanks,
Lorenzo
Quoted 9 lines · Hanjun Guo
Quoted 6 lines · Lorenzo Pieralisi
Quoted 3 lines · Robin Murphy
Quoted 2 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the
Why is it confusing ? Because we botched the kernel code :) ?
I think 'minus one' is not bringing any benefit :)
Well, in order to describe a 1:1 mapping of the entire possible ID space, the alternative would have to be to overload the otherwise-nonsensical value of 0 to mean 2^32, which I would argue is an even more non-obvious inconsistency. Encoding strictly positive values as 'value - 1' is a relatively common thing (at least in hardware design).
Robin.
Quoted 45 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
Quoted 7 lines
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platformsCc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>
Commit log rewritten below - please update the code (and check the log) as per this thread discussion.
Thanks,
Lorenzo
-- >8 --
From bc766b2913008519bdb59bbc38907451e8bac0d4 Mon Sep 17 00:00:00 2001
From: Hanjun Guo <[email protected]>
Date: Mon, 30 Dec 2019 20:27:04 +0800
Subject: [PATCH] ACPI/IORT: Fix 'Number of IDs' handling in iort_id_map()
The IORT specification [0] (Section 3, table 4, page 9) defines the 'Number of IDs' as 'The number of IDs in the range minus one'.
However, the IORT ID mapping function iort_id_map() treats the 'Number of IDs' field as if it were the full IDs mapping count, with the following check in place to detect out of boundary input IDs:
InputID >= Input base + Number of IDs
This check is flawed in that it considers the 'Number of IDs' field as the full number of IDs mapping and disregards the 'minus one' from the IDs count.
The correct check in iort_id_map() should be implemented as:
InputID > Input base + Number of IDs
this implements the specification correctly but unfortunately it breaks existing firmwares that erroneously set the 'Number of IDs' as the full IDs mapping count rather than IDs mapping count minus one.
e.g.
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS reference
PCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS reference
Two mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for InputID 0x1100 and with the correct InputID check in place in iort_id_map() the kernel would map the InputID to ITS 0xC4 not 0xD4 as it would be expected.
Therefore, to keep supporting existing flawed firmwares, introduce a workaround that instructs the kernel to use the old InputID range check logic in iort_id_map(), so that we can support both firmwares written with the flawed 'Number of IDs' logic and the correct one as defined in the specifications.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
--- drivers/acpi/arm64/iort.c | 55 ++++++++++++++++++++++++++++++++++++--- 1 file changed, 52 insertions(+), 3 deletions(-)
drivers/acpi/arm64/iort.c +52 -3
@@ -298,6 +298,42 @@ static acpi_status iort_match_node_callback(struct acpi_iort_node *node,return status; } +struct iort_workaround_oem_info { +char oem_id[ACPI_OEM_ID_SIZE +1 ]; +char oem_table_id[ACPI_OEM_TABLE_ID_SIZE +1 ]; +u32 oem_revision; +}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id ="HISI " , + .oem_table_id ="HIP07 " , + .oem_revision =0 , + }, { + .oem_id ="HISI " , + .oem_table_id ="HIP08 " , + .oem_revision =0 , + } +}; + +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +} +static int iort_id_map (struct acpi_iort_id_mapping *map,u8 type,u32 rid_in,u32 *rid_out) {
@@ -314,9 +350,21 @@ static int iort_id_map(struct acpi_iort_id_mapping *map, u8 type, u32 rid_in,return -ENXIO ; } -if (rid_in < map->input_base || - (rid_in >= map->input_base + map->id_count)) -return -ENXIO ; +/* +* IORT spec says Number of IDs = The number of IDs in the range minus +* one, but the IORT code ingored the "minus one", and some firmware +* did that too, so apply a workaround here to keep compatible with +* both new and old versions of the firmware. +*/ +if (apply_id_count_workaround) { +if (rid_in < map->input_base || + (rid_in >= map->input_base + map->id_count)) +return -ENXIO ; + }else { +if (rid_in < map->input_base || + (rid_in > map->input_base + map->id_count)) +return -ENXIO ; + } *rid_out = map->output_base + (rid_in - map->input_base);return 0 ;
@@ -1631,5 +1679,6 @@ void __init acpi_iort_init(void)return ; } +iort_check_id_count_workaround (iort_table);iort_init_platform_devices (); }
Signature · 1 line
2.17.1
Hanjun Guo
maintainerQuoted 50 lines · Lorenzo Pieralisi
Quoted 49 lines · Hanjun Guo
The IORT spec [0] says Number of IDs = The number of IDs in the range minus one, it is confusing but it was written down in the first version of the IORT spec. But the IORT ID mapping function
iort_id_map()did something wrong from the start, which bails out if:the request ID >= the input base + number of IDs
This is wrong because it ignored the "minus one", and breaks some valid usecases such as ID mapping to contain single device mapping without single mapping flag set.
Pankaj Bansal proposed a solution to fix the issue [1], which bails out if:
the request ID > the input base + number of IDs
This works as the spec defined, unfortunately some firmware didn't minus one for the number of IDs in the range, and the propoased solution will break those systems in this way:
PCI hostbridge mapping entry 1:
Input base: 0x1000 ID Count: 0x100 Output base: 0x1000 Output reference: 0xC4 //ITS referencePCI hostbridge mapping entry 2:
Input base: 0x1100 ID Count: 0x100 Output base: 0x2000 Output reference: 0xD4 //ITS referenceTwo mapping entries which the second entry's Input base = the first entry's Input base + ID count, so for requester ID
0x1100will map to ITS0xC4not0xD4if we update '>=' to '>'.So introduce a workaround to match the IORT's OEM information for the broken firmware, also update the logic of the ID mapping for firmwares report the number of IDs as the IORT spec defined, to make the code compatible for both kinds of system.
I checked the ACPI tables in the tianocore/edk2-platforms [2], only HiSilicon HIP07/08 did wrong, so just add HIP07/08 to the workaround info table, if we break other platforms, we can add that later.
[0]: http://infocenter.arm.com/help/topic/com.arm.doc.den0049d/DEN0049D_IO_Remapping_Table.pdf
[1]: https://patchwork.kernel.org/patch/11292823/
[2]: https://github.com/tianocore/edk2-platforms
Quoted 6 lines
Quoted 3 lines
Cc Pankaj Bansal <[email protected]>Cc Lorenzo Pieralisi <[email protected]>Signed-off-by Hanjun Guo <[email protected]>Commit log rewritten below - please update the code (and check the log) as per this thread discussion.
Thank you, I will resend it ASAP.
Thanks
Hanjun
Quoted 5 lines · Guohanjun (Hanjun Guo)
drivers/acpi/arm64/iort.c +5
-0+}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = {
drivers/acpi/arm64/iort.c static acpi_status iort_match_node_callback(struct acpi_iort_node *node, · line 319
Quoted 10 lines
drivers/acpi/arm64/iort.c +10
-0+ { + .oem_id = "HISI ", + .oem_table_id = "HIP07 ", + .oem_revision = 0, + }, { + .oem_id = "HISI ", + .oem_table_id = "HIP08 ", + .oem_revision = 0, + } +};
Am I right in saying that any future BIOS for these chipsets will have to continue to have buggy firmware? If so, it's unfortunate.
Thanks,
John
Quoted 16 lines
drivers/acpi/arm64/iort.c +16
-0+ +static void __init +iort_check_id_count_workaround (struct acpi_table_header *tbl) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (wa_info); i++) { +if (!memcmp (wa_info[i].oem_id, tbl->oem_id,ACPI_OEM_ID_SIZE ) && + !memcmp (wa_info[i].oem_table_id, tbl->oem_table_id,ACPI_OEM_TABLE_ID_SIZE ) && + wa_info[i].oem_revision == tbl->oem_revision) { + apply_id_count_workaround = true; +pr_warn (FW_BUG "ID count for ID mapping entry is wrong, applying workaround\n" ); +break ; + } + } +}
Hanjun Guo
maintainerQuoted 10 lines · John Garry
Quoted 9 lines · Guohanjun (Hanjun Guo)
drivers/acpi/arm64/iort.c +9
-0+}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id = "HISI ", + .oem_table_id = "HIP07 ", + .oem_revision = 0,
Quoted 8 lines
Quoted 6 lines
+ }, {
+ .oem_id = "HISI ",
+ .oem_table_id = "HIP08 ",
+ .oem_revision = 0,
+ }
+};Am I right in saying that any future BIOS for these chipsets will have to continue to have buggy firmware? If so, it's unfortunate.
For better compatibility, I would say yes :(
For example, if you fix that in the firmware, and update the IORT revision number, then it will run pretty good on new version of the kernel, but not on old version of kernel without the backporting of this patch.
Thanks
Hanjun
Quoted 20 lines · Hanjun Guo
Quoted 10 lines · John Garry
Quoted 9 lines · Guohanjun (Hanjun Guo)
drivers/acpi/arm64/iort.c +9
-0+}; + +static bool apply_id_count_workaround; + +static struct iort_workaround_oem_info wa_info[]__initdata = { + { + .oem_id = "HISI ", + .oem_table_id = "HIP07 ", + .oem_revision = 0,Quoted 8 lines
Quoted 6 lines
+ }, {
+ .oem_id = "HISI ",
+ .oem_table_id = "HIP08 ",
+ .oem_revision = 0,
+ }
+};Am I right in saying that any future BIOS for these chipsets will have to continue to have buggy firmware? If so, it's unfortunate.
Quoted 6 lines
For better compatibility, I would say yes :(
For example, if you fix that in the firmware, and update the IORT revision number, then it will run pretty good on new version of the kernel, but not on old version of kernel without the backporting of this patch.
ok, so that seems to be a trade off then. Having to backport introduces a risk.
So then it might be good to add a comment to ID count members in open-source edk2-platforms hip07 and hip08 IORTs to mention it is buggy, so not to be copied as a reference.
Cheers,
John