Enable injection using EINJv2 mode of operation.
[Tony: Mostly Zaid's original code. I just changed how the error ID
and syndrome bits are implemented. Also swapped out some camelcase
variable names]
Co-developed-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Zaid Alali <zaidal@os.amperecomputing.com>
---
drivers/acpi/apei/einj-core.c | 56 ++++++++++++++++++++++++++++-------
1 file changed, 45 insertions(+), 11 deletions(-)
diff --git a/drivers/acpi/apei/einj-core.c b/drivers/acpi/apei/einj-core.c
index 8d60e5f1785c..7741c2082f33 100644
--- a/drivers/acpi/apei/einj-core.c
+++ b/drivers/acpi/apei/einj-core.c
@@ -87,6 +87,7 @@ enum {
SETWA_FLAGS_APICID = 1,
SETWA_FLAGS_MEM = 2,
SETWA_FLAGS_PCIE_SBDF = 4,
+ SETWA_FLAGS_EINJV2 = 8,
};
/*
@@ -181,6 +182,7 @@ bool einj_initialized __ro_after_init;
static void __iomem *einj_param;
static u32 v5param_size;
+static bool is_v2;
static void einj_exec_ctx_init(struct apei_exec_context *ctx)
{
@@ -507,12 +509,20 @@ static int __einj_error_trigger(u64 trigger_paddr, u32 type,
return rc;
}
+static bool is_end_of_list(u8 *val)
+{
+ for (int i = 0; i < COMPONENT_LEN; ++i) {
+ if (val[i] != 0xFF)
+ return false;
+ }
+ return true;
+}
static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
u64 param3, u64 param4)
{
struct apei_exec_context ctx;
u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
- int rc;
+ int i, rc;
einj_exec_ctx_init(&ctx);
@@ -521,10 +531,10 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
return rc;
apei_exec_ctx_set_input(&ctx, type);
if (acpi5) {
- struct set_error_type_with_address *v5param, v5_struct;
+ struct set_error_type_with_address *v5param;
- v5param = &v5_struct;
- memcpy_fromio(v5param, einj_param, sizeof(*v5param));
+ v5param = kmalloc(v5param_size, GFP_KERNEL);
+ memcpy_fromio(v5param, einj_param, v5param_size);
v5param->type = type;
if (type & ACPI5_VENDOR_BIT) {
switch (vendor_flags) {
@@ -544,8 +554,21 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
v5param->flags = flags;
v5param->memory_address = param1;
v5param->memory_address_range = param2;
- v5param->apicid = param3;
- v5param->pcie_sbdf = param4;
+
+ if (is_v2) {
+ for (i = 0; i < max_nr_components; i++) {
+ if (is_end_of_list(syndrome_data[i].comp_id.acpi_id))
+ break;
+ v5param->einjv2_struct.component_arr[i].comp_id =
+ syndrome_data[i].comp_id;
+ v5param->einjv2_struct.component_arr[i].comp_synd =
+ syndrome_data[i].comp_synd;
+ }
+ v5param->einjv2_struct.component_arr_count = i;
+ } else {
+ v5param->apicid = param3;
+ v5param->pcie_sbdf = param4;
+ }
} else {
switch (type) {
case ACPI_EINJ_PROCESSOR_CORRECTABLE:
@@ -569,7 +592,8 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
break;
}
}
- memcpy_toio(einj_param, v5param, sizeof(*v5param));
+ memcpy_toio(einj_param, v5param, v5param_size);
+ kfree(v5param);
} else {
rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
if (rc)
@@ -631,10 +655,15 @@ int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2, u64 param3,
u64 base_addr, size;
/* If user manually set "flags", make sure it is legal */
- if (flags && (flags &
- ~(SETWA_FLAGS_APICID|SETWA_FLAGS_MEM|SETWA_FLAGS_PCIE_SBDF)))
+ if (flags && (flags & ~(SETWA_FLAGS_APICID | SETWA_FLAGS_MEM |
+ SETWA_FLAGS_PCIE_SBDF | SETWA_FLAGS_EINJV2)))
return -EINVAL;
+ /* check if type is a valid EINJv2 error type */
+ if (is_v2) {
+ if (!(type & available_error_type_v2))
+ return -EINVAL;
+ }
/*
* We need extra sanity checks for memory errors.
* Other types leap directly to injection.
@@ -743,7 +772,7 @@ static int available_error_type_show(struct seq_file *m, void *v)
if (available_error_type & einj_error_type_string[pos].mask)
seq_printf(m, "0x%08x\t%s\n", einj_error_type_string[pos].mask,
einj_error_type_string[pos].str);
- if (available_error_type & ACPI65_EINJV2_SUPP) {
+ if ((available_error_type & ACPI65_EINJV2_SUPP) && einj_v2_enabled) {
for (int pos = 0; pos < ARRAY_SIZE(einjv2_error_type_string); pos++) {
if (available_error_type_v2 & einjv2_error_type_string[pos].mask)
seq_printf(m, "V2_0x%08x\t%s\n", einjv2_error_type_string[pos].mask,
@@ -785,7 +814,7 @@ int einj_validate_error_type(u64 type)
if (tval & (tval - 1))
return -EINVAL;
if (!vendor)
- if (!(type & available_error_type))
+ if (!(type & (available_error_type | available_error_type_v2)))
return -EINVAL;
return 0;
@@ -804,9 +833,11 @@ static ssize_t error_type_set(struct file *file, const char __user *buf,
if (strncmp(einj_buf, "V2_", 3) == 0) {
if (!sscanf(einj_buf, "V2_%llx", &val))
return -EINVAL;
+ is_v2 = true;
} else {
if (!sscanf(einj_buf, "%llx", &val))
return -EINVAL;
+ is_v2 = false;
}
rc = einj_validate_error_type(val);
@@ -828,6 +859,9 @@ static int error_inject_set(void *data, u64 val)
if (!error_type)
return -EINVAL;
+ if (is_v2)
+ error_flags |= SETWA_FLAGS_EINJV2;
+
return einj_error_inject(error_type, error_flags, error_param1, error_param2,
error_param3, error_param4);
}
--
2.43.0Zaid Alali wrote:
> Enable injection using EINJv2 mode of operation.
>
> [Tony: Mostly Zaid's original code. I just changed how the error ID
> and syndrome bits are implemented. Also swapped out some camelcase
> variable names]
>
> Co-developed-by: Tony Luck <tony.luck@intel.com>
> Signed-off-by: Tony Luck <tony.luck@intel.com>
> Signed-off-by: Zaid Alali <zaidal@os.amperecomputing.com>
> ---
> drivers/acpi/apei/einj-core.c | 56 ++++++++++++++++++++++++++++-------
> 1 file changed, 45 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/acpi/apei/einj-core.c b/drivers/acpi/apei/einj-core.c
> index 8d60e5f1785c..7741c2082f33 100644
> --- a/drivers/acpi/apei/einj-core.c
> +++ b/drivers/acpi/apei/einj-core.c
> @@ -87,6 +87,7 @@ enum {
> SETWA_FLAGS_APICID = 1,
> SETWA_FLAGS_MEM = 2,
> SETWA_FLAGS_PCIE_SBDF = 4,
> + SETWA_FLAGS_EINJV2 = 8,
> };
>
> /*
> @@ -181,6 +182,7 @@ bool einj_initialized __ro_after_init;
>
> static void __iomem *einj_param;
> static u32 v5param_size;
> +static bool is_v2;
>
> static void einj_exec_ctx_init(struct apei_exec_context *ctx)
> {
> @@ -507,12 +509,20 @@ static int __einj_error_trigger(u64 trigger_paddr, u32 type,
> return rc;
> }
>
> +static bool is_end_of_list(u8 *val)
> +{
> + for (int i = 0; i < COMPONENT_LEN; ++i) {
Back in patch 3/7 these are defined using a hard coded value.
I think it might be better to at least use COMPONENT_LEN for those
definitions.
I'm also wondering if it would be better to have some type safety here...
but probably fine.
> + if (val[i] != 0xFF)
> + return false;
> + }
> + return true;
I'm unclear of the way this list is terminated. The cover letter does not
mention it. I read the documentation patch and it looks like you echo ''
to the id to terminate. How does that work here?
From the documentation patch.
# echo '' > component_id2 # Mark id2 invalid to terminate list
> +}
> static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> u64 param3, u64 param4)
> {
> struct apei_exec_context ctx;
> u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
> - int rc;
> + int i, rc;
>
> einj_exec_ctx_init(&ctx);
>
> @@ -521,10 +531,10 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> return rc;
> apei_exec_ctx_set_input(&ctx, type);
> if (acpi5) {
> - struct set_error_type_with_address *v5param, v5_struct;
> + struct set_error_type_with_address *v5param;
>
> - v5param = &v5_struct;
> - memcpy_fromio(v5param, einj_param, sizeof(*v5param));
> + v5param = kmalloc(v5param_size, GFP_KERNEL);
> + memcpy_fromio(v5param, einj_param, v5param_size);
> v5param->type = type;
> if (type & ACPI5_VENDOR_BIT) {
> switch (vendor_flags) {
> @@ -544,8 +554,21 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> v5param->flags = flags;
> v5param->memory_address = param1;
> v5param->memory_address_range = param2;
> - v5param->apicid = param3;
> - v5param->pcie_sbdf = param4;
> +
> + if (is_v2) {
> + for (i = 0; i < max_nr_components; i++) {
> + if (is_end_of_list(syndrome_data[i].comp_id.acpi_id))
> + break;
> + v5param->einjv2_struct.component_arr[i].comp_id =
> + syndrome_data[i].comp_id;
> + v5param->einjv2_struct.component_arr[i].comp_synd =
> + syndrome_data[i].comp_synd;
> + }
> + v5param->einjv2_struct.component_arr_count = i;
> + } else {
> + v5param->apicid = param3;
> + v5param->pcie_sbdf = param4;
> + }
> } else {
> switch (type) {
> case ACPI_EINJ_PROCESSOR_CORRECTABLE:
> @@ -569,7 +592,8 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> break;
> }
> }
> - memcpy_toio(einj_param, v5param, sizeof(*v5param));
> + memcpy_toio(einj_param, v5param, v5param_size);
> + kfree(v5param);
> } else {
> rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
> if (rc)
> @@ -631,10 +655,15 @@ int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2, u64 param3,
> u64 base_addr, size;
>
> /* If user manually set "flags", make sure it is legal */
> - if (flags && (flags &
> - ~(SETWA_FLAGS_APICID|SETWA_FLAGS_MEM|SETWA_FLAGS_PCIE_SBDF)))
> + if (flags && (flags & ~(SETWA_FLAGS_APICID | SETWA_FLAGS_MEM |
> + SETWA_FLAGS_PCIE_SBDF | SETWA_FLAGS_EINJV2)))
> return -EINVAL;
>
> + /* check if type is a valid EINJv2 error type */
> + if (is_v2) {
> + if (!(type & available_error_type_v2))
> + return -EINVAL;
> + }
> /*
> * We need extra sanity checks for memory errors.
> * Other types leap directly to injection.
> @@ -743,7 +772,7 @@ static int available_error_type_show(struct seq_file *m, void *v)
> if (available_error_type & einj_error_type_string[pos].mask)
> seq_printf(m, "0x%08x\t%s\n", einj_error_type_string[pos].mask,
> einj_error_type_string[pos].str);
> - if (available_error_type & ACPI65_EINJV2_SUPP) {
> + if ((available_error_type & ACPI65_EINJV2_SUPP) && einj_v2_enabled) {
> for (int pos = 0; pos < ARRAY_SIZE(einjv2_error_type_string); pos++) {
> if (available_error_type_v2 & einjv2_error_type_string[pos].mask)
> seq_printf(m, "V2_0x%08x\t%s\n", einjv2_error_type_string[pos].mask,
> @@ -785,7 +814,7 @@ int einj_validate_error_type(u64 type)
> if (tval & (tval - 1))
> return -EINVAL;
> if (!vendor)
> - if (!(type & available_error_type))
> + if (!(type & (available_error_type | available_error_type_v2)))
> return -EINVAL;
>
> return 0;
> @@ -804,9 +833,11 @@ static ssize_t error_type_set(struct file *file, const char __user *buf,
> if (strncmp(einj_buf, "V2_", 3) == 0) {
> if (!sscanf(einj_buf, "V2_%llx", &val))
> return -EINVAL;
> + is_v2 = true;
> } else {
> if (!sscanf(einj_buf, "%llx", &val))
> return -EINVAL;
> + is_v2 = false;
> }
>
> rc = einj_validate_error_type(val);
> @@ -828,6 +859,9 @@ static int error_inject_set(void *data, u64 val)
> if (!error_type)
> return -EINVAL;
>
> + if (is_v2)
> + error_flags |= SETWA_FLAGS_EINJV2;
> +
> +
Does this flag need to be cleared if a v1 error is being used?
Ira
> return einj_error_inject(error_type, error_flags, error_param1, error_param2,
> error_param3, error_param4);
> }
> --
> 2.43.0
>
On Fri, Jun 13, 2025 at 12:37:54PM -0500, Ira Weiny wrote:
> Zaid Alali wrote:
> > Enable injection using EINJv2 mode of operation.
> >
> > [Tony: Mostly Zaid's original code. I just changed how the error ID
> > and syndrome bits are implemented. Also swapped out some camelcase
> > variable names]
> >
> > Co-developed-by: Tony Luck <tony.luck@intel.com>
> > Signed-off-by: Tony Luck <tony.luck@intel.com>
> > Signed-off-by: Zaid Alali <zaidal@os.amperecomputing.com>
> > ---
> > drivers/acpi/apei/einj-core.c | 56 ++++++++++++++++++++++++++++-------
> > 1 file changed, 45 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/acpi/apei/einj-core.c b/drivers/acpi/apei/einj-core.c
> > index 8d60e5f1785c..7741c2082f33 100644
> > --- a/drivers/acpi/apei/einj-core.c
> > +++ b/drivers/acpi/apei/einj-core.c
> > @@ -87,6 +87,7 @@ enum {
> > SETWA_FLAGS_APICID = 1,
> > SETWA_FLAGS_MEM = 2,
> > SETWA_FLAGS_PCIE_SBDF = 4,
> > + SETWA_FLAGS_EINJV2 = 8,
> > };
> >
> > /*
> > @@ -181,6 +182,7 @@ bool einj_initialized __ro_after_init;
> >
> > static void __iomem *einj_param;
> > static u32 v5param_size;
> > +static bool is_v2;
> >
> > static void einj_exec_ctx_init(struct apei_exec_context *ctx)
> > {
> > @@ -507,12 +509,20 @@ static int __einj_error_trigger(u64 trigger_paddr, u32 type,
> > return rc;
> > }
> >
> > +static bool is_end_of_list(u8 *val)
> > +{
> > + for (int i = 0; i < COMPONENT_LEN; ++i) {
>
> Back in patch 3/7 these are defined using a hard coded value.
>
> I think it might be better to at least use COMPONENT_LEN for those
> definitions.
Thats a good idea! I will update this in the next revision.
>
> I'm also wondering if it would be better to have some type safety here...
> but probably fine.
>
> > + if (val[i] != 0xFF)
> > + return false;
> > + }
> > + return true;
>
> I'm unclear of the way this list is terminated. The cover letter does not
> mention it. I read the documentation patch and it looks like you echo ''
> to the id to terminate. How does that work here?
>
> From the documentation patch.
>
> # echo '' > component_id2 # Mark id2 invalid to terminate list
>
>
As shown in ptach 7/7 "Writing just a newline to any of these files sets
an invalid (all-ones) value."
> > +}
> > static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> > u64 param3, u64 param4)
> > {
> > struct apei_exec_context ctx;
> > u64 val, trigger_paddr, timeout = FIRMWARE_TIMEOUT;
> > - int rc;
> > + int i, rc;
> >
> > einj_exec_ctx_init(&ctx);
> >
> > @@ -521,10 +531,10 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> > return rc;
> > apei_exec_ctx_set_input(&ctx, type);
> > if (acpi5) {
> > - struct set_error_type_with_address *v5param, v5_struct;
> > + struct set_error_type_with_address *v5param;
> >
> > - v5param = &v5_struct;
> > - memcpy_fromio(v5param, einj_param, sizeof(*v5param));
> > + v5param = kmalloc(v5param_size, GFP_KERNEL);
> > + memcpy_fromio(v5param, einj_param, v5param_size);
> > v5param->type = type;
> > if (type & ACPI5_VENDOR_BIT) {
> > switch (vendor_flags) {
> > @@ -544,8 +554,21 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> > v5param->flags = flags;
> > v5param->memory_address = param1;
> > v5param->memory_address_range = param2;
> > - v5param->apicid = param3;
> > - v5param->pcie_sbdf = param4;
> > +
> > + if (is_v2) {
> > + for (i = 0; i < max_nr_components; i++) {
> > + if (is_end_of_list(syndrome_data[i].comp_id.acpi_id))
> > + break;
> > + v5param->einjv2_struct.component_arr[i].comp_id =
> > + syndrome_data[i].comp_id;
> > + v5param->einjv2_struct.component_arr[i].comp_synd =
> > + syndrome_data[i].comp_synd;
> > + }
> > + v5param->einjv2_struct.component_arr_count = i;
> > + } else {
> > + v5param->apicid = param3;
> > + v5param->pcie_sbdf = param4;
> > + }
> > } else {
> > switch (type) {
> > case ACPI_EINJ_PROCESSOR_CORRECTABLE:
> > @@ -569,7 +592,8 @@ static int __einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2,
> > break;
> > }
> > }
> > - memcpy_toio(einj_param, v5param, sizeof(*v5param));
> > + memcpy_toio(einj_param, v5param, v5param_size);
> > + kfree(v5param);
> > } else {
> > rc = apei_exec_run(&ctx, ACPI_EINJ_SET_ERROR_TYPE);
> > if (rc)
> > @@ -631,10 +655,15 @@ int einj_error_inject(u32 type, u32 flags, u64 param1, u64 param2, u64 param3,
> > u64 base_addr, size;
> >
> > /* If user manually set "flags", make sure it is legal */
> > - if (flags && (flags &
> > - ~(SETWA_FLAGS_APICID|SETWA_FLAGS_MEM|SETWA_FLAGS_PCIE_SBDF)))
> > + if (flags && (flags & ~(SETWA_FLAGS_APICID | SETWA_FLAGS_MEM |
> > + SETWA_FLAGS_PCIE_SBDF | SETWA_FLAGS_EINJV2)))
> > return -EINVAL;
> >
> > + /* check if type is a valid EINJv2 error type */
> > + if (is_v2) {
> > + if (!(type & available_error_type_v2))
> > + return -EINVAL;
> > + }
> > /*
> > * We need extra sanity checks for memory errors.
> > * Other types leap directly to injection.
> > @@ -743,7 +772,7 @@ static int available_error_type_show(struct seq_file *m, void *v)
> > if (available_error_type & einj_error_type_string[pos].mask)
> > seq_printf(m, "0x%08x\t%s\n", einj_error_type_string[pos].mask,
> > einj_error_type_string[pos].str);
> > - if (available_error_type & ACPI65_EINJV2_SUPP) {
> > + if ((available_error_type & ACPI65_EINJV2_SUPP) && einj_v2_enabled) {
> > for (int pos = 0; pos < ARRAY_SIZE(einjv2_error_type_string); pos++) {
> > if (available_error_type_v2 & einjv2_error_type_string[pos].mask)
> > seq_printf(m, "V2_0x%08x\t%s\n", einjv2_error_type_string[pos].mask,
> > @@ -785,7 +814,7 @@ int einj_validate_error_type(u64 type)
> > if (tval & (tval - 1))
> > return -EINVAL;
> > if (!vendor)
> > - if (!(type & available_error_type))
> > + if (!(type & (available_error_type | available_error_type_v2)))
> > return -EINVAL;
> >
> > return 0;
> > @@ -804,9 +833,11 @@ static ssize_t error_type_set(struct file *file, const char __user *buf,
> > if (strncmp(einj_buf, "V2_", 3) == 0) {
> > if (!sscanf(einj_buf, "V2_%llx", &val))
> > return -EINVAL;
> > + is_v2 = true;
> > } else {
> > if (!sscanf(einj_buf, "%llx", &val))
> > return -EINVAL;
> > + is_v2 = false;
> > }
> >
> > rc = einj_validate_error_type(val);
> > @@ -828,6 +859,9 @@ static int error_inject_set(void *data, u64 val)
> > if (!error_type)
> > return -EINVAL;
> >
> > + if (is_v2)
> > + error_flags |= SETWA_FLAGS_EINJV2;
> > +
> > +
>
> Does this flag need to be cleared if a v1 error is being used?
>
> Ira
The driver depends on "is_v2" to determine if the injection is v1 or v2, and the user
is supposed to set the flags for v1 and v2, but I think clearing the flag here is a good idea so
the user does not have to worry about that. I can see why someone would assume not needing
to set the flags when they see "V2_" prefix for v2 error types.
I'll also update this.
Zaid
>
> > return einj_error_inject(error_type, error_flags, error_param1, error_param2,
> > error_param3, error_param4);
> > }
> > --
> > 2.43.0
> >
© 2016 - 2025 Red Hat, Inc.