Skip to contentSkip to conversation

x86: acpi: Fix calculation of DSDT length

PATCH Superseded Wolfgang Wallner 3 participants 9 messages Sep 9, 2020 – Sep 16, 2020 x86

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

* acpi_inc()
* Calculate DSDT length
* acpi_align()

---

 arch/x86/lib/acpi_table.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/lib/acpi_table.c +3 -1
index 3a93fedfc3..6b827bfa3f 100644
@@ -427,7 +427,7 @@ ulong write_acpi_tables(ulong start_addr)
 	       (char *)&AmlCode + sizeof(struct acpi_table_header),
 	       dsdt->length - sizeof(struct acpi_table_header));
 
-	acpi_inc_align(ctx, dsdt->length - sizeof(struct acpi_table_header));
+	acpi_inc(ctx, dsdt->length - sizeof(struct acpi_table_header));
 
 	/* Pack GNVS into the ACPI table area */
 	for (i = 0; i < dsdt->length; i++) {
@@ -450,6 +450,8 @@ ulong write_acpi_tables(ulong start_addr)
 	dsdt->checksum = 0;
 	dsdt->checksum = table_compute_checksum((void *)dsdt, dsdt->length);
 
+	acpi_align(ctx);
+
 	/*
 	 * Fill in platform-specific global NVS variables. If this fails we
 	 * cannot return the error but this should only happen while debugging.
Signature · 1 line
2.28.0
#1 lore raw

PATCH 2/2 x86: acpi: Add memset to initialize SPCR table

Add a missing memset to acpi_create_spcr().

The other acpi_create_xxxx() functions perform a memset on their structures, acpi_create_spcr() does not and as a result the contents of this table are partly uninitialized (and thus random after every reset).

---

 arch/x86/lib/acpi_table.c | 3 +++
 1 file changed, 3 insertions(+)
arch/x86/lib/acpi_table.c +3 -0
index 6b827bfa3f..054235843e 100644
@@ -252,6 +252,8 @@ static void acpi_create_spcr(struct acpi_spcr *spcr)
 	int space_id;
 	int ret = -ENODEV;
 
+	memset((void *)spcr, 0, sizeof(struct acpi_spcr));
+
 	/* Fill out header fields */
 	acpi_fill_header(header, "SPCR");
 	header->length = sizeof(struct acpi_spcr);
@@ -359,6 +361,7 @@ static void acpi_create_spcr(struct acpi_spcr *spcr)
 		spcr->baud_rate = 0;
 
 	/* Fix checksum */
+
 	header->checksum = table_compute_checksum((void *)spcr, header->length);
 }
 
Signature · 1 line
2.28.0
#2 lore raw

PATCH 2/2 x86: acpi: Add memset to initialize SPCR table

Quoted 5 lines · Wolfgang Wallner

Add a missing memset to acpi_create_spcr().

The other acpi_create_xxxx() functions perform a memset on their structures, acpi_create_spcr() does not and as a result the contents of this table are partly uninitialized (and thus random after every reset).

Fixes: tag?
In any case, after addressing below,

Quoted 21 lines

---

 arch/x86/lib/acpi_table.c | 3 +++
 1 file changed, 3 insertions(+)
arch/x86/lib/acpi_table.c +2 -0
index 6b827bfa3f..054235843e 100644
@@ -252,6 +252,8 @@ static void acpi_create_spcr(struct acpi_spcr *spcr)
 	int space_id;
 	int ret = -ENODEV;
 
+	memset((void *)spcr, 0, sizeof(struct acpi_spcr));
+
 	/* Fill out header fields */
 	acpi_fill_header(header, "SPCR");
 	header->length = sizeof(struct acpi_spcr);

arch/x86/lib/acpi_table.c static void acpi_create_spcr(struct acpi_spcr *spcr) · line 366

Quoted 6 lines
arch/x86/lib/acpi_table.c +1 -0
@@ -359,6 +361,7 @@ static void acpi_create_spcr(struct acpi_spcr *spcr)
 		spcr->baud_rate = 0;
 
 	/* Fix checksum */
+
 	header->checksum = table_compute_checksum((void *)spcr, header->length);
 }

Unrelated change.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#3 lore raw

PATCH 2/2 x86: acpi: Add memset to initialize SPCR table

In reply to Wolfgang Wallner
Quoted 7 lines · Wolfgang Wallner

Add a missing memset to acpi_create_spcr().

The other acpi_create_xxxx() functions perform a memset on their structures, acpi_create_spcr() does not and as a result the contents of this table are partly uninitialized (and thus random after every reset).

Quoted 5 lines

---

 arch/x86/lib/acpi_table.c | 3 +++
 1 file changed, 3 insertions(+)

For further thought, I do wonder if acpi_fill_header() should do the memset, if sizeof() could be passed as a parameter?

#4 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

In reply to Wolfgang Wallner
Quoted 9 lines · Wolfgang Wallner

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

* acpi_inc()
* Calculate DSDT length
* acpi_align()

Perhaps Fixes: tag?

In any case,

Quoted 33 lines

---

 arch/x86/lib/acpi_table.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/lib/acpi_table.c +3 -1
index 3a93fedfc3..6b827bfa3f 100644
@@ -427,7 +427,7 @@ ulong write_acpi_tables(ulong start_addr)
 	       (char *)&AmlCode + sizeof(struct acpi_table_header),
 	       dsdt->length - sizeof(struct acpi_table_header));
 
-	acpi_inc_align(ctx, dsdt->length - sizeof(struct acpi_table_header));
+	acpi_inc(ctx, dsdt->length - sizeof(struct acpi_table_header));
 
 	/* Pack GNVS into the ACPI table area */
 	for (i = 0; i < dsdt->length; i++) {
@@ -450,6 +450,8 @@ ulong write_acpi_tables(ulong start_addr)
 	dsdt->checksum = 0;
 	dsdt->checksum = table_compute_checksum((void *)dsdt, dsdt->length);
 
+	acpi_align(ctx);
+
 	/*
 	 * Fill in platform-specific global NVS variables. If this fails we
 	 * cannot return the error but this should only happen while debugging.
Signature · 1 line
2.28.0
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#5 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

Quoted 4 lines · Andy Shevchenko
Quoted 3 lines · Wolfgang Wallner

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

Quoted 8 lines
Quoted 6 lines

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

* acpi_inc()
* Calculate DSDT length
* acpi_align()

Perhaps Fixes: tag?

Ah, it is against series not yet applied?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#6 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

Hi Andy,

-----"Andy Shevchenko" <[email protected]> schrieb: -----

Quoted 10 lines

Betreff: Re: [PATCH 1/2] x86: acpi: Fix calculation of DSDT length

Quoted 7 lines · Andy Shevchenko
Quoted 6 lines · Wolfgang Wallner

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

* acpi_inc()
* Calculate DSDT length
* acpi_align()

Perhaps Fixes: tag?

Ah, it is against series not yet applied?

No, the patch is against master.
But I could not point to a specific commit that is fixed, the code in question is a result of multiple commits.

Because of this I have also sent V2 without a Fixes tags.

regards, Wolfgang

#7 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

Quoted 17 lines · Wolfgang Wallner

-----"Andy Shevchenko" <[email protected]> schrieb: -----

Quoted 16 lines

Betreff: Re: [PATCH 1/2] x86: acpi: Fix calculation of DSDT length

Quoted 12 lines · Andy Shevchenko
Quoted 9 lines · Wolfgang Wallner

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

* acpi_inc()
* Calculate DSDT length
* acpi_align()

Perhaps Fixes: tag?

Quoted 7 lines
Quoted 1 line

Ah, it is against series not yet applied?

No, the patch is against master.
But I could not point to a specific commit that is fixed, the code in question is a result of multiple commits.

Because of this I have also sent V2 without a Fixes tags.

Hmm... Usually in kernel we put a chain of Fixes in such cases.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#8 lore raw

PATCH 1/2 x86: acpi: Fix calculation of DSDT length

In reply to Wolfgang Wallner
Quoted 11 lines · Wolfgang Wallner

Currently, the calculation for the length of the DSDT table includes any bytes that are added for alignment, but those bytes are not initialized.

This is because the DSDT length is calculated after a call to acpi_inc_align(). Split this up into the following sequence:

* acpi_inc()
* Calculate DSDT length
* acpi_align()

Quoted 6 lines

---

 arch/x86/lib/acpi_table.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)