diff mbox series

[v2] ACPICA: Replace fake flexible arrays with flexible array members

Message ID 20230127191621.gonna.262-kees@kernel.org
State New
Headers show
Series [v2] ACPICA: Replace fake flexible arrays with flexible array members | expand

Commit Message

Kees Cook Jan. 27, 2023, 7:16 p.m. UTC 
One-element arrays (and multi-element arrays being treated as
dynamically sized) are deprecated[1] and are being replaced with
flexible array members in support of the ongoing efforts to tighten the
FORTIFY_SOURCE routines on memcpy(), correctly instrument array indexing
with UBSAN_BOUNDS, and to globally enable -fstrict-flex-arrays=3.

Replace one-element array with flexible-array member in struct
acpi_resource_extended_irq. Replace 4-byte fixed-size array with 4-byte
padding in a union with a flexible-array member in struct
acpi_pci_routing_table.

This results in no differences in binary output.

Link: https://github.com/acpica/acpica/pull/813
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
---
v2: include stddef.h and switch to __DECLARE_FLEX_ARRAY()
v1: https://lore.kernel.org/lkml/20221118181538.never.225-kees@kernel.org/
---
 include/acpi/acrestyp.h | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

Comments

Guenter Roeck Feb. 13, 2023, 12:35 a.m. UTC | #1 
Hi,

On Fri, Jan 27, 2023 at 11:16:25AM -0800, Kees Cook wrote:
> One-element arrays (and multi-element arrays being treated as
> dynamically sized) are deprecated[1] and are being replaced with
> flexible array members in support of the ongoing efforts to tighten the
> FORTIFY_SOURCE routines on memcpy(), correctly instrument array indexing
> with UBSAN_BOUNDS, and to globally enable -fstrict-flex-arrays=3.
> 
> Replace one-element array with flexible-array member in struct
> acpi_resource_extended_irq. Replace 4-byte fixed-size array with 4-byte
> padding in a union with a flexible-array member in struct
> acpi_pci_routing_table.
> 
> This results in no differences in binary output.
> 
> Link: https://github.com/acpica/acpica/pull/813
> Signed-off-by: Kees Cook <keescook@chromium.org>
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

This patch results in boot failures of 32-bit images.
Reverting it fixes the problem.

On the failing boot tests, I see messages such as

ACPI: \_SB_.GSIA: Enabled at IRQ 117440528
ERROR: Unable to locate IOAPIC for GSI 117440528
ahci 0000:00:1f.2: PCI INT A: failed to register GSI

ACPI: \_SB_.GSIG: Enabled at IRQ 117440534
ERROR: Unable to locate IOAPIC for GSI 117440534
8139cp 0000:00:02.0: PCI INT A: failed to register GSI

Given that 117440534 == 0x7000016, that looks quite suspicious.
Indeed, after reverting this patch, the messages are different.

ACPI: \_SB_.GSIA: Enabled at IRQ 16
ahci 0000:00:1f.2: AHCI 0001.0000 32 slots 6 ports 1.5 Gbps 0x3f impl SATA mode

ACPI: \_SB_.GSIG: Enabled at IRQ 22
8139cp 0000:00:02.0 eth0: RTL-8139C+ at 0xd0804000, 52:54:00:12:34:56, IRQ 22

Guenter
Wysocki, Rafael J Feb. 13, 2023, 7:31 p.m. UTC | #2 
On 2/13/2023 1:35 AM, Guenter Roeck wrote:
> Hi,
>
> On Fri, Jan 27, 2023 at 11:16:25AM -0800, Kees Cook wrote:
>> One-element arrays (and multi-element arrays being treated as
>> dynamically sized) are deprecated[1] and are being replaced with
>> flexible array members in support of the ongoing efforts to tighten the
>> FORTIFY_SOURCE routines on memcpy(), correctly instrument array indexing
>> with UBSAN_BOUNDS, and to globally enable -fstrict-flex-arrays=3.
>>
>> Replace one-element array with flexible-array member in struct
>> acpi_resource_extended_irq. Replace 4-byte fixed-size array with 4-byte
>> padding in a union with a flexible-array member in struct
>> acpi_pci_routing_table.
>>
>> This results in no differences in binary output.
>>
>> Link: https://github.com/acpica/acpica/pull/813
>> Signed-off-by: Kees Cook <keescook@chromium.org>
>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> This patch results in boot failures of 32-bit images.
> Reverting it fixes the problem.
>
> On the failing boot tests, I see messages such as
>
> ACPI: \_SB_.GSIA: Enabled at IRQ 117440528
> ERROR: Unable to locate IOAPIC for GSI 117440528
> ahci 0000:00:1f.2: PCI INT A: failed to register GSI
>
> ACPI: \_SB_.GSIG: Enabled at IRQ 117440534
> ERROR: Unable to locate IOAPIC for GSI 117440534
> 8139cp 0000:00:02.0: PCI INT A: failed to register GSI
>
> Given that 117440534 == 0x7000016, that looks quite suspicious.
> Indeed, after reverting this patch, the messages are different.
>
> ACPI: \_SB_.GSIA: Enabled at IRQ 16
> ahci 0000:00:1f.2: AHCI 0001.0000 32 slots 6 ports 1.5 Gbps 0x3f impl SATA mode
>
> ACPI: \_SB_.GSIG: Enabled at IRQ 22
> 8139cp 0000:00:02.0 eth0: RTL-8139C+ at 0xd0804000, 52:54:00:12:34:56, IRQ 22

Thanks for the report, I have reverted this commit from my linux-next 
branch and will not include it into my 6.3-rc1 pull requests at all.

Thanks!
Kees Cook Feb. 13, 2023, 7:51 p.m. UTC | #3 
On February 12, 2023 4:35:46 PM PST, Guenter Roeck <linux@roeck-us.net> wrote:
>Hi,
>
>On Fri, Jan 27, 2023 at 11:16:25AM -0800, Kees Cook wrote:
>> One-element arrays (and multi-element arrays being treated as
>> dynamically sized) are deprecated[1] and are being replaced with
>> flexible array members in support of the ongoing efforts to tighten the
>> FORTIFY_SOURCE routines on memcpy(), correctly instrument array indexing
>> with UBSAN_BOUNDS, and to globally enable -fstrict-flex-arrays=3.
>> 
>> Replace one-element array with flexible-array member in struct
>> acpi_resource_extended_irq. Replace 4-byte fixed-size array with 4-byte
>> padding in a union with a flexible-array member in struct
>> acpi_pci_routing_table.
>> 
>> This results in no differences in binary output.
>> 
>> Link: https://github.com/acpica/acpica/pull/813
>> Signed-off-by: Kees Cook <keescook@chromium.org>
>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>
>This patch results in boot failures of 32-bit images.

Weird -- I didn't see any binary differences. I'll investigate. What compiler and arch?

-Kees

>Reverting it fixes the problem.
>
>On the failing boot tests, I see messages such as
>
>ACPI: \_SB_.GSIA: Enabled at IRQ 117440528
>ERROR: Unable to locate IOAPIC for GSI 117440528
>ahci 0000:00:1f.2: PCI INT A: failed to register GSI
>
>ACPI: \_SB_.GSIG: Enabled at IRQ 117440534
>ERROR: Unable to locate IOAPIC for GSI 117440534
>8139cp 0000:00:02.0: PCI INT A: failed to register GSI
>
>Given that 117440534 == 0x7000016, that looks quite suspicious.
>Indeed, after reverting this patch, the messages are different.

Yeah, seems like a high byte not getting cleared. Hmm.

>
>ACPI: \_SB_.GSIA: Enabled at IRQ 16
>ahci 0000:00:1f.2: AHCI 0001.0000 32 slots 6 ports 1.5 Gbps 0x3f impl SATA mode
>
>ACPI: \_SB_.GSIG: Enabled at IRQ 22
>8139cp 0000:00:02.0 eth0: RTL-8139C+ at 0xd0804000, 52:54:00:12:34:56, IRQ 22
>
>Guenter
Guenter Roeck Feb. 13, 2023, 8:05 p.m. UTC | #4 
On 2/13/23 11:51, Kees Cook wrote:
> On February 12, 2023 4:35:46 PM PST, Guenter Roeck <linux@roeck-us.net> wrote:
>> Hi,
>>
>> On Fri, Jan 27, 2023 at 11:16:25AM -0800, Kees Cook wrote:
>>> One-element arrays (and multi-element arrays being treated as
>>> dynamically sized) are deprecated[1] and are being replaced with
>>> flexible array members in support of the ongoing efforts to tighten the
>>> FORTIFY_SOURCE routines on memcpy(), correctly instrument array indexing
>>> with UBSAN_BOUNDS, and to globally enable -fstrict-flex-arrays=3.
>>>
>>> Replace one-element array with flexible-array member in struct
>>> acpi_resource_extended_irq. Replace 4-byte fixed-size array with 4-byte
>>> padding in a union with a flexible-array member in struct
>>> acpi_pci_routing_table.
>>>
>>> This results in no differences in binary output.
>>>
>>> Link: https://github.com/acpica/acpica/pull/813
>>> Signed-off-by: Kees Cook <keescook@chromium.org>
>>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>>
>> This patch results in boot failures of 32-bit images.
> 
> Weird -- I didn't see any binary differences. I'll investigate. What compiler and arch?
> 

gcc 11.3, binutils 2.39, x86 (32-bit build). This is pretty much defconfig
with ARCH=i386 and various debug options enabled.

64-bit builds do not have the problem.

Guenter

> -Kees
> 
>> Reverting it fixes the problem.
>>
>> On the failing boot tests, I see messages such as
>>
>> ACPI: \_SB_.GSIA: Enabled at IRQ 117440528
>> ERROR: Unable to locate IOAPIC for GSI 117440528
>> ahci 0000:00:1f.2: PCI INT A: failed to register GSI
>>
>> ACPI: \_SB_.GSIG: Enabled at IRQ 117440534
>> ERROR: Unable to locate IOAPIC for GSI 117440534
>> 8139cp 0000:00:02.0: PCI INT A: failed to register GSI
>>
>> Given that 117440534 == 0x7000016, that looks quite suspicious.
>> Indeed, after reverting this patch, the messages are different.
> 
> Yeah, seems like a high byte not getting cleared. Hmm.
> 
>>
>> ACPI: \_SB_.GSIA: Enabled at IRQ 16
>> ahci 0000:00:1f.2: AHCI 0001.0000 32 slots 6 ports 1.5 Gbps 0x3f impl SATA mode
>>
>> ACPI: \_SB_.GSIG: Enabled at IRQ 22
>> 8139cp 0000:00:02.0 eth0: RTL-8139C+ at 0xd0804000, 52:54:00:12:34:56, IRQ 22
>>
>> Guenter
> 
>
diff mbox series

Patch

diff --git a/include/acpi/acrestyp.h b/include/acpi/acrestyp.h
index a7fb8ddb3dc6..250046a7c870 100644
--- a/include/acpi/acrestyp.h
+++ b/include/acpi/acrestyp.h
@@ -10,6 +10,8 @@ 
 #ifndef __ACRESTYP_H__
 #define __ACRESTYP_H__
 
+#include <linux/stddef.h>
+
 /*
  * Definitions for Resource Attributes
  */
@@ -332,7 +334,7 @@  struct acpi_resource_extended_irq {
 	u8 wake_capable;
 	u8 interrupt_count;
 	struct acpi_resource_source resource_source;
-	u32 interrupts[1];
+	u32 interrupts[];
 };
 
 struct acpi_resource_generic_register {
@@ -679,7 +681,10 @@  struct acpi_pci_routing_table {
 	u32 pin;
 	u64 address;		/* here for 64-bit alignment */
 	u32 source_index;
-	char source[4];		/* pad to 64 bits so sizeof() works in all cases */
+	union {
+		char pad[4];	/* pad to 64 bits so sizeof() works in all cases */
+		__DECLARE_FLEX_ARRAY(char, source);
+	};
 };
 
 #endif				/* __ACRESTYP_H__ */