On Wed, May 07, 2025 at 01:59:12PM +0200, Rafael J. Wysocki wrote:
On Wed, May 7, 2025 at 1:53???PM Heyne, Maximilian mheyne@amazon.de wrote:
On Tue, May 06, 2025 at 03:11:20PM -0500, Jeremy Linton wrote:
Hi,
On 5/6/25 8:43 AM, Sudeep Holla wrote:
On Tue, May 06, 2025 at 01:13:02PM +0000, Heyne, Maximilian wrote:
Commit 7ab4f0e37a0f ("ACPI PPTT: Fix coding mistakes in a couple of sizeof() calls") corrects the processer entry size but unmasked a longer standing bug where the last entry in the structure can get skipped due to an off-by-one mistake if the last entry ends exactly at the end of the ACPI subtable.
Unless the firmware has populated an incorrect value for the header length, I don't see how this is possible. The table_end should point to the address immediately following the last byte of the table. None of the headers are only one byte long, so what am I missing that could explain this apparent off-by-one issue?.
More likely its because the sizeof() fix was merged without proper review and is wrong because the type isn't actually known on the object until the header is checked.
I agree that the type might not be known at this point but the condition
proc_sz = sizeof(struct acpi_pptt_processor); while((unsigned long)entry + proc_sz <= table_end)would make sure that there could potentially be a node of type acpi_pptt_processor because there is at least space for it. If the entry can't be of that size because it would go over table_end then it can't be an acpi_pptt_processor.
I don't follow.
If it is an acpi_pptt_processor entry, the original condition would be sufficient (assuming the correctness of the table header), wouldn't it?
On our hardware we have the following situation: For CPU 63, the acpi_pptt_processor starts at 0xCAA and it has a length of 0x14. The PPTT table has length 0xCBE. This is exactly the case I'm describing. You're traversing the entries in the table and arrive at the last entry but currently you exit the loop because 0xCAA + 0x14 == 0xCBE (it's not < 0xCBE which is table_end). So you skip over the last CPU node.
Therefore, I don't think the sizeof() fix is that wrong but we just need to adjust the while condition.
The sizeof() fix is correct, it makes the code work as designed.
Alternatively, we could at least make sure that we can safely access (without crossing table_end) the acpi_subtable_header to check the type.
Yes.
But the current approach seems cleaner to me.
Why do you think so?
Because the first line in the loop casts straight to acpi_pptt_processor. That's why I think it makes sense to check the entry at which you arrived can actually be a acpi_pptt_processor space-wise.
Amazon Web Services Development Center Germany GmbH Tamara-Danz-Str. 13 10243 Berlin Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B Sitz: Berlin Ust-ID: DE 365 538 597