x86, acpi: Handle lapic/x2apic entries in MADT

Commit Message

Marc Zyngier Aug. 26, 2015, 12:43 p.m. UTC

On 26/08/15 12:42, Lorenzo Pieralisi wrote:
> Hi Lukasz,
> 
> On Wed, Aug 26, 2015 at 11:43:04AM +0100, Marc Zyngier wrote:
>> Hi Lukasz,
>>
>> On 26/08/15 08:04, Anaczkowski, Lukasz wrote:
>>> On Monday, August 3, 2015 8:26 PM
>>> Lukasz Anaczkowski <[e-mail]> wrote:
>>>
>>>> v2: Fixed ARM64 syntax error
>>>
>>> Hi Marc,
>>>
>>> Does this patch look ok now?
> 
> No it does not, it seems to break arm64, I put together a fix
> below. I do not think the way you handle the count increment
> in acpi_parse_entries() is correct anyway, since you increment
> it only if max_entries != 0, which changes mainline behaviour.

Yeah, this is fundamentally flawed:

- count is only incremented when max_entries != 0, as you noticed
- With max_entries != 0, count now represent the sum of all matches
  Is that expected?
- The proc iteration stops after the first match. Why?
- The test for max_entries is done inside the proc loop. Why?

I came up with the following patch that restores arm64 to a booting state.

If the intention was to change the meaning of the acpi_parse_entries
return value, then this should be documented and agreed upon.


Thanks,

	M.

Comments

[e-mail] Aug. 26, 2015, 5:49 p.m. UTC | #1

Marc nad Lorenzo,

First of all appologies for breaking arm64 (again) and thank you for
debugging effort. I own you.

> - count is only incremented when max_entries != 0, as you noticed

You are right, sorry for that, it's fixed in v3.

> - With max_entries != 0, count now represent the sum of all matches
>  Is that expected?

I have no strong opinion on that one. All of the x86 ACPI entries
handling only checks for count < 0, or uses count from the
acpi_subtable_proc structure (and that's why I didn't noticed the
mainline breakage).
If you think it's not correct or less usable than other approach,
let me know.

> - The proc iteration stops after the first match. Why?

So, the initial implementation of the acpi_parse_entries accepted
single handler for the ACPI table. Now, with this change, assumption
is that different handlers for different tables/subtables are passed,
meaning only one can meet entry->type == proc[i].id condition.
mainline breakage). This approach saves one local varaible, but
I don't think this is ultimate argument :)

> - The test for max_entries is done inside the proc loop. Why?

That's obviously wrong in context of the overall wrong counting.

> [...] this should be documented and agreed upon.

I've added description with assumptions. Again, if you think it's
not correct, let me know.

Tomasz Nowicki wrote:
> should acpi_table_parse_entries suppose to be removed above?

Thanks for pointing this out. I've missed implementation of
acpi_table_parse_entries when was backporting initial patch.
I've added it back.

Cheers,
Lukasz

--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/acpi/tables.c b/drivers/acpi/tables.c
index 1217e41..f06327f 100644
--- a/drivers/acpi/tables.c
+++ b/drivers/acpi/tables.c
@@ -249,19 +249,24 @@  acpi_parse_entries(char *id, unsigned long table_size,

 	while (((unsigned long)entry) + sizeof(struct acpi_subtable_header) <
 	       table_end) {
+		bool match = false;
+
+		if (max_entries && count >= max_entries)
+			break;
 		for (i = 0; i < proc_num; i++) {
 			if (entry->type != proc[i].id)
 				continue;
-			if (max_entries && count++ >= max_entries)
-				continue;
 			if (proc[i].handler(entry, table_end)) {
 				proc[i].count = -EINVAL;
 				return -EINVAL;
 			}
 			proc[i].count++;
-			break;
+			match = true;
 		}

+		if (match)
+			count++;
+
 		/*
 		 * If entry->length is 0, break from this loop to avoid
 		 * infinite loop.