Re: [PATCH V7 1/8] ACPI: I/O Remapping Table (IORT) initial support

From: Tomasz Nowicki
Date: Wed Jun 22 2016 - 07:06:15 EST


On 22.06.2016 12:50, Marc Zyngier wrote:
On 21/06/16 08:12, Hanjun Guo wrote:
Hi Tomasz,

Sorry for jumping out late, just one comment below.

On 2016/6/20 19:02, Tomasz Nowicki wrote:
IORT shows representation of IO topology for ARM based systems.
It describes how various components are connected together on
parent-child basis e.g. PCI RC -> SMMU -> ITS. Also see IORT spec.

Initial support allows to detect IORT table presence and save its
root pointer obtained through acpi_get_table(). The pointer validity
depends on acpi_gbl_permanent_mmap because if acpi_gbl_permanent_mmap
is not set while using IORT nodes we would dereference unmapped pointers.

For the aforementioned reason call iort_table_detect() from acpi_init()
which guarantees acpi_gbl_permanent_mmap to be set at that point.

Add generic helpers which are helpful for scanning and retrieving
information from IORT table content. List of the most important helpers:
- iort_find_dev_node() finds IORT node for a given device
- iort_node_map_rid() maps device RID and returns IORT node which provides
final translation

Signed-off-by: Tomasz Nowicki <tn@xxxxxxxxxxxx>
---
drivers/acpi/Kconfig | 3 +
drivers/acpi/Makefile | 1 +
drivers/acpi/bus.c | 2 +
drivers/acpi/iort.c | 220 ++++++++++++++++++++++++++++++++++++++++++++++++++
include/linux/iort.h | 30 +++++++

[...]

+static acpi_status
+iort_match_node_callback(struct acpi_iort_node *node, void *context)
+{
+ struct device *dev = context;
+
+ switch (node->type) {
+ case ACPI_IORT_NODE_NAMED_COMPONENT: {
+ struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };

ACPI_ALLOCATE_BUFFER is used here, so ...

+ struct acpi_device *adev = to_acpi_device_node(dev->fwnode);
+ struct acpi_iort_named_component *ncomp;
+
+ if (!adev)
+ break;
+
+ ncomp = (struct acpi_iort_named_component *)node->node_data;
+
+ if (ACPI_FAILURE(acpi_get_name(adev->handle,
+ ACPI_FULL_PATHNAME, &buffer))) {
+ dev_warn(dev, "Can't get device full path name\n");
+ break;
+ }
+
+ if (!strcmp(ncomp->device_name, (char *)buffer.pointer))
+ return AE_OK;

... we need to kfree(buffer.pointer) before we return or break.

For the record, I've queued this patch on top:

diff --git a/drivers/acpi/iort.c b/drivers/acpi/iort.c
index 496dcf6..3b9e55b 100644
--- a/drivers/acpi/iort.c
+++ b/drivers/acpi/iort.c
@@ -158,11 +158,15 @@ iort_match_node_callback(struct acpi_iort_node *node, void *context)
if (ACPI_FAILURE(acpi_get_name(adev->handle,
ACPI_FULL_PATHNAME, &buffer))) {
dev_warn(dev, "Can't get device full path name\n");
- break;
- }
+ } else {
+ int match;

- if (!strcmp(ncomp->device_name, (char *)buffer.pointer))
- return AE_OK;
+ match = !strcmp(ncomp->device_name, buffer.pointer);
+ acpi_os_free(&buffer);
+
+ if (match)
+ return AE_OK;
+ }

break;
}


assuming that Rafael is OK with the general approach, of course.


Thanks Marc. As Lorenzo pointed out, we need another fix for iort_node_map_rid(). Do you want me to refresh the whole series or this patch only?

Tomasz