Hi Andre,
On 04/03/2017 09:28 PM, Andre Przywara wrote:
/* Scan the DT for any ITS nodes and create a list of host ITSes out of it. */
void gicv3_its_dt_init(const struct dt_device_node *node)
{
diff --git a/xen/arch/arm/gic-v3-lpi.c b/xen/arch/arm/gic-v3-lpi.c
index 8b3660a..d3ee141 100644
--- a/xen/arch/arm/gic-v3-lpi.c
+++ b/xen/arch/arm/gic-v3-lpi.c
@@ -20,25 +20,59 @@
[...]
+/*
+ * There could be a lot of LPIs on the host side, and they always go to
+ * a guest. So having a struct irq_desc for each of them would be wasteful
+ * and useless.
+ * Instead just store enough information to find the right VCPU to inject
+ * those LPIs into, which just requires the virtual LPI number.
+ * To avoid a global lock on this data structure, this is using a lockless
+ * approach relying on the architectural atomicty of native data types:
s/atomicty/atomicity/
[...]
+/*
+ * Allocate a block of 32 LPIs on the given host ITS for device "devid",
+ * starting with "eventid". Put them into the respective ITT by issuing a
+ * MAPTI command for each of them.
+ */
+int gicv3_allocate_host_lpi_block(struct domain *d, uint32_t *first_lpi)
+{
[...]
+ /* If we hit an unallocated chunk, we initialize it and use entry 0. */
+ if ( !lpi_data.host_lpis[chunk] )
+ {
+ union host_lpi *new_chunk;
+
+ /* TODO: NUMA locality for quicker IRQ path? */
+ new_chunk = xmalloc_bytes(PAGE_SIZE);
NIT: As suggested on v2, this could be xenheap_alloc_page(0);
+ if ( !new_chunk )
+ {
+ spin_unlock(&lpi_data.host_lpis_lock);
+ return -ENOMEM;
+ }
+
+ for ( i = 0; i < HOST_LPIS_PER_PAGE; i += LPI_BLOCK )
+ new_chunk[i].dom_id = DOMID_INVALID;
As host_lpis could be read without lock, I would add a barrier here to
make sure new_chunk[*].dom_id have been written before setting up the page.
A similar barrier would be needed in the do_lpi(...) path.
[...]
diff --git a/xen/include/asm-arm/gic.h b/xen/include/asm-arm/gic.h
index 836a103..d04bd04 100644
--- a/xen/include/asm-arm/gic.h
+++ b/xen/include/asm-arm/gic.h
@@ -220,6 +220,8 @@ enum gic_version {
GIC_V3,
};
+#define INVALID_LPI 0
Again, I think INVALID_LPI should be moved in irq.h to stay with the
definition of LPI_OFFSET. It was supposed to be fixed in v3...
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
