In commit fe8e93504ce8 ("irqchip/gic-v3-its: Use full range of LPIs"),
it removes the cap for lpi_id_bits. But it will cause more pointless
memory footprint.
There is a WARN_ON when my QDF2400 server boots up (pagesize is 4k)
============begin===============
[ 0.000000] GICv3: GIC: Using split EOI/Deactivate mode
[ 0.000000] GICv3: Distributor has no Range Selector support
[ 0.000000] GICv3: VLPI support, no direct LPI support
[ 0.000000] ACPI: SRAT not present
[ 0.000000] ITS [mem 0xff7efe0000-0xff7effffff]
[ 0.000000] ITS@0x000000ff7efe0000: Using ITS number 0
[ 0.000000] GIC: enabling workaround for ITS: QDF2400 erratum 0065
[ 0.000000] ITS@0x000000ff7efe0000: allocated 524288 Devices
@179f000000 (indirect, esz 8, psz 64K, shr 1)
[ 0.000000] ITS@0x000000ff7efe0000: allocated 8192 Interrupt
Collections @179f930000 (flat, esz 8, psz 64K, shr 1)
[ 0.000000] ITS@0x000000ff7efe0000: allocated 65536 Virtual CPUs
@179f980000 (flat, esz 8, psz 64K, shr 1)
[ 0.000000] WARNING: CPU: 0 PID: 0 at mm/page_alloc.c:4066
__alloc_pages_nodemask+0x2d8/0x1188
[ 0.000000] Modules linked in:
[ 0.000000] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.18.0+ #66
[ 0.000000] pstate: 20000085 (nzCv daIf -PAN -UAO)
[ 0.000000] pc : __alloc_pages_nodemask+0x2d8/0x1188
[ 0.000000] lr : __alloc_pages_nodemask+0x134/0x1188
[ 0.000000] sp : ffff0000091b3a30
[ 0.000000] x29: ffff0000091b3a30 x28: 0000000000000000
[ 0.000000] x27: ffff00000a045000 x26: 0000000000000000
[ 0.000000] x25: 0000000000000000 x24: ffff0000091c1000
[ 0.000000] x23: ffff0000091b3b98 x22: 000000000000000c
[ 0.000000] x21: ffff0000082dc130 x20: 0000000000000001
[ 0.000000] x19: 0000000000000000 x18: 000000000000003f
[ 0.000000] x17: 0000000000000000 x16: 0000000000000000
[ 0.000000] x15: ffffffffffffffff x14: 202c74616c662820
[ 0.000000] x13: ffff000009c5f9e0 x12: 0000000000000077
[ 0.000000] x11: 0000000000000078 x10: 0000000097110f47
[ 0.000000] x9 : 0000000000000000 x8 : 00000000009fff3f
[ 0.000000] x7 : 000000000000002b x6 : 000000000000000c
[ 0.000000] x5 : 0000000000000001 x4 : 0000000000000000
[ 0.000000] x3 : ffff000008fd1000 x2 : ffff000008fd1000
[ 0.000000] x1 : ffff0000091c1000 x0 : 0000000000000000
[ 0.000000] Call trace:
[ 0.000000] __alloc_pages_nodemask+0x2d8/0x1188
[ 0.000000] alloc_pages_current+0x8c/0xd8
[ 0.000000] its_allocate_prop_table+0x5c/0xb8
[ 0.000000] its_init+0x220/0x3c0
[ 0.000000] gic_init_bases+0x250/0x380
[ 0.000000] gic_acpi_init+0x16c/0x2a4
[ 0.000000] acpi_match_madt+0x50/0x88
[ 0.000000] acpi_table_parse_entries_array+0x180/0x204
[ 0.000000] acpi_table_parse_entries+0x60/0x84
[ 0.000000] acpi_table_parse_madt+0x40/0x4c
[ 0.000000] __acpi_probe_device_table+0x94/0xe8
[ 0.000000] irqchip_init+0x38/0x40
[ 0.000000] init_IRQ+0x70/0x9c
[ 0.000000] start_kernel+0x310/0x4c0
[ 0.000000] irq event stamp: 0
[ 0.000000] hardirqs last enabled at (0): [<0000000000000000>]
(null)
[ 0.000000] hardirqs last disabled at (0): [<0000000000000000>]
(null)
[ 0.000000] softirqs last enabled at (0): [<0000000000000000>]
(null)
[ 0.000000] softirqs last disabled at (0): [<0000000000000000>]
(null)
[ 0.000000] ---[ end trace 943781056d97862b ]---
============end============
In its_alloc_lpi_tables, lpi_id_bits is 24 in QDF2400. Then
its_allocate_prop_table will try to allocate 16M(order 12 if
pagesize=4k). Thus it causes the WARN_ON.
As said by Marc,
Capping lpi_id_bits at 16 (which is what we had before) is plenty,
will save a some memory, and gives some margin before we need to push
it up again.
This patch re-caps the lpi_id_bits.
Fixes: fe8e93504ce8 ("irqchip/gic-v3-its: Use full range of LPIs")
Signed-off-by: Jia He <jia...@hxt-semitech.com>
Suggested-by: Marc Zyngier <marc.zyng...@arm.com>
---
drivers/irqchip/irq-gic-v3-its.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 316a575..c2df341 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -1439,6 +1439,7 @@ static int its_irq_set_vcpu_affinity(struct irq_data *d,
void *vcpu_info)
* The consequence of the above is that allocation is cost is low, but
* freeing is expensive. We assumes that freeing rarely occurs.
*/
+#define ITS_MAX_LPI_NRBITS 16 /* 64K LPIs */
static DEFINE_MUTEX(lpi_range_lock);
static LIST_HEAD(lpi_range_list);
@@ -1625,7 +1626,8 @@ static int __init its_alloc_lpi_tables(void)
{
phys_addr_t paddr;
- lpi_id_bits = GICD_TYPER_ID_BITS(gic_rdists->gicd_typer);
+ lpi_id_bits = min_t(u32, GICD_TYPER_ID_BITS(gic_rdists->gicd_typer),
+ ITS_MAX_LPI_NRBITS);
gic_rdists->prop_page = its_allocate_prop_table(GFP_NOWAIT);
if (!gic_rdists->prop_page) {
pr_err("Failed to allocate PROPBASE\n");
--
1.8.3.1
