Received-SPF: pass (google.com: domain of linux-kernel+bounces-71847-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
From: Oliver Upton <oliver.upton@linux.dev>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: Marc Zyngier <maz@kernel.org>,
	Zenghui Yu <yuzenghui@huawei.com>,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org,
	kvmarm@lists.linux.dev,
	Jing Zhang <jingzhangos@google.com>,
	Oliver Upton <oliver.upton@linux.dev>,
	stable@vger.kernel.org,
	George Cherian <gcherian@marvell.com>
Subject: [PATCH 1/3] irqchip/gic-v3-its: Do not assume vPE tables are preallocated
Date: Mon, 19 Feb 2024 18:58:06 +0000
Message-ID: <20240219185809.286724-2-oliver.upton@linux.dev>
In-Reply-To: <20240219185809.286724-1-oliver.upton@linux.dev>
References: <20240219185809.286724-1-oliver.upton@linux.dev>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

We go through quite a bit of trouble to make sure we pick up any
preallocated LPI tables on the redistributors, as enabling LPIs is a
one-way switch. There is no such restriction for vLPIs, and for GICv4.1
we expect to allocate a new vPE table at boot.

This works as intended when initializing an ITS, however when setting up
a redistributor in its_cpu_init_lpis() the early return for preallocated
RD tables skips straight past GICv4 setup. This all comes to a head when
trying to kexec into a new kernel, as the new kernel silently fails to
set up GICv4, leading to a complete loss of SGIs and LPIs for KVM VMs
(ouch).

Slap a band-aid on the problem by ensuring its_cpu_init_lpis() always
initializes GICv4 on the way out, even if the other RD tables were
preallocated.

Cc: stable@vger.kernel.org
Fixes: 6479450f72c1 ("irqchip/gic-v4: Fix occasional VLPI drop")
Reported-by: George Cherian <gcherian@marvell.com>
Co-developed-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
---

I debated a bit on the fixes tag between the blamed commit and commit
5e5168461c22 ("irqchip/gic-v4.1: VPE table (aka GICR_VPROPBASER)
allocation"), although it would appear GICv4 could be left in an unknown
state after kexec too.

 drivers/irqchip/irq-gic-v3-its.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index d097001c1e3e..0022852ce494 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -3172,6 +3172,7 @@ static void its_cpu_init_lpis(void)
 	val |= GICR_CTLR_ENABLE_LPIS;
 	writel_relaxed(val, rbase + GICR_CTLR);
 
+out:
 	if (gic_rdists->has_vlpis && !gic_rdists->has_rvpeid) {
 		void __iomem *vlpi_base = gic_data_rdist_vlpi_base();
 
@@ -3207,7 +3208,6 @@ static void its_cpu_init_lpis(void)
 
 	/* Make sure the GIC has seen the above */
 	dsb(sy);
-out:
 	gic_data_rdist()->flags |= RD_LOCAL_LPI_ENABLED;
 	pr_info("GICv3: CPU%d: using %s LPI pending table @%pa\n",
 		smp_processor_id(),
-- 
2.44.0.rc0.258.g7320e95886-goog