diff mbox series

[PULL,10/20] Revert "hw/isa/vt82c686: Remove intermediate IRQ forwarder"

Message ID 20230307234711.55375-11-philmd@linaro.org
State Accepted
Commit 38200011319b5819ff268dadb1b175faa6b0764a
Headers show
Series [PULL,01/20] docs/system: Remove "mips" board from target-mips.rst | expand

Commit Message

Philippe Mathieu-Daudé March 7, 2023, 11:47 p.m. UTC
From: BALATON Zoltan <balaton@eik.bme.hu>

To be 'usable', QDev objects (which are QOM objects) must be
1/ initialized (at this point their properties can be modified), then
2/ realized (properties are consumed).
Some devices (objects) might depend on other devices. When creating
the 'QOM composition tree', parent objects can't be 'realized' until
all their children are. We might also have circular dependencies.
A common circular dependency occurs with IRQs. Device (A) has an
output IRQ wired to device (B), and device (B) has one to device (A).
When (A) is realized and connects its IRQ to an unrealized (B), the
IRQ handler on (B) is not yet created. QEMU pass IRQ between objects
as pointer. When (A) poll (B)'s IRQ, it is NULL. Later (B) is realized
and its IRQ pointers are populated, but (A) keeps a reference to a
NULL pointer.
A common pattern to bypass this circular limitation is to use 'proxy'
objects. Proxy (P) is created (and realized) before (A) and (B). Then
(A) and (B) can be created in different order, it doesn't matter: (P)
pointers are already populated.

Commit bb98e0f59cde ("hw/isa/vt82c686: Remove intermediate IRQ
forwarder") neglected the QOM/QDev circular dependency issue, and
removed the 'proxy' between the southbridge, its PCI functions and the
interrupt controller, resulting in PCI functions wiring output IRQs to
'NULL', leading to guest failures (IRQ never delivered) [1] [2].

Since we are entering feature freeze, it is safer to revert the
offending patch until we figure a way to strengthen our APIs.

[1] https://lore.kernel.org/qemu-devel/928a8552-ab62-9e6c-a492-d6453e338b9d@redhat.com/
[2] https://lore.kernel.org/qemu-devel/cover.1677628524.git.balaton@eik.bme.hu/

Signed-off-by: BALATON Zoltan <balaton@eik.bme.hu>
Tested-by: Rene Engel <ReneEngel80@emailn.de>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <cdfb3c5a42e505450f6803124f27856434c5b298.1677628524.git.balaton@eik.bme.hu>
[PMD: Reworded description]
Inspired-by: Bernhard Beschow <shentey@gmail.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
 hw/isa/vt82c686.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
diff mbox series

Patch

diff --git a/hw/isa/vt82c686.c b/hw/isa/vt82c686.c
index f4c40965cd..01e0148967 100644
--- a/hw/isa/vt82c686.c
+++ b/hw/isa/vt82c686.c
@@ -598,15 +598,23 @@  void via_isa_set_irq(PCIDevice *d, int n, int level)
     qemu_set_irq(s->isa_irqs_in[n], level);
 }
 
+static void via_isa_request_i8259_irq(void *opaque, int irq, int level)
+{
+    ViaISAState *s = opaque;
+    qemu_set_irq(s->cpu_intr, level);
+}
+
 static void via_isa_realize(PCIDevice *d, Error **errp)
 {
     ViaISAState *s = VIA_ISA(d);
     DeviceState *dev = DEVICE(d);
     PCIBus *pci_bus = pci_get_bus(d);
+    qemu_irq *isa_irq;
     ISABus *isa_bus;
     int i;
 
     qdev_init_gpio_out(dev, &s->cpu_intr, 1);
+    isa_irq = qemu_allocate_irqs(via_isa_request_i8259_irq, s, 1);
     isa_bus = isa_bus_new(dev, pci_address_space(d), pci_address_space_io(d),
                           errp);
 
@@ -614,7 +622,7 @@  static void via_isa_realize(PCIDevice *d, Error **errp)
         return;
     }
 
-    s->isa_irqs_in = i8259_init(isa_bus, s->cpu_intr);
+    s->isa_irqs_in = i8259_init(isa_bus, *isa_irq);
     isa_bus_register_input_irqs(isa_bus, s->isa_irqs_in);
     i8254_pit_init(isa_bus, 0x40, 0, NULL);
     i8257_dma_init(isa_bus, 0);