[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87tt6310hu.wl-maz@kernel.org>
Date: Fri, 02 May 2025 16:43:57 +0100
From: Marc Zyngier <maz@...nel.org>
To: Lorenzo Pieralisi <lpieralisi@...nel.org>
Cc: Thomas Gleixner <tglx@...utronix.de>,
Rob Herring <robh@...nel.org>,
Krzysztof Kozlowski <krzk+dt@...nel.org>,
Conor Dooley <conor+dt@...nel.org>,
Catalin Marinas <catalin.marinas@....com>,
Will Deacon <will@...nel.org>,
Arnd Bergmann <arnd@...db.de>,
Sascha Bischoff <sascha.bischoff@....com>,
Timothy Hayes <timothy.hayes@....com>,
"Liam R. Howlett" <Liam.Howlett@...cle.com>,
Mark Rutland <mark.rutland@....com>,
linux-arm-kernel@...ts.infradead.org,
linux-kernel@...r.kernel.org,
devicetree@...r.kernel.org
Subject: Re: [PATCH v2 21/22] irqchip/gic-v5: Add GICv5 IWB support
On Fri, 02 May 2025 08:59:42 +0100,
Lorenzo Pieralisi <lpieralisi@...nel.org> wrote:
>
> It looks like the msi_prepare() ITS callback (ie where the its_device is
> allocated) is called everytime an endpoint device driver requests a
> wired IRQ through:
>
> gicv5_its_msi_prepare+0x68c/0x6f8
> its_pmsi_prepare+0x16c/0x1b8
> __msi_domain_alloc_irqs+0x70/0x448
> __msi_domain_alloc_irq_at+0xf8/0x194
> msi_device_domain_alloc_wired+0x88/0x10c
> irq_create_fwspec_mapping+0x3a0/0x4c0
> irq_create_of_mapping+0xc0/0xe8
> of_irq_get+0xa0/0xe4
> platform_get_irq_optional+0x54/0x1c4
> platform_get_irq+0x1c/0x50
>
> so it becomes "shared" if multiple IWB wires are requested by endpoint
> drivers.
Right, I've reproduced on D05 with MBIGEN:
[ 5.505530] Reusing ITT for devID 40000
[ 5.505532] CPU: 36 UID: 0 PID: 557 Comm: (udev-worker) Not tainted 6.15.0-rc4-00079-geef147df4841-dirty #4403 PREEMPT
[ 5.505535] Hardware name: Huawei Taishan 2280 /D05, BIOS Hisilicon D05 IT21 Nemo 2.0 RC0 04/18/2018
[ 5.505536] Call trace:
[ 5.505537] show_stack+0x20/0x38 (C)
[ 5.505540] dump_stack_lvl+0x80/0xf8
[ 5.505543] dump_stack+0x18/0x28
[ 5.505546] its_msi_prepare+0xe4/0x1d0
[ 5.505549] its_pmsi_prepare+0x15c/0x1d0
[ 5.505552] __msi_domain_alloc_irqs+0x80/0x398
[ 5.505556] __msi_domain_alloc_irq_at+0x100/0x168
[ 5.505560] msi_device_domain_alloc_wired+0x9c/0x128
[ 5.505564] irq_create_fwspec_mapping+0x180/0x388
[ 5.505567] acpi_irq_get+0xac/0xe8
[ 5.505570] platform_get_irq_optional+0x1e8/0x208
[ 5.505574] devm_platform_get_irqs_affinity+0x58/0x298
[ 5.505578] hisi_sas_v2_interrupt_preinit+0x60/0xb0 [hisi_sas_v2_hw]
[ 5.505582] hisi_sas_probe+0x164/0x278 [hisi_sas_main]
[ 5.505588] hisi_sas_v2_probe+0x20/0x38 [hisi_sas_v2_hw]
[ 5.505591] platform_probe+0x70/0xd0
[ 5.505595] really_probe+0xc8/0x3a0
[ 5.505598] __driver_probe_device+0x84/0x170
[ 5.505600] driver_probe_device+0x44/0x120
[ 5.505603] __driver_attach+0xfc/0x210
[ 5.505606] bus_for_each_dev+0x7c/0xe8
[ 5.505608] driver_attach+0x2c/0x40
[ 5.505611] bus_add_driver+0x118/0x248
[ 5.505613] driver_register+0x68/0x138
[ 5.505616] __platform_driver_register+0x2c/0x40
[ 5.505619] hisi_sas_v2_driver_init+0x28/0xff8 [hisi_sas_v2_hw]
[ 5.505623] do_one_initcall+0x4c/0x2c0
[ 5.505626] do_init_module+0x60/0x230
[ 5.505629] load_module+0xa64/0xb30
[ 5.505631] init_module_from_file+0x8c/0xd8
[ 5.505634] idempotent_init_module+0x1c4/0x2b8
[ 5.505637] __arm64_sys_finit_module+0x74/0xe8
[ 5.505640] invoke_syscall+0x50/0x120
[ 5.505642] el0_svc_common.constprop.0+0x48/0xf0
[ 5.505644] do_el0_svc+0x24/0x38
[ 5.505646] el0_svc+0x34/0xf0
[ 5.505650] el0t_64_sync_handler+0x10c/0x138
[ 5.505654] el0t_64_sync+0x1ac/0x1b0
[ 5.505681] ID:78 pID:8382 vID:143
And that a few dozen times.
I'll have a think at how to unfsck this. This was previously avoided
by (IIRC) populating the domain upfront and letting the domain
matching code do its job. That behaviour seems to have been lost. On
the other hand, as long as you don't expect the ITT to *grow*, nothing
horrible should happen.
But I also get an interesting crash in msi_domain_debig_show(), so
there is more than just this corner case that is screwed.
M.
--
Jazz isn't dead. It just smells funny.
Powered by blists - more mailing lists