kernel/arch/powerpc/platforms/pseries/msi.c, branch linux-rolling-stable

powerpc/pseries: Correct MSI allocation tracking

2026-03-19T15:15:27Z

commit 35e4f2a17eb40288f9bcdb09549fa04a63a96279 upstream. The per-device MSI allocation calculation in pseries_irq_domain_alloc() is clearly wrong. It can still happen to work when nr_irqs is 1. Correct it. Fixes: c0215e2d72de ("powerpc/pseries: Fix MSI-X allocation failure when quota is exceeded") Cc: stable@vger.kernel.org Signed-off-by: Nam Cao Reviewed-by: Mahesh Salgaonkar Reviewed-by: Nilay Shroff [maddy: Fixed Nilay's reviewed-by tag] Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/20260302003948.1452016-1-namcao@linutronix.de Signed-off-by: Greg Kroah-Hartman

powerpc/pseries: Fix MSI-X allocation failure when quota is exceeded

2026-03-04T12:21:06Z

[ Upstream commit c0215e2d72debcd9cbc1c002fb012d50a3140387 ] Nilay reported that since commit daaa574aba6f ("powerpc/pseries/msi: Switch to msi_create_parent_irq_domain()"), the NVMe driver cannot enable MSI-X when the device's MSI-X table size is larger than the firmware's MSI quota for the device. This is because the commit changes how rtas_prepare_msi_irqs() is called: - Before, it is called when interrupts are allocated at the global interrupt domain with nvec_in being the number of allocated interrupts. rtas_prepare_msi_irqs() can return a positive number and the allocation will be retried. - Now, it is called at the creation of per-device interrupt domain with nvec_in being the number of interrupts that the device supports. If rtas_prepare_msi_irqs() returns positive, domain creation just fails. For Nilay's NVMe driver case, rtas_prepare_msi_irqs() returns a positive number (the quota). This causes per-device interrupt domain creation to fail and thus the NVMe driver cannot enable MSI-X. Rework to make this scenario works again: - pseries_msi_ops_prepare() only prepares as many interrupts as the quota permit. - pseries_irq_domain_alloc() fails if the device's quota is exceeded. Now, if the quota is exceeded, pseries_msi_ops_prepare() will only prepare as allowed by the quota. If device drivers attempt to allocate more interrupts than the quota permits, pseries_irq_domain_alloc() will return an error code and msi_handle_pci_fail() will allow device drivers a retry. Reported-by: Nilay Shroff Closes: https://lore.kernel.org/linuxppc-dev/6af2c4c2-97f6-4758-be33-256638ef39e5@linux.ibm.com/ Fixes: daaa574aba6f ("powerpc/pseries/msi: Switch to msi_create_parent_irq_domain()") Signed-off-by: Nam Cao Cc: stable@vger.kernel.org Tested-by: Nilay Shroff Acked-by: Nilay Shroff Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/20260107100230.1466093-1-namcao@linutronix.de Signed-off-by: Sasha Levin

powerpc/pseries/msi: Fix NULL pointer dereference at irq domain teardown

2025-10-13T04:09:02Z

pseries_msi_ops_teardown() reads pci_dev* from msi_alloc_info_t. However, pseries_msi_ops_prepare() does not populate this structure, thus it is all zeros. Consequently, pseries_msi_ops_teardown() triggers a NULL pointer dereference crash. struct pci_dev is available in struct irq_domain. Read it there instead. Reported-by: Venkat Rao Bagalkote Closes: https://lore.kernel.org/linuxppc-dev/878d7651-433a-46fe-a28b-1b7e893fcbe0@linux.ibm.com/ Tested-by: Venkat Rao Bagalkote Signed-off-by: Nam Cao Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/20251010120307.3281720-1-namcao@linutronix.de

powerpc/pseries/msi: Switch to msi_create_parent_irq_domain()

2025-09-06T12:03:19Z

Move away from the legacy MSI domain setup, switch to use msi_create_parent_irq_domain(). Signed-off-by: Nam Cao Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/c7a6d8f27fd217021dea4daad777e81a525ae460.1754903590.git.namcao@linutronix.de

powerpc/xive: Untangle xive from child interrupt controller drivers

2025-09-06T12:03:18Z

xive-specific data is stored in handler_data. This creates a mess, as xive has to rely on child interrupt controller drivers to clean up this data, as was done by 9a014f45688 ("powerpc/pseries/pci: Add a msi_free() handler to clear XIVE data"). Instead, store xive-specific data in chip_data and untangle the child drivers. Signed-off-by: Nam Cao Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/83968073022a4cc211dcbd0faccd20ec05e58c3e.1754903590.git.namcao@linutronix.de

powerpc/pseries/msi: Fix potential underflow and leak issue

2025-09-01T08:09:39Z

pseries_irq_domain_alloc() allocates interrupts at parent's interrupt domain. If it fails in the progress, all allocated interrupts are freed. The number of successfully allocated interrupts so far is stored "i". However, "i - 1" interrupts are freed. This is broken: - One interrupt is not be freed - If "i" is zero, "i - 1" wraps around Correct the number of freed interrupts to 'i'. Fixes: a5f3d2c17b07 ("powerpc/pseries/pci: Add MSI domains") Signed-off-by: Nam Cao Cc: stable@vger.kernel.org Reviewed-by: Cédric Le Goater Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/a980067f2b256bf716b4cd713bc1095966eed8cd.1754300646.git.namcao@linutronix.de

Merge tag 'irq-cleanups-2025-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2025-05-27T15:07:32Z

Pull irq cleanups from Thomas Gleixner: "A set of cleanups for the generic interrupt subsystem: - Consolidate on one set of functions for the interrupt domain code to get rid of pointlessly duplicated code with only marginal different semantics. - Update the documentation accordingly and consolidate the coding style of the irqdomain header" * tag 'irq-cleanups-2025-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (46 commits) irqdomain: Consolidate coding style irqdomain: Fix kernel-doc and add it to Documentation Documentation: irqdomain: Update it Documentation: irq-domain.rst: Simple improvements Documentation: irq/concepts: Minor improvements Documentation: irq/concepts: Add commas and reflow irqdomain: Improve kernel-docs of functions irqdomain: Make struct irq_domain_info variables const irqdomain: Use irq_domain_instantiate()'s return value as initializers irqdomain: Drop irq_linear_revmap() pinctrl: keembay: Switch to irq_find_mapping() irqchip/armada-370-xp: Switch to irq_find_mapping() gpu: ipu-v3: Switch to irq_find_mapping() gpio: idt3243x: Switch to irq_find_mapping() sh: Switch to irq_find_mapping() powerpc: Switch to irq_find_mapping() irqdomain: Drop irq_domain_add_*() functions powerpc: Switch irq_domain_add_nomap() to use fwnode thermal: Switch to irq_domain_create_linear() soc: Switch to irq_domain_create_*() ...

powerpc: Switch to of_fwnode_handle()

2025-05-16T19:06:08Z

of_node_to_fwnode() is irqdomain's reimplementation of the "officially" defined of_fwnode_handle(). The former is in the process of being removed, so use the latter instead. [ tglx: Fix up subject prefix ] Signed-off-by: Jiri Slaby (SUSE) Signed-off-by: Thomas Gleixner Link: https://lore.kernel.org/all/20250319092951.37667-9-jirislaby@kernel.org

powerpc/pseries/msi: Avoid reading PCI device registers in reduced power states

2025-04-29T05:37:45Z

When a system is being suspended to RAM, the PCI devices are also suspended and the PPC code ends up calling pseries_msi_compose_msg() and this triggers the BUG_ON() in __pci_read_msi_msg() because the device at this point is in reduced power state. In reduced power state, the memory mapped registers of the PCI device are not accessible. To replicate the bug: 1. Make sure deep sleep is selected # cat /sys/power/mem_sleep s2idle [deep] 2. Make sure console is not suspended (so that dmesg logs are visible) echo N > /sys/module/printk/parameters/console_suspend 3. Suspend the system echo mem > /sys/power/state To fix this behaviour, read the cached msi message of the device when the device is not in PCI_D0 power state instead of touching the hardware. Fixes: a5f3d2c17b07 ("powerpc/pseries/pci: Add MSI domains") Cc: stable@vger.kernel.org # v5.15+ Signed-off-by: Gautam Menghani Tested-by: Venkat Rao Bagalkote Reviewed-by: Vaibhav Jain Signed-off-by: Madhavan Srinivasan Link: https://patch.msgid.link/20250305090237.294633-1-gautam@linux.ibm.com

irqdomain: Rename irq_get_default_host() to irq_get_default_domain()

2025-04-04T14:39:10Z

Naming interrupt domains host is confusing at best and the irqdomain code uses both domain and host inconsistently. Therefore rename irq_get_default_host() to irq_get_default_domain(). Signed-off-by: Jiri Slaby (SUSE) Signed-off-by: Thomas Gleixner Link: https://lore.kernel.org/all/20250319092951.37667-4-jirislaby@kernel.org