Fix x2apic enablement and allow up to 32768 CPUs without IR where supported
Subsystems (22)
David Woodhouse
maintainercover Fix x2apic enablement and allow up to 32768 CPUs without IR where supported
Splitting out the simpler parts of my previous patch set. The full support for per-irqdomain affinity limits will take a bit more work but this part is quite simple.
Since we don't yet have per-irqdomain affinity, we currently attempt to avoid bringing CPUs online at all if they can't be targeted by external interrupts. Except we still let them get hotplugged later... which is moderately suboptimal.
Fix that, and support the hypervisor enlightenment which at least extends the range of targetable APIC IDs to 15 bits, as seen in the patch at https://patchwork.kernel.org/patch/11816693/ for qemu.
David Woodhouse (5):
x86/apic: Fix x2apic enablement without interrupt remapping x86/msi: Only use high bits of MSI address for DMAR unit x86/ioapic: Handle Extended Destination ID field in RTE x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Documentation/virt/kvm/cpuid.rst | 4 ++++ arch/x86/include/asm/apic.h | 1 + arch/x86/include/asm/io_apic.h | 3 ++- arch/x86/include/asm/mpspec.h | 1 + arch/x86/include/asm/x86_init.h | 2 ++ arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/apic/apic.c | 27 +++++++++++++++++++++------ arch/x86/kernel/apic/io_apic.c | 19 +++++++++++++------ arch/x86/kernel/apic/msi.c | 41 +++++++++++++++++++++++++++++++++++------ arch/x86/kernel/apic/x2apic_phys.c | 9 +++++++++ arch/x86/kernel/kvm.c | 6 ++++++ arch/x86/kernel/x86_init.c | 1 + 12 files changed, 96 insertions(+), 19 deletions(-)
David Woodhouse
maintainerPATCH 1/5 x86/apic: Fix x2apic enablement without interrupt remapping
From: David Woodhouse <[email protected]>
Currently, Linux as a hypervisor guest will enable x2apic only if there are no CPUs present at boot time with an APIC ID above 255.
Hotplugging a CPU later with a higher APIC ID would result in a CPU which cannot be targeted by external interrupts.
Add a filter in x2apic_apic_id_valid() which can be used to prevent such CPUs from coming online, and allow x2apic to be enabled even if they are present at boot time.
--- arch/x86/include/asm/apic.h | 1 + arch/x86/kernel/apic/apic.c | 14 ++++++++------ arch/x86/kernel/apic/x2apic_phys.c | 9 +++++++++ 3 files changed, 18 insertions(+), 6 deletions(-)
arch/x86/include/asm/apic.h +1 -0
@@ -259,6 +259,7 @@ static inline u64 native_x2apic_icr_read(void)extern int x2apic_mode;extern int x2apic_phys; +extern void __init x2apic_set_max_apicid (u32 apicid);extern void __init check_x2apic (void );extern void x2apic_setup (void );static inline int x2apic_enabled (void )
arch/x86/kernel/apic/apic.c +8 -6
@@ -1841,20 +1841,22 @@ static __init void try_to_enable_x2apic(int remap_mode)return ;if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) { -/* IR is required if there is APIC ID > 255 even when running -* under KVM +/* +* Using X2APIC without IR is not architecturally supported +* on bare metal but may be supported in guests. */ -if (max_physical_apicid >255 || - !x86_init.hyper.x2apic_available ()) { +if (!x86_init.hyper.x2apic_available ()) {pr_info ("x2apic: IRQ remapping doesn't support X2APIC mode\n" );x2apic_disable ();return ; }/* -* without IR all CPUs can be addressed by IOAPIC/MSI -* only in physical mode +* Without IR, all CPUs can be addressed by IOAPIC/MSI only +* in physical mode, and CPUs with an APIC ID that cannnot +* be addressed must not be brought online. */ +x2apic_set_max_apicid (255 ); x2apic_phys =1 ; }x2apic_enable ();
arch/x86/kernel/apic/x2apic_phys.c +9 -0
@@ -8,6 +8,12 @@int x2apic_phys;static struct apic apic_x2apic_phys; +static u32 x2apic_max_apicid; + +void __init x2apic_set_max_apicid (u32 apicid) +{ + x2apic_max_apicid = apicid; +}static int __init set_x2apic_phys_mode (char *arg) {
@@ -98,6 +104,9 @@ static int x2apic_phys_probe(void)/* Common x2apic functions, also used by x2apic_cluster */ int x2apic_apic_id_valid (u32 apicid) { +if (x2apic_max_apicid && apicid > x2apic_max_apicid) +return 0 ; +return 1 ; }
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 2/5 x86/msi: Only use high bits of MSI address for DMAR unit
From: David Woodhouse <[email protected]>
The Intel IOMMU has an MSI-like configuration for its interrupt, but it isn't really MSI. So it gets to abuse the high 32 bits of the address, and puts the high 24 bits of the extended APIC ID there.
This isn't something that can be used in the general case for real MSIs, since external devices using the high bits of the address would be performing writes to actual memory space above 4GiB, not targeted at the APIC.
Factor the hack out and allow it only to be used when appropriate, adding a WARN_ON_ONCE() if other MSIs are targeted at an unreachable APIC ID. In *theory* that should never happen since the compatibility MSI messages are not supposed to be used with Interrupt Remapping enabled. In practice, if IR is enabled but some devices aren't within scope of any given remapping unit, it might happen. But that's a longer story and this warning is the right thing to do in that case for the short term.
The x2apic_enabled() check isn't needed because Linux won't bring up CPUs with higher APIC IDs unless x2apic is enabled anyway.
--- arch/x86/kernel/apic/msi.c | 33 +++++++++++++++++++++++++++------ 1 file changed, 27 insertions(+), 6 deletions(-)
arch/x86/kernel/apic/msi.c +27 -6
@@ -23,13 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg) +static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) { msg->address_hi =MSI_ADDR_BASE_HI ; -if (x2apic_enabled ()) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); - msg->address_lo = MSI_ADDR_BASE_LO | ((apic->irq_dest_mode == 0) ?
@@ -43,18 +41,40 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)MSI_DATA_LEVEL_ASSERT |MSI_DATA_DELIVERY_FIXED |MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) { -__irq_msi_compose_msg (irqd_cfg (data), msg); +__irq_msi_compose_msg (irqd_cfg (data), msg, false); +} + +/* +* The Intel IOMMU (ab)uses the high bits of the MSI address to contain the +* high bits of the destination APIC ID. This can't be done in the general +* case for MSIs as it would be targeting real memory above 4GiB not the +* APIC. +*/ +static void dmar_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, true); }static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg); +__irq_msi_compose_msg (cfg, msg,0 );irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); }
@@ -288,6 +308,7 @@ static struct irq_chip dmar_msi_controller = { .irq_ack = irq_chip_ack_parent, .irq_set_affinity = msi_domain_set_affinity, .irq_retrigger = irq_chip_retrigger_hierarchy, + .irq_compose_msi_msg = dmar_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, .flags = IRQCHIP_SKIP_SET_WAKE, };
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 3/5 x86/ioapic: Handle Extended Destination ID field in RTE
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
The IOAPIC Redirection Table Entries contain an 8-bit Extended Destination ID field which maps to bits 11-4 of the MSI address.
The lowest bit is used to indicate remappable format, when interrupt remapping is in use. A hypervisor can use the other 7 bits to permit guests to address up to 15 bits of APIC IDs, thus allowing 32768 vCPUs before having to expose a vIOMMU and interrupt remapping to the guest.
No behavioural change in this patch, since nothing yet permits APIC IDs above 255 to be used with the non-IR IOAPIC domain. Except for the case where IR is enabled but there are IOAPICs which aren't in the scope of any IOMMU, which is totally hosed anyway and needs fixing independently of this change.
--- arch/x86/include/asm/io_apic.h | 3 ++- arch/x86/kernel/apic/io_apic.c | 19 +++++++++++++------ 2 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/io_apic.h +2 -1
@@ -78,7 +78,8 @@ struct IO_APIC_route_entry { mask :1 ,/* 0: enabled, 1: disabled */ __reserved_2 : 15; - __u32 __reserved_3 : 24, + __u32 __reserved_3 : 17, + ext_dest : 7, dest : 8; } __attribute__ ((packed));
arch/x86/kernel/apic/io_apic.c +13 -6
@@ -1239,10 +1239,10 @@ static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries) buf, (ir_entry->index2 <<15 ) | ir_entry->index, ir_entry->zero); else - printk(KERN_DEBUG "%s, %s, D(%02X), M(%1d)\n", + printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf, entry.dest_mode ==IOAPIC_DEST_MODE_LOGICAL ? -"logical " :"physical" , +"logical " :"physical" , entry.ext_dest, entry.dest, entry.delivery_mode); } }
@@ -1410,6 +1410,7 @@ void native_restore_boot_irq_mode(void)*/ if (ioapic_i8259.pin !=-1 ) {struct IO_APIC_route_entry entry; +u32 apic_id =read_apic_id ();memset (&entry,0 ,sizeof (entry)); entry.mask =IOAPIC_UNMASKED ;
@@ -1417,7 +1418,8 @@ void native_restore_boot_irq_mode(void) entry.polarity =IOAPIC_POL_HIGH ; entry.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry.delivery_mode = dest_ExtINT; - entry.dest =read_apic_id (); + entry.dest = apic_id &0xff ; + entry.ext_dest = apic_id >>8 ;/* * Add it to the IO-APIC irq-routing table:
@@ -1861,7 +1863,8 @@ static void ioapic_configure_entry(struct irq_data *irqd)* ioapic chip to verify that. */ if (irqd->chip == &ioapic_chip) { - mpd->entry.dest = cfg->dest_apicid; + mpd->entry.dest = cfg->dest_apicid &0xff ; + mpd->entry.ext_dest = cfg->dest_apicid >>8 ; mpd->entry.vector = cfg->vector; }for_each_irq_pin (entry, mpd->irq_2_pin)
@@ -2027,6 +2030,7 @@ static inline void __init unlock_ExtINT_logic(void)int apic, pin, i;struct IO_APIC_route_entry entry0, entry1;unsigned char save_control, save_freq_select; +u32 apic_id; pin =find_isa_irq_pin (8 , mp_INT);if (pin ==-1 ) {
@@ -2042,11 +2046,13 @@ static inline void __init unlock_ExtINT_logic(void) entry0 =ioapic_read_entry (apic, pin);clear_IO_APIC_pin (apic, pin); + apic_id =hard_smp_processor_id ();memset (&entry1,0 ,sizeof (entry1)); entry1.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry1.mask =IOAPIC_UNMASKED ; - entry1.dest =hard_smp_processor_id (); + entry1.dest = apic_id &0xff ; + entry1.ext_dest = apic_id >>8 ; entry1.delivery_mode = dest_ExtINT; entry1.polarity = entry0.polarity; entry1.trigger =IOAPIC_EDGE ;
@@ -2949,7 +2955,8 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data,memset (entry,0 ,sizeof (*entry)); entry->delivery_mode = apic->irq_delivery_mode; entry->dest_mode = apic->irq_dest_mode; - entry->dest = cfg->dest_apicid; + entry->dest = cfg->dest_apicid &0xff ; + entry->ext_dest = cfg->dest_apicid >>8 ; entry->vector = cfg->vector; entry->trigger = data->trigger; entry->polarity = data->polarity;
Signature · 1 line
2.26.2
Re: [PATCH 3/5] x86/ioapic: Handle Extended Destination ID field in RTE
arch/x86/kernel/apic/io_apic.c static void ioapic_configure_entry(struct irq_data *irqd) · line 1870
Quoted 10 lines · David Woodhouse
arch/x86/kernel/apic/io_apic.c +2
-1@@ -1861,7 +1863,8 @@ static void ioapic_configure_entry(struct irq_data *irqd)* ioapic chip to verify that. */ if (irqd->chip == &ioapic_chip) { - mpd->entry.dest = cfg->dest_apicid; + mpd->entry.dest = cfg->dest_apicid &0xff ; + mpd->entry.ext_dest = cfg->dest_apicid >>8 ; mpd->entry.vector = cfg->vector; }for_each_irq_pin (entry, mpd->irq_2_pin)
All the other sites did memset(0) before the assignment, and this the extra unconditional write of 0 to ext_dest is harmless.
This might be true for this site too, but it wasn't immediately obvious.
David Woodhouse
maintainerRe: [PATCH 3/5] x86/ioapic: Handle Extended Destination ID field in RTE
Quoted 15 lines · Peter Zijlstra
arch/x86/kernel/apic/io_apic.c static void ioapic_configure_entry(struct irq_data *irqd) · line 1870
Quoted 10 lines · David Woodhouse
arch/x86/kernel/apic/io_apic.c +2
-1@@ -1861,7 +1863,8 @@ static void ioapic_configure_entry(struct irq_data *irqd)* ioapic chip to verify that. */ if (irqd->chip == &ioapic_chip) { - mpd->entry.dest = cfg->dest_apicid; + mpd->entry.dest = cfg->dest_apicid &0xff ; + mpd->entry.ext_dest = cfg->dest_apicid >>8 ; mpd->entry.vector = cfg->vector; }for_each_irq_pin (entry, mpd->irq_2_pin)All the other sites did memset(0) before the assignment, and this the extra unconditional write of 0 to
ext_destis harmless.
Quoted 1 line
This might be true for this site too, but it wasn't immediately obvious.
Yeah. Really, that whole 16-bit field ought to be called 'msi_addr_bits_19_to_4' since there's no interpretation by the IOAPIC and it's *just* passed on in the MSI cycle. See my later patch which stops making them up for ourselves in the IOAPIC code and *just* swizzles them out of the MSI message which is composed for us by upstream.
In this case, since this piece of code is based on the knowledge that the upstream controller is accepting MSI in the x86 Compatibility Format, those seven bits are never going to have been used for anything else.
Re: [PATCH 3/5] x86/ioapic: Handle Extended Destination ID field in RTE
In reply to David WoodhouseQuoted 10 lines · David Woodhouse
From: David Woodhouse <[email protected]>
The IOAPIC Redirection Table Entries contain an 8-bit Extended Destination ID field which maps to bits 11-4 of the MSI address.
The lowest bit is used to indicate remappable format, when interrupt remapping is in use. A hypervisor can use the other 7 bits to permit guests to address up to 15 bits of APIC IDs, thus allowing 32768 vCPUs before having to expose a vIOMMU and interrupt remapping to the guest.
Quoted 5 lines
No behavioural change in this patch, since nothing yet permits APIC IDs above 255 to be used with the non-IR IOAPIC domain. Except for the case where IR is enabled but there are IOAPICs which aren't in the scope of any IOMMU, which is totally hosed anyway and needs fixing independently of this change.
Again: IOAPICs which are not covered by IR are detected and prevent IR enablement which makes the above a fairy tale. Changelogs are about facts.
arch/x86/include/asm/io_apic.h struct IO_APIC_route_entry { · line 82
Quoted 11 lines
arch/x86/include/asm/io_apic.h +2
-1@@ -78,7 +78,8 @@ struct IO_APIC_route_entry { mask :1 ,/* 0: enabled, 1: disabled */ __reserved_2 : 15; - __u32 __reserved_3 : 24, + __u32 __reserved_3 : 17, + ext_dest : 7,
This wants to be explicitely named 'virt_ext_dest'.
Thanks,
tglx
David Woodhouse
maintainerPATCH 4/5 x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
Some hypervisors can allow the guest to use the Extended Destination ID field in the IOAPIC RTE and MSI address to address up to 32768 CPUs.
--- arch/x86/include/asm/mpspec.h | 1 + arch/x86/include/asm/x86_init.h | 2 ++ arch/x86/kernel/apic/apic.c | 15 ++++++++++++++- arch/x86/kernel/apic/msi.c | 10 +++++++++- arch/x86/kernel/x86_init.c | 1 + 5 files changed, 27 insertions(+), 2 deletions(-)
arch/x86/include/asm/mpspec.h +1 -0
@@ -42,6 +42,7 @@ extern DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);extern unsigned int boot_cpu_physical_apicid;extern u8 boot_cpu_apic_version;extern unsigned long mp_lapic_addr; +extern int msi_ext_dest_id;#ifdef CONFIG_X86_LOCAL_APIC extern int smp_found_config;
arch/x86/include/asm/x86_init.h +2 -0
@@ -114,6 +114,7 @@ struct x86_init_pci {* @init_platform: platform setup * @guest_late_init: guest late init * @x2apic_available: X2APIC detection +* @msi_ext_dest_id: MSI and IOAPIC support 15-bit APIC IDs * @init_mem_mapping: setup early mappings during init_mem_mapping() * @init_after_bootmem: guest init after boot allocator is finished */
@@ -121,6 +122,7 @@ struct x86_hyper_init {void (*init_platform)(void );void (*guest_late_init)(void );bool (*x2apic_available)(void ); +bool (*msi_ext_dest_id)(void );void (*init_mem_mapping)(void );void (*init_after_bootmem)(void ); };
arch/x86/kernel/apic/apic.c +14 -1
@@ -1837,9 +1837,21 @@ static __init void x2apic_enable(void)static __init void try_to_enable_x2apic (int remap_mode) { +u32 apic_limit =255 ; +if (x2apic_state ==X2APIC_DISABLED )return ; +/* +* If the hypervisor supports extended destination ID in IOAPIC +* and MSI, that increases the maximum APIC ID that can be used +* for non-remapped IRQ domains. +*/ +if (x86_init.hyper.msi_ext_dest_id ()) { + msi_ext_dest_id =1 ; + apic_limit =32767 ; + } +if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) {/* * Using X2APIC without IR is not architecturally supported
@@ -1856,9 +1868,10 @@ static __init void try_to_enable_x2apic(int remap_mode)* in physical mode, and CPUs with an APIC ID that cannnot * be addressed must not be brought online. */ -x2apic_set_max_apicid (255 ); +x2apic_set_max_apicid (apic_limit); x2apic_phys =1 ; } +x2apic_enable (); }
arch/x86/kernel/apic/msi.c +9 -1
@@ -23,8 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; +int msi_ext_dest_id__ro_after_init ; +static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg,bool dmar) + { msg->address_hi =MSI_ADDR_BASE_HI ;
@@ -46,10 +49,15 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,* Only the IOMMU itself can use the trick of putting destination * APIC ID into the high bits of the address. Anything else would * just be writing to memory if it tried that, and needs IR to -* address higher APIC IDs. +* address APICs which can't be addressed in the normal 32-bit +* address range at 0xFFExxxxx. That is typically just 8 bits, but +* some hypervisors allow the extended destination ID field in bits +* 11-5 to be used, giving support for 15 bits of APIC IDs in total. */ if (dmar) msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else if (msi_ext_dest_id && cfg->dest_apicid <0x8000 ) + msg->address_lo |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid) >>3 ;else WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }
arch/x86/kernel/x86_init.c +1 -0
@@ -110,6 +110,7 @@ struct x86_init_ops x86_init __initdata = { .init_platform = x86_init_noop, .guest_late_init = x86_init_noop, .x2apic_available = bool_x86_init_noop, + .msi_ext_dest_id = bool_x86_init_noop, .init_mem_mapping = x86_init_noop, .init_after_bootmem = x86_init_noop, },
Signature · 1 line
2.26.2
Re: [PATCH 4/5] x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available
arch/x86/kernel/apic/apic.c static __init void x2apic_enable(void) · line 1853
Quoted 10 lines · David Woodhouse
arch/x86/kernel/apic/apic.c +9
-0+/* +* If the hypervisor supports extended destination ID in IOAPIC +* and MSI, that increases the maximum APIC ID that can be used +* for non-remapped IRQ domains. +*/ +if (x86_init.hyper.msi_ext_dest_id ()) { + msi_ext_dest_id =1 ; + apic_limit =32767 ; + }
This needs to be outside of the remap mode check because?
Quoted 21 lines
arch/x86/kernel/apic/apic.c +3
-1+if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) {/* * Using X2APIC without IR is not architecturally supported @@ -1856,9 +1868,10 @@ static __init void try_to_enable_x2apic(int remap_mode)* in physical mode, and CPUs with an APIC ID that cannnot * be addressed must not be brought online. */ -x2apic_set_max_apicid (255 ); +x2apic_set_max_apicid (apic_limit); x2apic_phys =1 ; } +x2apic_enable (); }arch/x86/kernel/apic/msi.c +0
-0
arch/x86/kernel/apic/msi.c line 26
Quoted 4 lines
arch/x86/kernel/apic/msi.c +1
-0@@ -23,8 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; +int msi_ext_dest_id__ro_after_init ;
bool please.
Aside of that this breaks the build for a kernel with CONFIG_PCI_MSI=n
Thanks,
tglx
Re: [PATCH 4/5] x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available
Quoted 7 lines · Thomas Gleixner
Quoted 6 lines · David Woodhouse
arch/x86/kernel/apic/msi.c +0
-0@@ -23,8 +23,11 @@
Quoted 7 lines
arch/x86/kernel/apic/msi.c line 26
Quoted 3 lines
arch/x86/kernel/apic/msi.c +1
-0struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; +int msi_ext_dest_id__ro_after_init ;bool please.
Aside of that this breaks the build for a kernel with
CONFIG_PCI_MSI=n
So this wants to be
bool virt_ext_dest_id __ro_after_init;
in apic.c and then please make the IO/APIC places depend on this as well so any change to the utilization of the reserved IO/APIC bits in the future is not going to end up in surprises.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 4/5] x86/apic: Support 15 bits of APIC ID in IOAPIC/MSI where available
In reply to Thomas GleixnerQuoted 13 lines · Thomas Gleixner
arch/x86/kernel/apic/apic.c static __init void x2apic_enable(void) · line 1853
Quoted 10 lines · David Woodhouse
arch/x86/kernel/apic/apic.c +9
-0+/* +* If the hypervisor supports extended destination ID in IOAPIC +* and MSI, that increases the maximum APIC ID that can be used +* for non-remapped IRQ domains. +*/ +if (x86_init.hyper.msi_ext_dest_id ()) { + msi_ext_dest_id =1 ; + apic_limit =32767 ; + }This needs to be outside of the remap mode check because?
Once upon a time, there was a later patch in the series which *also* used the apic_limit variable to generate a maximum affinity mask.
Now we've ditched that idea, I can put this back inside the remap mode check.
Quoted 23 lines
Quoted 23 lines
arch/x86/kernel/apic/apic.c +3
-1+if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) {/* * Using X2APIC without IR is not architecturally supported @@ -1856,9 +1868,10 @@ static __init void try_to_enable_x2apic(int remap_mode)* in physical mode, and CPUs with an APIC ID that cannnot * be addressed must not be brought online. */ -x2apic_set_max_apicid (255 ); +x2apic_set_max_apicid (apic_limit); x2apic_phys =1 ; } +x2apic_enable (); }arch/x86/kernel/apic/msi.c +0
-0@@ -23,8 +23,11 @@
Quoted 7 lines
arch/x86/kernel/apic/msi.c line 26
Quoted 3 lines
arch/x86/kernel/apic/msi.c +1
-0struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; +int msi_ext_dest_id__ro_after_init ;bool please.
Aside of that this breaks the build for a kernel with
CONFIG_PCI_MSI=n
Will fix (and rename).
David Woodhouse
maintainerPATCH 5/5 x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This allows the host to indicate that IOAPIC and MSI emulation supports 15-bit destination IDs, allowing up to 32768 CPUs without interrupt remapping.
cf. https://patchwork.kernel.org/patch/11816693/ for qemu
--- Documentation/virt/kvm/cpuid.rst | 4 ++++ arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/kvm.c | 6 ++++++ 3 files changed, 11 insertions(+)
Documentation/virt/kvm/cpuid.rst +4 -0
@@ -92,6 +92,10 @@ KVM_FEATURE_ASYNC_PF_INT 14 guest checks this feature bit async pf acknowledgment msr 0x4b564d07. +KVM_FEATURE_MSI_EXT_DEST_ID 15 guest checks this feature bit + before using extended destination + ID bits in MSI address bits 11-5. + KVM_FEATURE_CLOCSOURCE_STABLE_BIT 24 host will warn if no guest-side per-cpu warps are expeced in kvmclock
arch/x86/include/uapi/asm/kvm_para.h +1 -0
@@ -32,6 +32,7 @@#define KVM_FEATURE_POLL_CONTROL 12#define KVM_FEATURE_PV_SCHED_YIELD 13#define KVM_FEATURE_ASYNC_PF_INT 14 +#define KVM_FEATURE_MSI_EXT_DEST_ID 15#define KVM_HINTS_REALTIME 0
arch/x86/kernel/kvm.c +6 -0
@@ -743,12 +743,18 @@ static void __init kvm_init_platform(void) x86_platform.apic_post_init = kvm_apic_init; } +static bool __init kvm_msi_ext_dest_id (void ) +{ +return kvm_para_has_feature (KVM_FEATURE_MSI_EXT_DEST_ID ); +} +const __initconst struct hypervisor_x86 x86_hyper_kvm = { .name ="KVM" , .detect = kvm_detect, .type =X86_HYPER_KVM , .init.guest_late_init = kvm_guest_init, .init.x2apic_available = kvm_para_available, + .init.msi_ext_dest_id = kvm_msi_ext_dest_id, .init.init_platform = kvm_init_platform, };
Signature · 1 line
2.26.2
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 27 lines · David Woodhouse
From: David Woodhouse <[email protected]>
This allows the host to indicate that IOAPIC and MSI emulation supports 15-bit destination IDs, allowing up to 32768 CPUs without interrupt remapping.
cf. https://patchwork.kernel.org/patch/11816693/ for qemu
Signed-off-by David Woodhouse <[email protected]>Acked-by Paolo Bonzini <[email protected]>--- Documentation/virt/kvm/cpuid.rst | 4 ++++ arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/kvm.c | 6 ++++++ 3 files changed, 11 insertions(+)Documentation/virt/kvm/cpuid.rst +3
-0@@ -92,6 +92,10 @@ KVM_FEATURE_ASYNC_PF_INT 14 guest checks this feature bit async pf acknowledgment msr 0x4b564d07. +KVM_FEATURE_MSI_EXT_DEST_ID 15 guest checks this feature bit + before using extended destination + ID bits in MSI address
Quoted 1 line
bits 11-5.
Why MSI_EXT_DEST_ID? It's enabling that for MSI and IO/APIC. The underlying mechanism might be the same, but APIC_EXT_DEST_ID is more general and then you might also make the explanation of that bit match the changelog.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 4 lines · Thomas Gleixner
Why
MSI_EXT_DEST_ID? It's enabling that for MSI and IO/APIC. The underlying mechanism might be the same, butAPIC_EXT_DEST_IDis more general and then you might also make the explanation of that bit match the changelog.
It's enabling it for *everything* that generates MSI cycles — which includes IOAPIC, HPET, and MSI-capable PCI devices. Hell, and anything else which feels like generating a physical address cycle to 0xFEExxxxx addresses.
Again, the IOAPIC is just a device for turning pin-based interrupts into MSIs. Bits 19-4 of the address it writes to are taken directly from bits 63-48 of the IOAPIC RTE. There's complexity elsewhere but for *those* bits, It just uses the bits it's given, just like a PCI device or an HPET would.
When I implemented this in qemu I didn't even *touch* the IOAPIC support; it doesn't affect the IOAPIC at all, just like it doesn't affect the HPET or any of the MSI-capable PCI devices that qemu emulates. They just put the bits on the bus that they're told to, when they want to generate an interrupt.
This feature is an MSI feature.
Not an HPET feature.
Not a PCI feature.
Not an IOAPIC feature.
The fact that a *Linux* guest has special-case knowledge in its IOAPIC driver that duplicates what the MSI message composition does, is not a good justification for naming the feature bit bizarrely.
In fact I'm really tempted to make Linux's io_apic.c just use irq_chip_compose_msi_msg() and swizzle the bits out of the message identically for IR and non-IR alike (modulo the pin hack), and delete the IR_IO_APIC_route_entry struct completely.
That also completely removes the ext_dest_id trick from visibility in io_apic.c. And might avoid further confusion.
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 8 lines · David Woodhouse
In fact I'm really tempted to make Linux's
io_apic.cjust useirq_chip_compose_msi_msg()and swizzle the bits out of the message identically for IR and non-IR alike (modulo the pin hack), and delete theIR_IO_APIC_route_entrystruct completely.That also completely removes the
ext_dest_idtrick from visibility inio_apic.c. And might avoid further confusion.
That looks a bit like this, FWIW. Turns out it doesn't *entirely* remove the ext_dest_id trick from io_apic.c because it does still need to put together an entry manually for the ExtINT hackery and for restoring boot mode. But it does remove a lot of incestuous ioapic hackery from IOMMU drivers, and deletes more code than it adds.
It has the additional effect of enabling the WARN_ON for unreachable APIC IDs in __irq_msi_compose_msg(), for IOAPIC interrupts too.
(We'd want the x86_vector_domain to actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)
From 2fbc79588d4677ee1cc9df661162fcf1a7da57f0 Mon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Thu, 8 Oct 2020 15:44:42 +0100
Subject: [PATCH 6/5] x86/ioapic: Generate RTE directly from parent irqchip's MSI
message
The IOAPIC generates an MSI cycle with address/data bits taken from its Redirection Table Entry in some combination which used to make sense, but now is just a bunch of bits which get passed through in some seemingly arbitrary order.
Instead of making IRQ remapping drivers directly frob the IOAPIC RTE, let them just do their job and generate an MSI message. The bit swizzling to turn that MSI message into the IOAPIC's RTE is the same in all cases, since it's a function of the IOAPIC hardware. The IRQ remappers have no real need to get involved with that.
The only slight caveat is that the IOAPIC is interpreting some of those fields too, and it does want the 'vector' field to be unique to make EOI work. The AMD IOMMU happens to put its IRTE index in the bits that the IOAPIC thinks are the vector field, and accommodates this requirement by reserving the first 32 indices for the IOAPIC. The Intel IOMMU doesn't actually use the bits that the IOAPIC thinks are the vector field, so it fills in the 'pin' value there instead.
--- arch/x86/include/asm/hw_irq.h | 11 +++--- arch/x86/include/asm/msidef.h | 2 ++ arch/x86/kernel/apic/io_apic.c | 55 ++++++++++++++++++----------- drivers/iommu/amd/iommu.c | 14 -------- drivers/iommu/hyperv-iommu.c | 31 ---------------- drivers/iommu/intel/irq_remapping.c | 19 +++------- 6 files changed, 46 insertions(+), 86 deletions(-)
arch/x86/include/asm/hw_irq.h +5 -6
@@ -45,12 +45,11 @@ enum irq_alloc_type { };struct ioapic_alloc_info { -int pin; -int node; -u32 trigger :1 ; -u32 polarity :1 ; -u32 valid :1 ; -struct IO_APIC_route_entry *entry; +int pin; +int node; +u32 trigger :1 ; +u32 polarity :1 ; +u32 valid :1 ; };struct uv_alloc_info {
arch/x86/include/asm/msidef.h +2 -0
@@ -18,6 +18,7 @@#define MSI_DATA_DELIVERY_MODE_SHIFT 8#define MSI_DATA_DELIVERY_FIXED (0 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_DELIVERY_LOWPRI (1 << MSI_DATA_DELIVERY_MODE_SHIFT) +#define MSI_DATA_DELIVERY_MODE_MASK (3 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_LEVEL_SHIFT 14#define MSI_DATA_LEVEL_DEASSERT (0 << MSI_DATA_LEVEL_SHIFT)
@@ -37,6 +38,7 @@#define MSI_ADDR_DEST_MODE_SHIFT 2#define MSI_ADDR_DEST_MODE_PHYSICAL (0 << MSI_ADDR_DEST_MODE_SHIFT)#define MSI_ADDR_DEST_MODE_LOGICAL (1 << MSI_ADDR_DEST_MODE_SHIFT) +#define MSI_ADDR_DEST_MODE_MASK (1 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_ADDR_REDIRECTION_SHIFT 3#define MSI_ADDR_REDIRECTION_CPU (0 << MSI_ADDR_REDIRECTION_SHIFT)
arch/x86/kernel/apic/io_apic.c +34 -21
98 lines not on this page · raw message
drivers/iommu/amd/iommu.c +0 -14
@@ -3597,7 +3597,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, {struct irq_2_irte *irte_info = &data->irq_2_irte;struct msi_msg *msg = &data->msi_entry; -struct IO_APIC_route_entry *entry;struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];if (!iommu)
@@ -3611,19 +3610,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,switch (info->type) {case X86_IRQ_ALLOC_TYPE_IOAPIC : -/* Setup IOAPIC entry */ - entry = info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; -break ; -case X86_IRQ_ALLOC_TYPE_HPET :case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX :
drivers/iommu/hyperv-iommu.c +0 -31
@@ -40,7 +40,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data, {struct irq_data *parent = data->parent_data;struct irq_cfg *cfg =irqd_cfg (data); -struct IO_APIC_route_entry *entry;int ret;/* Return error If new irq affinity is out of ioapic_max_cpumask. */
@@ -51,9 +50,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data,if (ret <0 || ret ==IRQ_SET_MASK_OK_DONE )return ret; - entry = data->chip_data; - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector;send_cleanup_vector (cfg);return 0 ;
@@ -89,20 +85,6 @@ static int hyperv_irq_remapping_alloc(struct irq_domain *domain, irq_data->chip = &hyperv_ir_chip; -/* -* If there is interrupt remapping function of IOMMU, setting irq -* affinity only needs to change IRTE of IOMMU. But Hyper-V doesn't -* support interrupt remapping function, setting irq affinity of IO-APIC -* interrupts still needs to change IO-APIC registers. But ioapic_ -* configure_entry() will ignore value of cfg->vector and cfg-> -* dest_apicid when IO-APIC's parent irq domain is not the vector -* domain.(See ioapic_configure_entry()) In order to setting vector -* and dest_apicid to IO-APIC register, IO-APIC entry pointer is saved -* in the chip_data and hyperv_irq_remapping_activate()/hyperv_ir_set_ -* affinity() set vector and dest_apicid directly into IO-APIC entry. -*/ - irq_data->chip_data = info->ioapic.entry; -/* * Hypver-V IO APIC irq affinity should be in the scope of * ioapic_max_cpumask because no irq remapping support.
@@ -119,22 +101,9 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain,irq_domain_free_irqs_common (domain, virq, nr_irqs); } -static int hyperv_irq_remapping_activate (struct irq_domain *domain, -struct irq_data *irq_data,bool reserve) -{ -struct irq_cfg *cfg =irqd_cfg (irq_data); -struct IO_APIC_route_entry *entry = irq_data->chip_data; - - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector; - -return 0 ; -} -static const struct irq_domain_ops hyperv_ir_domain_ops = { .alloc = hyperv_irq_remapping_alloc, .free = hyperv_irq_remapping_free, - .activate = hyperv_irq_remapping_activate, };static int __init hyperv_prepare_irq_remapping (void )
drivers/iommu/intel/irq_remapping.c +5 -14
@@ -1265,7 +1265,6 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,struct irq_alloc_info *info, int index, int sub_handle) { - struct IR_IO_APIC_route_entry *entry; struct irte *irte = &data->irte_entry;struct msi_msg *msg = &data->msi_entry;
@@ -1281,23 +1280,15 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, irte->avail, irte->vector, irte->dest_id, irte->sid, irte->sq, irte->svt); - entry = (struct IR_IO_APIC_route_entry *)info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->index2 = (index >>15 ) &0x1 ; - entry->zero =0 ; - entry->format =1 ; - entry->index = (index &0x7fff );/* * IO-APIC RTE will be configured with virtual vector. * irq handler will do the explicit EOI to the io-apic. */ - entry->vector = info->ioapic.pin; - entry->mask =0 ;/* enable IRQ */ - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -if (info->ioapic.trigger) - entry->mask =1 ;/* Mask level triggered irqs. */ + msg->data = info->ioapic.pin; + msg->address_hi =MSI_ADDR_BASE_HI ; + msg->address_lo =MSI_ADDR_BASE_LO |MSI_ADDR_IR_EXT_INT | +MSI_ADDR_IR_INDEX1 (index) | +MSI_ADDR_IR_INDEX2 (index);break ;case X86_IRQ_ALLOC_TYPE_HPET :
Signature · 1 line
2.17.1
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 4 lines · David Woodhouse
On Thu, 2020-10-08 at 13:55 +0100, David Woodhouse wrote:
(We'd want the
x86_vector_domainto actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)
The compose function and the vector domain wrapper can simply move to vector.c
Quoted 26 lines
From
2fbc79588d4677ee1cc9df661162fcf1a7da57f0Mon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Thu, 8 Oct 2020 15:44:42 +0100
Subject: [PATCH 6/5] x86/ioapic: Generate RTE directly from parent irqchip's MSI
messageThe IOAPIC generates an MSI cycle with address/data bits taken from its Redirection Table Entry in some combination which used to make sense, but now is just a bunch of bits which get passed through in some seemingly arbitrary order.
Instead of making IRQ remapping drivers directly frob the IOAPIC RTE, let them just do their job and generate an MSI message. The bit swizzling to turn that MSI message into the IOAPIC's RTE is the same in all cases, since it's a function of the IOAPIC hardware. The IRQ remappers have no real need to get involved with that.
The only slight caveat is that the IOAPIC is interpreting some of those fields too, and it does want the 'vector' field to be unique to make EOI work. The AMD IOMMU happens to put its IRTE index in the bits that the IOAPIC thinks are the vector field, and accommodates this requirement by reserving the first 32 indices for the IOAPIC. The Intel IOMMU doesn't actually use the bits that the IOAPIC thinks are the vector field, so it fills in the 'pin' value there instead.
Signed-off-by David Woodhouse <[email protected]>
Quoted 8 lines
--- arch/x86/include/asm/hw_irq.h | 11 +++--- arch/x86/include/asm/msidef.h | 2 ++ arch/x86/kernel/apic/io_apic.c | 55 ++++++++++++++++++----------- drivers/iommu/amd/iommu.c | 14 -------- drivers/iommu/hyperv-iommu.c | 31 ---------------- drivers/iommu/intel/irq_remapping.c | 19 +++------- 6 files changed, 46 insertions(+), 86 deletions(-)
Nice :)
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1861
Quoted 6 lines
arch/x86/kernel/apic/io_apic.c +6
-0+static void mp_swizzle_msi_dest_bits (struct irq_data *irq_data,void *_entry) +{ +struct msi_msg msg; +u32 *entry = _entry; + +irq_chip_compose_msi_msg (irq_data, &msg);
Duh, for some stupid reason it never occured to me to do it that way.
Historically the MSI compose function was part of the MSI PCI chip and I just changed that recently when I reworked the code to make IMS support possible.
Historical blinders are pretty sticky :(
Quoted 14 lines
arch/x86/kernel/apic/io_apic.c +14
-0+/* +* They're in a bit of a random order for historical reasons, but +* the IO/APIC is just a device for turning interrupt lines into +* MSIs, and various bits of the MSI addr/data are just swizzled +* into/from the bits of Redirection Table Entry. +*/ + entry[0 ] &=0xfffff000 ; + entry[0 ] |= (msg.data & (MSI_DATA_DELIVERY_MODE_MASK | +MSI_DATA_VECTOR_MASK )); + entry[0 ] |= (msg.address_lo &MSI_ADDR_DEST_MODE_MASK ) <<9 ; + + entry[1 ] &=0xffff ; + entry[1 ] |= (msg.address_lo &MSI_ADDR_DEST_ID_MASK ) <<12 ; +}
....
Quoted 5 lines
drivers/iommu/amd/iommu.c +0
-3switch (info->type) {case X86_IRQ_ALLOC_TYPE_IOAPIC : -/* Setup IOAPIC entry */ - entry = info->ioapic.entry; - info->ioapic.entry =NULL ;
drivers/iommu/amd/iommu.c static void irq_remapping_prepare_irte(struct amd_ir_data *data, · line 3626
Quoted 10 lines
drivers/iommu/amd/iommu.c +0
-10-memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; -break ; -
Thanks for cleaning this up!
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 8 lines · Thomas Gleixner
Quoted 4 lines · David Woodhouse
On Thu, 2020-10-08 at 13:55 +0100, David Woodhouse wrote:
(We'd want the
x86_vector_domainto actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)The compose function and the vector domain wrapper can simply move to
vector.c
I ended up putting __irq_msi_compose_msg() into apic.c and that way I can make virt_ext_dest_id static in that file.
And then I can move all the HPET-MSI support into hpet.c too.
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/ext_dest_id
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 5 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
Quoted 3 lines
(We'd want the
x86_vector_domainto actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)
Quoted 7 lines
Quoted 2 lines
The compose function and the vector domain wrapper can simply move to
vector.cI ended up putting
__irq_msi_compose_msg()intoapic.cand that way I can makevirt_ext_dest_idstatic in that file.And then I can move all the HPET-MSI support into
hpet.ctoo.
Works for me.
For the next submission, can you please
- pick up the -ENODEV changes for HPET/IOAPIC which I posted earlier
- shuffle all that compose/IOAPIC cleanup around
before adding that extended dest id stuff.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 15 lines · Thomas Gleixner
Quoted 6 lines · David Woodhouse
Quoted 5 lines · Thomas Gleixner
Quoted 3 lines
(We'd want the
x86_vector_domainto actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)The compose function and the vector domain wrapper can simply move
to
Quoted 6 lines
Quoted 1 line
vector.cI ended up putting
__irq_msi_compose_msg()intoapic.cand that way I can makevirt_ext_dest_idstatic in that file.And then I can move all the HPET-MSI support into
hpet.ctoo.
Quoted 8 lines
Works for me.
Quoted 1 line
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/ext_dest_id
For the next submission, can you please
- pick up the
-ENODEVchanges for HPET/IOAPIC which I posted earlier
Ack.
Quoted 1 line
- shuffle all that compose/IOAPIC cleanup around
I'd prefer the MSI swizzling bit to stay last in the series. I want to stare hard at the hyperv-iommu part a bit more, and ideally even have it tested. I'd prefer the real functionality that I care about, not to depend on that cleanup.
If it actually let me remove all mention of ext_dest_id from the IOAPIC code and use *only* MSI swizzling, I'd be keener to reorder. But as noted, there are a couple of manual RTE constructions in there still anyway.
I can move __irq_compose_msi_msg() earlier in the series though, and then virt_ext_dest_id can be static in apic.c from its inception.
OK?
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
In reply to Thomas GleixnerQuoted 14 lines · Thomas Gleixner
Quoted 5 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
Quoted 3 lines
(We'd want the
x86_vector_domainto actually have an MSI compose function in the !CONFIG_PCI_MSI case if we did this, of course.)Quoted 7 lines
Quoted 2 lines
The compose function and the vector domain wrapper can simply move to
vector.cI ended up putting
__irq_msi_compose_msg()intoapic.cand that way I can makevirt_ext_dest_idstatic in that file.And then I can move all the HPET-MSI support into
hpet.ctoo.
Quoted 7 lines
Works for me.
For the next submission, can you please
- pick up the
-ENODEVchanges for HPET/IOAPIC which I posted earlier
I think the world will be a nicer place if HPET and IOAPIC have their own struct device and their drivers can just use dev_get_msi_domain().
The IRQ remapping drivers already plug into the device-add notifier and can fill in the appropriate MSI domain just like they do¹ for PCI and ACPI devices.
Using platform_add_bundle() for HPET looks trivial enough; I'll have a play with that and then do IOAPIC too if/when the initialisation order and hotplug handling all works out OK to install the correct msi_domain.
Signature · 6 lines
dwmw2 ¹ Yeah, I know they don't do it directly; it's done in pcibios_add_device(). But maybe they should? Because yeah, I also know they don't do it for ACPI devices either, because nothing does, and IRQ remapping on ANDD-listed devices doesn't work AFAICT.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 9 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
On Thu, Oct 08 2020 at 22:39, David Woodhouse wrote:
For the next submission, can you please- pick up the
-ENODEVchanges for HPET/IOAPIC which I posted earlierI think the world will be a nicer place if HPET and IOAPIC have their own struct device and their drivers can just use
dev_get_msi_domain().
Quoted 8 lines
The IRQ remapping drivers already plug into the device-add notifier and can fill in the appropriate MSI domain just like they do¹ for PCI and ACPI devices.
Using
platform_add_bundle()for HPET looks trivial enough; I'll have a play with that and then do IOAPIC too if/when the initialisation order and hotplug handling all works out OK to install the correctmsi_domain.
Yes, I was wondering about that when I made PCI at least use that mechanism, but had not had time to actually look at it.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 18 lines · Thomas Gleixner
Quoted 5 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
On Thu, Oct 08 2020 at 22:39, David Woodhouse wrote:
For the next submission, can you please- pick up the
-ENODEVchanges for HPET/IOAPIC which I postedearlier
Quoted 3 lines
I think the world will be a nicer place if HPET and IOAPIC have their own struct device and their drivers can just use
dev_get_msi_domain().Quoted 2 lines
The IRQ remapping drivers already plug into the device-add notifier
and
Quoted 4 lines
can fill in the appropriate MSI domain just like they do¹ for PCI and ACPI devices.
Using
platform_add_bundle()for HPET looks trivial enough; I'll have
Quoted 8 lines
a
Quoted 1 line
play with that and then do IOAPIC too if/when the initialisation
order
Quoted 2 lines
and hotplug handling all works out OK to install the correct
msi_domain.Yes, I was wondering about that when I made PCI at least use that mechanism, but had not had time to actually look at it.
Yeah. There's some muttering to be done for HPET about whether it's *its* MSI domain or whether it's the parent domain. But I'll have a play. I think we'll be able to drop the whole irq_remapping_get_irq_domain() thing.
Either way, it's a separate cleanup and the 15-bit APIC ID series I posted yesterday should be fine as it is.
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 11 lines · David Woodhouse
Quoted 10 lines · Thomas Gleixner
Quoted 7 lines · David Woodhouse
The IRQ remapping drivers already plug into the device-add notifier and can fill in the appropriate MSI domain just like they do¹ for PCI and ACPI devices.
Usingplatform_add_bundle()for HPET looks trivial enough; I'll have a play with that and then do IOAPIC too if/when the initialisation order and hotplug handling all works out OK to install the correctmsi_domain.
Quoted 7 lines
Quoted 2 lines
Yes, I was wondering about that when I made PCI at least use that mechanism, but had not had time to actually look at it.
Yeah. There's some muttering to be done for HPET about whether it's *its* MSI domain or whether it's the parent domain. But I'll have a play. I think we'll be able to drop the whole
irq_remapping_get_irq_domain()thing.
That would be really nice.
Quoted 2 lines
Either way, it's a separate cleanup and the 15-bit APIC ID series I posted yesterday should be fine as it is.
I go over it in the next days once more and stick it into my devel tree until rc1. Need to get some conflicts sorted with that Device MSI stuff.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 17 lines · Thomas Gleixner
Quoted 1 line · David Woodhouse
On 10 October 2020 12:44:10 BST, Thomas Gleixner <[email protected]>
Quoted 6 lines · wrote:
Quoted 6 lines · David Woodhouse
Quoted 5 lines
The IRQ remapping drivers already plug into the device-add notifier and can fill in the appropriate MSI domain just like they do¹ for PCI and ACPI devices.
Usingplatform_add_bundle()for HPET looks trivial enough; I'llhave
Quoted 7 lines
Quoted 6 lines
Quoted 3 lines
a play with that and then do IOAPIC too if/when the initialisation order and hotplug handling all works out OK to install the correct
msi_domain.Yes, I was wondering about that when I made PCI at least use that mechanism, but had not had time to actually look at it.
Quoted 6 lines
Quoted 4 lines
Yeah. There's some muttering to be done for HPET about whether it's *its* MSI domain or whether it's the parent domain. But I'll have a play. I think we'll be able to drop the whole
irq_remapping_get_irq_domain()thing.That would be really nice.
I can make it work for HPET if I fix up the point at which the IRQ remapping code registers a notifier on the platform bus. (At IRQ remap setup time is too early; when it registers the PCI bus notifier is too late.)
IOAPIC is harder though as the platform bus doesn't even exist that early. Maybe an early platform bus is possible but it would have to turn out particularly simple to do, or I'd need to find another use case too, to justify it. Will continue to play....
Quoted 6 lines
Quoted 2 lines
Either way, it's a separate cleanup and the 15-bit APIC ID series I posted yesterday should be fine as it is.
I go over it in the next days once more and stick it into my devel tree until rc1. Need to get some conflicts sorted with that Device MSI stuff.
While playing with HPET I noticed I need s/CONFIG_PCI_MSI/CONFIG_IRQ_GENERIC_MSI/ where the variables are declared at the top of msi.c to match the change I made later on. Can post v3 of the series or you can silently fix it up as you go; please advise.
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 16 lines · David Woodhouse
Quoted 9 lines · Thomas Gleixner
Quoted 1 line · David Woodhouse
On 10 October 2020 12:44:10 BST, Thomas Gleixner <[email protected]>
Quoted 4 lines · wrote:
Yeah. There's some muttering to be done for HPET about whether it's *its* MSI domain or whether it's the parent domain. But I'll have a play. I think we'll be able to drop the whole
irq_remapping_get_irq_domain()thing.That would be really nice.
I can make it work for HPET if I fix up the point at which the IRQ remapping code registers a notifier on the platform bus. (At IRQ remap setup time is too early; when it registers the PCI bus notifier is too late.)
Quoted 4 lines
IOAPIC is harder though as the platform bus doesn't even exist that early. Maybe an early platform bus is possible but it would have to turn out particularly simple to do, or I'd need to find another use case too, to justify it. Will continue to play....
You might want to look into using irq_find_matching_fwspec() instead for both HPET and IOAPIC. That needs a select() callback implemented in the remapping domains.
Quoted 9 lines
Quoted 3 lines
I go over it in the next days once more and stick it into my devel tree until rc1. Need to get some conflicts sorted with that Device MSI stuff.
While playing with HPET I noticed I need s/CONFIG_PCI_MSI/CONFIG_IRQ_GENERIC_MSI/ where the variables are declared at the top of
msi.cto match the change I made later on. Can post v3 of the series or you can silently fix it up as you go; please advise.
I think I might be able to handle that on my own :)
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 18 lines · Thomas Gleixner
Quoted 17 lines · David Woodhouse
Quoted 10 lines · Thomas Gleixner
Quoted 1 line · David Woodhouse
On 10 October 2020 12:44:10 BST, Thomas Gleixner <[email protected]>
Quoted 4 lines · wrote:
Yeah. There's some muttering to be done for HPET about whether it's *its* MSI domain or whether it's the parent domain. But I'll have a play. I think we'll be able to drop the whole
irq_remapping_get_irq_domain()thing.That would be really nice.
I can make it work for HPET if I fix up the point at which the IRQ remapping code registers a notifier on the platform bus. (At IRQ remap setup time is too early; when it registers the PCI bus notifier is too late.)
Quoted 8 lines
Quoted 4 lines
IOAPIC is harder though as the platform bus doesn't even exist that early. Maybe an early platform bus is possible but it would have to turn out particularly simple to do, or I'd need to find another use case too, to justify it. Will continue to play....
You might want to look into using
irq_find_matching_fwspec()instead for both HPET and IOAPIC. That needs aselect()callback implemented in the remapping domains.
That works. Pushed (along with that trivial HPET MSI fixup) to
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/ext_dest_id
Briefly tested with I/OAPIC with and without Intel remapping in qemu, not tested for HPET because I haven't worked out how to get qemu to do HPET-MSI yet.
David Woodhouse (8):
genirq/irqdomain: Implement get_name() method on irqchip fwnodes x86/apic: Add select() method on vector irqdomain iommu/amd: Implement select() method on remapping irqdomain iommu/vt-d: Implement select() method on remapping irqdomain iommu/hyper-v: Implement select() method on remapping irqdomain x86/hpet: Use irq_find_matching_fwspec() to find remapping irqdomain x86/ioapic: Use irq_find_matching_fwspec() to find remapping irqdomain
x86: Kill all traces of irq_remapping_get_irq_domain()
arch/x86/include/asm/hw_irq.h | 2 -- arch/x86/include/asm/irq_remapping.h | 9 ------ arch/x86/include/asm/irqdomain.h | 3 ++ arch/x86/kernel/apic/io_apic.c | 21 ++++++-------- arch/x86/kernel/apic/vector.c | 52 +++++++++++++++++++++++++++++++++++ arch/x86/kernel/hpet.c | 23 +++++++++------- drivers/iommu/amd/iommu.c | 53 +++++++++++++----------------------- drivers/iommu/hyperv-iommu.c | 18 ++++++------ drivers/iommu/intel/irq_remapping.c | 30 +++++++++----------- drivers/iommu/irq_remapping.c | 14 ---------- drivers/iommu/irq_remapping.h | 3 -- kernel/irq/irqdomain.c | 11 +++++++- 12 files changed, 128 insertions(+), 111 deletions(-)
arch/x86/include/asm/hw_irq.h +0 -2
@@ -40,8 +40,6 @@ enum irq_alloc_type {X86_IRQ_ALLOC_TYPE_PCI_MSIX ,X86_IRQ_ALLOC_TYPE_DMAR , X86_IRQ_ALLOC_TYPE_UV, -X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT , - X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT, }; struct ioapic_alloc_info {
arch/x86/include/asm/irq_remapping.h +0 -9
@@ -44,9 +44,6 @@ extern int irq_remapping_reenable(int);extern int irq_remap_enable_fault_handling (void );extern void panic_if_irq_remap (const char *msg); -extern struct irq_domain * -irq_remapping_get_irq_domain (struct irq_alloc_info *info); -/* Create PCI MSI/MSIx irqdomain, use @parent as the parent irqdomain. */ extern struct irq_domain *arch_create_remap_msi_irq_domain (struct irq_domain *par,const char *n,int id);
@@ -71,11 +68,5 @@ static inline void panic_if_irq_remap(const char *msg) { } -static inline struct irq_domain * -irq_remapping_get_irq_domain (struct irq_alloc_info *info) -{ -return NULL ; -} -#endif /* CONFIG_IRQ_REMAP */ #endif /* __X86_IRQ_REMAPPING_H */
arch/x86/include/asm/irqdomain.h +3 -0
@@ -12,6 +12,9 @@ enum { X86_IRQ_ALLOC_LEGACY = 0x2, }; +int x86_fwspec_is_ioapic (struct irq_fwspec *fwspec); +int x86_fwspec_is_hpet (struct irq_fwspec *fwspec); +extern struct irq_domain *x86_vector_domain; extern void init_irq_alloc_info(struct irq_alloc_info *info,
arch/x86/kernel/apic/io_apic.c +9 -12
@@ -2305,36 +2305,33 @@ static inline void __init check_timer(void)static int mp_irqdomain_create (int ioapic) { -struct irq_alloc_info info;struct irq_domain *parent;int hwirqs =mp_ioapic_pin_count (ioapic);struct ioapic *ip = &ioapics[ioapic];struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;struct mp_ioapic_gsi *gsi_cfg =mp_ioapic_gsi_routing (ioapic);struct fwnode_handle *fn; -char *name ="IO-APIC" ; +struct irq_fwspec fwspec;if (cfg->type ==IOAPIC_DOMAIN_INVALID )return 0 ; -init_irq_alloc_info (&info,NULL ); - info.type =X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT ; - info.devid =mpc_ioapic_id (ioapic); - parent =irq_remapping_get_irq_domain (&info); -if (!parent) - parent = x86_vector_domain; -else - name ="IO-APIC-IR" ; -/* Handle device tree enumerated APICs proper */ if (cfg->dev) { fn =of_node_to_fwnode (cfg->dev); }else { - fn =irq_domain_alloc_named_id_fwnode (name, ioapic); + fn =irq_domain_alloc_named_id_fwnode ("IO-APIC" , ioapic);if (!fn)return -ENOMEM ; } + fwspec.fwnode = fn; + fwspec.param_count =1 ; + fwspec.param[0 ] = ioapic; + parent =irq_find_matching_fwspec (&fwspec,DOMAIN_BUS_ANY ); +if (!parent) +return -ENODEV ; + ip->irqdomain =irq_domain_create_linear (fn, hwirqs, cfg->ops, (void *)(long )ioapic);
arch/x86/kernel/apic/vector.c +52 -0
59 lines not on this page · raw message
arch/x86/kernel/hpet.c +13 -10
@@ -543,8 +543,8 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id) {struct msi_domain_info *domain_info;struct irq_domain *parent, *d; -struct irq_alloc_info info;struct fwnode_handle *fn; +struct irq_fwspec fwspec;if (x86_vector_domain ==NULL )return NULL ;
@@ -556,15 +556,6 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id) *domain_info = hpet_msi_domain_info; domain_info->data = (void *)(long )hpet_id; -init_irq_alloc_info (&info,NULL ); - info.type =X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT ; - info.devid = hpet_id; - parent =irq_remapping_get_irq_domain (&info); -if (parent ==NULL ) - parent = x86_vector_domain; -else - hpet_msi_controller.name ="IR-HPET-MSI" ; - fn =irq_domain_alloc_named_id_fwnode (hpet_msi_controller.name, hpet_id);if (!fn) {
@@ -572,6 +563,18 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id)return NULL ; } + fwspec.fwnode = fn; + fwspec.param_count =1 ; + fwspec.param[0 ] = hpet_id; + parent =irq_find_matching_fwspec (&fwspec,DOMAIN_BUS_ANY ); +if (!parent) { +irq_domain_free_fwnode (fn); +kfree (domain_info); +return NULL ; + } +if (parent != x86_vector_domain) + hpet_msi_controller.name ="IR-HPET-MSI" ; + d =msi_create_irq_domain (fn, domain_info, parent);if (!d) {irq_domain_free_fwnode (fn);
drivers/iommu/amd/iommu.c +19 -34
81 lines not on this page · raw message
drivers/iommu/hyperv-iommu.c +9 -9
@@ -61,6 +61,14 @@ static struct irq_chip hyperv_ir_chip = { .irq_set_affinity = hyperv_ir_set_affinity, }; +static int hyperv_irq_remapping_select (struct irq_domain *d, +struct irq_fwspec *fwspec, +enum irq_domain_bus_token bus_token) +{ +/* Claim only the first (and only) I/OAPIC */ +return x86_fwspec_is_ioapic (fwspec) && fwspec->param[0 ] ==0 ; +} +static int hyperv_irq_remapping_alloc (struct irq_domain *domain,unsigned int virq,unsigned int nr_irqs,void *arg)
@@ -102,6 +110,7 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain, }static const struct irq_domain_ops hyperv_ir_domain_ops = { + .select = hyperv_irq_remapping_select, .alloc = hyperv_irq_remapping_alloc, .free = hyperv_irq_remapping_free, };
@@ -151,18 +160,9 @@ static int __init hyperv_enable_irq_remapping(void) return IRQ_REMAP_X2APIC_MODE; } -static struct irq_domain *hyperv_get_irq_domain (struct irq_alloc_info *info) -{ -if (info->type ==X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT ) -return ioapic_ir_domain; -else -return NULL ; -} -struct irq_remap_ops hyperv_irq_remap_ops = { .prepare = hyperv_prepare_irq_remapping, .enable = hyperv_enable_irq_remapping, - .get_irq_domain = hyperv_get_irq_domain, };#endif
drivers/iommu/intel/irq_remapping.c +13 -17
@@ -1128,29 +1128,12 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest) irte->redir_hint =1 ; } -static struct irq_domain *intel_get_irq_domain (struct irq_alloc_info *info) -{ -if (!info) -return NULL ; - -switch (info->type) { -case X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT : -return map_ioapic_to_ir (info->devid); -case X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT : -return map_hpet_to_ir (info->devid); -default : -WARN_ON_ONCE (1 ); -return NULL ; - } -} -struct irq_remap_ops intel_irq_remap_ops = { .prepare = intel_prepare_irq_remapping, .enable = intel_enable_irq_remapping, .disable = disable_irq_remapping, .reenable = reenable_irq_remapping, .enable_faulting = enable_drhd_fault_handling, - .get_irq_domain = intel_get_irq_domain, };static void intel_ir_reconfigure_irte (struct irq_data *irqd,bool force)
@@ -1435,7 +1418,20 @@ static void intel_irq_remapping_deactivate(struct irq_domain *domain,modify_irte (&data->irq_2_iommu, &entry); } +static int intel_irq_remapping_select (struct irq_domain *d, +struct irq_fwspec *fwspec, +enum irq_domain_bus_token bus_token) +{ +if (x86_fwspec_is_ioapic (fwspec)) +return d ==map_ioapic_to_ir (fwspec->param[0 ]); +else if (x86_fwspec_is_hpet (fwspec)) +return d ==map_hpet_to_ir (fwspec->param[0 ]); + +return 0 ; +} +static const struct irq_domain_ops intel_ir_domain_ops = { + .select = intel_irq_remapping_select, .alloc = intel_irq_remapping_alloc, .free = intel_irq_remapping_free, .activate = intel_irq_remapping_activate,
drivers/iommu/irq_remapping.c +0 -14
@@ -158,17 +158,3 @@ void panic_if_irq_remap(const char *msg)if (irq_remapping_enabled)panic (msg); } - -/** -* irq_remapping_get_irq_domain - Get the irqdomain serving the request @info -* @info: interrupt allocation information, used to identify the IOMMU device -* -* Returns pointer to IRQ domain, or NULL on failure. -*/ -struct irq_domain *irq_remapping_get_irq_domain (struct irq_alloc_info *info) -{ -if (!remap_ops || !remap_ops->get_irq_domain) -return NULL ; - -return remap_ops->get_irq_domain (info); -}
drivers/iommu/irq_remapping.h +0 -3
@@ -42,9 +42,6 @@ struct irq_remap_ops {/* Enable fault handling */ int (*enable_faulting)(void ); - -/* Get the irqdomain associated to IOMMU device */ -struct irq_domain *(*get_irq_domain)(struct irq_alloc_info *); };extern struct irq_remap_ops intel_irq_remap_ops;
kernel/irq/irqdomain.c +10 -1
@@ -42,7 +42,16 @@ static inline void debugfs_add_domain_dir(struct irq_domain *d) { }static inline void debugfs_remove_domain_dir (struct irq_domain *d) { }#endif -const struct fwnode_operations irqchip_fwnode_ops; +static const char *irqchip_fwnode_get_name (const struct fwnode_handle *fwnode) +{ + struct irqchip_fwid *fwid = container_of(fwnode, struct irqchip_fwid, fwnode); + +return fwid->name; +} + +const struct fwnode_operations irqchip_fwnode_ops = { + .get_name = irqchip_fwnode_get_name, +};EXPORT_SYMBOL_GPL (irqchip_fwnode_ops);/**
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 6 lines · David Woodhouse
Quoted 3 lines · Thomas Gleixner
You might want to look into using
irq_find_matching_fwspec()instead for both HPET and IOAPIC. That needs aselect()callback implemented in the remapping domains.That works.
:)
Nasty, but way better than what we have now. Now I start to wonder whether we can get rid of a few other things as well especially the non-standard alloc_info thingy.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 8 lines · Thomas Gleixner
Quoted 7 lines · David Woodhouse
Quoted 5 lines · Thomas Gleixner
You might want to look into using
irq_find_matching_fwspec()instead for
both HPET and IOAPIC. That needs aselect()callback implemented in the
remapping domains.
Quoted 5 lines
Quoted 1 line
That works.
:)
Nasty, but way better than what we have now.
Want me to send that out in email or is the git tree enough for now?
I've cleaned it up a little and fixed a bug in the I/OAPIC error path.
Still not entirely convinced about the apic->apic_id_valid(32768) thing but it should work well enough, and doesn't require exporting any extra state from apic.c that way.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 6 lines · David Woodhouse
Quoted 1 line · Thomas Gleixner
Nasty, but way better than what we have now.
Want me to send that out in email or is the git tree enough for now?
I've cleaned it up a little and fixed a bug in the I/OAPIC error path.
Mail would be nice once you are confident with the pile.
Quoted 3 lines
Still not entirely convinced about the
apic->apic_id_valid(32768)thing but it should work well enough, and doesn't require exporting any extra state fromapic.cthat way.
Yeah, that part is odd.
I really dislike the way how irq_find_matching_fwspec() works. The 'rc' value is actually boolean despite being type 'int' and if 'rc' is not 0 then it returns the domain even if 'rc' is an error code.
But that does not allow to return error codes from a domain match() / select() callback which is what we really want to express that there is something fishy.
Something like the below perhaps? Needs more thought obviously.
Marc, any opinion ?
Thanks,
tglx
8<-------------
arch/x86/kernel/apic/vector.c | 18 ++++++++++++++++++
include/linux/irqdomain.h | 1 + kernel/irq/irqdomain.c | 2 +- 3 files changed, 20 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/vector.c +18 -0
@@ -593,6 +593,24 @@ static int x86_vector_alloc_irqs(structreturn err; } +struct irq_domain *x86_select_parent_domain (struct irq_fwspec *fwspec) +{ +struct irq_domain *dom; + +/* +* If Interrupt Remapping is enabled then the match function +* returns either the remapping domain or an error code if the +* device is not registered with the remapping unit. +* +* If remapping is not enabled, then the function returns NULL. +*/ + dom =irq_find_matching_fwspec (fwspec,DOMAIN_BUS_IR ); +if (dom) +return IS_ERR (dom) ?NULL : dom; + +return x86_vector_domain; +} +#ifdef CONFIG_GENERIC_IRQ_DEBUGFS static void x86_vector_debug_show (struct seq_file *m,struct irq_domain *d,struct irq_data *irqd,int ind)
include/linux/irqdomain.h +1 -0
@@ -85,6 +85,7 @@ enum irq_domain_bus_token {DOMAIN_BUS_TI_SCI_INTA_MSI ,DOMAIN_BUS_WAKEUP ,DOMAIN_BUS_VMD_MSI , + DOMAIN_BUS_IR, };/**
kernel/irq/irqdomain.c +1 -1
@@ -395,7 +395,7 @@ struct irq_domain *irq_find_matching_fws (h->bus_token == bus_token)));if (rc) { - found = h; + found = rc <0 ?ERR_PTR (rc) : h;break ; } }
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 9 lines · Thomas Gleixner
Quoted 6 lines · David Woodhouse
Quoted 1 line · Thomas Gleixner
Nasty, but way better than what we have now.
Want me to send that out in email or is the git tree enough for now?
I've cleaned it up a little and fixed a bug in the I/OAPIC error path.
Mail would be nice once you are confident with the pile.
After the next cup of coffee. Will send it as an incremental series on top of my previous ext_dest_id series.
Quoted 6 lines
Quoted 3 lines
Still not entirely convinced about the
apic->apic_id_valid(32768)thing but it should work well enough, and doesn't require exporting any extra state fromapic.cthat way.Yeah, that part is odd.
Quoted 7 lines
I really dislike the way how
irq_find_matching_fwspec()works. The 'rc' value is actually boolean despite being type 'int' and if 'rc' is not 0 then it returns the domain even if 'rc' is an error code.But that does not allow to return error codes from a domain
match()/select()callback which is what we really want to express that there is something fishy.
I don't know that we need to return an explicit error. I made x86_vector_select() refuse to match if IRQ remapping is enabled, which means that the irq_find_matching_fwspec() call will fail by returning NULL in precisely the cases it should. (Which should never happen; qv).
That failure means HPET will refuse to set up MSI, or I/OAPIC will refuse to initialise (later causing a BUG and a failure to boot, which is probably the correct thing to do).
It's all fine.
+ dom = irq_find_matching_fwspec(fwspec, DOMAIN_BUS_IR);
+ if (dom)
+ return IS_ERR(dom) ? NULL : dom;
+
+ return x86_vector_domain;
+}
Ick. There's no need for that.
Eliminating that awful "if not found then slip the x86_vector_domain in as a special case" was the whole *point* of using irq_find_matching_fwspec() in the first place.
David Woodhouse
maintainercover Remove irq_remapping_get_irq_domain()
I didn't much like the I/OAPIC and HPET drivers having magical knowledge that they had to substitute x86_vector_domain if their call to irq_remapping_get_irq_domain() returned NULL.
When Thomas tried to make it handle error returns from …get_irq_domain() distinctly from the NULL case too, it made me even sadder. So I killed it with fire.
Now they just use irq_find_matching_fwspec() to find an appropriate irqdomain. Each remapping irqdomain just needs to say 'yep, that's me' for the HPETs or I/OAPICs which are within their scope, while the x86_vector_domain accepts them all but only if interrupt remapping is *disabled*. No more special knowledge in the caller.
If IR is enabled and there's a child device which escapes the scope of all remapping units, it gets NULL for its parent irqdomain and will fail to initialise, which is the correct thing to do in that "should never happen" case. For HPET that'll mean that it just doesn't support MSI, while I/OAPIC will refuse to initialise and trigger a BUG_ON because Linux quite likes it when *all* the I/OAPICs it knows about get initialised successfully.
This is on top of the previous 'ext_dest_id' series at
https://patchwork.kernel.org/project/kvm/list/?series=362037
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/ext_dest_id
David Woodhouse (9):
genirq/irqdomain: Implement get_name() method on irqchip fwnodes x86/apic: Add select() method on vector irqdomain iommu/amd: Implement select() method on remapping irqdomain iommu/vt-d: Implement select() method on remapping irqdomain iommu/hyper-v: Implement select() method on remapping irqdomain x86/hpet: Use irq_find_matching_fwspec() to find remapping irqdomain x86/ioapic: Use irq_find_matching_fwspec() to find remapping irqdomain
x86: Kill all traces of irq_remapping_get_irq_domain() iommu/vt-d: Simplify intel_irq_remapping_select()
arch/x86/include/asm/hw_irq.h | 2 -- arch/x86/include/asm/irq_remapping.h | 9 --------- arch/x86/include/asm/irqdomain.h | 3 +++ arch/x86/kernel/apic/io_apic.c | 24 ++++++++++++------------ arch/x86/kernel/apic/vector.c | 43 +++++++++++++++++++++++++++++++++++++++++++ arch/x86/kernel/hpet.c | 23 +++++++++++++---------- drivers/iommu/amd/iommu.c | 53 +++++++++++++++++++---------------------------------- drivers/iommu/hyperv-iommu.c | 18 +++++++++--------- drivers/iommu/intel/irq_remapping.c | 43 +++++++++++++++++-------------------------- drivers/iommu/irq_remapping.c | 14 -------------- drivers/iommu/irq_remapping.h | 3 --- kernel/irq/irqdomain.c | 11 ++++++++++- 12 files changed, 126 insertions(+), 120 deletions(-)
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
David Woodhouse
maintainerPATCH 1/9 genirq/irqdomain: Implement get_name() method on irqchip fwnodes
From: David Woodhouse <[email protected]>
--- kernel/irq/irqdomain.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/irq/irqdomain.c +10 -1
@@ -42,7 +42,16 @@ static inline void debugfs_add_domain_dir(struct irq_domain *d) { }static inline void debugfs_remove_domain_dir (struct irq_domain *d) { }#endif -const struct fwnode_operations irqchip_fwnode_ops; +static const char *irqchip_fwnode_get_name (const struct fwnode_handle *fwnode) +{ + struct irqchip_fwid *fwid = container_of(fwnode, struct irqchip_fwid, fwnode); + +return fwid->name; +} + +const struct fwnode_operations irqchip_fwnode_ops = { + .get_name = irqchip_fwnode_get_name, +};EXPORT_SYMBOL_GPL (irqchip_fwnode_ops);/**
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
David Woodhouse
maintainerPATCH 2/9 x86/apic: Add select() method on vector irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This will be used to select the irqdomain for I/OAPIC and HPET.
--- arch/x86/include/asm/irqdomain.h | 3 +++ arch/x86/kernel/apic/vector.c | 43 ++++++++++++++++++++++++++++++++ 2 files changed, 46 insertions(+)
arch/x86/include/asm/irqdomain.h +3 -0
@@ -12,6 +12,9 @@ enum { X86_IRQ_ALLOC_LEGACY = 0x2, }; +extern int x86_fwspec_is_ioapic (struct irq_fwspec *fwspec); +extern int x86_fwspec_is_hpet (struct irq_fwspec *fwspec); +extern struct irq_domain *x86_vector_domain; extern void init_irq_alloc_info(struct irq_alloc_info *info,
arch/x86/kernel/apic/vector.c +43 -0
50 lines not on this page · raw message
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 3/9 iommu/amd: Implement select() method on remapping irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- drivers/iommu/amd/iommu.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
drivers/iommu/amd/iommu.c +19 -0
@@ -3590,6 +3590,24 @@ struct irq_remap_ops amd_iommu_irq_ops = { .get_irq_domain = get_irq_domain, }; +static int irq_remapping_select (struct irq_domain *d,struct irq_fwspec *fwspec, +enum irq_domain_bus_token bus_token) +{ +struct amd_iommu *iommu; +int devid =-1 ; + +if (x86_fwspec_is_ioapic (fwspec)) + devid =get_ioapic_devid (fwspec->param[0 ]); +else if (x86_fwspec_is_ioapic (fwspec)) + devid =get_hpet_devid (fwspec->param[0 ]); + +if (devid <0 ) +return 0 ; + + iommu = amd_iommu_rlookup_table[devid]; +return iommu && iommu->ir_domain == d; +} + static void irq_remapping_prepare_irte(struct amd_ir_data *data, struct irq_cfg *irq_cfg, struct irq_alloc_info *info,
@@ -3813,6 +3831,7 @@ static void irq_remapping_deactivate(struct irq_domain *domain, }static const struct irq_domain_ops amd_ir_domain_ops = { + .select = irq_remapping_select, .alloc = irq_remapping_alloc, .free = irq_remapping_free, .activate = irq_remapping_activate,
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 4/9 iommu/vt-d: Implement select() method on remapping irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- drivers/iommu/intel/irq_remapping.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
drivers/iommu/intel/irq_remapping.c +13 -0
@@ -1435,7 +1435,20 @@ static void intel_irq_remapping_deactivate(struct irq_domain *domain,modify_irte (&data->irq_2_iommu, &entry); } +static int intel_irq_remapping_select (struct irq_domain *d, +struct irq_fwspec *fwspec, +enum irq_domain_bus_token bus_token) +{ +if (x86_fwspec_is_ioapic (fwspec)) +return d ==map_ioapic_to_ir (fwspec->param[0 ]); +else if (x86_fwspec_is_hpet (fwspec)) +return d ==map_hpet_to_ir (fwspec->param[0 ]); + +return 0 ; +} +static const struct irq_domain_ops intel_ir_domain_ops = { + .select = intel_irq_remapping_select, .alloc = intel_irq_remapping_alloc, .free = intel_irq_remapping_free, .activate = intel_irq_remapping_activate,
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
David Woodhouse
maintainerPATCH 5/9 iommu/hyper-v: Implement select() method on remapping irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- drivers/iommu/hyperv-iommu.c | 9 +++++++++ 1 file changed, 9 insertions(+)
drivers/iommu/hyperv-iommu.c +9 -0
@@ -61,6 +61,14 @@ static struct irq_chip hyperv_ir_chip = { .irq_set_affinity = hyperv_ir_set_affinity, }; +static int hyperv_irq_remapping_select (struct irq_domain *d, +struct irq_fwspec *fwspec, +enum irq_domain_bus_token bus_token) +{ +/* Claim only the first (and only) I/OAPIC */ +return x86_fwspec_is_ioapic (fwspec) && fwspec->param[0 ] ==0 ; +} +static int hyperv_irq_remapping_alloc (struct irq_domain *domain,unsigned int virq,unsigned int nr_irqs,void *arg)
@@ -102,6 +110,7 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain, }static const struct irq_domain_ops hyperv_ir_domain_ops = { + .select = hyperv_irq_remapping_select, .alloc = hyperv_irq_remapping_alloc, .free = hyperv_irq_remapping_free, };
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
David Woodhouse
maintainerPATCH 6/9 x86/hpet: Use irq_find_matching_fwspec() to find remapping irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- arch/x86/kernel/hpet.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)
arch/x86/kernel/hpet.c +13 -10
@@ -543,8 +543,8 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id) {struct msi_domain_info *domain_info;struct irq_domain *parent, *d; -struct irq_alloc_info info;struct fwnode_handle *fn; +struct irq_fwspec fwspec;if (x86_vector_domain ==NULL )return NULL ;
@@ -556,15 +556,6 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id) *domain_info = hpet_msi_domain_info; domain_info->data = (void *)(long )hpet_id; -init_irq_alloc_info (&info,NULL ); - info.type =X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT ; - info.devid = hpet_id; - parent =irq_remapping_get_irq_domain (&info); -if (parent ==NULL ) - parent = x86_vector_domain; -else - hpet_msi_controller.name ="IR-HPET-MSI" ; - fn =irq_domain_alloc_named_id_fwnode (hpet_msi_controller.name, hpet_id);if (!fn) {
@@ -572,6 +563,18 @@ static struct irq_domain *hpet_create_irq_domain(int hpet_id)return NULL ; } + fwspec.fwnode = fn; + fwspec.param_count =1 ; + fwspec.param[0 ] = hpet_id; + parent =irq_find_matching_fwspec (&fwspec,DOMAIN_BUS_ANY ); +if (!parent) { +irq_domain_free_fwnode (fn); +kfree (domain_info); +return NULL ; + } +if (parent != x86_vector_domain) + hpet_msi_controller.name ="IR-HPET-MSI" ; + d =msi_create_irq_domain (fn, domain_info, parent);if (!d) {irq_domain_free_fwnode (fn);
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 7/9 x86/ioapic: Use irq_find_matching_fwspec() to find remapping irqdomain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- arch/x86/kernel/apic/io_apic.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
arch/x86/kernel/apic/io_apic.c +12 -12
@@ -2305,36 +2305,36 @@ static inline void __init check_timer(void)static int mp_irqdomain_create (int ioapic) { -struct irq_alloc_info info;struct irq_domain *parent;int hwirqs =mp_ioapic_pin_count (ioapic);struct ioapic *ip = &ioapics[ioapic];struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;struct mp_ioapic_gsi *gsi_cfg =mp_ioapic_gsi_routing (ioapic);struct fwnode_handle *fn; -char *name ="IO-APIC" ; +struct irq_fwspec fwspec;if (cfg->type ==IOAPIC_DOMAIN_INVALID )return 0 ; -init_irq_alloc_info (&info,NULL ); - info.type =X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT ; - info.devid =mpc_ioapic_id (ioapic); - parent =irq_remapping_get_irq_domain (&info); -if (!parent) - parent = x86_vector_domain; -else - name ="IO-APIC-IR" ; -/* Handle device tree enumerated APICs proper */ if (cfg->dev) { fn =of_node_to_fwnode (cfg->dev); }else { - fn =irq_domain_alloc_named_id_fwnode (name, ioapic); + fn =irq_domain_alloc_named_id_fwnode ("IO-APIC" , ioapic);if (!fn)return -ENOMEM ; } + fwspec.fwnode = fn; + fwspec.param_count =1 ; + fwspec.param[0 ] = ioapic; + parent =irq_find_matching_fwspec (&fwspec,DOMAIN_BUS_ANY ); +if (!parent) { +if (!cfg->dev) +irq_domain_free_fwnode (fn); +return -ENODEV ; + } + ip->irqdomain =irq_domain_create_linear (fn, hwirqs, cfg->ops, (void *)(long )ioapic);
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 8/9 x86: Kill all traces of irq_remapping_get_irq_domain()
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
--- arch/x86/include/asm/hw_irq.h | 2 -- arch/x86/include/asm/irq_remapping.h | 9 -------- drivers/iommu/amd/iommu.c | 34 ---------------------------- drivers/iommu/hyperv-iommu.c | 9 -------- drivers/iommu/intel/irq_remapping.c | 17 -------------- drivers/iommu/irq_remapping.c | 14 ------------ drivers/iommu/irq_remapping.h | 3 --- 7 files changed, 88 deletions(-)
arch/x86/include/asm/hw_irq.h +0 -2
@@ -40,8 +40,6 @@ enum irq_alloc_type {X86_IRQ_ALLOC_TYPE_PCI_MSIX ,X86_IRQ_ALLOC_TYPE_DMAR , X86_IRQ_ALLOC_TYPE_UV, -X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT , - X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT, }; struct ioapic_alloc_info {
arch/x86/include/asm/irq_remapping.h +0 -9
@@ -44,9 +44,6 @@ extern int irq_remapping_reenable(int);extern int irq_remap_enable_fault_handling (void );extern void panic_if_irq_remap (const char *msg); -extern struct irq_domain * -irq_remapping_get_irq_domain (struct irq_alloc_info *info); -/* Create PCI MSI/MSIx irqdomain, use @parent as the parent irqdomain. */ extern struct irq_domain *arch_create_remap_msi_irq_domain (struct irq_domain *par,const char *n,int id);
@@ -71,11 +68,5 @@ static inline void panic_if_irq_remap(const char *msg) { } -static inline struct irq_domain * -irq_remapping_get_irq_domain (struct irq_alloc_info *info) -{ -return NULL ; -} -#endif /* CONFIG_IRQ_REMAP */ #endif /* __X86_IRQ_REMAPPING_H */
drivers/iommu/amd/iommu.c +0 -34
@@ -3536,10 +3536,8 @@ static int get_devid(struct irq_alloc_info *info) {switch (info->type) {case X86_IRQ_ALLOC_TYPE_IOAPIC : -case X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT :return get_ioapic_devid (info->devid);case X86_IRQ_ALLOC_TYPE_HPET : -case X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT :return get_hpet_devid (info->devid);case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX :
@@ -3550,44 +3548,12 @@ static int get_devid(struct irq_alloc_info *info) } } -static struct irq_domain *get_irq_domain_for_devid (struct irq_alloc_info *info, -int devid) -{ -struct amd_iommu *iommu = amd_iommu_rlookup_table[devid]; - -if (!iommu) -return NULL ; - -switch (info->type) { -case X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT : -case X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT : -return iommu->ir_domain; -default : -WARN_ON_ONCE (1 ); -return NULL ; - } -} - -static struct irq_domain *get_irq_domain (struct irq_alloc_info *info) -{ -int devid; - -if (!info) -return NULL ; - - devid =get_devid (info); -if (devid <0 ) -return NULL ; -return get_irq_domain_for_devid (info, devid); -} -struct irq_remap_ops amd_iommu_irq_ops = { .prepare = amd_iommu_prepare, .enable = amd_iommu_enable, .disable = amd_iommu_disable, .reenable = amd_iommu_reenable, .enable_faulting = amd_iommu_enable_faulting, - .get_irq_domain = get_irq_domain, }; static int irq_remapping_select(struct irq_domain *d, struct irq_fwspec *fwspec,
drivers/iommu/hyperv-iommu.c +0 -9
@@ -160,18 +160,9 @@ static int __init hyperv_enable_irq_remapping(void) return IRQ_REMAP_X2APIC_MODE; } -static struct irq_domain *hyperv_get_irq_domain (struct irq_alloc_info *info) -{ -if (info->type ==X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT ) -return ioapic_ir_domain; -else -return NULL ; -} -struct irq_remap_ops hyperv_irq_remap_ops = { .prepare = hyperv_prepare_irq_remapping, .enable = hyperv_enable_irq_remapping, - .get_irq_domain = hyperv_get_irq_domain, };#endif
drivers/iommu/intel/irq_remapping.c +0 -17
@@ -1128,29 +1128,12 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest) irte->redir_hint =1 ; } -static struct irq_domain *intel_get_irq_domain (struct irq_alloc_info *info) -{ -if (!info) -return NULL ; - -switch (info->type) { -case X86_IRQ_ALLOC_TYPE_IOAPIC_GET_PARENT : -return map_ioapic_to_ir (info->devid); -case X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT : -return map_hpet_to_ir (info->devid); -default : -WARN_ON_ONCE (1 ); -return NULL ; - } -} -struct irq_remap_ops intel_irq_remap_ops = { .prepare = intel_prepare_irq_remapping, .enable = intel_enable_irq_remapping, .disable = disable_irq_remapping, .reenable = reenable_irq_remapping, .enable_faulting = enable_drhd_fault_handling, - .get_irq_domain = intel_get_irq_domain, };static void intel_ir_reconfigure_irte (struct irq_data *irqd,bool force)
drivers/iommu/irq_remapping.c +0 -14
@@ -158,17 +158,3 @@ void panic_if_irq_remap(const char *msg)if (irq_remapping_enabled)panic (msg); } - -/** -* irq_remapping_get_irq_domain - Get the irqdomain serving the request @info -* @info: interrupt allocation information, used to identify the IOMMU device -* -* Returns pointer to IRQ domain, or NULL on failure. -*/ -struct irq_domain *irq_remapping_get_irq_domain (struct irq_alloc_info *info) -{ -if (!remap_ops || !remap_ops->get_irq_domain) -return NULL ; - -return remap_ops->get_irq_domain (info); -}
drivers/iommu/irq_remapping.h +0 -3
@@ -42,9 +42,6 @@ struct irq_remap_ops {/* Enable fault handling */ int (*enable_faulting)(void ); - -/* Get the irqdomain associated to IOMMU device */ -struct irq_domain *(*get_irq_domain)(struct irq_alloc_info *); };extern struct irq_remap_ops intel_irq_remap_ops;
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH 9/9 iommu/vt-d: Simplify intel_irq_remapping_select()
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
Now that the old get_irq_domain() method has gone, we can consolidate on just the map_XXX_to_iommu() functions.
--- drivers/iommu/intel/irq_remapping.c | 19 +++++++------------ 1 file changed, 7 insertions(+), 12 deletions(-)
drivers/iommu/intel/irq_remapping.c +7 -12
@@ -204,13 +204,13 @@ static int modify_irte(struct irq_2_iommu *irq_iommu,return rc; } -static struct irq_domain *map_hpet_to_ir (u8 hpet_id) +static struct intel_iommu *map_hpet_to_iommu (u8 hpet_id) {int i;for (i =0 ; i <MAX_HPET_TBS ; i++) {if (ir_hpet[i].id == hpet_id && ir_hpet[i].iommu) -return ir_hpet[i].iommu->ir_domain; +return ir_hpet[i].iommu; }return NULL ; }
@@ -226,13 +226,6 @@ static struct intel_iommu *map_ioapic_to_iommu(int apic)return NULL ; } -static struct irq_domain *map_ioapic_to_ir (int apic) -{ -struct intel_iommu *iommu =map_ioapic_to_iommu (apic); - -return iommu ? iommu->ir_domain :NULL ; -} -static struct irq_domain *map_dev_to_ir (struct pci_dev *dev) {struct dmar_drhd_unit *drhd =dmar_find_matched_drhd_unit (dev);
@@ -1422,12 +1415,14 @@ static int intel_irq_remapping_select(struct irq_domain *d,struct irq_fwspec *fwspec, enum irq_domain_bus_token bus_token) { + struct intel_iommu *iommu = NULL; +if (x86_fwspec_is_ioapic (fwspec)) - return d == map_ioapic_to_ir(fwspec->param[0]); + iommu =map_ioapic_to_iommu (fwspec->param[0 ]);else if (x86_fwspec_is_hpet (fwspec)) - return d == map_hpet_to_ir(fwspec->param[0]); + iommu =map_hpet_to_iommu (fwspec->param[0 ]); -return 0 ; +return iommu && d == iommu->ir_domain; } static const struct irq_domain_ops intel_ir_domain_ops = {
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
In reply to David WoodhouseQuoted 8 lines · David Woodhouse
On Tue, 2020-10-13 at 00:13 +0200, Thomas Gleixner wrote:
arch/x86/kernel/apic/vector.c +6
-0+ dom =irq_find_matching_fwspec (fwspec,DOMAIN_BUS_IR ); +if (dom) +return IS_ERR (dom) ?NULL : dom; + +return x86_vector_domain; +}
Quoted 5 lines
Ick. There's no need for that.
Eliminating that awful "if not found then slip the
x86_vector_domainin as a special case" was the whole *point* of usingirq_find_matching_fwspec()in the first place.
The point was to get rid of irq_remapping_get_irq_domain().
And TBH,
if (apicid_valid(32768))
is just another way to slip the vector domain in. It's just differently awful.
Having an explicit answer from the search for IR:
- Here is the domain
- Your device is not registered properly
- IR not enabled or not supported
is way more obvious than the above disguised is_remapping_enabled() check.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 9 lines · Thomas Gleixner
Quoted 8 lines · David Woodhouse
On Tue, 2020-10-13 at 00:13 +0200, Thomas Gleixner wrote:
arch/x86/kernel/apic/vector.c +6
-0+ dom =irq_find_matching_fwspec (fwspec,DOMAIN_BUS_IR ); +if (dom) +return IS_ERR (dom) ?NULL : dom; + +return x86_vector_domain; +}
Quoted 7 lines
Quoted 5 lines
Ick. There's no need for that.
Eliminating that awful "if not found then slip the
x86_vector_domainin as a special case" was the whole *point* of usingirq_find_matching_fwspec()in the first place.The point was to get rid of
irq_remapping_get_irq_domain().
My reason for doing it was to get rid of irq_remapping_get_irq_domain() *because* I hated the special-casing and magical slipping in of x86_vector_domain when it returned NULL.
Quoted 6 lines
And TBH,
if (apicid_valid(32768))is just another way to slip the vector domain in. It's just differently awful.
For me, that's very much not just "slipping the vector domain in". That's the vector domain returning true in its *own* ->select() function, in the circumstances where it wants to be used.
The key difference is that nobody needs an external magic pointer to the x86_vector_domain. In a true irqdomain hierarchy system, shouldn't we be trying to eliminate *all* those magic pointers to specific domains, if we can?
And sure, the apicid_valid(32768) as a proxy for irq_remapping_enabled is a bit of an ugly trick. I suppose we can explicitly expose irq_remapping_enabled from drivers/iommu if we wanted to.
Quoted 8 lines
Having an explicit answer from the search for IR:
- Here is the domain
- Your device is not registered properly
- IR not enabled or not supportedis way more obvious than the above disguised
is_remapping_enabled()check.
I just don't even like thinking of it as a 'search for IR'.
HPET shouldn't be caring about IR any more than PCI devices do. It just wants its parent irqdomain, that's all.
For I/OAPIC there's the slight complexity that it does actually ack level-triggered interrupts differently when it's behind IR. But I don't think we need a whole separate irq_chip for that; surely it could be handled internally in ioapic_ack_level() ?
Either way, even with that slight hack it's nicer to think of mp_irqdomain_create() just wanting to find its parent domain, without any special knowledge of IR and falling back to x86_parent_domain. The hack for IR level-ack is then self-contained.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
In reply to Thomas GleixnerQuoted 24 lines · Thomas Gleixner
Quoted 8 lines · David Woodhouse
On Tue, 2020-10-13 at 00:13 +0200, Thomas Gleixner wrote:
arch/x86/kernel/apic/vector.c +6
-0+ dom =irq_find_matching_fwspec (fwspec,DOMAIN_BUS_IR ); +if (dom) +return IS_ERR (dom) ?NULL : dom; + +return x86_vector_domain; +}Quoted 5 lines
Ick. There's no need for that.
Eliminating that awful "if not found then slip the
x86_vector_domainin as a special case" was the whole *point* of usingirq_find_matching_fwspec()in the first place.The point was to get rid of
irq_remapping_get_irq_domain().And TBH,
if (apicid_valid(32768))is just another way to slip the vector domain in. It's just differently awful.
Quoted 8 lines
Having an explicit answer from the search for IR:
- Here is the domain
- Your device is not registered properly
- IR not enabled or not supportedis way more obvious than the above disguised
is_remapping_enabled()check.
And after becoming more awake, that wont work anyway because there is more than one IR domain, so there is no way to return an error "You forgot to register" obviously.
But the APIC id (32768) valid check is also broken because IR can be enabled even without X2APIC.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 6 lines · Thomas Gleixner
And after becoming more awake, that wont work anyway because there is more than one IR domain, so there is no way to return an error "You forgot to register" obviously.
But the APIC id (32768) valid check is also broken because IR can be enabled even without X2APIC.
Nope, it's perfectly OK to allow HPET and I/OAPIC to be parented in the x86_vector_domain in that case, regardless of IR.
The *actual* criterion for x86_vector_select() returning zero to say "don't use me" is "could there be CPUs in the system which can't be reached through x86_vector_msi_compose_msg()?". It's not really about IR at all.
The apic_id_valid(32768) check is checking precisely the right thing.
David Woodhouse
maintainerRe: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 11 lines · David Woodhouse
Quoted 6 lines · Thomas Gleixner
And after becoming more awake, that wont work anyway because there is more than one IR domain, so there is no way to return an error "You forgot to register" obviously.
But the APIC id (32768) valid check is also broken because IR can be enabled even without X2APIC.
Nope, it's perfectly OK to allow HPET and I/OAPIC to be parented in the
x86_vector_domainin that case, regardless of IR.
Quoted 6 lines
The *actual* criterion for
x86_vector_select()returning zero to say "don't use me" is "could there be CPUs in the system which can't be reached throughx86_vector_msi_compose_msg()?". It's not really about IR at all.The
apic_id_valid(32768)check is checking precisely the right thing.
With that realisation, I've fixed the comment in my ext_dest_id branch to remove all mention of IRQ remapping. It now looks like this:
static int x86_vector_select(struct irq_domain *d, struct irq_fwspec *fwspec,
enum irq_domain_bus_token bus_token)
{
/*
* HPET and I/OAPIC drivers use irq_find_matching_irqdomain()
* to find their parent irqdomain. For x86_vector_domain to be
* suitable, all CPUs in the system must be reachable by its
* x86_vector_msi_compose_msg() function. Which is only true
* in !x2apic mode, or in x2apic physical mode if APIC IDs were
* restricted to 8 or 15 bits at boot time. In those cases,
* 1<<15 will *not* be a valid APIC ID.
*/
if (apic->apic_id_valid(1<<15))
return 0;
return x86_fwspec_is_ioapic(fwspec) || x86_fwspec_is_hpet(fwspec);
}
That makes it clearer that this isn't just some incestuous interaction with IRQ remapping — that APIC ID limit really is the basis on which this irqdomain, all by itself, makes the decision about whether it's capable of being the parent irqdomain to the requesting device.
It just so happens that *if* you allow higher APIC IDs in the system and *if* no other irqdomains exist which take ownership of a given HPET or I/OAPIC, then that child device will fail to find a suitable parent irqdomain and will fail to initialise. And that's just how we want it.
Sure, we have that special case in x2apic startup where we ensure that if we don't have IR then we limit the APIC IDs. But that's there, and *this* code is perfectly self-contained and isolated.
Re: [PATCH 5/5] x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
Quoted 18 lines · David Woodhouse
With that realisation, I've fixed the comment in my
ext_dest_idbranch to remove all mention of IRQ remapping. It now looks like this:static int x86_vector_select(struct
irq_domain*d, structirq_fwspec*fwspec,
enumirq_domain_bus_tokenbus_token)
{
/*
* HPET and I/OAPIC drivers useirq_find_matching_irqdomain()
* to find their parent irqdomain. Forx86_vector_domainto be
* suitable, all CPUs in the system must be reachable by its
*x86_vector_msi_compose_msg()function. Which is only true
* in !x2apic mode, or in x2apic physical mode if APIC IDs were
* restricted to 8 or 15 bits at boot time. In those cases,
* 1<<15 will *not* be a valid APIC ID.
*/
if (apic->apic_id_valid(1<<15))
return 0;
Quoted 7 lines
return
x86_fwspec_is_ioapic(fwspec)||x86_fwspec_is_hpet(fwspec);
}That makes it clearer that this isn't just some incestuous interaction with IRQ remapping — that APIC ID limit really is the basis on which this irqdomain, all by itself, makes the decision about whether it's capable of being the parent irqdomain to the requesting device.
Yes, that makes sense now.
Thanks,
tglx
Re: [PATCH 1/5] x86/apic: Fix x2apic enablement without interrupt remapping
In reply to David WoodhouseQuoted 4 lines · David Woodhouse
From: David Woodhouse <[email protected]>
static struct apic
apic_x2apic_phys;
+static u32x2apic_max_apicid;
__ro_after_init?
Thanks,
tglx
David Woodhouse
maintainerv2 cover Fix x2apic enablement and allow up to 32768 CPUs without IR where supported
In reply to David WoodhouseFix the conditions for enabling x2apic on guests without interrupt remapping, and support 15-bit Extended Destination ID to allow 32768 CPUs without IR on hypervisors that support it.
The last patch in the series now makes io_apic.c generate its RTE from the MSI message created by the parent irqchip, and removes all the nasty hackery where IRQ remapping drivers would frob I/OAPIC RTEs for themselves directly. It's last because I'd quite like to see it tested especially with Hyper-V and it doesn't actually eliminate the need for io_apic.c to know about the 15-bit extension anyway.
v2:
• Minor cleanups.
• Move __irq_msi_compose_msg() to apic.c, make virt_ext_dest_id static.
• Generate I/OAPIC RTE directly from parent irqchip's MSI messages.
• Clean up HPET MSI support into hpet.c now that we can.
David Woodhouse (8):
x86/apic: Fix x2apic enablement without interrupt remapping x86/msi: Only use high bits of MSI address for DMAR unit x86/apic: Always provide irq_compose_msi_msg() method for vector domain x86/ioapic: Handle Extended Destination ID field in RTE x86/apic: Support 15 bits of APIC ID in MSI where available x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
x86/hpet: Move MSI support into hpet.c
x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Documentation/virt/kvm/cpuid.rst | 4 + arch/x86/include/asm/apic.h | 9 +-- arch/x86/include/asm/hpet.h | 11 --- arch/x86/include/asm/hw_irq.h | 11 ++- arch/x86/include/asm/io_apic.h | 3 +- arch/x86/include/asm/msidef.h | 2 + arch/x86/include/asm/x86_init.h | 2 + arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/apic/apic.c | 68 ++++++++++++++-- arch/x86/kernel/apic/io_apic.c | 66 +++++++++------ arch/x86/kernel/apic/msi.c | 152 +++-------------------------------- arch/x86/kernel/apic/vector.c | 6 ++ arch/x86/kernel/apic/x2apic_phys.c | 9 +++ arch/x86/kernel/hpet.c | 116 ++++++++++++++++++++++++-- arch/x86/kernel/kvm.c | 6 ++ arch/x86/kernel/x86_init.c | 1 + drivers/iommu/amd/iommu.c | 14 ---- drivers/iommu/hyperv-iommu.c | 31 ------- drivers/iommu/intel/irq_remapping.c | 19 ++--- 19 files changed, 276 insertions(+), 255 deletions(-)
David Woodhouse
maintainerPATCH v2 1/8 x86/apic: Fix x2apic enablement without interrupt remapping
From: David Woodhouse <[email protected]>
Currently, Linux as a hypervisor guest will enable x2apic only if there are no CPUs present at boot time with an APIC ID above 255.
Hotplugging a CPU later with a higher APIC ID would result in a CPU which cannot be targeted by external interrupts.
Add a filter in x2apic_apic_id_valid() which can be used to prevent such CPUs from coming online, and allow x2apic to be enabled even if they are present at boot time.
--- arch/x86/include/asm/apic.h | 1 + arch/x86/kernel/apic/apic.c | 14 ++++++++------ arch/x86/kernel/apic/x2apic_phys.c | 9 +++++++++ 3 files changed, 18 insertions(+), 6 deletions(-)
arch/x86/include/asm/apic.h +1 -0
@@ -259,6 +259,7 @@ static inline u64 native_x2apic_icr_read(void)extern int x2apic_mode;extern int x2apic_phys; +extern void __init x2apic_set_max_apicid (u32 apicid);extern void __init check_x2apic (void );extern void x2apic_setup (void );static inline int x2apic_enabled (void )
arch/x86/kernel/apic/apic.c +8 -6
@@ -1841,20 +1841,22 @@ static __init void try_to_enable_x2apic(int remap_mode)return ;if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) { -/* IR is required if there is APIC ID > 255 even when running -* under KVM +/* +* Using X2APIC without IR is not architecturally supported +* on bare metal but may be supported in guests. */ -if (max_physical_apicid >255 || - !x86_init.hyper.x2apic_available ()) { +if (!x86_init.hyper.x2apic_available ()) {pr_info ("x2apic: IRQ remapping doesn't support X2APIC mode\n" );x2apic_disable ();return ; }/* -* without IR all CPUs can be addressed by IOAPIC/MSI -* only in physical mode +* Without IR, all CPUs can be addressed by IOAPIC/MSI only +* in physical mode, and CPUs with an APIC ID that cannnot +* be addressed must not be brought online. */ +x2apic_set_max_apicid (255 ); x2apic_phys =1 ; }x2apic_enable ();
arch/x86/kernel/apic/x2apic_phys.c +9 -0
@@ -8,6 +8,12 @@int x2apic_phys;static struct apic apic_x2apic_phys; +static u32 x2apic_max_apicid__ro_after_init ; + +void __init x2apic_set_max_apicid (u32 apicid) +{ + x2apic_max_apicid = apicid; +}static int __init set_x2apic_phys_mode (char *arg) {
@@ -98,6 +104,9 @@ static int x2apic_phys_probe(void)/* Common x2apic functions, also used by x2apic_cluster */ int x2apic_apic_id_valid (u32 apicid) { +if (x2apic_max_apicid && apicid > x2apic_max_apicid) +return 0 ; +return 1 ; }
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 2/8 x86/msi: Only use high bits of MSI address for DMAR unit
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
The Intel IOMMU has an MSI-like configuration for its interrupt, but it isn't really MSI. So it gets to abuse the high 32 bits of the address, and puts the high 24 bits of the extended APIC ID there.
This isn't something that can be used in the general case for real MSIs, since external devices using the high bits of the address would be performing writes to actual memory space above 4GiB, not targeted at the APIC.
Factor the hack out and allow it only to be used when appropriate, adding a WARN_ON_ONCE() if other MSIs are targeted at an unreachable APIC ID. That should never happen since the compatibility MSI messages are not used when Interrupt Remapping is enabled.
The x2apic_enabled() check isn't needed because Linux won't bring up CPUs with higher APIC IDs unless IR and x2apic are enabled anyway.
--- arch/x86/kernel/apic/msi.c | 33 +++++++++++++++++++++++++++------ 1 file changed, 27 insertions(+), 6 deletions(-)
arch/x86/kernel/apic/msi.c +27 -6
@@ -23,13 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg) +static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) { msg->address_hi =MSI_ADDR_BASE_HI ; -if (x2apic_enabled ()) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); - msg->address_lo = MSI_ADDR_BASE_LO | ((apic->irq_dest_mode == 0) ?
@@ -43,18 +41,29 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)MSI_DATA_LEVEL_ASSERT |MSI_DATA_DELIVERY_FIXED |MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) { -__irq_msi_compose_msg (irqd_cfg (data), msg); +__irq_msi_compose_msg (irqd_cfg (data), msg, false); }static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg); +__irq_msi_compose_msg (cfg, msg, false);irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); }
@@ -276,6 +285,17 @@ struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,#endif #ifdef CONFIG_DMAR_TABLE +/* +* The Intel IOMMU (ab)uses the high bits of the MSI address to contain the +* high bits of the destination APIC ID. This can't be done in the general +* case for MSIs as it would be targeting real memory above 4GiB not the +* APIC. +*/ +static void dmar_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, true); +} +static void dmar_msi_write_msg (struct irq_data *data,struct msi_msg *msg) {dmar_msi_write (data->irq, msg);
@@ -288,6 +308,7 @@ static struct irq_chip dmar_msi_controller = { .irq_ack = irq_chip_ack_parent, .irq_set_affinity = msi_domain_set_affinity, .irq_retrigger = irq_chip_retrigger_hierarchy, + .irq_compose_msi_msg = dmar_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, .flags = IRQCHIP_SKIP_SET_WAKE, };
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 3/8 x86/apic: Always provide irq_compose_msi_msg() method for vector domain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This shouldn't be dependent on PCI_MSI. HPET and I/OAPIC can deliver interrupts through MSI without having any PCI in the system at all.
--- arch/x86/include/asm/apic.h | 8 +++----- arch/x86/kernel/apic/apic.c | 32 +++++++++++++++++++++++++++++++ arch/x86/kernel/apic/msi.c | 36 ----------------------------------- arch/x86/kernel/apic/vector.c | 6 ++++++ 4 files changed, 41 insertions(+), 41 deletions(-)
arch/x86/include/asm/apic.h +3 -5
@@ -521,12 +521,10 @@ static inline void apic_smt_update(void) { }#endif struct msi_msg ; +struct irq_cfg ; -#ifdef CONFIG_PCI_MSI -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg); -#else -# define x86_vector_msi_compose_msg NULL -#endif +extern void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar);extern void ioapic_zap_locks (void );
arch/x86/kernel/apic/apic.c +32 -0
@@ -50,6 +50,7 @@#include <asm/io_apic.h> #include <asm/desc.h> #include <asm/hpet.h> +#include <asm/msidef.h> #include <asm/mtrr.h> #include <asm/time.h> #include <asm/smp.h>
@@ -2480,6 +2481,37 @@ int hard_smp_processor_id(void)return read_apic_id (); } +void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) +{ + msg->address_hi =MSI_ADDR_BASE_HI ; + + msg->address_lo = +MSI_ADDR_BASE_LO | + ((apic->irq_dest_mode ==0 ) ? +MSI_ADDR_DEST_MODE_PHYSICAL : +MSI_ADDR_DEST_MODE_LOGICAL ) | +MSI_ADDR_REDIRECTION_CPU | +MSI_ADDR_DEST_ID (cfg->dest_apicid); + + msg->data = +MSI_DATA_TRIGGER_EDGE | +MSI_DATA_LEVEL_ASSERT | +MSI_DATA_DELIVERY_FIXED | +MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); +} +/* * Override the generic EOI implementation with an optimized version. * Only called during early boot when only one CPU is active and with
arch/x86/kernel/apic/msi.c +0 -36
@@ -23,42 +23,6 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, -bool dmar) -{ - msg->address_hi =MSI_ADDR_BASE_HI ; - - msg->address_lo = -MSI_ADDR_BASE_LO | - ((apic->irq_dest_mode ==0 ) ? -MSI_ADDR_DEST_MODE_PHYSICAL : -MSI_ADDR_DEST_MODE_LOGICAL ) | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (cfg->dest_apicid); - - msg->data = -MSI_DATA_TRIGGER_EDGE | -MSI_DATA_LEVEL_ASSERT | -MSI_DATA_DELIVERY_FIXED | -MSI_DATA_VECTOR (cfg->vector); - -/* -* Only the IOMMU itself can use the trick of putting destination -* APIC ID into the high bits of the address. Anything else would -* just be writing to memory if it tried that, and needs IR to -* address higher APIC IDs. -*/ -if (dmar) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); -else -WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); -} - -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) -{ -__irq_msi_compose_msg (irqd_cfg (data), msg, false); -} -static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, };
arch/x86/kernel/apic/vector.c +6 -0
@@ -818,6 +818,12 @@ void apic_ack_edge(struct irq_data *irqd)apic_ack_irq (irqd); } +static void x86_vector_msi_compose_msg (struct irq_data *data, +struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, false); +} +static struct irq_chip lapic_controller = { .name ="APIC" , .irq_ack = apic_ack_edge,
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 4/8 x86/ioapic: Handle Extended Destination ID field in RTE
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
Bits 63-48 of the I/OAPIC Redirection Table Entry map directly to bits 19-4 of the address used in the resulting MSI cycle.
Historically, the x86 MSI format only used the top 8 of those 16 bits as the destination APIC ID, and the "Extended Destination ID" in the lower 8 bits was unused.
With interrupt remapping, the lowest bit of the Extended Destination ID (bit 48 of RTE, bit 4 of MSI address) is now used to indicate a remappable format MSI.
A hypervisor can use the other 7 bits of the Extended Destination ID to permit guests to address up to 15 bits of APIC IDs, thus allowing 32768 vCPUs before having to expose a vIOMMU and interrupt remapping to the guest.
This enlightenment could theoretically be transparent to the I/OAPIC code if it were always generating its RTE from an MSI message created by the parent irqchip. That cleanup will happen separately but doesn't cover all cases — for the ExtINT hackery and restoring boot mode, RTEs are still generated locally. So we have to teach the I/OAPIC about the ext_dest bits anyway.
No behavioural change in this patch, since nothing yet permits APIC IDs above 255 to be used with the non-IR I/OAPIC domain.
--- arch/x86/include/asm/io_apic.h | 3 ++- arch/x86/kernel/apic/io_apic.c | 19 +++++++++++++------ 2 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/io_apic.h +2 -1
@@ -78,7 +78,8 @@ struct IO_APIC_route_entry { mask :1 ,/* 0: enabled, 1: disabled */ __reserved_2 : 15; - __u32 __reserved_3 : 24, + __u32 __reserved_3 : 17, + virt_ext_dest : 7, dest : 8; } __attribute__ ((packed));
arch/x86/kernel/apic/io_apic.c +13 -6
@@ -1239,10 +1239,10 @@ static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries) buf, (ir_entry->index2 <<15 ) | ir_entry->index, ir_entry->zero); else - printk(KERN_DEBUG "%s, %s, D(%02X), M(%1d)\n", + printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf, entry.dest_mode ==IOAPIC_DEST_MODE_LOGICAL ? -"logical " :"physical" , +"logical " :"physical" , entry.virt_ext_dest, entry.dest, entry.delivery_mode); } }
@@ -1410,6 +1410,7 @@ void native_restore_boot_irq_mode(void)*/ if (ioapic_i8259.pin !=-1 ) {struct IO_APIC_route_entry entry; +u32 apic_id =read_apic_id ();memset (&entry,0 ,sizeof (entry)); entry.mask =IOAPIC_UNMASKED ;
@@ -1417,7 +1418,8 @@ void native_restore_boot_irq_mode(void) entry.polarity =IOAPIC_POL_HIGH ; entry.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry.delivery_mode = dest_ExtINT; - entry.dest =read_apic_id (); + entry.dest = apic_id &0xff ; + entry.virt_ext_dest = apic_id >>8 ;/* * Add it to the IO-APIC irq-routing table:
@@ -1861,7 +1863,8 @@ static void ioapic_configure_entry(struct irq_data *irqd)* ioapic chip to verify that. */ if (irqd->chip == &ioapic_chip) { - mpd->entry.dest = cfg->dest_apicid; + mpd->entry.dest = cfg->dest_apicid &0xff ; + mpd->entry.virt_ext_dest = cfg->dest_apicid >>8 ; mpd->entry.vector = cfg->vector; }for_each_irq_pin (entry, mpd->irq_2_pin)
@@ -2027,6 +2030,7 @@ static inline void __init unlock_ExtINT_logic(void)int apic, pin, i;struct IO_APIC_route_entry entry0, entry1;unsigned char save_control, save_freq_select; +u32 apic_id; pin =find_isa_irq_pin (8 , mp_INT);if (pin ==-1 ) {
@@ -2042,11 +2046,13 @@ static inline void __init unlock_ExtINT_logic(void) entry0 =ioapic_read_entry (apic, pin);clear_IO_APIC_pin (apic, pin); + apic_id =hard_smp_processor_id ();memset (&entry1,0 ,sizeof (entry1)); entry1.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry1.mask =IOAPIC_UNMASKED ; - entry1.dest =hard_smp_processor_id (); + entry1.dest = apic_id &0xff ; + entry1.virt_ext_dest = apic_id >>8 ; entry1.delivery_mode = dest_ExtINT; entry1.polarity = entry0.polarity; entry1.trigger =IOAPIC_EDGE ;
@@ -2949,7 +2955,8 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data,memset (entry,0 ,sizeof (*entry)); entry->delivery_mode = apic->irq_delivery_mode; entry->dest_mode = apic->irq_dest_mode; - entry->dest = cfg->dest_apicid; + entry->dest = cfg->dest_apicid &0xff ; + entry->virt_ext_dest = cfg->dest_apicid >>8 ; entry->vector = cfg->vector; entry->trigger = data->trigger; entry->polarity = data->polarity;
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 5/8 x86/apic: Support 15 bits of APIC ID in MSI where available
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
Some hypervisors can allow the guest to use the Extended Destination ID field in the MSI address to address up to 32768 CPUs.
This applies to all downstream devices which generate MSI cycles, including HPET, I/OAPIC and PCI MSI.
HPET and PCI MSI use the same __irq_msi_compose_msg() function, while I/OAPIC generates its own and had support for the extended bits added in a previous commit.
--- arch/x86/include/asm/x86_init.h | 2 ++ arch/x86/kernel/apic/apic.c | 26 ++++++++++++++++++++++++-- arch/x86/kernel/x86_init.c | 1 + 3 files changed, 27 insertions(+), 2 deletions(-)
arch/x86/include/asm/x86_init.h +2 -0
@@ -114,6 +114,7 @@ struct x86_init_pci {* @init_platform: platform setup * @guest_late_init: guest late init * @x2apic_available: X2APIC detection +* @msi_ext_dest_id: MSI supports 15-bit APIC IDs * @init_mem_mapping: setup early mappings during init_mem_mapping() * @init_after_bootmem: guest init after boot allocator is finished */
@@ -121,6 +122,7 @@ struct x86_hyper_init {void (*init_platform)(void );void (*guest_late_init)(void );bool (*x2apic_available)(void ); +bool (*msi_ext_dest_id)(void );void (*init_mem_mapping)(void );void (*init_after_bootmem)(void ); };
arch/x86/kernel/apic/apic.c +24 -2
@@ -94,6 +94,11 @@ static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;*/ static int apic_extnmi__ro_after_init =APIC_EXTNMI_BSP ; +/* +* Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID +*/ +static bool virt_ext_dest_id__ro_after_init ; +/* * Map cpu index to physical APIC ID */
@@ -1842,6 +1847,8 @@ static __init void try_to_enable_x2apic(int remap_mode)return ;if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) { +u32 apic_limit =255 ; +/* * Using X2APIC without IR is not architecturally supported * on bare metal but may be supported in guests.
@@ -1852,12 +1859,22 @@ static __init void try_to_enable_x2apic(int remap_mode)return ; } +/* +* If the hypervisor supports extended destination ID in +* MSI, that increases the maximum APIC ID that can be +* used for non-remapped IRQ domains. +*/ +if (x86_init.hyper.msi_ext_dest_id ()) { + virt_ext_dest_id =1 ; + apic_limit =32767 ; + } +/* * Without IR, all CPUs can be addressed by IOAPIC/MSI only * in physical mode, and CPUs with an APIC ID that cannnot * be addressed must not be brought online. */ -x2apic_set_max_apicid (255 ); +x2apic_set_max_apicid (apic_limit); x2apic_phys =1 ; }x2apic_enable ();
@@ -2504,10 +2521,15 @@ void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,* Only the IOMMU itself can use the trick of putting destination * APIC ID into the high bits of the address. Anything else would * just be writing to memory if it tried that, and needs IR to -* address higher APIC IDs. +* address APICs which can't be addressed in the normal 32-bit +* address range at 0xFFExxxxx. That is typically just 8 bits, but +* some hypervisors allow the extended destination ID field in bits +* 11-5 to be used, giving support for 15 bits of APIC IDs in total. */ if (dmar) msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else if (virt_ext_dest_id && cfg->dest_apicid <0x8000 ) + msg->address_lo |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid) >>3 ;else WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }
arch/x86/kernel/x86_init.c +1 -0
@@ -110,6 +110,7 @@ struct x86_init_ops x86_init __initdata = { .init_platform = x86_init_noop, .guest_late_init = x86_init_noop, .x2apic_available = bool_x86_init_noop, + .msi_ext_dest_id = bool_x86_init_noop, .init_mem_mapping = x86_init_noop, .init_after_bootmem = x86_init_noop, },
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 6/8 x86/kvm: Add KVM_FEATURE_MSI_EXT_DEST_ID
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This allows the host to indicate that MSI emulation supports 15-bit destination IDs, allowing up to 32768 CPUs without interrupt remapping.
cf. https://patchwork.kernel.org/patch/11816693/ for qemu
--- Documentation/virt/kvm/cpuid.rst | 4 ++++ arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/kvm.c | 6 ++++++ 3 files changed, 11 insertions(+)
Documentation/virt/kvm/cpuid.rst +4 -0
@@ -92,6 +92,10 @@ KVM_FEATURE_ASYNC_PF_INT 14 guest checks this feature bit async pf acknowledgment msr 0x4b564d07. +KVM_FEATURE_MSI_EXT_DEST_ID 15 guest checks this feature bit + before using extended destination + ID bits in MSI address bits 11-5. + KVM_FEATURE_CLOCSOURCE_STABLE_BIT 24 host will warn if no guest-side per-cpu warps are expeced in kvmclock
arch/x86/include/uapi/asm/kvm_para.h +1 -0
@@ -32,6 +32,7 @@#define KVM_FEATURE_POLL_CONTROL 12#define KVM_FEATURE_PV_SCHED_YIELD 13#define KVM_FEATURE_ASYNC_PF_INT 14 +#define KVM_FEATURE_MSI_EXT_DEST_ID 15#define KVM_HINTS_REALTIME 0
arch/x86/kernel/kvm.c +6 -0
@@ -743,12 +743,18 @@ static void __init kvm_init_platform(void) x86_platform.apic_post_init = kvm_apic_init; } +static bool __init kvm_msi_ext_dest_id (void ) +{ +return kvm_para_has_feature (KVM_FEATURE_MSI_EXT_DEST_ID ); +} +const __initconst struct hypervisor_x86 x86_hyper_kvm = { .name ="KVM" , .detect = kvm_detect, .type =X86_HYPER_KVM , .init.guest_late_init = kvm_guest_init, .init.x2apic_available = kvm_para_available, + .init.msi_ext_dest_id = kvm_msi_ext_dest_id, .init.init_platform = kvm_init_platform, };
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 7/8 x86/hpet: Move MSI support into hpet.c
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This isn't really dependent on PCI MSI; it's just generic MSI which is now supported by the generic x86_vector_domain. Move the HPET MSI support back into hpet.c with the rest of the HPET support.
--- arch/x86/include/asm/hpet.h | 11 ---- arch/x86/kernel/apic/msi.c | 111 ---------------------------------- arch/x86/kernel/hpet.c | 116 ++++++++++++++++++++++++++++++++++-- 3 files changed, 111 insertions(+), 127 deletions(-)
arch/x86/include/asm/hpet.h +0 -11
@@ -74,17 +74,6 @@ extern void hpet_disable(void);extern unsigned int hpet_readl (unsigned int a);extern void force_hpet_resume (void ); -struct irq_data ; -struct hpet_channel ; -struct irq_domain ; - -extern void hpet_msi_unmask (struct irq_data *data); -extern void hpet_msi_mask (struct irq_data *data); -extern void hpet_msi_write (struct hpet_channel *hc,struct msi_msg *msg); -extern struct irq_domain *hpet_create_irq_domain (int hpet_id); -extern int hpet_assign_irq (struct irq_domain *domain, -struct hpet_channel *hc,int dev_num); -#ifdef CONFIG_HPET_EMULATE_RTC #include <linux/interrupt.h>
arch/x86/kernel/apic/msi.c +0 -111
114 lines not on this page · raw message
arch/x86/kernel/hpet.c +111 -5
145 lines not on this page · raw message
Signature · 1 line
2.26.2
David Woodhouse
maintainerPATCH v2 8/8 x86/ioapic: Generate RTE directly from parent irqchip's MSI message
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
The I/OAPIC generates an MSI cycle with address/data bits taken from its Redirection Table Entry in some combination which used to make sense, but now is just a bunch of bits which get passed through in some seemingly arbitrary order.
Instead of making IRQ remapping drivers directly frob the I/OAPIC RTE, let them just do their job and generate an MSI message. The bit swizzling to turn that MSI message into the IOAPIC's RTE is the same in all cases, since it's a function of the I/OAPIC hardware. The IRQ remappers have no real need to get involved with that.
The only slight caveat is that the I/OAPIC is interpreting some of those fields too, and it does want the 'vector' field to be unique to make EOI work. The AMD IOMMU happens to put its IRTE index in the bits that the I/OAPIC thinks are the vector field, and accommodates this requirement by reserving the first 32 indices for the I/OAPIC. The Intel IOMMU doesn't actually use the bits that the I/OAPIC thinks are the vector field, so it fills in the 'pin' value there instead.
--- arch/x86/include/asm/hw_irq.h | 11 +++--- arch/x86/include/asm/msidef.h | 2 ++ arch/x86/kernel/apic/io_apic.c | 55 ++++++++++++++++++----------- drivers/iommu/amd/iommu.c | 14 -------- drivers/iommu/hyperv-iommu.c | 31 ---------------- drivers/iommu/intel/irq_remapping.c | 19 +++------- 6 files changed, 46 insertions(+), 86 deletions(-)
arch/x86/include/asm/hw_irq.h +5 -6
@@ -45,12 +45,11 @@ enum irq_alloc_type { };struct ioapic_alloc_info { -int pin; -int node; -u32 trigger :1 ; -u32 polarity :1 ; -u32 valid :1 ; -struct IO_APIC_route_entry *entry; +int pin; +int node; +u32 trigger :1 ; +u32 polarity :1 ; +u32 valid :1 ; };struct uv_alloc_info {
arch/x86/include/asm/msidef.h +2 -0
@@ -18,6 +18,7 @@#define MSI_DATA_DELIVERY_MODE_SHIFT 8#define MSI_DATA_DELIVERY_FIXED (0 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_DELIVERY_LOWPRI (1 << MSI_DATA_DELIVERY_MODE_SHIFT) +#define MSI_DATA_DELIVERY_MODE_MASK (3 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_LEVEL_SHIFT 14#define MSI_DATA_LEVEL_DEASSERT (0 << MSI_DATA_LEVEL_SHIFT)
@@ -37,6 +38,7 @@#define MSI_ADDR_DEST_MODE_SHIFT 2#define MSI_ADDR_DEST_MODE_PHYSICAL (0 << MSI_ADDR_DEST_MODE_SHIFT)#define MSI_ADDR_DEST_MODE_LOGICAL (1 << MSI_ADDR_DEST_MODE_SHIFT) +#define MSI_ADDR_DEST_MODE_MASK (1 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_ADDR_REDIRECTION_SHIFT 3#define MSI_ADDR_REDIRECTION_CPU (0 << MSI_ADDR_REDIRECTION_SHIFT)
arch/x86/kernel/apic/io_apic.c +34 -21
98 lines not on this page · raw message
drivers/iommu/amd/iommu.c +0 -14
@@ -3597,7 +3597,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, {struct irq_2_irte *irte_info = &data->irq_2_irte;struct msi_msg *msg = &data->msi_entry; -struct IO_APIC_route_entry *entry;struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];if (!iommu)
@@ -3611,19 +3610,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,switch (info->type) {case X86_IRQ_ALLOC_TYPE_IOAPIC : -/* Setup IOAPIC entry */ - entry = info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; -break ; -case X86_IRQ_ALLOC_TYPE_HPET :case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX :
drivers/iommu/hyperv-iommu.c +0 -31
@@ -40,7 +40,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data, {struct irq_data *parent = data->parent_data;struct irq_cfg *cfg =irqd_cfg (data); -struct IO_APIC_route_entry *entry;int ret;/* Return error If new irq affinity is out of ioapic_max_cpumask. */
@@ -51,9 +50,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data,if (ret <0 || ret ==IRQ_SET_MASK_OK_DONE )return ret; - entry = data->chip_data; - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector;send_cleanup_vector (cfg);return 0 ;
@@ -89,20 +85,6 @@ static int hyperv_irq_remapping_alloc(struct irq_domain *domain, irq_data->chip = &hyperv_ir_chip; -/* -* If there is interrupt remapping function of IOMMU, setting irq -* affinity only needs to change IRTE of IOMMU. But Hyper-V doesn't -* support interrupt remapping function, setting irq affinity of IO-APIC -* interrupts still needs to change IO-APIC registers. But ioapic_ -* configure_entry() will ignore value of cfg->vector and cfg-> -* dest_apicid when IO-APIC's parent irq domain is not the vector -* domain.(See ioapic_configure_entry()) In order to setting vector -* and dest_apicid to IO-APIC register, IO-APIC entry pointer is saved -* in the chip_data and hyperv_irq_remapping_activate()/hyperv_ir_set_ -* affinity() set vector and dest_apicid directly into IO-APIC entry. -*/ - irq_data->chip_data = info->ioapic.entry; -/* * Hypver-V IO APIC irq affinity should be in the scope of * ioapic_max_cpumask because no irq remapping support.
@@ -119,22 +101,9 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain,irq_domain_free_irqs_common (domain, virq, nr_irqs); } -static int hyperv_irq_remapping_activate (struct irq_domain *domain, -struct irq_data *irq_data,bool reserve) -{ -struct irq_cfg *cfg =irqd_cfg (irq_data); -struct IO_APIC_route_entry *entry = irq_data->chip_data; - - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector; - -return 0 ; -} -static const struct irq_domain_ops hyperv_ir_domain_ops = { .alloc = hyperv_irq_remapping_alloc, .free = hyperv_irq_remapping_free, - .activate = hyperv_irq_remapping_activate, };static int __init hyperv_prepare_irq_remapping (void )
drivers/iommu/intel/irq_remapping.c +5 -14
@@ -1265,7 +1265,6 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,struct irq_alloc_info *info, int index, int sub_handle) { - struct IR_IO_APIC_route_entry *entry; struct irte *irte = &data->irte_entry;struct msi_msg *msg = &data->msi_entry;
@@ -1281,23 +1280,15 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, irte->avail, irte->vector, irte->dest_id, irte->sid, irte->sq, irte->svt); - entry = (struct IR_IO_APIC_route_entry *)info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->index2 = (index >>15 ) &0x1 ; - entry->zero =0 ; - entry->format =1 ; - entry->index = (index &0x7fff );/* * IO-APIC RTE will be configured with virtual vector. * irq handler will do the explicit EOI to the io-apic. */ - entry->vector = info->ioapic.pin; - entry->mask =0 ;/* enable IRQ */ - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -if (info->ioapic.trigger) - entry->mask =1 ;/* Mask level triggered irqs. */ + msg->data = info->ioapic.pin; + msg->address_hi =MSI_ADDR_BASE_HI ; + msg->address_lo =MSI_ADDR_BASE_LO |MSI_ADDR_IR_EXT_INT | +MSI_ADDR_IR_INDEX1 (index) | +MSI_ADDR_IR_INDEX2 (index);break ;case X86_IRQ_ALLOC_TYPE_HPET :
Signature · 1 line
2.26.2
Re: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
On Fri, Oct 09 2020 at 11:46, David Woodhouse wrote:
@@ -45,12 +45,11 @@ enum irq_alloc_type {
};
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1859
Quoted 4 lines
arch/x86/kernel/apic/io_apic.c +4
-0+static void mp_swizzle_msi_dest_bits (struct irq_data *irq_data,void *_entry) +{ +struct msi_msg msg; +u32 *entry = _entry;
Why is this a void * argument and then converting it to a u32 *? Just to make that function completely unreadable?
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1861
Quoted 2 lines
arch/x86/kernel/apic/io_apic.c +2
-0+ +irq_chip_compose_msi_msg (irq_data, &msg);
Lacks a comment. Also mp_swizzle... is a misnomer as this invokes the msi compose function which is not what the function name suggests.
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1875
Quoted 13 lines
arch/x86/kernel/apic/io_apic.c +13
-0+/* +* They're in a bit of a random order for historical reasons, but +* the IO/APIC is just a device for turning interrupt lines into +* MSIs, and various bits of the MSI addr/data are just swizzled +* into/from the bits of Redirection Table Entry. +*/ + entry[0 ] &=0xfffff000 ; + entry[0 ] |= (msg.data & (MSI_DATA_DELIVERY_MODE_MASK | +MSI_DATA_VECTOR_MASK )); + entry[0 ] |= (msg.address_lo &MSI_ADDR_DEST_MODE_MASK ) <<9 ; + + entry[1 ] &=0xffff ; + entry[1 ] |= (msg.address_lo &MSI_ADDR_DEST_ID_MASK ) <<12 ;
Sorry, but this is unreviewable gunk. The whole msi_msg setup sucks with this unholy macro maze. I have a half finished series which allows architectures to provide shadow members for data, address_* so this can be done proper with bitfields.
Aside of that it works magically because polarity,trigger and mask bit have been set up before. But of course a comment about this is completely overrated.
Thanks,
tglx
Re: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 4 lines · Thomas Gleixner
On Fri, Oct 09 2020 at 11:46, David Woodhouse wrote:
Aside of that it works magically because polarity,trigger and mask bit have been set up before. But of course a comment about this is completely overrated.
Also this part:
Quoted 8 lines
arch/x86/kernel/apic/io_apic.c +3
-3-static void mp_setup_entry (struct irq_cfg *cfg,struct mp_chip_data *data, -struct IO_APIC_route_entry *entry) +static void mp_setup_entry (struct irq_data *irq_data,struct mp_chip_data *data) { +struct IO_APIC_route_entry *entry = &data->entry; +memset (entry,0 ,sizeof (*entry)); - entry->delivery_mode = apic->irq_delivery_mode;
arch/x86/kernel/apic/io_apic.c static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data, · line 2978
Quoted 10 lines
arch/x86/kernel/apic/io_apic.c +3
-4- entry->dest_mode = apic->irq_dest_mode; - entry->dest = cfg->dest_apicid &0xff ; - entry->virt_ext_dest = cfg->dest_apicid >>8 ; - entry->vector = cfg->vector; + +mp_swizzle_msi_dest_bits (irq_data, entry); + entry->trigger = data->trigger; entry->polarity = data->polarity;/*
does not make sense. It did not make sense before either, but now it does even make less sense.
During allocation this only needs to setup the I/O-APIC specific bits (trigger, polarity, mask). The rest is filled in when the actual activation happens. Nothing writes that entry _before_ activation.
/me goes to mop up more
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 24 lines · Thomas Gleixner
Quoted 10 lines
-static void mp_setup_entry(struct
irq_cfg*cfg, structmp_chip_data*data,(quoted diff) +1
-1- struct IO_APIC_route_entry *entry) +static void mp_setup_entry(struct irq_data *irq_data, struct
mp_chip_data*data)
{(quoted diff) +2
-1+ struct IO_APIC_route_entry *entry = &data->entry; + memset(entry, 0, sizeof(*entry)); - entry->delivery_mode = apic->irq_delivery_mode;Review
Quoted 10 lines
(quoted diff) +3
-4- entry->dest_mode = apic->irq_dest_mode; - entry->dest = cfg->dest_apicid & 0xff; - entry->virt_ext_dest = cfg->dest_apicid >> 8; - entry->vector = cfg->vector; + + mp_swizzle_msi_dest_bits(irq_data, entry); + entry->trigger = data->trigger; entry->polarity = data->polarity; /*does not make sense. It did not make sense before either, but now it does even make less sense.
Quoted 5 lines
During allocation this only needs to setup the I/O-APIC specific bits (trigger, polarity, mask). The rest is filled in when the actual activation happens. Nothing writes that entry _before_ activation.
/me goes to mop up more
Yeah... that code was indeed a pile of crap before I looked at it, wasn't it? And I indeed failed to spot it and mop it up as I touched it.
Here's the version I've just pushed to my tree, which I'll test properly both with and without IR over the weekend before posting v3.
There is no way that bit swizzling is every going to be anything short of fugly. That's just the reality of the hardware. Even doing it with bitfields is just going to be masking the issue by having structure definitions that don't actually match the I/OAPIC documentation. Better to be up front about it. I've added more words though; more words always help...
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/ext_dest_id
From f912b52996d381fd8a631dd10c713772c2ade478 Mon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Thu, 8 Oct 2020 15:44:42 +0100
Subject: [PATCH 10/19] x86/ioapic: Generate RTE directly from parent irqchip's
MSI message
The I/OAPIC generates an MSI cycle with address/data bits taken from its Redirection Table Entry in some combination which used to make sense, but now is just a bunch of bits which get passed through in some seemingly arbitrary order.
Instead of making IRQ remapping drivers directly frob the I/OAPIC RTE, let them just do their job and generate an MSI message. The bit swizzling to turn that MSI message into the IOAPIC's RTE is the same in all cases, since it's a function of the I/OAPIC hardware. The IRQ remappers have no real need to get involved with that.
The only slight caveat is that the I/OAPIC is interpreting some of those fields too, and it does want the 'vector' field to be unique to make EOI work. The AMD IOMMU happens to put its IRTE index in the bits that the I/OAPIC thinks are the vector field, and accommodates this requirement by reserving the first 32 indices for the I/OAPIC. The Intel IOMMU doesn't actually use the bits that the I/OAPIC thinks are the vector field, so it fills in the 'pin' value there instead.
--- arch/x86/include/asm/hw_irq.h | 11 ++-- arch/x86/include/asm/msidef.h | 2 + arch/x86/kernel/apic/io_apic.c | 81 ++++++++++++++++++++++------- drivers/iommu/amd/iommu.c | 14 ----- drivers/iommu/hyperv-iommu.c | 31 ----------- drivers/iommu/intel/irq_remapping.c | 19 ++----- 6 files changed, 74 insertions(+), 84 deletions(-)
arch/x86/include/asm/hw_irq.h +5 -6
@@ -45,12 +45,11 @@ enum irq_alloc_type { };struct ioapic_alloc_info { -int pin; -int node; -u32 trigger :1 ; -u32 polarity :1 ; -u32 valid :1 ; -struct IO_APIC_route_entry *entry; +int pin; +int node; +u32 trigger :1 ; +u32 polarity :1 ; +u32 valid :1 ; };struct uv_alloc_info {
arch/x86/include/asm/msidef.h +2 -0
@@ -18,6 +18,7 @@#define MSI_DATA_DELIVERY_MODE_SHIFT 8#define MSI_DATA_DELIVERY_FIXED (0 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_DELIVERY_LOWPRI (1 << MSI_DATA_DELIVERY_MODE_SHIFT) +#define MSI_DATA_DELIVERY_MODE_MASK (3 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_DATA_LEVEL_SHIFT 14#define MSI_DATA_LEVEL_DEASSERT (0 << MSI_DATA_LEVEL_SHIFT)
@@ -37,6 +38,7 @@#define MSI_ADDR_DEST_MODE_SHIFT 2#define MSI_ADDR_DEST_MODE_PHYSICAL (0 << MSI_ADDR_DEST_MODE_SHIFT)#define MSI_ADDR_DEST_MODE_LOGICAL (1 << MSI_ADDR_DEST_MODE_SHIFT) +#define MSI_ADDR_DEST_MODE_MASK (1 << MSI_DATA_DELIVERY_MODE_SHIFT)#define MSI_ADDR_REDIRECTION_SHIFT 3#define MSI_ADDR_REDIRECTION_CPU (0 << MSI_ADDR_REDIRECTION_SHIFT)
arch/x86/kernel/apic/io_apic.c +62 -19
126 lines not on this page · raw message
drivers/iommu/amd/iommu.c +0 -14
@@ -3597,7 +3597,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, {struct irq_2_irte *irte_info = &data->irq_2_irte;struct msi_msg *msg = &data->msi_entry; -struct IO_APIC_route_entry *entry;struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];if (!iommu)
@@ -3611,19 +3610,6 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,switch (info->type) {case X86_IRQ_ALLOC_TYPE_IOAPIC : -/* Setup IOAPIC entry */ - entry = info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; -break ; -case X86_IRQ_ALLOC_TYPE_HPET :case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX :
drivers/iommu/hyperv-iommu.c +0 -31
@@ -40,7 +40,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data, {struct irq_data *parent = data->parent_data;struct irq_cfg *cfg =irqd_cfg (data); -struct IO_APIC_route_entry *entry;int ret;/* Return error If new irq affinity is out of ioapic_max_cpumask. */
@@ -51,9 +50,6 @@ static int hyperv_ir_set_affinity(struct irq_data *data,if (ret <0 || ret ==IRQ_SET_MASK_OK_DONE )return ret; - entry = data->chip_data; - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector;send_cleanup_vector (cfg);return 0 ;
@@ -89,20 +85,6 @@ static int hyperv_irq_remapping_alloc(struct irq_domain *domain, irq_data->chip = &hyperv_ir_chip; -/* -* If there is interrupt remapping function of IOMMU, setting irq -* affinity only needs to change IRTE of IOMMU. But Hyper-V doesn't -* support interrupt remapping function, setting irq affinity of IO-APIC -* interrupts still needs to change IO-APIC registers. But ioapic_ -* configure_entry() will ignore value of cfg->vector and cfg-> -* dest_apicid when IO-APIC's parent irq domain is not the vector -* domain.(See ioapic_configure_entry()) In order to setting vector -* and dest_apicid to IO-APIC register, IO-APIC entry pointer is saved -* in the chip_data and hyperv_irq_remapping_activate()/hyperv_ir_set_ -* affinity() set vector and dest_apicid directly into IO-APIC entry. -*/ - irq_data->chip_data = info->ioapic.entry; -/* * Hypver-V IO APIC irq affinity should be in the scope of * ioapic_max_cpumask because no irq remapping support.
@@ -119,22 +101,9 @@ static void hyperv_irq_remapping_free(struct irq_domain *domain,irq_domain_free_irqs_common (domain, virq, nr_irqs); } -static int hyperv_irq_remapping_activate (struct irq_domain *domain, -struct irq_data *irq_data,bool reserve) -{ -struct irq_cfg *cfg =irqd_cfg (irq_data); -struct IO_APIC_route_entry *entry = irq_data->chip_data; - - entry->dest = cfg->dest_apicid; - entry->vector = cfg->vector; - -return 0 ; -} -static const struct irq_domain_ops hyperv_ir_domain_ops = { .alloc = hyperv_irq_remapping_alloc, .free = hyperv_irq_remapping_free, - .activate = hyperv_irq_remapping_activate, };static int __init hyperv_prepare_irq_remapping (void )
drivers/iommu/intel/irq_remapping.c +5 -14
@@ -1265,7 +1265,6 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,struct irq_alloc_info *info, int index, int sub_handle) { - struct IR_IO_APIC_route_entry *entry; struct irte *irte = &data->irte_entry;struct msi_msg *msg = &data->msi_entry;
@@ -1281,23 +1280,15 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, irte->avail, irte->vector, irte->dest_id, irte->sid, irte->sq, irte->svt); - entry = (struct IR_IO_APIC_route_entry *)info->ioapic.entry; - info->ioapic.entry =NULL ; -memset (entry,0 ,sizeof (*entry)); - entry->index2 = (index >>15 ) &0x1 ; - entry->zero =0 ; - entry->format =1 ; - entry->index = (index &0x7fff );/* * IO-APIC RTE will be configured with virtual vector. * irq handler will do the explicit EOI to the io-apic. */ - entry->vector = info->ioapic.pin; - entry->mask =0 ;/* enable IRQ */ - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -if (info->ioapic.trigger) - entry->mask =1 ;/* Mask level triggered irqs. */ + msg->data = info->ioapic.pin; + msg->address_hi =MSI_ADDR_BASE_HI ; + msg->address_lo =MSI_ADDR_BASE_LO |MSI_ADDR_IR_EXT_INT | +MSI_ADDR_IR_INDEX1 (index) | +MSI_ADDR_IR_INDEX2 (index);break ;case X86_IRQ_ALLOC_TYPE_HPET :
Signature · 1 line
2.17.1
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
In reply to Thomas GleixnerQuoted 6 lines · Thomas Gleixner
On Fri, Oct 09 2020 at 11:46, David Woodhouse wrote:
(quoted diff) +0
-0@@ -45,12 +45,11 @@ enum irq_alloc_type { };Quoted 1 line
+static void mp_swizzle_msi_dest_bits(struct
irq_data*irq_data, void
Quoted 8 lines
*_entry)arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1859
Quoted 3 lines
arch/x86/kernel/apic/io_apic.c +3
-0+{ +struct msi_msg msg; +u32 *entry = _entry;Why is this a void * argument and then converting it to a u32 *? Just to
make that function completely unreadable?
It makes the callers slightly more readable, not having to cast to uint32_t* from the struct.
I did ponder defining a new struct with bitfields named along the lines of 'msi_addr_bits_19_to_4', but that seemed like overkill.
Quoted 5 lines
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1861
Quoted 2 lines
arch/x86/kernel/apic/io_apic.c +2
-0+ +irq_chip_compose_msi_msg (irq_data, &msg);Lacks a comment. Also
mp_swizzle... is a misnomer as this invokes the msi compose function which is not what the function name suggests.
It's got a four-line comment right there after it as we use the bits we got from it.
Quoted 7 lines
Quoted 7 lines
arch/x86/kernel/apic/io_apic.c +7
-0+/* +* They're in a bit of a random order for historical reasons, but +* the IO/APIC is just a device for turning interrupt lines into +* MSIs, and various bits of the MSI addr/data are just swizzled +* into/from the bits of Redirection Table Entry. +*/ + entry[0 ] &=0xfffff000 ;
Quoted 8 lines
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1875
Quoted 6 lines
arch/x86/kernel/apic/io_apic.c +6
-0+ entry[0 ] |= (msg.data & (MSI_DATA_DELIVERY_MODE_MASK | +MSI_DATA_VECTOR_MASK )); + entry[0 ] |= (msg.address_lo &MSI_ADDR_DEST_MODE_MASK ) <<9 ; + + entry[1 ] &=0xffff ; + entry[1 ] |= (msg.address_lo &MSI_ADDR_DEST_ID_MASK ) <<12 ;Sorry, but this is unreviewable gunk.
Crap. Sure, want to look at the I/OAPIC and MSI documentation and observe that it's just shifting bits around and "these bits go there, those bits go here..." but there's no magic which will make that go away. At some point you end up swizzling bits around with seemingly random bitmasks and shifts which you have to have worked out from the docs.
Now that I killed off the various IOMMU bits which also horrifically touch the RTE directly, perhaps there is more scope for faffing around with it differently, but it won't really change much. It's just urinating into the atmospheric disturbance.
Quoted 3 lines
Aside of that it works magically because polarity,trigger and mask bit have been set up before. But of course a comment about this is completely overrated.
Well yes, there's a lot about the I/OAPIC code which is a bit horrid but this patch is doing just one thing: making the bits get from e.g. cfg->dest_apicid and cfg->vector into the RTE in a generic fashion which does the right thing for IR too. Other cleanups are somewhat orthogonal, but yes it's a good spot that one of these was somewhat redundant in the first place. We could fix that up in a separate patch which comes first perhaps.
If we're starting to clean up I/OAPIC code I'd like to take a long hard look at the level ack code paths, and the fact that we have separate ack_apic_irq and apic_ack_irq functions (or something like that; on phone now) which do different things. Perhaps the order (or the capitals on APIC which one of them has) makes it sane and meaningful for them both to exist and do different things?
I also note the Hyper-V "remapping" takes the IR code path and I'm not sure that's OK. Because of the complete lack of overall documentation on what it's all doing and why.
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
Re: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 5 lines · David Woodhouse
On 22 October 2020 22:43:52 BST, Thomas Gleixner <[email protected]> wrote:
It makes the callers slightly more readable, not having to cast to uint32_t* from the struct.I did ponder defining a new struct with bitfields named along the lines of '
msi_addr_bits_19_to_4', but that seemed like overkill.
I did something like this in the meantime, because all of this just sucks.
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git x86/apic
Hot of the press and completely untested.
Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
arch/x86/kernel/apic/io_apic.c +7
-0+/* +* They're in a bit of a random order for historical reasons, but +* the IO/APIC is just a device for turning interrupt lines into +* MSIs, and various bits of the MSI addr/data are just swizzled +* into/from the bits of Redirection Table Entry. +*/ + entry[0 ] &=0xfffff000 ;Quoted 8 lines
arch/x86/kernel/apic/io_apic.c static void ioapic_ir_ack_level(struct irq_data *irq_data) · line 1875
Quoted 6 lines
arch/x86/kernel/apic/io_apic.c +6
-0+ entry[0 ] |= (msg.data & (MSI_DATA_DELIVERY_MODE_MASK | +MSI_DATA_VECTOR_MASK )); + entry[0 ] |= (msg.address_lo &MSI_ADDR_DEST_MODE_MASK ) <<9 ; + + entry[1 ] &=0xffff ; + entry[1 ] |= (msg.address_lo &MSI_ADDR_DEST_ID_MASK ) <<12 ;Sorry, but this is unreviewable gunk.
Quoted 6 lines
Crap. Sure, want to look at the I/OAPIC and MSI documentation and observe that it's just shifting bits around and "these bits go there, those bits go here..." but there's no magic which will make that go away. At some point you end up swizzling bits around with seemingly random bitmasks and shifts which you have to have worked out from the docs.
Yes, we can't avoid the bit swizzling at all. But it can be made more readable.
Quoted 4 lines
Now that I killed off the various IOMMU bits which also horrifically touch the RTE directly, perhaps there is more scope for faffing around with it differently, but it won't really change much. It's just urinating into the atmospheric disturbance.
Well, you can look at it this way, but I surely have better things to do than taking pencil and paper and drawing up mappings when reading through that code or a patch against it.
Quoted 11 lines
Quoted 3 lines
Aside of that it works magically because polarity,trigger and mask bit have been set up before. But of course a comment about this is completely overrated.
Well yes, there's a lot about the I/OAPIC code which is a bit horrid but this patch is doing just one thing: making the bits get from e.g.
cfg->dest_apicidandcfg->vectorinto the RTE in a generic fashion which does the right thing for IR too. Other cleanups are somewhat orthogonal, but yes it's a good spot that one of these was somewhat redundant in the first place. We could fix that up in a separate patch which comes first perhaps.
Yes, that code is horrid, but adding a comment to that effect when changing it is not asked too much, right?
Quoted 6 lines
If we're starting to clean up I/OAPIC code I'd like to take a long hard look at the level ack code paths, and the fact that we have separate
ack_apic_irqandapic_ack_irqfunctions (or something like that; on phone now) which do different things. Perhaps the order (or the capitals on APIC which one of them has) makes it sane and meaningful for them both to exist and do different things?
The naming conventions are definitely not sane.
Quoted 3 lines
I also note the Hyper-V "remapping" takes the IR code path and I'm not sure that's OK. Because of the complete lack of overall documentation on what it's all doing and why.
I stared into that today as well. There is a fundamental difference between real hardware remapping and this.
Especially the affinity setting mode changes with remapping and hyper-v goes into the remap path, but that hyperv driver does not issue "irq_set_status_flags(virq, IRQ_MOVE_PCNTXT)", which means that you can't change the affinity of the I/O-APIC interrupts in that hyperv mode at all.
If that flag is not set then affinity changes are done in actual interrupt context in ioapic_ack_level() which is only used for the non-IR chip ...
I'm still wrapping my head around getting rid of this thing completely because now it's just a subset of your KVM case with the only restriction that I/O-APIC cannot be affined to any CPU with a APIC id greater than 255.
Thanks,
tglx
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 7 lines · Thomas Gleixner
Quoted 5 lines · David Woodhouse
On 22 October 2020 22:43:52 BST, Thomas Gleixner <[email protected]> wrote:
It makes the callers slightly more readable, not having to cast to uint32_t* from the struct.I did ponder defining a new struct with bitfields named along the lines of '
msi_addr_bits_19_to_4', but that seemed like overkill.
Quoted 6 lines
I did something like this in the meantime, because all of this just sucks.
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.gitx86/apicHot of the press and completely untested.
Hm, your struct IO_APIC_route_entry isn't actually a union; you've defined a 128-bit structure with the IR fields *following* the non-IR fields. But there *is* a union in io_apic.c, of that 128-bit structure and two uint32_ts. Suspect that wasn't quite what you intended. I'll prod at it this morning and turn it into a single union of the three, and give it some testing.
Also, my "Move MSI support into hpet.c" patch¹ got updated to s/CONFIG_PCI_MSI/CONFIG_GENERIC_MSI_IRQ/ at about line 53 for the MSI- related variable declarations, which was going to be in the next version I posted.
I was also hoping Paolo was going to take the patch which just defines the KVM_FEATURE_MSI_EXT_DEST_ID bit² ASAP, so that we end up with a second patch³ that *just* wires it up to x86_init.msi_ext_dest_id() for KVM.
¹ https://git.infradead.org/users/dwmw2/linux.git/commitdiff/734719c1f4
² https://git.infradead.org/users/dwmw2/linux.git/commitdiff/3f371d6749
³ https://git.infradead.org/users/dwmw2/linux.git/commitdiff/8399e14eb5
Quoted 2 lines
Yes, we can't avoid the bit swizzling at all. But it can be made more readable.
Hm, I was about to concede that your version is a bit more readable.
But then I got to your new __ipi_msi_compose_msg() and realised that it isn't working because it's setting the 0xFEE base address in the _low_ bits, somehow...
msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
printk("1 Compose MSI message %x/%x\n", msg->address_lo, msg->data);
msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
printk("2 Compose MSI message %x/%x\n", msg->address_lo, msg->data);
msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
printk("3 Compose MSI message %x/%x\n", msg->address_lo, msg->data);[ 1.793874] 1 Compose MSI message fee/0 [ 1.794310] 2 Compose MSI message fee/0 [ 1.794768] 3 Compose MSI message f02/0
And now I wish it was just a simple shift instead of an unholy maze of overlapping unions of bitfields. But I'll make more coffee and stare at it harder...
Quoted 2 lines
Yes, that code is horrid, but adding a comment to that effect when changing it is not asked too much, right?
Sure. I just actually hadn't noticed that setting the dest/vector bits right there was entirely redundant in the first place.
Quoted 4 lines
I'm still wrapping my head around getting rid of this thing completely because now it's just a subset of your KVM case with the only restriction that I/O-APIC cannot be affined to any CPU with a APIC id greater than 255.
It was only ever that restriction anyway, wasn't it? Hyper-V PCI has its own MSI handling, and there's no HPET so it was only ever the I/OAPIC which was problematic there.
There are Hyper-V VM sizes with 416 vCPUs which depend on this today, and which don't have the 15-bit MSI extension. Removing hyperv-iommu would prevent us from using all the vCPUs on those. You *could* make hyperv-iommu decline to initialise if x86_init.msi_ext_dest_id() returns true though⁴.
⁴ https://git.infradead.org/users/dwmw2/linux.git/commitdiff/633ccf0d42
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 4 lines · David Woodhouse
And now I wish it was just a simple shift instead of an unholy maze of overlapping unions of bitfields. But I'll make more coffee and stare at it harder...
Hah, it really *was* unions of the bitfields. This boots...
arch/x86/include/asm/io_apic.h +6 -1
@@ -65,6 +65,8 @@ union IO_APIC_reg_03 { }; struct IO_APIC_route_entry { + union { + struct { u64 vector : 8, delivery_mode : 3, dest_mode_logical : 1,
@@ -77,7 +79,8 @@ struct IO_APIC_route_entry { reserved_1 : 17, virt_destid_8_14 : 7, destid_0_7 : 8; - + }; +struct {u64 ir_shared_0 :8 , ir_zero :3 , ir_index_15 : 1,
@@ -85,6 +88,8 @@ struct IO_APIC_route_entry { ir_reserved_0 : 31, ir_format : 1, ir_index_0_14 : 15; + }; + }; } __attribute__ ((packed));struct irq_alloc_info ;
arch/x86/include/asm/msi.h +4 -1
@@ -25,6 +25,7 @@ typedef struct x86_msi_data { typedef struct x86_msi_addr_lo { union { + struct { u32 reserved_0 : 2, dest_mode_logical : 1, redirect_hint : 1,
@@ -32,13 +33,15 @@ typedef struct x86_msi_addr_lo { virt_destid_8_14 : 7, destid_0_7 : 8, base_address : 12; - + }; +struct {u32 dmar_reserved_0 :2 , dmar_index_15 :1 , dmar_subhandle_valid :1 , dmar_format :1 , dmar_index_0_14 :15 , dmar_base_address :12 ; + }; }; } __attribute__ ((packed)) arch_msi_msg_addr_lo_t;#define arch_msi_msg_addr_lo x86_msi_addr_lo
Paolo Bonzini
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
In reply to David WoodhouseQuoted 8 lines · David Woodhouse
I was also hoping Paolo was going to take the patch which just defines the
KVM_FEATURE_MSI_EXT_DEST_IDbit² ASAP, so that we end up with a second patch³ that *just* wires it up tox86_init.msi_ext_dest_id()for KVM.¹ https://git.infradead.org/users/dwmw2/linux.git/commitdiff/734719c1f4
² https://git.infradead.org/users/dwmw2/linux.git/commitdiff/3f371d6749
³ https://git.infradead.org/users/dwmw2/linux.git/commitdiff/8399e14eb5
Yes, I am going to take it.
I was already sort of playing with fire with the 5.10 pull request (and with me being lousy in general during the 5.10 development period, to be honest), so I left it for rc2 or rc3. It's just docs and it happened to conflict with another documentation patch that had gone in through Jon Corbet's tree.
Paolo
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 17 lines · Paolo Bonzini
Quoted 1 line · David Woodhouse
I was also hoping Paolo was going to take the patch which just
defines
Quoted 2 lines
the
KVM_FEATURE_MSI_EXT_DEST_IDbit² ASAP, so that we end up with a second patch³ that *just* wires it up tox86_init.msi_ext_dest_id()for
Quoted 3 lines
KVM.
¹
https://git.infradead.org/users/dwmw2/linux.git/commitdiff/734719c1f4
Quoted 1 line
²
https://git.infradead.org/users/dwmw2/linux.git/commitdiff/3f371d6749
Quoted 1 line
³
https://git.infradead.org/users/dwmw2/linux.git/commitdiff/8399e14eb5
Yes, I am going to take it.
Quoted 7 lines
I was already sort of playing with fire with the 5.10 pull request (and with me being lousy in general during the 5.10 development period, to be
honest), so I left it for rc2 or rc3. It's just docs and it happened to
conflict with another documentation patch that had gone in through Jon Corbet's tree.
OK, thanks. I'll rework Thomas's tree with that first and the other changes I'd mentioned in my parts, as well as fixing up that unholy chimæra of struct/union in which we set some bitfields from each side of the union, test and push it out later today.
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
David Woodhouse
maintainerRe: [PATCH v2 8/8] x86/ioapic: Generate RTE directly from parent irqchip's MSI message
Quoted 4 lines · David Woodhouse
OK, thanks. I'll rework Thomas's tree with that first and the other changes I'd mentioned in my parts, as well as fixing up that unholy chimæra of struct/union in which we set some bitfields from each side of the union, test and push it out later today.
OK, pushed out to
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/x86/apic]
It's Thomas's tree plus the struct/union fixes and other things I mentioned earlier, a few comment fixes in the 'Generate RTE directly from parent irqchip' patch, but the most interesting part is finishing the job of the 'Cleanup IO/APIC route entry structs' patch...
From 54b623fc2b03eadb76485b4ca0ade3e79acf6c27 Mon Sep 17 00:00:00 2001
From: Thomas Gleixner <[email protected]>
Date: Thu, 22 Oct 2020 14:48:18 +0200
Subject: [PATCH 124/137] x86/ioapic: Cleanup IO/APIC route entry structs
Having two seperate structs for the I/O-APIC RTE entries (non-remapped and DMAR remapped) requires type casts and makes it hard to map.
Combine them in IO_APIC_routing_entry by defining a union of two 64bit bitfields. Use naming which reflects which bits are shared and which bits are actually different for the operating modes.
[dwmw2: Fix it up and finish the job, pulling the 32-bit w1,w2 words for
register access into the same union and eliminating a few more places where bits were accessed through masks and shifts.]
--- arch/x86/include/asm/io_apic.h | 78 ++++++--------- arch/x86/kernel/apic/io_apic.c | 144 +++++++++++++--------------- drivers/iommu/amd/iommu.c | 8 +- drivers/iommu/hyperv-iommu.c | 4 +- drivers/iommu/intel/irq_remapping.c | 19 ++-- 5 files changed, 108 insertions(+), 145 deletions(-)
arch/x86/include/asm/io_apic.h +28 -50
@@ -13,15 +13,6 @@* Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar */ -/* I/O Unit Redirection Table */ -#define IO_APIC_REDIR_VECTOR_MASK 0x000FF -#define IO_APIC_REDIR_DEST_LOGICAL 0x00800 -#define IO_APIC_REDIR_DEST_PHYSICAL 0x00000 -#define IO_APIC_REDIR_SEND_PENDING (1 << 12) -#define IO_APIC_REDIR_REMOTE_IRR (1 << 14) -#define IO_APIC_REDIR_LEVEL_TRIGGER (1 << 15) -#define IO_APIC_REDIR_MASKED (1 << 16) -/* * The structure of the IO-APIC: */
@@ -65,52 +56,39 @@ union IO_APIC_reg_03 { };struct IO_APIC_route_entry { -__u32 vector :8 , - delivery_mode :3 ,/* 000: FIXED -* 001: lowest prio -* 111: ExtINT -*/ - dest_mode :1 ,/* 0: physical, 1: logical */ - delivery_status :1 , - polarity :1 , - irr :1 , - trigger :1 ,/* 0: edge, 1: level */ - mask :1 ,/* 0: enabled, 1: disabled */ - __reserved_2 :15 ; - -__u32 __reserved_3 :24 , - dest :8 ; -} __attribute__ ((packed)); - -struct IR_IO_APIC_route_entry { -__u64 vector :8 , - zero :3 , - index2 :1 , - delivery_status :1 , - polarity :1 , - irr :1 , - trigger :1 , - mask :1 , - reserved :31 , - format :1 , - index :15 ; +union { +struct { +u64 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + delivery_status :1 , + active_low :1 , + irr :1 , + is_level :1 , + masked :1 , + reserved_0 :15 , + reserved_1 :24 , + destid_0_7 :8 ; + }; +struct { +u64 ir_shared_0 :8 , + ir_zero :3 , + ir_index_15 :1 , + ir_shared_1 :5 , + ir_reserved_0 :31 , + ir_format :1 , + ir_index_0_14 :15 ; + }; +struct { +u64 w1 :32 , + w2 :32 ; + }; + }; } __attribute__ ((packed));struct irq_alloc_info ;struct ioapic_domain_cfg ; -#define IOAPIC_EDGE 0 -#define IOAPIC_LEVEL 1 - -#define IOAPIC_MASKED 1 -#define IOAPIC_UNMASKED 0 - -#define IOAPIC_POL_HIGH 0 -#define IOAPIC_POL_LOW 1 - -#define IOAPIC_DEST_MODE_PHYSICAL 0 -#define IOAPIC_DEST_MODE_LOGICAL 1 -#define IOAPIC_MAP_ALLOC 0x1#define IOAPIC_MAP_CHECK 0x2
arch/x86/kernel/apic/io_apic.c +65 -79
298 lines not on this page · raw message
drivers/iommu/amd/iommu.c +4 -4
@@ -3687,11 +3687,11 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, entry = info->ioapic.entry; info->ioapic.entry =NULL ;memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->trigger = info->ioapic.is_level; - entry->polarity = info->ioapic.active_low; + entry->vector = index; + entry->is_level = info->ioapic.is_level; + entry->active_low = info->ioapic.active_low;/* Mask level triggered irqs. */ - entry->mask = info->ioapic.is_level; + entry->masked = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
drivers/iommu/hyperv-iommu.c +2 -2
@@ -52,7 +52,7 @@ static int hyperv_ir_set_affinity(struct irq_data *data,return ret; entry = data->chip_data; - entry->dest = cfg->dest_apicid; + entry->destid_0_7 = cfg->dest_apicid; entry->vector = cfg->vector;send_cleanup_vector (cfg);
@@ -125,7 +125,7 @@ static int hyperv_irq_remapping_activate(struct irq_domain *domain,struct irq_cfg *cfg =irqd_cfg (irq_data);struct IO_APIC_route_entry *entry = irq_data->chip_data; - entry->dest = cfg->dest_apicid; + entry->destid_0_7 = cfg->dest_apicid; entry->vector = cfg->vector;return 0 ;
drivers/iommu/intel/irq_remapping.c +9 -10
@@ -1279,8 +1279,8 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,struct irq_alloc_info *info, int index, int sub_handle) { - struct IR_IO_APIC_route_entry *entry; struct irte *irte = &data->irte_entry; +struct IO_APIC_route_entry *entry;prepare_irte (irte, irq_cfg->vector, irq_cfg->dest_apicid);switch (info->type) {
@@ -1294,22 +1294,21 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, irte->avail, irte->vector, irte->dest_id, irte->sid, irte->sq, irte->svt); - entry = (struct IR_IO_APIC_route_entry *)info->ioapic.entry; + entry = info->ioapic.entry; info->ioapic.entry =NULL ;memset (entry,0 ,sizeof (*entry)); - entry->index2 = (index >>15 ) &0x1 ; - entry->zero =0 ; - entry->format =1 ; - entry->index = (index &0x7fff ); + entry->ir_index_15 = !!(index &0x8000 ); + entry->ir_format = true; + entry->ir_index_0_14 = index &0x7fff ;/* * IO-APIC RTE will be configured with virtual vector. * irq handler will do the explicit EOI to the io-apic. */ - entry->vector = info->ioapic.pin; - entry->trigger = info->ioapic.is_level; - entry->polarity = info->ioapic.active_low; + entry->vector = info->ioapic.pin; + entry->is_level = info->ioapic.is_level; + entry->active_low = info->ioapic.active_low;/* Mask level triggered irqs. */ - entry->mask = info->ioapic.is_level; + entry->masked = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
Signature · 1 line
2.17.1
David Woodhouse
maintainerv3 cover Fix x2apic enablement and allow more CPUs, clean up I/OAPIC and MSI bitfields
Fix the conditions for enabling x2apic on guests without interrupt remapping, and support 15-bit Extended Destination ID to allow 32768 CPUs without IR on hypervisors that support it.
Make the I/OAPIC code generate its RTE directly from the MSI message created by the parent irqchip, and fix up a bunch of magic mask/shift macros to use bitfields for MSI messages and I/OAPIC RTEs while we're at it.
v3:
• Lots of bitfield cleanups from Thomas.
• Disable hyperv-iommu if 15-bit extension is present.
• Fix inconsistent CONFIG_PCI_MSI/CONFIG_GENERIC_MSI_IRQ in hpet.c
• Split KVM_FEATURE_MSI_EXT_DEST_ID patch, half of which is going upstream through KVM tree (and the other half needs to wait, or have an #ifdef) so is left at the top of the tree.
v2:
• Minor cleanups.
• Move __irq_msi_compose_msg() to apic.c, make virt_ext_dest_id static.
• Generate I/OAPIC RTE directly from parent irqchip's MSI messages.
• Clean up HPET MSI support into hpet.c now that we can.
David Woodhouse (19):
x86/apic: Fix x2apic enablement without interrupt remapping x86/msi: Only use high bits of MSI address for DMAR unit x86/apic: Always provide irq_compose_msi_msg() method for vector domain x86/hpet: Move MSI support into hpet.c
x86/ioapic: Generate RTE directly from parent irqchip's MSI message genirq/irqdomain: Implement get_name() method on irqchip fwnodes x86/apic: Add select() method on vector irqdomain
iommu/amd: Implement select() method on remapping irqdomain iommu/vt-d: Implement select() method on remapping irqdomain iommu/hyper-v: Implement select() method on remapping irqdomain x86/hpet: Use irq_find_matching_fwspec() to find remapping irqdomain x86/ioapic: Use irq_find_matching_fwspec() to find remapping irqdomain
x86: Kill all traces of irq_remapping_get_irq_domain() iommu/vt-d: Simplify intel_irq_remapping_select()
x86/ioapic: Handle Extended Destination ID field in RTE x86/apic: Support 15 bits of APIC ID in MSI where available iommu/hyper-v: Disable IRQ pseudo-remapping if 15 bit APIC IDs are available x86/kvm: Reserve KVM_FEATURE_MSI_EXT_DEST_ID
x86/kvm: Enable 15-bit extension when KVM_FEATURE_MSI_EXT_DEST_ID detected
Thomas Gleixner (16):
x86/apic/uv: Fix inconsistent destination mode x86/devicetree: Fix the ioapic interrupt type table x86/apic: Cleanup delivery mode defines x86/apic: Replace pointless apic::dest_logical usage x86/apic: Get rid of apic::dest_logical x86/apic: Cleanup destination mode
genirq/msi: Allow shadow declarations of msi_msg::$member x86/msi: Provide msi message shadow structs iommu/intel: Use msi_msg shadow structs iommu/amd: Use msi_msg shadow structs
PCI: vmd: Use msi_msg shadow structs
x86/kvm: Use msi_msg shadow structs x86/pci/xen: Use msi_msg shadow structs x86/msi: Remove msidef.h
x86/io_apic: Cleanup trigger/polarity helpers x86/ioapic: Cleanup IO/APIC route entry structs
Documentation/virt/kvm/cpuid.rst | 4 + arch/x86/include/asm/apic.h | 16 +- arch/x86/include/asm/apicdef.h | 16 +- arch/x86/include/asm/hpet.h | 11 - arch/x86/include/asm/hw_irq.h | 13 +- arch/x86/include/asm/io_apic.h | 79 ++---- arch/x86/include/asm/irq_remapping.h | 9 - arch/x86/include/asm/irqdomain.h | 3 + arch/x86/include/asm/msi.h | 50 ++++ arch/x86/include/asm/msidef.h | 57 ---- arch/x86/include/asm/x86_init.h | 2 + arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/apic/apic.c | 73 ++++- arch/x86/kernel/apic/apic_flat_64.c | 18 +- arch/x86/kernel/apic/apic_noop.c | 10 +- arch/x86/kernel/apic/apic_numachip.c | 16 +- arch/x86/kernel/apic/bigsmp_32.c | 9 +- arch/x86/kernel/apic/io_apic.c | 503 ++++++++++++++++++---------------- arch/x86/kernel/apic/ipi.c | 6 +- arch/x86/kernel/apic/msi.c | 153 +---------- arch/x86/kernel/apic/probe_32.c | 9 +- arch/x86/kernel/apic/vector.c | 49 ++++ arch/x86/kernel/apic/x2apic_cluster.c | 10 +- arch/x86/kernel/apic/x2apic_phys.c | 17 +- arch/x86/kernel/apic/x2apic_uv_x.c | 12 +- arch/x86/kernel/devicetree.c | 30 +- arch/x86/kernel/hpet.c | 122 ++++++++- arch/x86/kernel/kvm.c | 6 + arch/x86/kernel/smpboot.c | 8 +- arch/x86/kernel/x86_init.c | 1 + arch/x86/kvm/irq_comm.c | 31 +-- arch/x86/pci/intel_mid_pci.c | 8 +- arch/x86/pci/xen.c | 26 +- arch/x86/platform/uv/uv_irq.c | 4 +- arch/x86/xen/apic.c | 7 +- drivers/iommu/amd/amd_iommu_types.h | 2 +- drivers/iommu/amd/init.c | 46 ++-- drivers/iommu/amd/iommu.c | 93 +++---- drivers/iommu/hyperv-iommu.c | 44 +-- drivers/iommu/intel/irq_remapping.c | 102 +++---- drivers/iommu/irq_remapping.c | 14 - drivers/iommu/irq_remapping.h | 3 - drivers/pci/controller/pci-hyperv.c | 6 +- drivers/pci/controller/vmd.c | 9 +- include/asm-generic/msi.h | 4 + include/linux/msi.h | 46 +++- kernel/irq/irqdomain.c | 11 +- 47 files changed, 890 insertions(+), 879 deletions(-)
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
David Woodhouse
maintainerPATCH v3 1/35 x86/apic: Fix x2apic enablement without interrupt remapping
From: David Woodhouse <[email protected]>
Currently, Linux as a hypervisor guest will enable x2apic only if there are no CPUs present at boot time with an APIC ID above 255.
Hotplugging a CPU later with a higher APIC ID would result in a CPU which cannot be targeted by external interrupts.
Add a filter in x2apic_apic_id_valid() which can be used to prevent such CPUs from coming online, and allow x2apic to be enabled even if they are present at boot time.
--- arch/x86/include/asm/apic.h | 1 + arch/x86/kernel/apic/apic.c | 14 ++++++++------ arch/x86/kernel/apic/x2apic_phys.c | 9 +++++++++ 3 files changed, 18 insertions(+), 6 deletions(-)
arch/x86/include/asm/apic.h +1 -0
@@ -259,6 +259,7 @@ static inline u64 native_x2apic_icr_read(void)extern int x2apic_mode;extern int x2apic_phys; +extern void __init x2apic_set_max_apicid (u32 apicid);extern void __init check_x2apic (void );extern void x2apic_setup (void );static inline int x2apic_enabled (void )
arch/x86/kernel/apic/apic.c +8 -6
@@ -1841,20 +1841,22 @@ static __init void try_to_enable_x2apic(int remap_mode)return ;if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) { -/* IR is required if there is APIC ID > 255 even when running -* under KVM +/* +* Using X2APIC without IR is not architecturally supported +* on bare metal but may be supported in guests. */ -if (max_physical_apicid >255 || - !x86_init.hyper.x2apic_available ()) { +if (!x86_init.hyper.x2apic_available ()) {pr_info ("x2apic: IRQ remapping doesn't support X2APIC mode\n" );x2apic_disable ();return ; }/* -* without IR all CPUs can be addressed by IOAPIC/MSI -* only in physical mode +* Without IR, all CPUs can be addressed by IOAPIC/MSI only +* in physical mode, and CPUs with an APIC ID that cannnot +* be addressed must not be brought online. */ +x2apic_set_max_apicid (255 ); x2apic_phys =1 ; }x2apic_enable ();
arch/x86/kernel/apic/x2apic_phys.c +9 -0
@@ -8,6 +8,12 @@int x2apic_phys;static struct apic apic_x2apic_phys; +static u32 x2apic_max_apicid__ro_after_init ; + +void __init x2apic_set_max_apicid (u32 apicid) +{ + x2apic_max_apicid = apicid; +}static int __init set_x2apic_phys_mode (char *arg) {
@@ -98,6 +104,9 @@ static int x2apic_phys_probe(void)/* Common x2apic functions, also used by x2apic_cluster */ int x2apic_apic_id_valid (u32 apicid) { +if (x2apic_max_apicid && apicid > x2apic_max_apicid) +return 0 ; +return 1 ; }
Signature · 1 line
2.26.2
[tip: x86/apic] x86/apic: Fix x2apic enablement without interrupt remapping
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 26573a97746c7a99f394f9d398ce91a8853b3b89 Gitweb: https://git.kernel.org/tip/26573a97746c7a99f394f9d398ce91a8853b3b89 Author: David Woodhouse <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:01 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:23 +01:00
x86/apic: Fix x2apic enablement without interrupt remapping
Currently, Linux as a hypervisor guest will enable x2apic only if there are no CPUs present at boot time with an APIC ID above 255.
Hotplugging a CPU later with a higher APIC ID would result in a CPU which cannot be targeted by external interrupts.
Add a filter in x2apic_apic_id_valid() which can be used to prevent such CPUs from coming online, and allow x2apic to be enabled even if they are present at boot time.
--- arch/x86/include/asm/apic.h | 1 + arch/x86/kernel/apic/apic.c | 14 ++++++++------ arch/x86/kernel/apic/x2apic_phys.c | 9 +++++++++ 3 files changed, 18 insertions(+), 6 deletions(-)
arch/x86/include/asm/apic.h +1 -0
@@ -259,6 +259,7 @@ static inline u64 native_x2apic_icr_read(void)extern int x2apic_mode;extern int x2apic_phys; +extern void __init x2apic_set_max_apicid (u32 apicid);extern void __init check_x2apic (void );extern void x2apic_setup (void );static inline int x2apic_enabled (void )
arch/x86/kernel/apic/apic.c +8 -6
@@ -1841,20 +1841,22 @@ static __init void try_to_enable_x2apic(int remap_mode)return ;if (remap_mode !=IRQ_REMAP_X2APIC_MODE ) { -/* IR is required if there is APIC ID > 255 even when running -* under KVM +/* +* Using X2APIC without IR is not architecturally supported +* on bare metal but may be supported in guests. */ -if (max_physical_apicid >255 || - !x86_init.hyper.x2apic_available ()) { +if (!x86_init.hyper.x2apic_available ()) {pr_info ("x2apic: IRQ remapping doesn't support X2APIC mode\n" );x2apic_disable ();return ; }/* -* without IR all CPUs can be addressed by IOAPIC/MSI -* only in physical mode +* Without IR, all CPUs can be addressed by IOAPIC/MSI only +* in physical mode, and CPUs with an APIC ID that cannnot +* be addressed must not be brought online. */ +x2apic_set_max_apicid (255 ); x2apic_phys =1 ; }x2apic_enable ();
arch/x86/kernel/apic/x2apic_phys.c +9 -0
@@ -8,6 +8,12 @@int x2apic_phys;static struct apic apic_x2apic_phys; +static u32 x2apic_max_apicid__ro_after_init ; + +void __init x2apic_set_max_apicid (u32 apicid) +{ + x2apic_max_apicid = apicid; +}static int __init set_x2apic_phys_mode (char *arg) {
@@ -98,6 +104,9 @@ static int x2apic_phys_probe(void)/* Common x2apic functions, also used by x2apic_cluster */ int x2apic_apic_id_valid (u32 apicid) { +if (x2apic_max_apicid && apicid > x2apic_max_apicid) +return 0 ; +return 1 ; }
David Woodhouse
maintainerPATCH v3 2/35 x86/msi: Only use high bits of MSI address for DMAR unit
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
The Intel IOMMU has an MSI-like configuration for its interrupt, but it isn't really MSI. So it gets to abuse the high 32 bits of the address, and puts the high 24 bits of the extended APIC ID there.
This isn't something that can be used in the general case for real MSIs, since external devices using the high bits of the address would be performing writes to actual memory space above 4GiB, not targeted at the APIC.
Factor the hack out and allow it only to be used when appropriate, adding a WARN_ON_ONCE() if other MSIs are targeted at an unreachable APIC ID. That should never happen since the compatibility MSI messages are not used when Interrupt Remapping is enabled.
The x2apic_enabled() check isn't needed because Linux won't bring up CPUs with higher APIC IDs unless IR and x2apic are enabled anyway.
--- arch/x86/kernel/apic/msi.c | 33 +++++++++++++++++++++++++++------ 1 file changed, 27 insertions(+), 6 deletions(-)
arch/x86/kernel/apic/msi.c +27 -6
@@ -23,13 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg) +static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) { msg->address_hi =MSI_ADDR_BASE_HI ; -if (x2apic_enabled ()) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); - msg->address_lo = MSI_ADDR_BASE_LO | ((apic->irq_dest_mode == 0) ?
@@ -43,18 +41,29 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)MSI_DATA_LEVEL_ASSERT |MSI_DATA_DELIVERY_FIXED |MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) { -__irq_msi_compose_msg (irqd_cfg (data), msg); +__irq_msi_compose_msg (irqd_cfg (data), msg, false); }static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg); +__irq_msi_compose_msg (cfg, msg, false);irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); }
@@ -276,6 +285,17 @@ struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,#endif #ifdef CONFIG_DMAR_TABLE +/* +* The Intel IOMMU (ab)uses the high bits of the MSI address to contain the +* high bits of the destination APIC ID. This can't be done in the general +* case for MSIs as it would be targeting real memory above 4GiB not the +* APIC. +*/ +static void dmar_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, true); +} +static void dmar_msi_write_msg (struct irq_data *data,struct msi_msg *msg) {dmar_msi_write (data->irq, msg);
@@ -288,6 +308,7 @@ static struct irq_chip dmar_msi_controller = { .irq_ack = irq_chip_ack_parent, .irq_set_affinity = msi_domain_set_affinity, .irq_retrigger = irq_chip_retrigger_hierarchy, + .irq_compose_msi_msg = dmar_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, .flags = IRQCHIP_SKIP_SET_WAKE, };
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/msi: Only use high bits of MSI address for DMAR unit
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 47bea873cf809f490cfac0c4e88533fd7fed6064 Gitweb: https://git.kernel.org/tip/47bea873cf809f490cfac0c4e88533fd7fed6064 Author: David Woodhouse <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:02 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/msi: Only use high bits of MSI address for DMAR unit
The Intel IOMMU has an MSI-like configuration for its interrupt, but it isn't really MSI. So it gets to abuse the high 32 bits of the address, and puts the high 24 bits of the extended APIC ID there.
This isn't something that can be used in the general case for real MSIs, since external devices using the high bits of the address would be performing writes to actual memory space above 4GiB, not targeted at the APIC.
Factor the hack out and allow it only to be used when appropriate, adding a WARN_ON_ONCE() if other MSIs are targeted at an unreachable APIC ID. That should never happen since the compatibility MSI messages are not used when Interrupt Remapping is enabled.
The x2apic_enabled() check isn't needed because Linux won't bring up CPUs with higher APIC IDs unless IR and x2apic are enabled anyway.
--- arch/x86/kernel/apic/msi.c | 33 +++++++++++++++++++++++++++------ 1 file changed, 27 insertions(+), 6 deletions(-)
arch/x86/kernel/apic/msi.c +27 -6
@@ -23,13 +23,11 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg) +static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) { msg->address_hi =MSI_ADDR_BASE_HI ; -if (x2apic_enabled ()) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); - msg->address_lo = MSI_ADDR_BASE_LO | ((apic->irq_dest_mode == 0) ?
@@ -43,18 +41,29 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)MSI_DATA_LEVEL_ASSERT |MSI_DATA_DELIVERY_FIXED |MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); }void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) { -__irq_msi_compose_msg (irqd_cfg (data), msg); +__irq_msi_compose_msg (irqd_cfg (data), msg, false); }static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg); +__irq_msi_compose_msg (cfg, msg, false);irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); }
@@ -276,6 +285,17 @@ struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,#endif #ifdef CONFIG_DMAR_TABLE +/* +* The Intel IOMMU (ab)uses the high bits of the MSI address to contain the +* high bits of the destination APIC ID. This can't be done in the general +* case for MSIs as it would be targeting real memory above 4GiB not the +* APIC. +*/ +static void dmar_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, true); +} +static void dmar_msi_write_msg (struct irq_data *data,struct msi_msg *msg) {dmar_msi_write (data->irq, msg);
@@ -288,6 +308,7 @@ static struct irq_chip dmar_msi_controller = { .irq_ack = irq_chip_ack_parent, .irq_set_affinity = msi_domain_set_affinity, .irq_retrigger = irq_chip_retrigger_hierarchy, + .irq_compose_msi_msg = dmar_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, .flags = IRQCHIP_SKIP_SET_WAKE, };
David Woodhouse
maintainerPATCH v3 3/35 x86/apic/uv: Fix inconsistent destination mode
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
The UV x2apic is strictly using physical destination mode, but apic::dest_logical is initialized with APIC_DEST_LOGICAL.
This does not matter much because UV does not use any of the generic functions which use apic::dest_logical, but is still inconsistent.
--- arch/x86/kernel/apic/x2apic_uv_x.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/apic/x2apic_uv_x.c +1 -1
@@ -811,7 +811,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .irq_dest_mode = 0, /* Physical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, + .dest_logical = APIC_DEST_PHYSICAL, .check_apicid_used = NULL, .init_apic_ldr = uv_init_apic_ldr,
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/apic/uv: Fix inconsistent destination mode
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 93b7a3d6a1f0f159d390959de7a1b9ad863d6b26 Gitweb: https://git.kernel.org/tip/93b7a3d6a1f0f159d390959de7a1b9ad863d6b26 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:03 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/apic/uv: Fix inconsistent destination mode
The UV x2apic is strictly using physical destination mode, but apic::dest_logical is initialized with APIC_DEST_LOGICAL.
This does not matter much because UV does not use any of the generic functions which use apic::dest_logical, but is still inconsistent.
--- arch/x86/kernel/apic/x2apic_uv_x.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/apic/x2apic_uv_x.c +1 -1
@@ -811,7 +811,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .irq_dest_mode = 0, /* Physical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, + .dest_logical = APIC_DEST_PHYSICAL, .check_apicid_used = NULL, .init_apic_ldr = uv_init_apic_ldr,
David Woodhouse
maintainerPATCH v3 4/35 x86/devicetree: Fix the ioapic interrupt type table
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
The ioapic interrupt type table is wrong as it assumes that polarity in IO/APIC context means active high when set. But the IO/APIC polarity is working the other way round. This works because the ordering of the entries is consistent with the device tree and the type information is not used by the IO/APIC interrupt chip.
The whole trigger and polarity business of IO/APIC is misleading and the corresponding constants which are defined as 0/1 are not used consistently and are going to be removed.
Rename the type table members to 'is_level' and 'active_low' and adjust the type information for consistency sake.
No functional change.
--- arch/x86/kernel/devicetree.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)
arch/x86/kernel/devicetree.c +15 -15
@@ -184,31 +184,31 @@ static unsigned int ioapic_id;struct of_ioapic_type {u32 out_type; -u32 trigger; -u32 polarity; +u32 is_level; +u32 active_low; };static struct of_ioapic_type of_ioapic_type[] = { { - .out_type =IRQ_TYPE_EDGE_RISING , - .trigger =IOAPIC_EDGE , - .polarity =1 , + .out_type =IRQ_TYPE_EDGE_FALLING , + .is_level =0 , + .active_low =1 , }, { - .out_type =IRQ_TYPE_LEVEL_LOW , - .trigger =IOAPIC_LEVEL , - .polarity =0 , + .out_type =IRQ_TYPE_LEVEL_HIGH , + .is_level =1 , + .active_low =0 , }, { - .out_type =IRQ_TYPE_LEVEL_HIGH , - .trigger =IOAPIC_LEVEL , - .polarity =1 , + .out_type =IRQ_TYPE_LEVEL_LOW , + .is_level =1 , + .active_low =1 , }, { - .out_type =IRQ_TYPE_EDGE_FALLING , - .trigger =IOAPIC_EDGE , - .polarity =0 , + .out_type =IRQ_TYPE_EDGE_RISING , + .is_level =0 , + .active_low =0 , }, };
@@ -228,7 +228,7 @@ static int dt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,return -EINVAL ; it = &of_ioapic_type[type_index]; -ioapic_set_alloc_attr (&tmp,NUMA_NO_NODE , it->trigger, it->polarity); +ioapic_set_alloc_attr (&tmp,NUMA_NO_NODE , it->is_level, it->active_low); tmp.devid =mpc_ioapic_id (mp_irqdomain_ioapic_idx (domain)); tmp.ioapic.pin = fwspec->param[0 ];
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/devicetree: Fix the ioapic interrupt type table
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 2e730cb56b2cd1626fecaf23ef1537fb24721ef2 Gitweb: https://git.kernel.org/tip/2e730cb56b2cd1626fecaf23ef1537fb24721ef2 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:04 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/devicetree: Fix the ioapic interrupt type table
The ioapic interrupt type table is wrong as it assumes that polarity in IO/APIC context means active high when set. But the IO/APIC polarity is working the other way round. This works because the ordering of the entries is consistent with the device tree and the type information is not used by the IO/APIC interrupt chip.
The whole trigger and polarity business of IO/APIC is misleading and the corresponding constants which are defined as 0/1 are not used consistently and are going to be removed.
Rename the type table members to 'is_level' and 'active_low' and adjust the type information for consistency sake.
No functional change.
--- arch/x86/kernel/devicetree.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)
arch/x86/kernel/devicetree.c +15 -15
@@ -184,31 +184,31 @@ static unsigned int ioapic_id;struct of_ioapic_type {u32 out_type; -u32 trigger; -u32 polarity; +u32 is_level; +u32 active_low; };static struct of_ioapic_type of_ioapic_type[] = { { - .out_type =IRQ_TYPE_EDGE_RISING , - .trigger =IOAPIC_EDGE , - .polarity =1 , + .out_type =IRQ_TYPE_EDGE_FALLING , + .is_level =0 , + .active_low =1 , }, { - .out_type =IRQ_TYPE_LEVEL_LOW , - .trigger =IOAPIC_LEVEL , - .polarity =0 , + .out_type =IRQ_TYPE_LEVEL_HIGH , + .is_level =1 , + .active_low =0 , }, { - .out_type =IRQ_TYPE_LEVEL_HIGH , - .trigger =IOAPIC_LEVEL , - .polarity =1 , + .out_type =IRQ_TYPE_LEVEL_LOW , + .is_level =1 , + .active_low =1 , }, { - .out_type =IRQ_TYPE_EDGE_FALLING , - .trigger =IOAPIC_EDGE , - .polarity =0 , + .out_type =IRQ_TYPE_EDGE_RISING , + .is_level =0 , + .active_low =0 , }, };
@@ -228,7 +228,7 @@ static int dt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,return -EINVAL ; it = &of_ioapic_type[type_index]; -ioapic_set_alloc_attr (&tmp,NUMA_NO_NODE , it->trigger, it->polarity); +ioapic_set_alloc_attr (&tmp,NUMA_NO_NODE , it->is_level, it->active_low); tmp.devid =mpc_ioapic_id (mp_irqdomain_ioapic_idx (domain)); tmp.ioapic.pin = fwspec->param[0 ];
David Woodhouse
maintainerPATCH v3 5/35 x86/apic: Cleanup delivery mode defines
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
The enum ioapic_irq_destination_types and the enumerated constants starting with 'dest_' are gross misnomers because they describe the delivery mode.
Rename then enum and the constants so they actually make sense.
--- arch/x86/include/asm/apic.h | 3 ++- arch/x86/include/asm/apicdef.h | 16 +++++++--------- arch/x86/kernel/apic/apic_flat_64.c | 4 ++-- arch/x86/kernel/apic/apic_noop.c | 2 +- arch/x86/kernel/apic/apic_numachip.c | 4 ++-- arch/x86/kernel/apic/bigsmp_32.c | 2 +- arch/x86/kernel/apic/io_apic.c | 11 ++++++----- arch/x86/kernel/apic/probe_32.c | 2 +- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- arch/x86/kernel/apic/x2apic_phys.c | 2 +- arch/x86/kernel/apic/x2apic_uv_x.c | 6 +++--- arch/x86/platform/uv/uv_irq.c | 2 +- drivers/iommu/amd/iommu.c | 4 ++-- drivers/iommu/intel/irq_remapping.c | 2 +- drivers/pci/controller/pci-hyperv.c | 6 +++--- 15 files changed, 34 insertions(+), 34 deletions(-)
arch/x86/include/asm/apic.h +2 -1
@@ -309,7 +309,8 @@ struct apic {/* dest_logical is used by the IPI functions */ u32 dest_logical;u32 disable_esr; -u32 irq_delivery_mode; + +enum apic_delivery_modes delivery_mode;u32 irq_dest_mode;u32 (*calc_dest_apicid)(unsigned int cpu);
arch/x86/include/asm/apicdef.h +7 -9
@@ -432,15 +432,13 @@ struct local_apic {#define BAD_APICID 0xFFFFu#endif -enum ioapic_irq_destination_types { - dest_Fixed =0 , - dest_LowestPrio =1 , - dest_SMI =2 , - dest__reserved_1 =3 , - dest_NMI =4 , - dest_INIT =5 , - dest__reserved_2 =6 , - dest_ExtINT =7 +enum apic_delivery_modes { +APIC_DELIVERY_MODE_FIXED =0 , +APIC_DELIVERY_MODE_LOWESTPRIO =1 , +APIC_DELIVERY_MODE_SMI =2 , +APIC_DELIVERY_MODE_NMI =4 , +APIC_DELIVERY_MODE_INIT =5 , +APIC_DELIVERY_MODE_EXTINT =7 , };#endif /* _ASM_X86_APICDEF_H */
arch/x86/kernel/apic/apic_flat_64.c +2 -2
@@ -113,7 +113,7 @@ static struct apic apic_flat __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = flat_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 1, /* logical */ .disable_esr = 0,
@@ -206,7 +206,7 @@ static struct apic apic_physflat __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = flat_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/apic_noop.c +1 -1
@@ -95,7 +95,7 @@ struct apic apic_noop __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = noop_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* logical delivery broadcast to all CPUs: */ .irq_dest_mode = 1,
arch/x86/kernel/apic/apic_numachip.c +2 -2
@@ -246,7 +246,7 @@ static const struct apic apic_numachip1 __refconst = { .apic_id_valid = numachip_apic_id_valid, .apic_id_registered = numachip_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
@@ -295,7 +295,7 @@ static const struct apic apic_numachip2 __refconst = { .apic_id_valid = numachip_apic_id_valid, .apic_id_registered = numachip_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/bigsmp_32.c +1 -1
@@ -127,7 +127,7 @@ static struct apic apic_bigsmp __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = bigsmp_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* phys delivery to target CPU: */ .irq_dest_mode = 0,
arch/x86/kernel/apic/io_apic.c +6 -5
@@ -535,7 +535,7 @@ static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)/* Check delivery_mode to be sure we're not clearing an SMI pin */ entry =ioapic_read_entry (apic, pin); -if (entry.delivery_mode == dest_SMI) +if (entry.delivery_mode ==APIC_DELIVERY_MODE_SMI )return ;/*
@@ -1368,7 +1368,8 @@ void __init enable_IO_APIC(void)/* If the interrupt line is enabled and in ExtInt mode * I have found the pin where the i8259 is connected. */ -if ((entry.mask ==0 ) && (entry.delivery_mode == dest_ExtINT)) { +if ((entry.mask ==0 ) && + (entry.delivery_mode ==APIC_DELIVERY_MODE_EXTINT )) { ioapic_i8259.apic = apic; ioapic_i8259.pin = pin; goto found_i8259;
@@ -1416,7 +1417,7 @@ void native_restore_boot_irq_mode(void) entry.trigger =IOAPIC_EDGE ; entry.polarity =IOAPIC_POL_HIGH ; entry.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; - entry.delivery_mode = dest_ExtINT; + entry.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; entry.dest =read_apic_id ();/*
@@ -2047,7 +2048,7 @@ static inline void __init unlock_ExtINT_logic(void) entry1.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry1.mask =IOAPIC_UNMASKED ; entry1.dest =hard_smp_processor_id (); - entry1.delivery_mode = dest_ExtINT; + entry1.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; entry1.polarity = entry0.polarity; entry1.trigger =IOAPIC_EDGE ; entry1.vector =0 ;
@@ -2948,7 +2949,7 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, struct IO_APIC_route_entry *entry) {memset (entry,0 ,sizeof (*entry)); - entry->delivery_mode = apic->irq_delivery_mode; + entry->delivery_mode = apic->delivery_mode; entry->dest_mode = apic->irq_dest_mode; entry->dest = cfg->dest_apicid; entry->vector = cfg->vector;
arch/x86/kernel/apic/probe_32.c +1 -1
@@ -69,7 +69,7 @@ static struct apic apic_default __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = default_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* logical delivery broadcast to all CPUs: */ .irq_dest_mode = 1,
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -184,7 +184,7 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .apic_id_valid = x2apic_apic_id_valid, .apic_id_registered = x2apic_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 1, /* logical */ .disable_esr = 0,
arch/x86/kernel/apic/x2apic_phys.c +1 -1
@@ -157,7 +157,7 @@ static struct apic apic_x2apic_phys __ro_after_init = { .apic_id_valid = x2apic_apic_id_valid, .apic_id_registered = x2apic_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/x2apic_uv_x.c +3 -3
@@ -703,9 +703,9 @@ static void uv_send_IPI_one(int cpu, int vector)unsigned long dmode, val;if (vector ==NMI_VECTOR ) - dmode = dest_NMI; + dmode =APIC_DELIVERY_MODE_NMI ;else - dmode = dest_Fixed; + dmode =APIC_DELIVERY_MODE_FIXED ; val = (1UL <<UVH_IPI_INT_SEND_SHFT ) | (apicid << UVH_IPI_INT_APIC_ID_SHFT) |
@@ -807,7 +807,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .apic_id_valid = uv_apic_id_valid, .apic_id_registered = uv_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* Physical */ .disable_esr = 0,
arch/x86/platform/uv/uv_irq.c +1 -1
@@ -35,7 +35,7 @@ static void uv_program_mmr(struct irq_cfg *cfg, struct uv_irq_2_mmr_pnode *info) mmr_value =0 ; entry = (struct uv_IO_APIC_route_entry *)&mmr_value; entry->vector = cfg->vector; - entry->delivery_mode = apic->irq_delivery_mode; + entry->delivery_mode = apic->delivery_mode; entry->dest_mode = apic->irq_dest_mode; entry->polarity =0 ; entry->trigger =0 ;
drivers/iommu/amd/iommu.c +2 -2
@@ -3671,7 +3671,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, data->irq_2_irte.devid = devid; data->irq_2_irte.index = index + sub_handle; - iommu->irte_ops->prepare (data->entry, apic->irq_delivery_mode, + iommu->irte_ops->prepare (data->entry, apic->delivery_mode, apic->irq_dest_mode, irq_cfg->vector, irq_cfg->dest_apicid, devid);
@@ -3944,7 +3944,7 @@ int amd_iommu_deactivate_guest_mode(void *data) entry->lo.fields_remap.valid = valid; entry->lo.fields_remap.dm = apic->irq_dest_mode; - entry->lo.fields_remap.int_type = apic->irq_delivery_mode; + entry->lo.fields_remap.int_type = apic->delivery_mode; entry->hi.fields.vector = cfg->vector; entry->lo.fields_remap.destination =APICID_TO_IRTE_DEST_LO (cfg->dest_apicid);
drivers/iommu/intel/irq_remapping.c +1 -1
@@ -1122,7 +1122,7 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest)* irq migration in the presence of interrupt-remapping. */ irte->trigger_mode =0 ; - irte->dlvry_mode = apic->irq_delivery_mode; + irte->dlvry_mode = apic->delivery_mode; irte->vector = vector; irte->dest_id =IRTE_DEST (dest); irte->redir_hint =1 ;
drivers/pci/controller/pci-hyperv.c +3 -3
@@ -1226,7 +1226,7 @@ static void hv_irq_unmask(struct irq_data *data) params->int_target.vector = cfg->vector;/* -* Honoring apic->irq_delivery_mode set to dest_Fixed by +* Honoring apic->delivery_mode set to APIC_DELIVERY_MODE_FIXED by * setting the HV_DEVICE_INTERRUPT_TARGET_MULTICAST flag results in a * spurious interrupt storm. Not doing so does not seem to have a * negative effect (yet?).
@@ -1310,7 +1310,7 @@ static u32 hv_compose_msi_req_v1( int_pkt->wslot.slot = slot; int_pkt->int_desc.vector = vector; int_pkt->int_desc.vector_count =1 ; - int_pkt->int_desc.delivery_mode = dest_Fixed; + int_pkt->int_desc.delivery_mode =APIC_DELIVERY_MODE_FIXED ;/* * Create MSI w/ dummy vCPU set, overwritten by subsequent retarget in
@@ -1331,7 +1331,7 @@ static u32 hv_compose_msi_req_v2( int_pkt->wslot.slot = slot; int_pkt->int_desc.vector = vector; int_pkt->int_desc.vector_count =1 ; - int_pkt->int_desc.delivery_mode = dest_Fixed; + int_pkt->int_desc.delivery_mode =APIC_DELIVERY_MODE_FIXED ;/* * Create MSI w/ dummy vCPU set targeting just one vCPU, overwritten
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/apic: Cleanup delivery mode defines
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 721612994f53ed600b39a80d912b10f51960e2e3 Gitweb: https://git.kernel.org/tip/721612994f53ed600b39a80d912b10f51960e2e3 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:05 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/apic: Cleanup delivery mode defines
The enum ioapic_irq_destination_types and the enumerated constants starting with 'dest_' are gross misnomers because they describe the delivery mode.
Rename then enum and the constants so they actually make sense.
--- arch/x86/include/asm/apic.h | 3 ++- arch/x86/include/asm/apicdef.h | 16 +++++++--------- arch/x86/kernel/apic/apic_flat_64.c | 4 ++-- arch/x86/kernel/apic/apic_noop.c | 2 +- arch/x86/kernel/apic/apic_numachip.c | 4 ++-- arch/x86/kernel/apic/bigsmp_32.c | 2 +- arch/x86/kernel/apic/io_apic.c | 11 ++++++----- arch/x86/kernel/apic/probe_32.c | 2 +- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- arch/x86/kernel/apic/x2apic_phys.c | 2 +- arch/x86/kernel/apic/x2apic_uv_x.c | 6 +++--- arch/x86/platform/uv/uv_irq.c | 2 +- drivers/iommu/amd/iommu.c | 4 ++-- drivers/iommu/intel/irq_remapping.c | 2 +- drivers/pci/controller/pci-hyperv.c | 6 +++--- 15 files changed, 34 insertions(+), 34 deletions(-)
arch/x86/include/asm/apic.h +2 -1
@@ -309,7 +309,8 @@ struct apic {/* dest_logical is used by the IPI functions */ u32 dest_logical;u32 disable_esr; -u32 irq_delivery_mode; + +enum apic_delivery_modes delivery_mode;u32 irq_dest_mode;u32 (*calc_dest_apicid)(unsigned int cpu);
arch/x86/include/asm/apicdef.h +7 -9
@@ -432,15 +432,13 @@ struct local_apic {#define BAD_APICID 0xFFFFu#endif -enum ioapic_irq_destination_types { - dest_Fixed =0 , - dest_LowestPrio =1 , - dest_SMI =2 , - dest__reserved_1 =3 , - dest_NMI =4 , - dest_INIT =5 , - dest__reserved_2 =6 , - dest_ExtINT =7 +enum apic_delivery_modes { +APIC_DELIVERY_MODE_FIXED =0 , +APIC_DELIVERY_MODE_LOWESTPRIO =1 , +APIC_DELIVERY_MODE_SMI =2 , +APIC_DELIVERY_MODE_NMI =4 , +APIC_DELIVERY_MODE_INIT =5 , +APIC_DELIVERY_MODE_EXTINT =7 , };#endif /* _ASM_X86_APICDEF_H */
arch/x86/kernel/apic/apic_flat_64.c +2 -2
@@ -113,7 +113,7 @@ static struct apic apic_flat __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = flat_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 1, /* logical */ .disable_esr = 0,
@@ -206,7 +206,7 @@ static struct apic apic_physflat __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = flat_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/apic_noop.c +1 -1
@@ -95,7 +95,7 @@ struct apic apic_noop __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = noop_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* logical delivery broadcast to all CPUs: */ .irq_dest_mode = 1,
arch/x86/kernel/apic/apic_numachip.c +2 -2
@@ -246,7 +246,7 @@ static const struct apic apic_numachip1 __refconst = { .apic_id_valid = numachip_apic_id_valid, .apic_id_registered = numachip_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
@@ -295,7 +295,7 @@ static const struct apic apic_numachip2 __refconst = { .apic_id_valid = numachip_apic_id_valid, .apic_id_registered = numachip_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/bigsmp_32.c +1 -1
@@ -127,7 +127,7 @@ static struct apic apic_bigsmp __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = bigsmp_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* phys delivery to target CPU: */ .irq_dest_mode = 0,
arch/x86/kernel/apic/io_apic.c +6 -5
@@ -535,7 +535,7 @@ static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)/* Check delivery_mode to be sure we're not clearing an SMI pin */ entry =ioapic_read_entry (apic, pin); -if (entry.delivery_mode == dest_SMI) +if (entry.delivery_mode ==APIC_DELIVERY_MODE_SMI )return ;/*
@@ -1368,7 +1368,8 @@ void __init enable_IO_APIC(void)/* If the interrupt line is enabled and in ExtInt mode * I have found the pin where the i8259 is connected. */ -if ((entry.mask ==0 ) && (entry.delivery_mode == dest_ExtINT)) { +if ((entry.mask ==0 ) && + (entry.delivery_mode ==APIC_DELIVERY_MODE_EXTINT )) { ioapic_i8259.apic = apic; ioapic_i8259.pin = pin; goto found_i8259;
@@ -1416,7 +1417,7 @@ void native_restore_boot_irq_mode(void) entry.trigger =IOAPIC_EDGE ; entry.polarity =IOAPIC_POL_HIGH ; entry.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; - entry.delivery_mode = dest_ExtINT; + entry.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; entry.dest =read_apic_id ();/*
@@ -2047,7 +2048,7 @@ static inline void __init unlock_ExtINT_logic(void) entry1.dest_mode =IOAPIC_DEST_MODE_PHYSICAL ; entry1.mask =IOAPIC_UNMASKED ; entry1.dest =hard_smp_processor_id (); - entry1.delivery_mode = dest_ExtINT; + entry1.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; entry1.polarity = entry0.polarity; entry1.trigger =IOAPIC_EDGE ; entry1.vector =0 ;
@@ -2948,7 +2949,7 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, struct IO_APIC_route_entry *entry) {memset (entry,0 ,sizeof (*entry)); - entry->delivery_mode = apic->irq_delivery_mode; + entry->delivery_mode = apic->delivery_mode; entry->dest_mode = apic->irq_dest_mode; entry->dest = cfg->dest_apicid; entry->vector = cfg->vector;
arch/x86/kernel/apic/probe_32.c +1 -1
@@ -69,7 +69,7 @@ static struct apic apic_default __ro_after_init = { .apic_id_valid = default_apic_id_valid, .apic_id_registered = default_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED,/* logical delivery broadcast to all CPUs: */ .irq_dest_mode = 1,
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -184,7 +184,7 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .apic_id_valid = x2apic_apic_id_valid, .apic_id_registered = x2apic_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 1, /* logical */ .disable_esr = 0,
arch/x86/kernel/apic/x2apic_phys.c +1 -1
@@ -157,7 +157,7 @@ static struct apic apic_x2apic_phys __ro_after_init = { .apic_id_valid = x2apic_apic_id_valid, .apic_id_registered = x2apic_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* physical */ .disable_esr = 0,
arch/x86/kernel/apic/x2apic_uv_x.c +3 -3
@@ -703,9 +703,9 @@ static void uv_send_IPI_one(int cpu, int vector)unsigned long dmode, val;if (vector ==NMI_VECTOR ) - dmode = dest_NMI; + dmode =APIC_DELIVERY_MODE_NMI ;else - dmode = dest_Fixed; + dmode =APIC_DELIVERY_MODE_FIXED ; val = (1UL <<UVH_IPI_INT_SEND_SHFT ) | (apicid << UVH_IPI_INT_APIC_ID_SHFT) |
@@ -807,7 +807,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .apic_id_valid = uv_apic_id_valid, .apic_id_registered = uv_apic_id_registered, - .irq_delivery_mode = dest_Fixed, + .delivery_mode = APIC_DELIVERY_MODE_FIXED, .irq_dest_mode = 0, /* Physical */ .disable_esr = 0,
arch/x86/platform/uv/uv_irq.c +1 -1
@@ -35,7 +35,7 @@ static void uv_program_mmr(struct irq_cfg *cfg, struct uv_irq_2_mmr_pnode *info) mmr_value =0 ; entry = (struct uv_IO_APIC_route_entry *)&mmr_value; entry->vector = cfg->vector; - entry->delivery_mode = apic->irq_delivery_mode; + entry->delivery_mode = apic->delivery_mode; entry->dest_mode = apic->irq_dest_mode; entry->polarity =0 ; entry->trigger =0 ;
drivers/iommu/amd/iommu.c +2 -2
@@ -3671,7 +3671,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, data->irq_2_irte.devid = devid; data->irq_2_irte.index = index + sub_handle; - iommu->irte_ops->prepare (data->entry, apic->irq_delivery_mode, + iommu->irte_ops->prepare (data->entry, apic->delivery_mode, apic->irq_dest_mode, irq_cfg->vector, irq_cfg->dest_apicid, devid);
@@ -3944,7 +3944,7 @@ int amd_iommu_deactivate_guest_mode(void *data) entry->lo.fields_remap.valid = valid; entry->lo.fields_remap.dm = apic->irq_dest_mode; - entry->lo.fields_remap.int_type = apic->irq_delivery_mode; + entry->lo.fields_remap.int_type = apic->delivery_mode; entry->hi.fields.vector = cfg->vector; entry->lo.fields_remap.destination =APICID_TO_IRTE_DEST_LO (cfg->dest_apicid);
drivers/iommu/intel/irq_remapping.c +1 -1
@@ -1122,7 +1122,7 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest)* irq migration in the presence of interrupt-remapping. */ irte->trigger_mode =0 ; - irte->dlvry_mode = apic->irq_delivery_mode; + irte->dlvry_mode = apic->delivery_mode; irte->vector = vector; irte->dest_id =IRTE_DEST (dest); irte->redir_hint =1 ;
drivers/pci/controller/pci-hyperv.c +3 -3
@@ -1226,7 +1226,7 @@ static void hv_irq_unmask(struct irq_data *data) params->int_target.vector = cfg->vector;/* -* Honoring apic->irq_delivery_mode set to dest_Fixed by +* Honoring apic->delivery_mode set to APIC_DELIVERY_MODE_FIXED by * setting the HV_DEVICE_INTERRUPT_TARGET_MULTICAST flag results in a * spurious interrupt storm. Not doing so does not seem to have a * negative effect (yet?).
@@ -1324,7 +1324,7 @@ static u32 hv_compose_msi_req_v1( int_pkt->wslot.slot = slot; int_pkt->int_desc.vector = vector; int_pkt->int_desc.vector_count =1 ; - int_pkt->int_desc.delivery_mode = dest_Fixed; + int_pkt->int_desc.delivery_mode =APIC_DELIVERY_MODE_FIXED ;/* * Create MSI w/ dummy vCPU set, overwritten by subsequent retarget in
@@ -1345,7 +1345,7 @@ static u32 hv_compose_msi_req_v2( int_pkt->wslot.slot = slot; int_pkt->int_desc.vector = vector; int_pkt->int_desc.vector_count =1 ; - int_pkt->int_desc.delivery_mode = dest_Fixed; + int_pkt->int_desc.delivery_mode =APIC_DELIVERY_MODE_FIXED ;/* * Create MSI w/ dummy vCPU set targeting just one vCPU, overwritten
David Woodhouse
maintainerPATCH v3 6/35 x86/apic: Replace pointless apic::dest_logical usage
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
All these functions are only used for logical destination mode. So reading the destination mode mask from the apic structure is a pointless exercise. Just hand in the proper constant: APIC_DEST_LOGICAL.
--- arch/x86/kernel/apic/apic_flat_64.c | 2 +- arch/x86/kernel/apic/ipi.c | 6 +++--- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- 3 files changed, 5 insertions(+), 5 deletions(-)
arch/x86/kernel/apic/apic_flat_64.c +1 -1
@@ -53,7 +53,7 @@ static void _flat_send_IPI_mask(unsigned long mask, int vector)unsigned long flags;local_irq_save (flags); -__default_send_IPI_dest_field (mask, vector, apic->dest_logical); +__default_send_IPI_dest_field (mask, vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
arch/x86/kernel/apic/ipi.c +3 -3
@@ -260,7 +260,7 @@ void default_send_IPI_mask_sequence_logical(const struct cpumask *mask,for_each_cpu (query_cpu, mask)__default_send_IPI_dest_field (early_per_cpu (x86_cpu_to_logical_apicid, query_cpu), - vector, apic->dest_logical); + vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
@@ -279,7 +279,7 @@ void default_send_IPI_mask_allbutself_logical(const struct cpumask *mask,continue ;__default_send_IPI_dest_field (early_per_cpu (x86_cpu_to_logical_apicid, query_cpu), - vector, apic->dest_logical); + vector,APIC_DEST_LOGICAL ); }local_irq_restore (flags); }
@@ -297,7 +297,7 @@ void default_send_IPI_mask_logical(const struct cpumask *cpumask, int vector)local_irq_save (flags);WARN_ON (mask & ~cpumask_bits (cpu_online_mask)[0 ]); -__default_send_IPI_dest_field (mask, vector, apic->dest_logical); +__default_send_IPI_dest_field (mask, vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -61,7 +61,7 @@ __x2apic_send_IPI_mask(const struct cpumask *mask, int vector, int apic_dest)if (!dest)continue ; -__x2apic_send_IPI_dest (dest, vector, apic->dest_logical); +__x2apic_send_IPI_dest (dest, vector,APIC_DEST_LOGICAL );/* Remove cluster CPUs from tmpmask */ cpumask_andnot (tmpmsk, tmpmsk, &cmsk->mask); }
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/apic: Replace pointless apic:: Dest_logical usage
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 22e0db42097b30a49771744e514350b7e9dd26f2 Gitweb: https://git.kernel.org/tip/22e0db42097b30a49771744e514350b7e9dd26f2 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:06 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/apic: Replace pointless apic:: Dest_logical usage
All these functions are only used for logical destination mode. So reading the destination mode mask from the apic structure is a pointless exercise. Just hand in the proper constant: APIC_DEST_LOGICAL.
--- arch/x86/kernel/apic/apic_flat_64.c | 2 +- arch/x86/kernel/apic/ipi.c | 6 +++--- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- 3 files changed, 5 insertions(+), 5 deletions(-)
arch/x86/kernel/apic/apic_flat_64.c +1 -1
@@ -53,7 +53,7 @@ static void _flat_send_IPI_mask(unsigned long mask, int vector)unsigned long flags;local_irq_save (flags); -__default_send_IPI_dest_field (mask, vector, apic->dest_logical); +__default_send_IPI_dest_field (mask, vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
arch/x86/kernel/apic/ipi.c +3 -3
@@ -260,7 +260,7 @@ void default_send_IPI_mask_sequence_logical(const struct cpumask *mask,for_each_cpu (query_cpu, mask)__default_send_IPI_dest_field (early_per_cpu (x86_cpu_to_logical_apicid, query_cpu), - vector, apic->dest_logical); + vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
@@ -279,7 +279,7 @@ void default_send_IPI_mask_allbutself_logical(const struct cpumask *mask,continue ;__default_send_IPI_dest_field (early_per_cpu (x86_cpu_to_logical_apicid, query_cpu), - vector, apic->dest_logical); + vector,APIC_DEST_LOGICAL ); }local_irq_restore (flags); }
@@ -297,7 +297,7 @@ void default_send_IPI_mask_logical(const struct cpumask *cpumask, int vector)local_irq_save (flags);WARN_ON (mask & ~cpumask_bits (cpu_online_mask)[0 ]); -__default_send_IPI_dest_field (mask, vector, apic->dest_logical); +__default_send_IPI_dest_field (mask, vector,APIC_DEST_LOGICAL );local_irq_restore (flags); }
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -61,7 +61,7 @@ __x2apic_send_IPI_mask(const struct cpumask *mask, int vector, int apic_dest)if (!dest)continue ; -__x2apic_send_IPI_dest (dest, vector, apic->dest_logical); +__x2apic_send_IPI_dest (dest, vector,APIC_DEST_LOGICAL );/* Remove cluster CPUs from tmpmask */ cpumask_andnot (tmpmsk, tmpmsk, &cmsk->mask); }
David Woodhouse
maintainerPATCH v3 7/35 x86/apic: Get rid of apic::dest_logical
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
struct apic has two members which store information about the destination
mode: dest_logical and irq_dest_mode.
dest_logical contains a mask which was historically used to set the destination mode in IPI messages. Over time the usage was reduced and the logical/physical functions were seperated.
There are only a few places which still use 'dest_logical' but they can ues 'irq_dest_mode' instead.
irq_dest_mode is actually a boolean where 0 means physical destination mode and 1 means logical destination mode. Of course the name does not reflect the functionality. This will be cleaned up in a subsequent change.
Remove apic::dest_logical and fixup the remaining users.
--- arch/x86/include/asm/apic.h | 2 -- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/apic/apic_flat_64.c | 8 ++------ arch/x86/kernel/apic/apic_noop.c | 4 +--- arch/x86/kernel/apic/apic_numachip.c | 8 ++------ arch/x86/kernel/apic/bigsmp_32.c | 4 +--- arch/x86/kernel/apic/probe_32.c | 4 +--- arch/x86/kernel/apic/x2apic_cluster.c | 4 +--- arch/x86/kernel/apic/x2apic_phys.c | 4 +--- arch/x86/kernel/apic/x2apic_uv_x.c | 4 +--- arch/x86/kernel/smpboot.c | 5 +++-- arch/x86/xen/apic.c | 4 +--- 12 files changed, 15 insertions(+), 38 deletions(-)
arch/x86/include/asm/apic.h +0 -2
@@ -306,8 +306,6 @@ struct apic {void (*send_IPI_all)(int vector);void (*send_IPI_self)(int vector); -/* dest_logical is used by the IPI functions */ -u32 dest_logical;u32 disable_esr;enum apic_delivery_modes delivery_mode;
arch/x86/kernel/apic/apic.c +1 -1
@@ -1591,7 +1591,7 @@ static void setup_local_APIC(void) apic->init_apic_ldr ();#ifdef CONFIG_X86_32 -if (apic->dest_logical) { +if (apic->irq_dest_mode ==1 ) {int logical_apicid, ldr_apicid;/*
arch/x86/kernel/apic/apic_flat_64.c +2 -6
@@ -117,11 +117,9 @@ static struct apic apic_flat __ro_after_init = { .irq_dest_mode = 1, /* logical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
@@ -210,11 +208,9 @@ static struct apic apic_physflat __ro_after_init = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = physflat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/apic_noop.c +1 -3
@@ -100,11 +100,9 @@ struct apic apic_noop __ro_after_init = { .irq_dest_mode = 1, .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = default_check_apicid_used, + .check_apicid_used = default_check_apicid_used, .init_apic_ldr = noop_init_apic_ldr, - .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = NULL,
arch/x86/kernel/apic/apic_numachip.c +2 -6
@@ -250,11 +250,9 @@ static const struct apic apic_numachip1 __refconst = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
@@ -299,11 +297,9 @@ static const struct apic apic_numachip2 __refconst = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/bigsmp_32.c +1 -3
@@ -132,11 +132,9 @@ static struct apic apic_bigsmp __ro_after_init = { .irq_dest_mode = 0, .disable_esr = 1, - .dest_logical = 0, - .check_apicid_used = bigsmp_check_apicid_used, + .check_apicid_used = bigsmp_check_apicid_used, .init_apic_ldr = bigsmp_init_apic_ldr, - .ioapic_phys_id_map = bigsmp_ioapic_phys_id_map, .setup_apic_routing = bigsmp_setup_apic_routing, .cpu_present_to_apicid = bigsmp_cpu_present_to_apicid,
arch/x86/kernel/apic/probe_32.c +1 -3
@@ -74,11 +74,9 @@ static struct apic apic_default __ro_after_init = { .irq_dest_mode = 1, .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = default_check_apicid_used, + .check_apicid_used = default_check_apicid_used, .init_apic_ldr = default_init_apic_ldr, - .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = setup_apic_flat_routing, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_cluster.c +1 -3
@@ -188,11 +188,9 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .irq_dest_mode = 1, /* logical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = init_x2apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_phys.c +1 -3
@@ -161,11 +161,9 @@ static struct apic apic_x2apic_phys __ro_after_init = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = init_x2apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_uv_x.c +1 -3
@@ -811,11 +811,9 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .irq_dest_mode = 0, /* Physical */ .disable_esr = 0, - .dest_logical = APIC_DEST_PHYSICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = uv_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/smpboot.c +3 -2
@@ -747,13 +747,14 @@ static void __init smp_quirk_init_udelay(void)int wakeup_secondary_cpu_via_nmi (int apicid,unsigned long start_eip) { +u32 dm = apic->irq_dest_mode ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ;unsigned long send_status, accept_status =0 ;int maxlvt;/* Target chip */ /* Boot on the stack */ /* Kick the second */ -apic_icr_write (APIC_DM_NMI | apic->dest_logical, apicid); +apic_icr_write (APIC_DM_NMI | dm, apicid);pr_debug ("Waiting for send to finish...\n" ); send_status =safe_apic_wait_icr_idle ();
@@ -980,7 +981,7 @@ wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,if (!boot_error) { enable_start_cpu0 =1 ; *cpu0_nmi_registered =1 ; -if (apic->dest_logical ==APIC_DEST_LOGICAL ) +if (apic->irq_dest_mode) id = cpu0_logical_apicid;else id = apicid;
arch/x86/xen/apic.c +1 -3
@@ -152,11 +152,9 @@ static struct apic xen_pv_apic = {/* .irq_dest_mode - used in native_compose_msi_msg only */ .disable_esr = 0, -/* .dest_logical - default_send_IPI_ use it but we use our own. */ - .check_apicid_used = default_check_apicid_used, /* Used on 32-bit */ + .check_apicid_used = default_check_apicid_used, /* Used on 32-bit */ .init_apic_ldr = xen_noop, /* setup_local_APIC calls it */ - .ioapic_phys_id_map = default_ioapic_phys_id_map, /* Used on 32-bit */ .setup_apic_routing = NULL, .cpu_present_to_apicid = xen_cpu_present_to_apicid,
Signature · 1 line
2.26.2
[tip: x86/apic] x86/apic: Get rid of apic:: Dest_logical
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: e57d04e5fa00f7649d4c00796f8d12054799be4a Gitweb: https://git.kernel.org/tip/e57d04e5fa00f7649d4c00796f8d12054799be4a Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:07 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:24 +01:00
x86/apic: Get rid of apic:: Dest_logical
struct apic has two members which store information about the destination
mode: dest_logical and irq_dest_mode.
dest_logical contains a mask which was historically used to set the destination mode in IPI messages. Over time the usage was reduced and the logical/physical functions were seperated.
There are only a few places which still use 'dest_logical' but they can use 'irq_dest_mode' instead.
irq_dest_mode is actually a boolean where 0 means physical destination mode and 1 means logical destination mode. Of course the name does not reflect the functionality. This will be cleaned up in a subsequent change.
Remove apic::dest_logical and fixup the remaining users.
--- arch/x86/include/asm/apic.h | 2 -- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/apic/apic_flat_64.c | 8 ++------ arch/x86/kernel/apic/apic_noop.c | 4 +--- arch/x86/kernel/apic/apic_numachip.c | 8 ++------ arch/x86/kernel/apic/bigsmp_32.c | 4 +--- arch/x86/kernel/apic/probe_32.c | 4 +--- arch/x86/kernel/apic/x2apic_cluster.c | 4 +--- arch/x86/kernel/apic/x2apic_phys.c | 4 +--- arch/x86/kernel/apic/x2apic_uv_x.c | 4 +--- arch/x86/kernel/smpboot.c | 5 +++-- arch/x86/xen/apic.c | 4 +--- 12 files changed, 15 insertions(+), 38 deletions(-)
arch/x86/include/asm/apic.h +0 -2
@@ -306,8 +306,6 @@ struct apic {void (*send_IPI_all)(int vector);void (*send_IPI_self)(int vector); -/* dest_logical is used by the IPI functions */ -u32 dest_logical;u32 disable_esr;enum apic_delivery_modes delivery_mode;
arch/x86/kernel/apic/apic.c +1 -1
@@ -1591,7 +1591,7 @@ static void setup_local_APIC(void) apic->init_apic_ldr ();#ifdef CONFIG_X86_32 -if (apic->dest_logical) { +if (apic->irq_dest_mode ==1 ) {int logical_apicid, ldr_apicid;/*
arch/x86/kernel/apic/apic_flat_64.c +2 -6
@@ -117,11 +117,9 @@ static struct apic apic_flat __ro_after_init = { .irq_dest_mode = 1, /* logical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
@@ -210,11 +208,9 @@ static struct apic apic_physflat __ro_after_init = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = physflat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/apic_noop.c +1 -3
@@ -100,11 +100,9 @@ struct apic apic_noop __ro_after_init = { .irq_dest_mode = 1, .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = default_check_apicid_used, + .check_apicid_used = default_check_apicid_used, .init_apic_ldr = noop_init_apic_ldr, - .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = NULL,
arch/x86/kernel/apic/apic_numachip.c +2 -6
@@ -250,11 +250,9 @@ static const struct apic apic_numachip1 __refconst = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
@@ -299,11 +297,9 @@ static const struct apic apic_numachip2 __refconst = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = flat_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/bigsmp_32.c +1 -3
@@ -132,11 +132,9 @@ static struct apic apic_bigsmp __ro_after_init = { .irq_dest_mode = 0, .disable_esr = 1, - .dest_logical = 0, - .check_apicid_used = bigsmp_check_apicid_used, + .check_apicid_used = bigsmp_check_apicid_used, .init_apic_ldr = bigsmp_init_apic_ldr, - .ioapic_phys_id_map = bigsmp_ioapic_phys_id_map, .setup_apic_routing = bigsmp_setup_apic_routing, .cpu_present_to_apicid = bigsmp_cpu_present_to_apicid,
arch/x86/kernel/apic/probe_32.c +1 -3
@@ -74,11 +74,9 @@ static struct apic apic_default __ro_after_init = { .irq_dest_mode = 1, .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = default_check_apicid_used, + .check_apicid_used = default_check_apicid_used, .init_apic_ldr = default_init_apic_ldr, - .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = setup_apic_flat_routing, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_cluster.c +1 -3
@@ -188,11 +188,9 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .irq_dest_mode = 1, /* logical */ .disable_esr = 0, - .dest_logical = APIC_DEST_LOGICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = init_x2apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_phys.c +1 -3
@@ -161,11 +161,9 @@ static struct apic apic_x2apic_phys __ro_after_init = { .irq_dest_mode = 0, /* physical */ .disable_esr = 0, - .dest_logical = 0, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = init_x2apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/apic/x2apic_uv_x.c +1 -3
@@ -811,11 +811,9 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .irq_dest_mode = 0, /* Physical */ .disable_esr = 0, - .dest_logical = APIC_DEST_PHYSICAL, - .check_apicid_used = NULL, + .check_apicid_used = NULL, .init_apic_ldr = uv_init_apic_ldr, - .ioapic_phys_id_map = NULL, .setup_apic_routing = NULL, .cpu_present_to_apicid = default_cpu_present_to_apicid,
arch/x86/kernel/smpboot.c +3 -2
@@ -747,13 +747,14 @@ static void __init smp_quirk_init_udelay(void)int wakeup_secondary_cpu_via_nmi (int apicid,unsigned long start_eip) { +u32 dm = apic->irq_dest_mode ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ;unsigned long send_status, accept_status =0 ;int maxlvt;/* Target chip */ /* Boot on the stack */ /* Kick the second */ -apic_icr_write (APIC_DM_NMI | apic->dest_logical, apicid); +apic_icr_write (APIC_DM_NMI | dm, apicid);pr_debug ("Waiting for send to finish...\n" ); send_status =safe_apic_wait_icr_idle ();
@@ -980,7 +981,7 @@ wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,if (!boot_error) { enable_start_cpu0 =1 ; *cpu0_nmi_registered =1 ; -if (apic->dest_logical ==APIC_DEST_LOGICAL ) +if (apic->irq_dest_mode) id = cpu0_logical_apicid;else id = apicid;
arch/x86/xen/apic.c +1 -3
@@ -152,11 +152,9 @@ static struct apic xen_pv_apic = {/* .irq_dest_mode - used in native_compose_msi_msg only */ .disable_esr = 0, -/* .dest_logical - default_send_IPI_ use it but we use our own. */ - .check_apicid_used = default_check_apicid_used, /* Used on 32-bit */ + .check_apicid_used = default_check_apicid_used, /* Used on 32-bit */ .init_apic_ldr = xen_noop, /* setup_local_APIC calls it */ - .ioapic_phys_id_map = default_ioapic_phys_id_map, /* Used on 32-bit */ .setup_apic_routing = NULL, .cpu_present_to_apicid = xen_cpu_present_to_apicid,
David Woodhouse
maintainerPATCH v3 8/35 x86/apic: Cleanup destination mode
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
apic::irq_dest_mode is actually a boolean, but defined as u32 and named in a way which does not explain what it means.
Make it a boolean and rename it to 'dest_mode_logical'
--- arch/x86/include/asm/apic.h | 2 +- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/apic/apic_flat_64.c | 4 ++-- arch/x86/kernel/apic/apic_noop.c | 4 +--- arch/x86/kernel/apic/apic_numachip.c | 4 ++-- arch/x86/kernel/apic/bigsmp_32.c | 3 +-- arch/x86/kernel/apic/io_apic.c | 2 +- arch/x86/kernel/apic/msi.c | 6 +++--- arch/x86/kernel/apic/probe_32.c | 3 +-- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- arch/x86/kernel/apic/x2apic_phys.c | 2 +- arch/x86/kernel/apic/x2apic_uv_x.c | 2 +- arch/x86/kernel/smpboot.c | 7 ++----- arch/x86/platform/uv/uv_irq.c | 2 +- arch/x86/xen/apic.c | 3 +-- drivers/iommu/amd/amd_iommu_types.h | 2 +- drivers/iommu/amd/iommu.c | 8 ++++---- drivers/iommu/intel/irq_remapping.c | 2 +- 18 files changed, 26 insertions(+), 34 deletions(-)
arch/x86/include/asm/apic.h +1 -1
@@ -309,7 +309,7 @@ struct apic {u32 disable_esr;enum apic_delivery_modes delivery_mode; -u32 irq_dest_mode; +bool dest_mode_logical;u32 (*calc_dest_apicid)(unsigned int cpu);
arch/x86/kernel/apic/apic.c +1 -1
@@ -1591,7 +1591,7 @@ static void setup_local_APIC(void) apic->init_apic_ldr ();#ifdef CONFIG_X86_32 -if (apic->irq_dest_mode ==1 ) { +if (apic->dest_mode_logical) {int logical_apicid, ldr_apicid;/*
arch/x86/kernel/apic/apic_flat_64.c +2 -2
@@ -114,7 +114,7 @@ static struct apic apic_flat __ro_after_init = { .apic_id_registered = flat_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 1, /* logical */ + .dest_mode_logical = true, .disable_esr = 0,
@@ -205,7 +205,7 @@ static struct apic apic_physflat __ro_after_init = { .apic_id_registered = flat_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/apic_noop.c +1 -3
@@ -96,8 +96,7 @@ struct apic apic_noop __ro_after_init = { .apic_id_registered = noop_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* logical delivery broadcast to all CPUs: */ - .irq_dest_mode = 1, + .dest_mode_logical = true, .disable_esr = 0,
@@ -105,7 +104,6 @@ struct apic apic_noop __ro_after_init = { .init_apic_ldr = noop_init_apic_ldr, .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = NULL, - .cpu_present_to_apicid = default_cpu_present_to_apicid, .apicid_to_cpu_present = physid_set_mask_of_physid,
arch/x86/kernel/apic/apic_numachip.c +2 -2
@@ -247,7 +247,7 @@ static const struct apic apic_numachip1 __refconst = { .apic_id_registered = numachip_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
@@ -294,7 +294,7 @@ static const struct apic apic_numachip2 __refconst = { .apic_id_registered = numachip_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/bigsmp_32.c +1 -2
@@ -128,8 +128,7 @@ static struct apic apic_bigsmp __ro_after_init = { .apic_id_registered = bigsmp_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* phys delivery to target CPU: */ - .irq_dest_mode = 0, + .dest_mode_logical = false, .disable_esr = 1,
arch/x86/kernel/apic/io_apic.c +1 -1
@@ -2950,7 +2950,7 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, {memset (entry,0 ,sizeof (*entry)); entry->delivery_mode = apic->delivery_mode; - entry->dest_mode = apic->irq_dest_mode; + entry->dest_mode = apic->dest_mode_logical; entry->dest = cfg->dest_apicid; entry->vector = cfg->vector; entry->trigger = data->trigger;
arch/x86/kernel/apic/msi.c +3 -3
@@ -30,9 +30,9 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg, msg->address_lo =MSI_ADDR_BASE_LO | - ((apic->irq_dest_mode ==0 ) ? -MSI_ADDR_DEST_MODE_PHYSICAL : -MSI_ADDR_DEST_MODE_LOGICAL ) | + (apic->dest_mode_logical ? +MSI_ADDR_DEST_MODE_LOGICAL : +MSI_ADDR_DEST_MODE_PHYSICAL ) |MSI_ADDR_REDIRECTION_CPU |MSI_ADDR_DEST_ID (cfg->dest_apicid);
arch/x86/kernel/apic/probe_32.c +1 -2
@@ -70,8 +70,7 @@ static struct apic apic_default __ro_after_init = { .apic_id_registered = default_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* logical delivery broadcast to all CPUs: */ - .irq_dest_mode = 1, + .dest_mode_logical = true, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -185,7 +185,7 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .apic_id_registered = x2apic_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 1, /* logical */ + .dest_mode_logical = true, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_phys.c +1 -1
@@ -158,7 +158,7 @@ static struct apic apic_x2apic_phys __ro_after_init = { .apic_id_registered = x2apic_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_uv_x.c +1 -1
@@ -808,7 +808,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .apic_id_registered = uv_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* Physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/smpboot.c +2 -5
@@ -747,7 +747,7 @@ static void __init smp_quirk_init_udelay(void)int wakeup_secondary_cpu_via_nmi (int apicid,unsigned long start_eip) { -u32 dm = apic->irq_dest_mode ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ; +u32 dm = apic->dest_mode_logical ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ;unsigned long send_status, accept_status =0 ;int maxlvt;
@@ -981,10 +981,7 @@ wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,if (!boot_error) { enable_start_cpu0 =1 ; *cpu0_nmi_registered =1 ; -if (apic->irq_dest_mode) - id = cpu0_logical_apicid; -else - id = apicid; + id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid; boot_error =wakeup_secondary_cpu_via_nmi (id, start_ip); }
arch/x86/platform/uv/uv_irq.c +1 -1
@@ -36,7 +36,7 @@ static void uv_program_mmr(struct irq_cfg *cfg, struct uv_irq_2_mmr_pnode *info) entry = (struct uv_IO_APIC_route_entry *)&mmr_value; entry->vector = cfg->vector; entry->delivery_mode = apic->delivery_mode; - entry->dest_mode = apic->irq_dest_mode; + entry->dest_mode = apic->dest_mode_logical; entry->polarity =0 ; entry->trigger =0 ; entry->mask =0 ;
arch/x86/xen/apic.c +1 -2
@@ -148,8 +148,7 @@ static struct apic xen_pv_apic = { .apic_id_valid = xen_id_always_valid, .apic_id_registered = xen_id_always_registered, -/* .irq_delivery_mode - used in native_compose_msi_msg only */ -/* .irq_dest_mode - used in native_compose_msi_msg only */ +/* .delivery_mode and .dest_mode_logical not used by XENPV */ .disable_esr = 0,
drivers/iommu/amd/amd_iommu_types.h +1 -1
@@ -893,7 +893,7 @@ struct amd_ir_data { };struct amd_irte_ops { -void (*prepare)(void *,u32 ,u32 ,u8 ,u32 ,int ); +void (*prepare)(void *,u32 ,bool ,u8 ,u32 ,int );void (*activate)(void *,u16 ,u16 );void (*deactivate)(void *,u16 ,u16 );void (*set_affinity)(void *,u16 ,u16 ,u8 ,u32 );
drivers/iommu/amd/iommu.c +4 -4
@@ -3466,7 +3466,7 @@ static void free_irte(u16 devid, int index) }static void irte_prepare (void *entry, -u32 delivery_mode,u32 dest_mode, +u32 delivery_mode,bool dest_mode,u8 vector,u32 dest_apicid,int devid) {union irte *irte = (union irte *) entry;
@@ -3480,7 +3480,7 @@ static void irte_prepare(void *entry, }static void irte_ga_prepare (void *entry, -u32 delivery_mode,u32 dest_mode, +u32 delivery_mode,bool dest_mode,u8 vector,u32 dest_apicid,int devid) {struct irte_ga *irte = (struct irte_ga *) entry;
@@ -3672,7 +3672,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, data->irq_2_irte.devid = devid; data->irq_2_irte.index = index + sub_handle; iommu->irte_ops->prepare (data->entry, apic->delivery_mode, - apic->irq_dest_mode, irq_cfg->vector, + apic->dest_mode_logical, irq_cfg->vector, irq_cfg->dest_apicid, devid);switch (info->type) {
@@ -3943,7 +3943,7 @@ int amd_iommu_deactivate_guest_mode(void *data) entry->hi.val =0 ; entry->lo.fields_remap.valid = valid; - entry->lo.fields_remap.dm = apic->irq_dest_mode; + entry->lo.fields_remap.dm = apic->dest_mode_logical; entry->lo.fields_remap.int_type = apic->delivery_mode; entry->hi.fields.vector = cfg->vector; entry->lo.fields_remap.destination =
drivers/iommu/intel/irq_remapping.c +1 -1
@@ -1113,7 +1113,7 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest)memset (irte,0 ,sizeof (*irte)); irte->present =1 ; - irte->dst_mode = apic->irq_dest_mode; + irte->dst_mode = apic->dest_mode_logical;/* * Trigger mode in the IRTE will always be edge, and for IO-APIC, the * actual level or edge trigger will be setup in the IO-APIC
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/apic: Cleanup destination mode
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 8c44963b603db76e3e5f57d90d027657ba43c1fe Gitweb: https://git.kernel.org/tip/8c44963b603db76e3e5f57d90d027657ba43c1fe Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:08 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
x86/apic: Cleanup destination mode
apic::irq_dest_mode is actually a boolean, but defined as u32 and named in a way which does not explain what it means.
Make it a boolean and rename it to 'dest_mode_logical'
--- arch/x86/include/asm/apic.h | 2 +- arch/x86/kernel/apic/apic.c | 2 +- arch/x86/kernel/apic/apic_flat_64.c | 4 ++-- arch/x86/kernel/apic/apic_noop.c | 4 +--- arch/x86/kernel/apic/apic_numachip.c | 4 ++-- arch/x86/kernel/apic/bigsmp_32.c | 3 +-- arch/x86/kernel/apic/io_apic.c | 2 +- arch/x86/kernel/apic/msi.c | 6 +++--- arch/x86/kernel/apic/probe_32.c | 3 +-- arch/x86/kernel/apic/x2apic_cluster.c | 2 +- arch/x86/kernel/apic/x2apic_phys.c | 2 +- arch/x86/kernel/apic/x2apic_uv_x.c | 2 +- arch/x86/kernel/smpboot.c | 7 ++----- arch/x86/platform/uv/uv_irq.c | 2 +- arch/x86/xen/apic.c | 3 +-- drivers/iommu/amd/amd_iommu_types.h | 2 +- drivers/iommu/amd/iommu.c | 8 ++++---- drivers/iommu/intel/irq_remapping.c | 2 +- 18 files changed, 26 insertions(+), 34 deletions(-)
arch/x86/include/asm/apic.h +1 -1
@@ -309,7 +309,7 @@ struct apic {u32 disable_esr;enum apic_delivery_modes delivery_mode; -u32 irq_dest_mode; +bool dest_mode_logical;u32 (*calc_dest_apicid)(unsigned int cpu);
arch/x86/kernel/apic/apic.c +1 -1
@@ -1591,7 +1591,7 @@ static void setup_local_APIC(void) apic->init_apic_ldr ();#ifdef CONFIG_X86_32 -if (apic->irq_dest_mode ==1 ) { +if (apic->dest_mode_logical) {int logical_apicid, ldr_apicid;/*
arch/x86/kernel/apic/apic_flat_64.c +2 -2
@@ -114,7 +114,7 @@ static struct apic apic_flat __ro_after_init = { .apic_id_registered = flat_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 1, /* logical */ + .dest_mode_logical = true, .disable_esr = 0,
@@ -205,7 +205,7 @@ static struct apic apic_physflat __ro_after_init = { .apic_id_registered = flat_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/apic_noop.c +1 -3
@@ -96,8 +96,7 @@ struct apic apic_noop __ro_after_init = { .apic_id_registered = noop_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* logical delivery broadcast to all CPUs: */ - .irq_dest_mode = 1, + .dest_mode_logical = true, .disable_esr = 0,
@@ -105,7 +104,6 @@ struct apic apic_noop __ro_after_init = { .init_apic_ldr = noop_init_apic_ldr, .ioapic_phys_id_map = default_ioapic_phys_id_map, .setup_apic_routing = NULL, - .cpu_present_to_apicid = default_cpu_present_to_apicid, .apicid_to_cpu_present = physid_set_mask_of_physid,
arch/x86/kernel/apic/apic_numachip.c +2 -2
@@ -247,7 +247,7 @@ static const struct apic apic_numachip1 __refconst = { .apic_id_registered = numachip_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
@@ -294,7 +294,7 @@ static const struct apic apic_numachip2 __refconst = { .apic_id_registered = numachip_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/bigsmp_32.c +1 -2
@@ -128,8 +128,7 @@ static struct apic apic_bigsmp __ro_after_init = { .apic_id_registered = bigsmp_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* phys delivery to target CPU: */ - .irq_dest_mode = 0, + .dest_mode_logical = false, .disable_esr = 1,
arch/x86/kernel/apic/io_apic.c +1 -1
@@ -2950,7 +2950,7 @@ static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data, {memset (entry,0 ,sizeof (*entry)); entry->delivery_mode = apic->delivery_mode; - entry->dest_mode = apic->irq_dest_mode; + entry->dest_mode = apic->dest_mode_logical; entry->dest = cfg->dest_apicid; entry->vector = cfg->vector; entry->trigger = data->trigger;
arch/x86/kernel/apic/msi.c +3 -3
@@ -30,9 +30,9 @@ static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg, msg->address_lo =MSI_ADDR_BASE_LO | - ((apic->irq_dest_mode ==0 ) ? -MSI_ADDR_DEST_MODE_PHYSICAL : -MSI_ADDR_DEST_MODE_LOGICAL ) | + (apic->dest_mode_logical ? +MSI_ADDR_DEST_MODE_LOGICAL : +MSI_ADDR_DEST_MODE_PHYSICAL ) |MSI_ADDR_REDIRECTION_CPU |MSI_ADDR_DEST_ID (cfg->dest_apicid);
arch/x86/kernel/apic/probe_32.c +1 -2
@@ -70,8 +70,7 @@ static struct apic apic_default __ro_after_init = { .apic_id_registered = default_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, -/* logical delivery broadcast to all CPUs: */ - .irq_dest_mode = 1, + .dest_mode_logical = true, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_cluster.c +1 -1
@@ -185,7 +185,7 @@ static struct apic apic_x2apic_cluster __ro_after_init = { .apic_id_registered = x2apic_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 1, /* logical */ + .dest_mode_logical = true, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_phys.c +1 -1
@@ -158,7 +158,7 @@ static struct apic apic_x2apic_phys __ro_after_init = { .apic_id_registered = x2apic_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/apic/x2apic_uv_x.c +1 -1
@@ -808,7 +808,7 @@ static struct apic apic_x2apic_uv_x __ro_after_init = { .apic_id_registered = uv_apic_id_registered, .delivery_mode = APIC_DELIVERY_MODE_FIXED, - .irq_dest_mode = 0, /* Physical */ + .dest_mode_logical = false, .disable_esr = 0,
arch/x86/kernel/smpboot.c +2 -5
@@ -747,7 +747,7 @@ static void __init smp_quirk_init_udelay(void)int wakeup_secondary_cpu_via_nmi (int apicid,unsigned long start_eip) { -u32 dm = apic->irq_dest_mode ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ; +u32 dm = apic->dest_mode_logical ?APIC_DEST_LOGICAL :APIC_DEST_PHYSICAL ;unsigned long send_status, accept_status =0 ;int maxlvt;
@@ -981,10 +981,7 @@ wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,if (!boot_error) { enable_start_cpu0 =1 ; *cpu0_nmi_registered =1 ; -if (apic->irq_dest_mode) - id = cpu0_logical_apicid; -else - id = apicid; + id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid; boot_error =wakeup_secondary_cpu_via_nmi (id, start_ip); }
arch/x86/platform/uv/uv_irq.c +1 -1
@@ -36,7 +36,7 @@ static void uv_program_mmr(struct irq_cfg *cfg, struct uv_irq_2_mmr_pnode *info) entry = (struct uv_IO_APIC_route_entry *)&mmr_value; entry->vector = cfg->vector; entry->delivery_mode = apic->delivery_mode; - entry->dest_mode = apic->irq_dest_mode; + entry->dest_mode = apic->dest_mode_logical; entry->polarity =0 ; entry->trigger =0 ; entry->mask =0 ;
arch/x86/xen/apic.c +1 -2
@@ -148,8 +148,7 @@ static struct apic xen_pv_apic = { .apic_id_valid = xen_id_always_valid, .apic_id_registered = xen_id_always_registered, -/* .irq_delivery_mode - used in native_compose_msi_msg only */ -/* .irq_dest_mode - used in native_compose_msi_msg only */ +/* .delivery_mode and .dest_mode_logical not used by XENPV */ .disable_esr = 0,
drivers/iommu/amd/amd_iommu_types.h +1 -1
@@ -893,7 +893,7 @@ struct amd_ir_data { };struct amd_irte_ops { -void (*prepare)(void *,u32 ,u32 ,u8 ,u32 ,int ); +void (*prepare)(void *,u32 ,bool ,u8 ,u32 ,int );void (*activate)(void *,u16 ,u16 );void (*deactivate)(void *,u16 ,u16 );void (*set_affinity)(void *,u16 ,u16 ,u8 ,u32 );
drivers/iommu/amd/iommu.c +4 -4
@@ -3466,7 +3466,7 @@ static void free_irte(u16 devid, int index) }static void irte_prepare (void *entry, -u32 delivery_mode,u32 dest_mode, +u32 delivery_mode,bool dest_mode,u8 vector,u32 dest_apicid,int devid) {union irte *irte = (union irte *) entry;
@@ -3480,7 +3480,7 @@ static void irte_prepare(void *entry, }static void irte_ga_prepare (void *entry, -u32 delivery_mode,u32 dest_mode, +u32 delivery_mode,bool dest_mode,u8 vector,u32 dest_apicid,int devid) {struct irte_ga *irte = (struct irte_ga *) entry;
@@ -3672,7 +3672,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, data->irq_2_irte.devid = devid; data->irq_2_irte.index = index + sub_handle; iommu->irte_ops->prepare (data->entry, apic->delivery_mode, - apic->irq_dest_mode, irq_cfg->vector, + apic->dest_mode_logical, irq_cfg->vector, irq_cfg->dest_apicid, devid);switch (info->type) {
@@ -3943,7 +3943,7 @@ int amd_iommu_deactivate_guest_mode(void *data) entry->hi.val =0 ; entry->lo.fields_remap.valid = valid; - entry->lo.fields_remap.dm = apic->irq_dest_mode; + entry->lo.fields_remap.dm = apic->dest_mode_logical; entry->lo.fields_remap.int_type = apic->delivery_mode; entry->hi.fields.vector = cfg->vector; entry->lo.fields_remap.destination =
drivers/iommu/intel/irq_remapping.c +1 -1
@@ -1113,7 +1113,7 @@ static void prepare_irte(struct irte *irte, int vector, unsigned int dest)memset (irte,0 ,sizeof (*irte)); irte->present =1 ; - irte->dst_mode = apic->irq_dest_mode; + irte->dst_mode = apic->dest_mode_logical;/* * Trigger mode in the IRTE will always be edge, and for IO-APIC, the * actual level or edge trigger will be setup in the IO-APIC
David Woodhouse
maintainerPATCH v3 9/35 x86/apic: Always provide irq_compose_msi_msg() method for vector domain
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This shouldn't be dependent on PCI_MSI. HPET and I/O-APIC can deliver interrupts through MSI without having any PCI in the system at all.
--- arch/x86/include/asm/apic.h | 8 +++----- arch/x86/kernel/apic/apic.c | 32 ++++++++++++++++++++++++++++++ arch/x86/kernel/apic/msi.c | 37 ----------------------------------- arch/x86/kernel/apic/vector.c | 6 ++++++ 4 files changed, 41 insertions(+), 42 deletions(-)
arch/x86/include/asm/apic.h +3 -5
@@ -520,12 +520,10 @@ static inline void apic_smt_update(void) { }#endif struct msi_msg ; +struct irq_cfg ; -#ifdef CONFIG_PCI_MSI -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg); -#else -# define x86_vector_msi_compose_msg NULL -#endif +extern void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar);extern void ioapic_zap_locks (void );
arch/x86/kernel/apic/apic.c +32 -0
@@ -50,6 +50,7 @@#include <asm/io_apic.h> #include <asm/desc.h> #include <asm/hpet.h> +#include <asm/msidef.h> #include <asm/mtrr.h> #include <asm/time.h> #include <asm/smp.h>
@@ -2480,6 +2481,37 @@ int hard_smp_processor_id(void)return read_apic_id (); } +void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) +{ + msg->address_hi =MSI_ADDR_BASE_HI ; + + msg->address_lo = +MSI_ADDR_BASE_LO | + (apic->dest_mode_logical ? +MSI_ADDR_DEST_MODE_LOGICAL : +MSI_ADDR_DEST_MODE_PHYSICAL ) | +MSI_ADDR_REDIRECTION_CPU | +MSI_ADDR_DEST_ID (cfg->dest_apicid); + + msg->data = +MSI_DATA_TRIGGER_EDGE | +MSI_DATA_LEVEL_ASSERT | +MSI_DATA_DELIVERY_FIXED | +MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); +} +/* * Override the generic EOI implementation with an optimized version. * Only called during early boot when only one CPU is active and with
arch/x86/kernel/apic/msi.c +0 -37
@@ -15,7 +15,6 @@#include <linux/hpet.h> #include <linux/msi.h> #include <asm/irqdomain.h> -#include <asm/msidef.h> #include <asm/hpet.h> #include <asm/hw_irq.h> #include <asm/apic.h>
@@ -23,42 +22,6 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, -bool dmar) -{ - msg->address_hi =MSI_ADDR_BASE_HI ; - - msg->address_lo = -MSI_ADDR_BASE_LO | - (apic->dest_mode_logical ? -MSI_ADDR_DEST_MODE_LOGICAL : -MSI_ADDR_DEST_MODE_PHYSICAL ) | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (cfg->dest_apicid); - - msg->data = -MSI_DATA_TRIGGER_EDGE | -MSI_DATA_LEVEL_ASSERT | -MSI_DATA_DELIVERY_FIXED | -MSI_DATA_VECTOR (cfg->vector); - -/* -* Only the IOMMU itself can use the trick of putting destination -* APIC ID into the high bits of the address. Anything else would -* just be writing to memory if it tried that, and needs IR to -* address higher APIC IDs. -*/ -if (dmar) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); -else -WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); -} - -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) -{ -__irq_msi_compose_msg (irqd_cfg (data), msg, false); -} -static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, };
arch/x86/kernel/apic/vector.c +6 -0
@@ -818,6 +818,12 @@ void apic_ack_edge(struct irq_data *irqd)apic_ack_irq (irqd); } +static void x86_vector_msi_compose_msg (struct irq_data *data, +struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, false); +} +static struct irq_chip lapic_controller = { .name ="APIC" , .irq_ack = apic_ack_edge,
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/apic: Always provide irq_compose_msi_msg() method for vector domain
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: f598181acfb36f67e1de138cbe80a7db497f7d8c Gitweb: https://git.kernel.org/tip/f598181acfb36f67e1de138cbe80a7db497f7d8c Author: David Woodhouse <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:09 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
x86/apic: Always provide irq_compose_msi_msg() method for vector domain
This shouldn't be dependent on PCI_MSI. HPET and I/O-APIC can deliver interrupts through MSI without having any PCI in the system at all.
--- arch/x86/include/asm/apic.h | 8 ++----- arch/x86/kernel/apic/apic.c | 32 +++++++++++++++++++++++++++++- arch/x86/kernel/apic/msi.c | 37 +---------------------------------- arch/x86/kernel/apic/vector.c | 6 ++++++- 4 files changed, 41 insertions(+), 42 deletions(-)
arch/x86/include/asm/apic.h +3 -5
@@ -520,12 +520,10 @@ static inline void apic_smt_update(void) { }#endif struct msi_msg ; +struct irq_cfg ; -#ifdef CONFIG_PCI_MSI -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg); -#else -# define x86_vector_msi_compose_msg NULL -#endif +extern void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar);extern void ioapic_zap_locks (void );
arch/x86/kernel/apic/apic.c +32 -0
@@ -50,6 +50,7 @@#include <asm/io_apic.h> #include <asm/desc.h> #include <asm/hpet.h> +#include <asm/msidef.h> #include <asm/mtrr.h> #include <asm/time.h> #include <asm/smp.h>
@@ -2480,6 +2481,37 @@ int hard_smp_processor_id(void)return read_apic_id (); } +void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, +bool dmar) +{ + msg->address_hi =MSI_ADDR_BASE_HI ; + + msg->address_lo = +MSI_ADDR_BASE_LO | + (apic->dest_mode_logical ? +MSI_ADDR_DEST_MODE_LOGICAL : +MSI_ADDR_DEST_MODE_PHYSICAL ) | +MSI_ADDR_REDIRECTION_CPU | +MSI_ADDR_DEST_ID (cfg->dest_apicid); + + msg->data = +MSI_DATA_TRIGGER_EDGE | +MSI_DATA_LEVEL_ASSERT | +MSI_DATA_DELIVERY_FIXED | +MSI_DATA_VECTOR (cfg->vector); + +/* +* Only the IOMMU itself can use the trick of putting destination +* APIC ID into the high bits of the address. Anything else would +* just be writing to memory if it tried that, and needs IR to +* address higher APIC IDs. +*/ +if (dmar) + msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); +else +WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); +} +/* * Override the generic EOI implementation with an optimized version. * Only called during early boot when only one CPU is active and with
arch/x86/kernel/apic/msi.c +0 -37
@@ -15,7 +15,6 @@#include <linux/hpet.h> #include <linux/msi.h> #include <asm/irqdomain.h> -#include <asm/msidef.h> #include <asm/hpet.h> #include <asm/hw_irq.h> #include <asm/apic.h>
@@ -23,42 +22,6 @@struct irq_domain *x86_pci_msi_default_domain__ro_after_init ; -static void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, -bool dmar) -{ - msg->address_hi =MSI_ADDR_BASE_HI ; - - msg->address_lo = -MSI_ADDR_BASE_LO | - (apic->dest_mode_logical ? -MSI_ADDR_DEST_MODE_LOGICAL : -MSI_ADDR_DEST_MODE_PHYSICAL ) | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (cfg->dest_apicid); - - msg->data = -MSI_DATA_TRIGGER_EDGE | -MSI_DATA_LEVEL_ASSERT | -MSI_DATA_DELIVERY_FIXED | -MSI_DATA_VECTOR (cfg->vector); - -/* -* Only the IOMMU itself can use the trick of putting destination -* APIC ID into the high bits of the address. Anything else would -* just be writing to memory if it tried that, and needs IR to -* address higher APIC IDs. -*/ -if (dmar) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); -else -WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); -} - -void x86_vector_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) -{ -__irq_msi_compose_msg (irqd_cfg (data), msg, false); -} -static void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) {struct msi_msg msg[2 ] = { [1 ] = { }, };
arch/x86/kernel/apic/vector.c +6 -0
@@ -818,6 +818,12 @@ void apic_ack_edge(struct irq_data *irqd)apic_ack_irq (irqd); } +static void x86_vector_msi_compose_msg (struct irq_data *data, +struct msi_msg *msg) +{ +__irq_msi_compose_msg (irqd_cfg (data), msg, false); +} +static struct irq_chip lapic_controller = { .name ="APIC" , .irq_ack = apic_ack_edge,
David Woodhouse
maintainerPATCH v3 10/35 x86/hpet: Move MSI support into hpet.c
In reply to David WoodhouseFrom: David Woodhouse <[email protected]>
This isn't really dependent on PCI MSI; it's just generic MSI which is now supported by the generic x86_vector_domain. Move the HPET MSI support back into hpet.c with the rest of the HPET support.
--- arch/x86/include/asm/hpet.h | 11 ---- arch/x86/kernel/apic/msi.c | 111 --------------------------------- arch/x86/kernel/hpet.c | 118 ++++++++++++++++++++++++++++++++++-- 3 files changed, 112 insertions(+), 128 deletions(-)
arch/x86/include/asm/hpet.h +0 -11
@@ -74,17 +74,6 @@ extern void hpet_disable(void);extern unsigned int hpet_readl (unsigned int a);extern void force_hpet_resume (void ); -struct irq_data ; -struct hpet_channel ; -struct irq_domain ; - -extern void hpet_msi_unmask (struct irq_data *data); -extern void hpet_msi_mask (struct irq_data *data); -extern void hpet_msi_write (struct hpet_channel *hc,struct msi_msg *msg); -extern struct irq_domain *hpet_create_irq_domain (int hpet_id); -extern int hpet_assign_irq (struct irq_domain *domain, -struct hpet_channel *hc,int dev_num); -#ifdef CONFIG_HPET_EMULATE_RTC #include <linux/interrupt.h>
arch/x86/kernel/apic/msi.c +0 -111
114 lines not on this page · raw message
arch/x86/kernel/hpet.c +112 -6
153 lines not on this page · raw message
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/hpet: Move MSI support into hpet.c
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 3d7295eb3003aea9f89de35304b3a88ae4d5036b Gitweb: https://git.kernel.org/tip/3d7295eb3003aea9f89de35304b3a88ae4d5036b Author: David Woodhouse <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:10 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
x86/hpet: Move MSI support into hpet.c
This isn't really dependent on PCI MSI; it's just generic MSI which is now supported by the generic x86_vector_domain. Move the HPET MSI support back into hpet.c with the rest of the HPET support.
--- arch/x86/include/asm/hpet.h | 11 +--- arch/x86/kernel/apic/msi.c | 111 +--------------------------------- arch/x86/kernel/hpet.c | 118 +++++++++++++++++++++++++++++++++-- 3 files changed, 112 insertions(+), 128 deletions(-)
arch/x86/include/asm/hpet.h +0 -11
@@ -74,17 +74,6 @@ extern void hpet_disable(void);extern unsigned int hpet_readl (unsigned int a);extern void force_hpet_resume (void ); -struct irq_data ; -struct hpet_channel ; -struct irq_domain ; - -extern void hpet_msi_unmask (struct irq_data *data); -extern void hpet_msi_mask (struct irq_data *data); -extern void hpet_msi_write (struct hpet_channel *hc,struct msi_msg *msg); -extern struct irq_domain *hpet_create_irq_domain (int hpet_id); -extern int hpet_assign_irq (struct irq_domain *domain, -struct hpet_channel *hc,int dev_num); -#ifdef CONFIG_HPET_EMULATE_RTC #include <linux/interrupt.h>
arch/x86/kernel/apic/msi.c +0 -111
114 lines not on this page · raw message
arch/x86/kernel/hpet.c +112 -6
153 lines not on this page · raw message
David Woodhouse
maintainerPATCH v3 11/35 genirq/msi: Allow shadow declarations of msi_msg::$member
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Architectures like x86 have their MSI messages in various bits of the data, address_lo and address_hi field. Composing or decomposing these messages with bitmasks and shifts is possible, but unreadable gunk.
Allow architectures to provide an architecture specific representation for each member of msi_msg. Provide empty defaults for each and stick them into an union.
--- include/asm-generic/msi.h | 4 ++++ include/linux/msi.h | 46 +++++++++++++++++++++++++++++++++++---- 2 files changed, 46 insertions(+), 4 deletions(-)
include/asm-generic/msi.h +4 -0
@@ -4,6 +4,8 @@#include <linux/types.h> +#ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN +#ifndef NUM_MSI_ALLOC_SCRATCHPAD_REGS # define NUM_MSI_ALLOC_SCRATCHPAD_REGS 2#endif
@@ -30,4 +32,6 @@ typedef struct msi_alloc_info {#define GENERIC_MSI_DOMAIN_OPS 1 +#endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */ +#endif
include/linux/msi.h +42 -4
60 lines not on this page · raw message
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] genirq/msi: Allow shadow declarations of msi_msg:: $member
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 8073c1ac82c12aaf1b475a3ce5328d43b3eaa4ae Gitweb: https://git.kernel.org/tip/8073c1ac82c12aaf1b475a3ce5328d43b3eaa4ae Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:11 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
genirq/msi: Allow shadow declarations of msi_msg:: $member
Architectures like x86 have their MSI messages in various bits of the data, address_lo and address_hi field. Composing or decomposing these messages with bitmasks and shifts is possible, but unreadable gunk.
Allow architectures to provide an architecture specific representation for each member of msi_msg. Provide empty defaults for each and stick them into an union.
--- include/asm-generic/msi.h | 4 +++- include/linux/msi.h | 46 ++++++++++++++++++++++++++++++++++---- 2 files changed, 46 insertions(+), 4 deletions(-)
include/asm-generic/msi.h +4 -0
@@ -4,6 +4,8 @@#include <linux/types.h> +#ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN +#ifndef NUM_MSI_ALLOC_SCRATCHPAD_REGS # define NUM_MSI_ALLOC_SCRATCHPAD_REGS 2#endif
@@ -30,4 +32,6 @@ typedef struct msi_alloc_info {#define GENERIC_MSI_DOMAIN_OPS 1 +#endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */ +#endif
include/linux/msi.h +42 -4
60 lines not on this page · raw message
David Woodhouse
maintainerPATCH v3 12/35 x86/msi: Provide msi message shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Create shadow structs with named bitfields for msi_msg data, address_lo and address_hi and use them in the MSI message composer.
Provide a function to retrieve the destination ID. This could be inline, but that'd create a circular header dependency.
--- arch/x86/include/asm/msi.h | 49 +++++++++++++++++++++++++++++++++++++ arch/x86/kernel/apic/apic.c | 35 ++++++++++++++------------ 2 files changed, 68 insertions(+), 16 deletions(-)
arch/x86/include/asm/msi.h +49 -0
@@ -9,4 +9,53 @@ typedef struct irq_alloc_info msi_alloc_info_t;int pci_msi_prepare (struct irq_domain *domain,struct device *dev,int nvec,msi_alloc_info_t *arg); +/* Structs and defines for the X86 specific MSI message format */ + +typedef struct x86_msi_data { +u32 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + reserved :2 , + active_low :1 , + is_level :1 ; + +u32 dmar_subhandle; +} __attribute__ ((packed))arch_msi_msg_data_t ; +#define arch_msi_msg_data x86_msi_data + +typedef struct x86_msi_addr_lo { +union { +struct { +u32 reserved_0 :2 , + dest_mode_logical :1 , + redirect_hint :1 , + reserved_1 :8 , + destid_0_7 :8 , + base_address :12 ; + }; +struct { +u32 dmar_reserved_0 :2 , + dmar_index_15 :1 , + dmar_subhandle_valid :1 , + dmar_format :1 , + dmar_index_0_14 :15 , + dmar_base_address :12 ; + }; + }; +} __attribute__ ((packed))arch_msi_msg_addr_lo_t ; +#define arch_msi_msg_addr_lo x86_msi_addr_lo + +#define X86_MSI_BASE_ADDRESS_LOW (0xfee00000 >> 20) + +typedef struct x86_msi_addr_hi { +u32 reserved :8 , + destid_8_31 :24 ; +} __attribute__ ((packed))arch_msi_msg_addr_hi_t ; +#define arch_msi_msg_addr_hi x86_msi_addr_hi + +#define X86_MSI_BASE_ADDRESS_HIGH (0) + +struct msi_msg ; +u32 x86_msi_msg_get_destid (struct msi_msg *msg,bool extid); +#endif /* _ASM_X86_MSI_H */
arch/x86/kernel/apic/apic.c +19 -16
@@ -50,7 +50,6 @@#include <asm/io_apic.h> #include <asm/desc.h> #include <asm/hpet.h> -#include <asm/msidef.h> #include <asm/mtrr.h> #include <asm/time.h> #include <asm/smp.h>
@@ -2484,22 +2483,16 @@ int hard_smp_processor_id(void)void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg,bool dmar) { - msg->address_hi =MSI_ADDR_BASE_HI ; +memset (msg,0 ,sizeof (*msg)); - msg->address_lo = -MSI_ADDR_BASE_LO | - (apic->dest_mode_logical ? -MSI_ADDR_DEST_MODE_LOGICAL : -MSI_ADDR_DEST_MODE_PHYSICAL ) | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (cfg->dest_apicid); + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical; + msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid &0xFF ; - msg->data = -MSI_DATA_TRIGGER_EDGE | -MSI_DATA_LEVEL_ASSERT | -MSI_DATA_DELIVERY_FIXED | -MSI_DATA_VECTOR (cfg->vector); + msg->arch_data.delivery_mode =APIC_DELIVERY_MODE_FIXED ; + msg->arch_data.vector = cfg->vector; + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ;/* * Only the IOMMU itself can use the trick of putting destination * APIC ID into the high bits of the address. Anything else would
@@ -2507,11 +2500,21 @@ void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,* address higher APIC IDs. */ if (dmar) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); + msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >>8 ;else -WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); +WARN_ON_ONCE (cfg->dest_apicid >0xFF ); } +u32 x86_msi_msg_get_destid (struct msi_msg *msg,bool extid) +{ +u32 dest = msg->arch_addr_lo.destid_0_7; + +if (extid) + dest |= msg->arch_addr_hi.destid_8_31 <<8 ; +return dest; +} +EXPORT_SYMBOL_GPL (x86_msi_msg_get_destid); +/* * Override the generic EOI implementation with an optimized version. * Only called during early boot when only one CPU is active and with
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/msi: Provide msi message shadow structs
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 6285aa507366729c618d5295fb540b24a956088a Gitweb: https://git.kernel.org/tip/6285aa507366729c618d5295fb540b24a956088a Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:12 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
x86/msi: Provide msi message shadow structs
Create shadow structs with named bitfields for msi_msg data, address_lo and address_hi and use them in the MSI message composer.
Provide a function to retrieve the destination ID. This could be inline, but that'd create a circular header dependency.
[dwmw2: fix bitfields not all to be a union]
--- arch/x86/include/asm/msi.h | 49 ++++++++++++++++++++++++++++++++++++- arch/x86/kernel/apic/apic.c | 35 ++++++++++++++------------ 2 files changed, 68 insertions(+), 16 deletions(-)
arch/x86/include/asm/msi.h +49 -0
@@ -9,4 +9,53 @@ typedef struct irq_alloc_info msi_alloc_info_t;int pci_msi_prepare (struct irq_domain *domain,struct device *dev,int nvec,msi_alloc_info_t *arg); +/* Structs and defines for the X86 specific MSI message format */ + +typedef struct x86_msi_data { +u32 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + reserved :2 , + active_low :1 , + is_level :1 ; + +u32 dmar_subhandle; +} __attribute__ ((packed))arch_msi_msg_data_t ; +#define arch_msi_msg_data x86_msi_data + +typedef struct x86_msi_addr_lo { +union { +struct { +u32 reserved_0 :2 , + dest_mode_logical :1 , + redirect_hint :1 , + reserved_1 :8 , + destid_0_7 :8 , + base_address :12 ; + }; +struct { +u32 dmar_reserved_0 :2 , + dmar_index_15 :1 , + dmar_subhandle_valid :1 , + dmar_format :1 , + dmar_index_0_14 :15 , + dmar_base_address :12 ; + }; + }; +} __attribute__ ((packed))arch_msi_msg_addr_lo_t ; +#define arch_msi_msg_addr_lo x86_msi_addr_lo + +#define X86_MSI_BASE_ADDRESS_LOW (0xfee00000 >> 20) + +typedef struct x86_msi_addr_hi { +u32 reserved :8 , + destid_8_31 :24 ; +} __attribute__ ((packed))arch_msi_msg_addr_hi_t ; +#define arch_msi_msg_addr_hi x86_msi_addr_hi + +#define X86_MSI_BASE_ADDRESS_HIGH (0) + +struct msi_msg ; +u32 x86_msi_msg_get_destid (struct msi_msg *msg,bool extid); +#endif /* _ASM_X86_MSI_H */
arch/x86/kernel/apic/apic.c +19 -16
@@ -50,7 +50,6 @@#include <asm/io_apic.h> #include <asm/desc.h> #include <asm/hpet.h> -#include <asm/msidef.h> #include <asm/mtrr.h> #include <asm/time.h> #include <asm/smp.h>
@@ -2484,22 +2483,16 @@ int hard_smp_processor_id(void)void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg,bool dmar) { - msg->address_hi =MSI_ADDR_BASE_HI ; +memset (msg,0 ,sizeof (*msg)); - msg->address_lo = -MSI_ADDR_BASE_LO | - (apic->dest_mode_logical ? -MSI_ADDR_DEST_MODE_LOGICAL : -MSI_ADDR_DEST_MODE_PHYSICAL ) | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (cfg->dest_apicid); + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical; + msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid &0xFF ; - msg->data = -MSI_DATA_TRIGGER_EDGE | -MSI_DATA_LEVEL_ASSERT | -MSI_DATA_DELIVERY_FIXED | -MSI_DATA_VECTOR (cfg->vector); + msg->arch_data.delivery_mode =APIC_DELIVERY_MODE_FIXED ; + msg->arch_data.vector = cfg->vector; + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ;/* * Only the IOMMU itself can use the trick of putting destination * APIC ID into the high bits of the address. Anything else would
@@ -2507,11 +2500,21 @@ void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,* address higher APIC IDs. */ if (dmar) - msg->address_hi |=MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid); + msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >>8 ;else -WARN_ON_ONCE (MSI_ADDR_EXT_DEST_ID (cfg->dest_apicid)); +WARN_ON_ONCE (cfg->dest_apicid >0xFF ); } +u32 x86_msi_msg_get_destid (struct msi_msg *msg,bool extid) +{ +u32 dest = msg->arch_addr_lo.destid_0_7; + +if (extid) + dest |= msg->arch_addr_hi.destid_8_31 <<8 ; +return dest; +} +EXPORT_SYMBOL_GPL (x86_msi_msg_get_destid); +/* * Override the generic EOI implementation with an optimized version. * Only called during early boot when only one CPU is active and with
Re: [PATCH v3 12/35] x86/msi: Provide msi message shadow structs
In reply to David WoodhouseWhile working on some out-of-tree patches, we noticed that assignment to dmar_subhandle of struct x86_msi_data lead to a QEMU warning about reserved bits in MSI data being set:
qemu-system-x86_64: vtd_interrupt_remap_msi: invalid IR MSI (sid=256, address=0xfee003d8, data=0x10000)
This message originates from hw/i386/intel_iommu.c in QEMU:
#define VTD_IR_MSI_DATA_RESERVED (0xffff0000) if (origin->data & VTD_IR_MSI_DATA_RESERVED) { ... }
Looking at struct x86_msi_data, it appears that it is actually 48-bits in size since the bitfield containing the vector, delivery_mode etc is 2 bytes wide followed by dmar_subhandle which is 32 bits. Thus assignment to dmar_subhandle leads to bits > 16 being set.
If I am not mistaken, the MSI data field should be 32-bits wide for all platforms (struct msi_msg, include/linux/msi.h). Is this analysis correct or did I miss something wrt. handling of dmar_subhandle?
The attached patch fixes the issue for us.
Regards,
- reto
PATCH 1/1 x86/msi: Fix msi message data shadow struct
The x86 MSI message data is 32 bits in total and is either in compatibility or remappable format, see Intel Virtualization Technology for Directed I/O, section 5.1.2.
--- arch/x86/include/asm/msi.h | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-)
arch/x86/include/asm/msi.h +11 -8
@@ -12,14 +12,17 @@ int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,/* Structs and defines for the X86 specific MSI message format */ typedef struct x86_msi_data { -u32 vector :8 , - delivery_mode :3 , - dest_mode_logical :1 , - reserved :2 , - active_low :1 , - is_level :1 ; - -u32 dmar_subhandle; +union { +struct { +u32 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + reserved :2 , + active_low :1 , + is_level :1 ; + }; +u32 dmar_subhandle; + }; } __attribute__ ((packed))arch_msi_msg_data_t ;#define arch_msi_msg_data x86_msi_data
Signature · 1 line
2.30.2
Re: [PATCH] x86/msi: Fix msi message data shadow struct
Reto,
Quoted 7 lines · Reto Buerki
The x86 MSI message data is 32 bits in total and is either in compatibility or remappable format, see Intel Virtualization Technology for Directed I/O, section 5.1.2.
Fixes 6285aa50736 ("x86/msi: Provide msi message shadow structs")Signed-off-by Reto Buerki <[email protected]>Signed-off-by Adrian-Ken Rueegsegger <[email protected]>
This signed-off by chain is incorrect. It would be correct if Adrian-Ken would have sent the patch.
If you both worked on that then please use the Co-developed-by tag according to Documentation/process/
Thanks,
tglx
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH 1/1 x86/msi: Fix msi message data shadow struct
The x86 MSI message data is 32 bits in total and is either in compatibility or remappable format, see Intel Virtualization Technology for Directed I/O, section 5.1.2.
--- arch/x86/include/asm/msi.h | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-)
arch/x86/include/asm/msi.h +11 -8
@@ -12,14 +12,17 @@ int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,/* Structs and defines for the X86 specific MSI message format */ typedef struct x86_msi_data { -u32 vector :8 , - delivery_mode :3 , - dest_mode_logical :1 , - reserved :2 , - active_low :1 , - is_level :1 ; - -u32 dmar_subhandle; +union { +struct { +u32 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + reserved :2 , + active_low :1 , + is_level :1 ; + }; +u32 dmar_subhandle; + }; } __attribute__ ((packed))arch_msi_msg_data_t ;#define arch_msi_msg_data x86_msi_data
Signature · 6 lines
2.30.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/urgent] x86/msi: Fix msi message data shadow struct
The following commit has been merged into the x86/urgent branch of tip:
The following commit has been merged into the x86/urgent branch of tip:
Commit-ID: 59b18a1e65b7a2134814106d0860010e10babe18 Gitweb: https://git.kernel.org/tip/59b18a1e65b7a2134814106d0860010e10babe18 Author: Reto Buerki <[email protected]> AuthorDate: Thu, 07 Apr 2022 13:06:47 +02:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Thu, 07 Apr 2022 15:19:32 +02:00
x86/msi: Fix msi message data shadow struct
The x86 MSI message data is 32 bits in total and is either in compatibility or remappable format, see Intel Virtualization Technology for Directed I/O, section 5.1.2.
--- arch/x86/include/asm/msi.h | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-)
arch/x86/include/asm/msi.h +11 -8
@@ -12,14 +12,17 @@ int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,/* Structs and defines for the X86 specific MSI message format */ typedef struct x86_msi_data { -u32 vector :8 , - delivery_mode :3 , - dest_mode_logical :1 , - reserved :2 , - active_low :1 , - is_level :1 ; - -u32 dmar_subhandle; +union { +struct { +u32 vector :8 , + delivery_mode :3 , + dest_mode_logical :1 , + reserved :2 , + active_low :1 , + is_level :1 ; + }; +u32 dmar_subhandle; + }; } __attribute__ ((packed))arch_msi_msg_data_t ;#define arch_msi_msg_data x86_msi_data
Re: [PATCH v3 12/35] x86/msi: Provide msi message shadow structs
In reply to Reto BuerkiQuoted 11 lines · Reto Buerki
While working on some out-of-tree patches, we noticed that assignment to
dmar_subhandleof structx86_msi_datalead to a QEMU warning about reserved bits in MSI data being set:qemu-system-x86_64:
vtd_interrupt_remap_msi: invalid IR MSI (sid=256, address=0xfee003d8, data=0x10000)This message originates from
hw/i386/intel_iommu.cin QEMU:#define
VTD_IR_MSI_DATA_RESERVED(0xffff0000) if (origin->data&VTD_IR_MSI_DATA_RESERVED) { ... }
Quoted 8 lines
Looking at struct
x86_msi_data, it appears that it is actually 48-bits in size since the bitfield containing the vector,delivery_modeetc is 2 bytes wide followed bydmar_subhandlewhich is 32 bits. Thus assignment todmar_subhandleleads to bits > 16 being set.If I am not mistaken, the MSI data field should be 32-bits wide for all platforms (struct
msi_msg,include/linux/msi.h). Is this analysis correct or did I miss something wrt. handling ofdmar_subhandle?
It's correct and I'm completely surprised that this went unnoticed for more than a year. Where is that brown paperbag...
Thanks,
tglx
David Woodhouse
maintainerPATCH v3 13/35 iommu/intel: Use msi_msg shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Use the bitfields in the x86 shadow struct to compose the MSI message.
--- drivers/iommu/intel/irq_remapping.c | 24 ++++++++++++++++-------- 1 file changed, 16 insertions(+), 8 deletions(-)
drivers/iommu/intel/irq_remapping.c +16 -8
@@ -20,7 +20,6 @@#include <asm/cpu.h> #include <asm/irq_remapping.h> #include <asm/pci-direct.h> -#include <asm/msidef.h> #include "../irq_remapping.h"
@@ -1260,6 +1259,21 @@ static struct irq_chip intel_ir_chip = { .irq_set_vcpu_affinity = intel_ir_set_vcpu_affinity, }; +static void fill_msi_msg (struct msi_msg *msg,u32 index,u32 subhandle) +{ +memset (msg,0 ,sizeof (*msg)); + + msg->arch_addr_lo.dmar_base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.dmar_subhandle_valid = true; + msg->arch_addr_lo.dmar_format = true; + msg->arch_addr_lo.dmar_index_0_14 = index &0x7FFF ; + msg->arch_addr_lo.dmar_index_15 = !!(index &0x8000 ); + + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + + msg->arch_data.dmar_subhandle = subhandle; +} + static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, struct irq_cfg *irq_cfg, struct irq_alloc_info *info,
@@ -1267,7 +1281,6 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, {struct IR_IO_APIC_route_entry *entry;struct irte *irte = &data->irte_entry; -struct msi_msg *msg = &data->msi_entry;prepare_irte (irte, irq_cfg->vector, irq_cfg->dest_apicid);switch (info->type) {
@@ -1308,12 +1321,7 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,else set_msi_sid (irte,msi_desc_to_pci_dev (info->desc)); - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->data = sub_handle; - msg->address_lo =MSI_ADDR_BASE_LO |MSI_ADDR_IR_EXT_INT | -MSI_ADDR_IR_SHV | -MSI_ADDR_IR_INDEX1 (index) | -MSI_ADDR_IR_INDEX2 (index); +fill_msi_msg (&data->msi_entry, index, sub_handle);break ;default :
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] iommu/intel: Use msi_msg shadow structs
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 5c0d0e2cc6e0e7a96c25351fd67c775e7b1f11f0 Gitweb: https://git.kernel.org/tip/5c0d0e2cc6e0e7a96c25351fd67c775e7b1f11f0 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:13 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:25 +01:00
iommu/intel: Use msi_msg shadow structs
Use the bitfields in the x86 shadow struct to compose the MSI message.
--- drivers/iommu/intel/irq_remapping.c | 24 ++++++++++++++++-------- 1 file changed, 16 insertions(+), 8 deletions(-)
drivers/iommu/intel/irq_remapping.c +16 -8
@@ -20,7 +20,6 @@#include <asm/cpu.h> #include <asm/irq_remapping.h> #include <asm/pci-direct.h> -#include <asm/msidef.h> #include "../irq_remapping.h"
@@ -1260,6 +1259,21 @@ static struct irq_chip intel_ir_chip = { .irq_set_vcpu_affinity = intel_ir_set_vcpu_affinity, }; +static void fill_msi_msg (struct msi_msg *msg,u32 index,u32 subhandle) +{ +memset (msg,0 ,sizeof (*msg)); + + msg->arch_addr_lo.dmar_base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.dmar_subhandle_valid = true; + msg->arch_addr_lo.dmar_format = true; + msg->arch_addr_lo.dmar_index_0_14 = index &0x7FFF ; + msg->arch_addr_lo.dmar_index_15 = !!(index &0x8000 ); + + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + + msg->arch_data.dmar_subhandle = subhandle; +} + static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, struct irq_cfg *irq_cfg, struct irq_alloc_info *info,
@@ -1267,7 +1281,6 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data, {struct IR_IO_APIC_route_entry *entry;struct irte *irte = &data->irte_entry; -struct msi_msg *msg = &data->msi_entry;prepare_irte (irte, irq_cfg->vector, irq_cfg->dest_apicid);switch (info->type) {
@@ -1308,12 +1321,7 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,else set_msi_sid (irte,msi_desc_to_pci_dev (info->desc)); - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->data = sub_handle; - msg->address_lo =MSI_ADDR_BASE_LO |MSI_ADDR_IR_EXT_INT | -MSI_ADDR_IR_SHV | -MSI_ADDR_IR_INDEX1 (index) | -MSI_ADDR_IR_INDEX2 (index); +fill_msi_msg (&data->msi_entry, index, sub_handle);break ;default :
David Woodhouse
maintainerPATCH v3 14/35 iommu/amd: Use msi_msg shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Get rid of the macro mess and use the shadow structs for the x86 specific MSI message format. Convert the intcapxt setup to use named bitfields as well while touching it anyway.
--- drivers/iommu/amd/init.c | 46 ++++++++++++++++++++++----------------- drivers/iommu/amd/iommu.c | 14 +++++++----- 2 files changed, 35 insertions(+), 25 deletions(-)
drivers/iommu/amd/init.c +26 -20
@@ -23,7 +23,6 @@#include <asm/pci-direct.h> #include <asm/iommu.h> #include <asm/apic.h> -#include <asm/msidef.h> #include <asm/gart.h> #include <asm/x86_init.h> #include <asm/iommu_table.h>
@@ -1966,10 +1965,16 @@ static int iommu_setup_msi(struct amd_iommu *iommu)return 0 ; } -#define XT_INT_DEST_MODE (x) (((x) & 0x1ULL) << 2) -#define XT_INT_DEST_LO (x) (((x) & 0xFFFFFFULL) << 8) -#define XT_INT_VEC (x) (((x) & 0xFFULL) << 32) -#define XT_INT_DEST_HI (x) ((((x) >> 24) & 0xFFULL) << 56) +union intcapxt { +u64 capxt; +u64 reserved_0 :2 , + dest_mode_logical :1 , + reserved_1 :5 , + destid_0_23 :24 , + vector :8 , + reserved_2 :16 , + destid_24_31 :8 ; +} __attribute__ ((packed));/* * Setup the IntCapXT registers with interrupt routing information
@@ -1978,28 +1983,29 @@ static int iommu_setup_msi(struct amd_iommu *iommu)*/ static void iommu_update_intcapxt (struct amd_iommu *iommu) { -u64 val; -u32 addr_lo =readl (iommu->mmio_base +MMIO_MSI_ADDR_LO_OFFSET ); -u32 addr_hi =readl (iommu->mmio_base +MMIO_MSI_ADDR_HI_OFFSET ); -u32 data =readl (iommu->mmio_base +MMIO_MSI_DATA_OFFSET ); -bool dm = (addr_lo >>MSI_ADDR_DEST_MODE_SHIFT ) &0x1 ; -u32 dest = ((addr_lo >>MSI_ADDR_DEST_ID_SHIFT ) &0xFF ); +struct msi_msg msg; +union intcapxt xt; +u32 destid; -if (x2apic_enabled ()) - dest |=MSI_ADDR_EXT_DEST_ID (addr_hi); + msg.address_lo =readl (iommu->mmio_base +MMIO_MSI_ADDR_LO_OFFSET ); + msg.address_hi =readl (iommu->mmio_base +MMIO_MSI_ADDR_HI_OFFSET ); + msg.data =readl (iommu->mmio_base +MMIO_MSI_DATA_OFFSET ); - val =XT_INT_VEC (data &0xFF ) | -XT_INT_DEST_MODE (dm) | -XT_INT_DEST_LO (dest) | -XT_INT_DEST_HI (dest); + destid =x86_msi_msg_get_destid (&msg,x2apic_enabled ()); + + xt.capxt =0ULL ; + xt.dest_mode_logical = msg.arch_data.dest_mode_logical; + xt.vector = msg.arch_data.vector; + xt.destid_0_23 = destid &GENMASK (23 ,0 ); + xt.destid_24_31 = destid >>24 ;/** * Current IOMMU implemtation uses the same IRQ for all * 3 IOMMU interrupts. */ -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_EVT_OFFSET ); -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_PPR_OFFSET ); -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_GALOG_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_EVT_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_PPR_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_GALOG_OFFSET ); } static void _irq_notifier_notify(struct irq_affinity_notify *notify,
drivers/iommu/amd/iommu.c +9 -5
@@ -35,7 +35,6 @@#include <asm/io_apic.h> #include <asm/apic.h> #include <asm/hw_irq.h> -#include <asm/msidef.h> #include <asm/proto.h> #include <asm/iommu.h> #include <asm/gart.h>
@@ -3656,13 +3655,20 @@ struct irq_remap_ops amd_iommu_irq_ops = { .get_irq_domain = get_irq_domain, }; +static void fill_msi_msg (struct msi_msg *msg,u32 index) +{ + msg->data = index; + msg->address_lo =0 ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; +} +static void irq_remapping_prepare_irte (struct amd_ir_data *data,struct irq_cfg *irq_cfg,struct irq_alloc_info *info,int devid,int index,int sub_handle) {struct irq_2_irte *irte_info = &data->irq_2_irte; -struct msi_msg *msg = &data->msi_entry;struct IO_APIC_route_entry *entry;struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
@@ -3693,9 +3699,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,case X86_IRQ_ALLOC_TYPE_HPET :case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX : - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->address_lo =MSI_ADDR_BASE_LO ; - msg->data = irte_info->index; +fill_msi_msg (&data->msi_entry, irte_info->index);break ;default :
Signature · 1 line
2.26.2
[tip: x86/apic] iommu/amd: Use msi_msg shadow structs
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: b5c3786ee3704bd8cd5b29ae168526f2b1af4557 Gitweb: https://git.kernel.org/tip/b5c3786ee3704bd8cd5b29ae168526f2b1af4557 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:14 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
iommu/amd: Use msi_msg shadow structs
Get rid of the macro mess and use the shadow structs for the x86 specific MSI message format. Convert the intcapxt setup to use named bitfields as well while touching it anyway.
--- drivers/iommu/amd/init.c | 46 +++++++++++++++++++++----------------- drivers/iommu/amd/iommu.c | 14 +++++++----- 2 files changed, 35 insertions(+), 25 deletions(-)
drivers/iommu/amd/init.c +26 -20
@@ -23,7 +23,6 @@#include <asm/pci-direct.h> #include <asm/iommu.h> #include <asm/apic.h> -#include <asm/msidef.h> #include <asm/gart.h> #include <asm/x86_init.h> #include <asm/iommu_table.h>
@@ -1966,10 +1965,16 @@ static int iommu_setup_msi(struct amd_iommu *iommu)return 0 ; } -#define XT_INT_DEST_MODE (x) (((x) & 0x1ULL) << 2) -#define XT_INT_DEST_LO (x) (((x) & 0xFFFFFFULL) << 8) -#define XT_INT_VEC (x) (((x) & 0xFFULL) << 32) -#define XT_INT_DEST_HI (x) ((((x) >> 24) & 0xFFULL) << 56) +union intcapxt { +u64 capxt; +u64 reserved_0 :2 , + dest_mode_logical :1 , + reserved_1 :5 , + destid_0_23 :24 , + vector :8 , + reserved_2 :16 , + destid_24_31 :8 ; +} __attribute__ ((packed));/* * Setup the IntCapXT registers with interrupt routing information
@@ -1978,28 +1983,29 @@ static int iommu_setup_msi(struct amd_iommu *iommu)*/ static void iommu_update_intcapxt (struct amd_iommu *iommu) { -u64 val; -u32 addr_lo =readl (iommu->mmio_base +MMIO_MSI_ADDR_LO_OFFSET ); -u32 addr_hi =readl (iommu->mmio_base +MMIO_MSI_ADDR_HI_OFFSET ); -u32 data =readl (iommu->mmio_base +MMIO_MSI_DATA_OFFSET ); -bool dm = (addr_lo >>MSI_ADDR_DEST_MODE_SHIFT ) &0x1 ; -u32 dest = ((addr_lo >>MSI_ADDR_DEST_ID_SHIFT ) &0xFF ); +struct msi_msg msg; +union intcapxt xt; +u32 destid; -if (x2apic_enabled ()) - dest |=MSI_ADDR_EXT_DEST_ID (addr_hi); + msg.address_lo =readl (iommu->mmio_base +MMIO_MSI_ADDR_LO_OFFSET ); + msg.address_hi =readl (iommu->mmio_base +MMIO_MSI_ADDR_HI_OFFSET ); + msg.data =readl (iommu->mmio_base +MMIO_MSI_DATA_OFFSET ); - val =XT_INT_VEC (data &0xFF ) | -XT_INT_DEST_MODE (dm) | -XT_INT_DEST_LO (dest) | -XT_INT_DEST_HI (dest); + destid =x86_msi_msg_get_destid (&msg,x2apic_enabled ()); + + xt.capxt =0ULL ; + xt.dest_mode_logical = msg.arch_data.dest_mode_logical; + xt.vector = msg.arch_data.vector; + xt.destid_0_23 = destid &GENMASK (23 ,0 ); + xt.destid_24_31 = destid >>24 ;/** * Current IOMMU implemtation uses the same IRQ for all * 3 IOMMU interrupts. */ -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_EVT_OFFSET ); -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_PPR_OFFSET ); -writeq (val, iommu->mmio_base +MMIO_INTCAPXT_GALOG_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_EVT_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_PPR_OFFSET ); +writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_GALOG_OFFSET ); } static void _irq_notifier_notify(struct irq_affinity_notify *notify,
drivers/iommu/amd/iommu.c +9 -5
@@ -35,7 +35,6 @@#include <asm/io_apic.h> #include <asm/apic.h> #include <asm/hw_irq.h> -#include <asm/msidef.h> #include <asm/proto.h> #include <asm/iommu.h> #include <asm/gart.h>
@@ -3656,13 +3655,20 @@ struct irq_remap_ops amd_iommu_irq_ops = { .get_irq_domain = get_irq_domain, }; +static void fill_msi_msg (struct msi_msg *msg,u32 index) +{ + msg->data = index; + msg->address_lo =0 ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; +} +static void irq_remapping_prepare_irte (struct amd_ir_data *data,struct irq_cfg *irq_cfg,struct irq_alloc_info *info,int devid,int index,int sub_handle) {struct irq_2_irte *irte_info = &data->irq_2_irte; -struct msi_msg *msg = &data->msi_entry;struct IO_APIC_route_entry *entry;struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
@@ -3693,9 +3699,7 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data,case X86_IRQ_ALLOC_TYPE_HPET :case X86_IRQ_ALLOC_TYPE_PCI_MSI :case X86_IRQ_ALLOC_TYPE_PCI_MSIX : - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->address_lo =MSI_ADDR_BASE_LO ; - msg->data = irte_info->index; +fill_msi_msg (&data->msi_entry, irte_info->index);break ;default :
David Woodhouse
maintainerPATCH v3 15/35 PCI: vmd: Use msi_msg shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Use the x86 shadow structs in msi_msg instead of the macros.
--- drivers/pci/controller/vmd.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/pci/controller/vmd.c +4 -5
@@ -18,7 +18,6 @@#include <asm/irqdomain.h> #include <asm/device.h> #include <asm/msi.h> -#include <asm/msidef.h> #define VMD_CFGBAR 0#define VMD_MEMBAR1 2
@@ -131,10 +130,10 @@ static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)struct vmd_irq_list *irq = vmdirq->irq;struct vmd_dev *vmd =irq_data_get_irq_handler_data (data); - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->address_lo =MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_ID (index_from_irqs (vmd, irq)); - msg->data =0 ; + memset(&msg, 0, sizeof(*msg); + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.destid_0_7 =index_from_irqs (vmd, irq); }/*
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH v3 15/35] PCI: vmd: Use msi_msg shadow structs
Quoted 23 lines · David Woodhouse
From: Thomas Gleixner <[email protected]>
Use the x86 shadow structs in
msi_msginstead of the macros.Signed-off-by Thomas Gleixner <[email protected]>Signed-off-by David Woodhouse <[email protected]>--- drivers/pci/controller/vmd.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)drivers/pci/controller/vmd.c +0
-1@@ -18,7 +18,6 @@#include <asm/irqdomain.h> #include <asm/device.h> #include <asm/msi.h> -#include <asm/msidef.h> #define VMD_CFGBAR 0#define VMD_MEMBAR1 2
drivers/pci/controller/vmd.c static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg) · line 133
Quoted 8 lines
drivers/pci/controller/vmd.c +1
-4@@ -131,10 +130,10 @@ static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)struct vmd_irq_list *irq = vmdirq->irq;struct vmd_dev *vmd =irq_data_get_irq_handler_data (data); - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->address_lo =MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_ID (index_from_irqs (vmd, irq)); - msg->data =0 ; + memset(&msg, 0, sizeof(*msg);
This should be:
+ memset(msg, 0, sizeof(*msg);
Quoted 9 lines
drivers/pci/controller/vmd.c +3
-0+ msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.destid_0_7 =index_from_irqs (vmd, irq); }/* Signature · 1 line
2.26.2
Signature · 5 lines
Kees Cook _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH v3 15/35] PCI: vmd: Use msi_msg shadow structs
Review
Quoted 6 lines · Kees Cook
drivers/pci/controller/vmd.c static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg) · line 133
Quoted 1 line · David Woodhouse
drivers/pci/controller/vmd.c +1
-0+ memset(&msg, 0, sizeof(*msg);This should be:
(quoted diff) +1
-0+ memset(msg, 0, sizeof(*msg);
memset(msg, 0, sizeof(*msg));
Then it compiles _and_ is correct :)
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH v3 15/35] PCI: vmd: Use msi_msg shadow structs
Quoted 11 lines · Thomas Gleixner
Review
Quoted 6 lines · Kees Cook
drivers/pci/controller/vmd.c static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg) · line 133
Quoted 1 line · David Woodhouse
drivers/pci/controller/vmd.c +1
-0+ memset(&msg, 0, sizeof(*msg);This should be:
(quoted diff) +1
-0+ memset(msg, 0, sizeof(*msg);memset(msg, 0, sizeof(*msg));Then it compiles _and_ is correct :)
\o/ ;)
Signature · 5 lines
Kees Cook _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] PCI: vmd: Use msi_msg shadow structs
In reply to David WoodhouseThe following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: e16c8058a10ba8e38d0d1ad0b64e444b245ffdbd Gitweb: https://git.kernel.org/tip/e16c8058a10ba8e38d0d1ad0b64e444b245ffdbd Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:15 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
PCI: vmd: Use msi_msg shadow structs
Use the x86 shadow structs in msi_msg instead of the macros.
--- drivers/pci/controller/vmd.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/pci/controller/vmd.c +4 -5
@@ -18,7 +18,6 @@#include <asm/irqdomain.h> #include <asm/device.h> #include <asm/msi.h> -#include <asm/msidef.h> #define VMD_CFGBAR 0#define VMD_MEMBAR1 2
@@ -131,10 +130,10 @@ static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)struct vmd_irq_list *irq = vmdirq->irq;struct vmd_dev *vmd =irq_data_get_irq_handler_data (data); - msg->address_hi =MSI_ADDR_BASE_HI ; - msg->address_lo =MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_ID (index_from_irqs (vmd, irq)); - msg->data =0 ; +memset (msg,0 ,sizeof (*msg)); + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_addr_lo.destid_0_7 =index_from_irqs (vmd, irq); }/*
David Woodhouse
maintainerPATCH v3 16/35 x86/kvm: Use msi_msg shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Use the bitfields in the x86 shadow structs instead of decomposing the 32bit value with macros.
--- arch/x86/kvm/irq_comm.c | 31 +++++++++++++------------------ 1 file changed, 13 insertions(+), 18 deletions(-)
arch/x86/kvm/irq_comm.c +13 -18
@@ -16,8 +16,6 @@#include <trace/events/kvm.h> -#include <asm/msidef.h> -#include "irq.h" #include "ioapic.h"
@@ -104,22 +102,19 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,void kvm_set_msi_irq (struct kvm *kvm,struct kvm_kernel_irq_routing_entry *e,struct kvm_lapic_irq *irq) { -trace_kvm_msi_set_irq (e->msi.address_lo | (kvm->arch.x2apic_format ? - (u64 )e->msi.address_hi <<32 :0 ), - e->msi.data); - - irq->dest_id = (e->msi.address_lo & -MSI_ADDR_DEST_ID_MASK ) >>MSI_ADDR_DEST_ID_SHIFT ; -if (kvm->arch.x2apic_format) - irq->dest_id |=MSI_ADDR_EXT_DEST_ID (e->msi.address_hi); - irq->vector = (e->msi.data & -MSI_DATA_VECTOR_MASK ) >>MSI_DATA_VECTOR_SHIFT ; - irq->dest_mode =kvm_lapic_irq_dest_mode ( - !!((1 <<MSI_ADDR_DEST_MODE_SHIFT ) & e->msi.address_lo)); - irq->trig_mode = (1 <<MSI_DATA_TRIGGER_SHIFT ) & e->msi.data; - irq->delivery_mode = e->msi.data &0x700 ; - irq->msi_redir_hint = ((e->msi.address_lo - &MSI_ADDR_REDIRECTION_LOWPRI ) >0 ); +struct msi_msg msg = { .address_lo = e->msi.address_lo, + .address_hi = e->msi.address_hi, + .data = e->msi.data }; + +trace_kvm_msi_set_irq (msg.address_lo | (kvm->arch.x2apic_format ? + (u64 )msg.address_hi <<32 :0 ), msg.data); + + irq->dest_id =x86_msi_msg_get_destid (&msg, kvm->arch.x2apic_format); + irq->vector = msg.arch_data.vector; + irq->dest_mode =kvm_lapic_irq_dest_mode (msg.arch_addr_lo.dest_mode_logical); + irq->trig_mode = msg.arch_data.is_level; + irq->delivery_mode = msg.arch_data.delivery_mode <<8 ; + irq->msi_redir_hint = msg.arch_addr_lo.redirect_hint; irq->level =1 ; irq->shorthand =APIC_DEST_NOSHORT ; }
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/kvm: Use msi_msg shadow structs
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 485940e0e691d6d7874fe1fe3b9453c5af41aace Gitweb: https://git.kernel.org/tip/485940e0e691d6d7874fe1fe3b9453c5af41aace Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:16 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
x86/kvm: Use msi_msg shadow structs
Use the bitfields in the x86 shadow structs instead of decomposing the 32bit value with macros.
--- arch/x86/kvm/irq_comm.c | 31 +++++++++++++------------------ 1 file changed, 13 insertions(+), 18 deletions(-)
arch/x86/kvm/irq_comm.c +13 -18
@@ -16,8 +16,6 @@#include <trace/events/kvm.h> -#include <asm/msidef.h> -#include "irq.h" #include "ioapic.h"
@@ -104,22 +102,19 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,void kvm_set_msi_irq (struct kvm *kvm,struct kvm_kernel_irq_routing_entry *e,struct kvm_lapic_irq *irq) { -trace_kvm_msi_set_irq (e->msi.address_lo | (kvm->arch.x2apic_format ? - (u64 )e->msi.address_hi <<32 :0 ), - e->msi.data); - - irq->dest_id = (e->msi.address_lo & -MSI_ADDR_DEST_ID_MASK ) >>MSI_ADDR_DEST_ID_SHIFT ; -if (kvm->arch.x2apic_format) - irq->dest_id |=MSI_ADDR_EXT_DEST_ID (e->msi.address_hi); - irq->vector = (e->msi.data & -MSI_DATA_VECTOR_MASK ) >>MSI_DATA_VECTOR_SHIFT ; - irq->dest_mode =kvm_lapic_irq_dest_mode ( - !!((1 <<MSI_ADDR_DEST_MODE_SHIFT ) & e->msi.address_lo)); - irq->trig_mode = (1 <<MSI_DATA_TRIGGER_SHIFT ) & e->msi.data; - irq->delivery_mode = e->msi.data &0x700 ; - irq->msi_redir_hint = ((e->msi.address_lo - &MSI_ADDR_REDIRECTION_LOWPRI ) >0 ); +struct msi_msg msg = { .address_lo = e->msi.address_lo, + .address_hi = e->msi.address_hi, + .data = e->msi.data }; + +trace_kvm_msi_set_irq (msg.address_lo | (kvm->arch.x2apic_format ? + (u64 )msg.address_hi <<32 :0 ), msg.data); + + irq->dest_id =x86_msi_msg_get_destid (&msg, kvm->arch.x2apic_format); + irq->vector = msg.arch_data.vector; + irq->dest_mode =kvm_lapic_irq_dest_mode (msg.arch_addr_lo.dest_mode_logical); + irq->trig_mode = msg.arch_data.is_level; + irq->delivery_mode = msg.arch_data.delivery_mode <<8 ; + irq->msi_redir_hint = msg.arch_addr_lo.redirect_hint; irq->level =1 ; irq->shorthand =APIC_DEST_NOSHORT ; }
David Woodhouse
maintainerPATCH v3 17/35 x86/pci/xen: Use msi_msg shadow structs
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Use the msi_msg shadow structs and compose the message with named bitfields instead of the unreadable macro maze.
--- arch/x86/pci/xen.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)
arch/x86/pci/xen.c +11 -15
@@ -152,7 +152,6 @@ static int acpi_register_gsi_xen(struct device *dev, u32 gsi,#if defined(CONFIG_PCI_MSI )#include <linux/msi.h> -#include <asm/msidef.h> struct xen_pci_frontend_ops *xen_pci_frontend;EXPORT_SYMBOL_GPL (xen_pci_frontend);
@@ -210,23 +209,20 @@ static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)return ret; } -#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \ - MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0)) -static void xen_msi_compose_msg (struct pci_dev *pdev,unsigned int pirq,struct msi_msg *msg) { -/* We set vector == 0 to tell the hypervisor we don't care about it, -* but we want a pirq setup instead. -* We use the dest_id field to pass the pirq that we want. */ - msg->address_hi =MSI_ADDR_BASE_HI |MSI_ADDR_EXT_DEST_ID (pirq); - msg->address_lo = -MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_MODE_PHYSICAL | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (pirq); - - msg->data =XEN_PIRQ_MSI_DATA ; +/* +* We set vector == 0 to tell the hypervisor we don't care about +* it, but we want a pirq setup instead. We use the dest_id fields +* to pass the pirq that we want. +*/ +memset (msg,0 ,sizeof (*msg)); + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_hi.destid_8_31 = pirq >>8 ; + msg->arch_addr_lo.destid_0_7 = pirq &0xFF ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_data.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; }static int xen_hvm_setup_msi_irqs (struct pci_dev *dev,int nvec,int type)
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
RE: [PATCH v3 17/35] x86/pci/xen: Use msi_msg shadow structs
From: David Woodhouse
Quoted 48 lines
Sent: 24 October 2020 22:35
From: Thomas Gleixner <[email protected]>
Use the
msi_msgshadow structs and compose the message with named bitfields instead of the unreadable macro maze.Signed-off-by Thomas Gleixner <[email protected]>Signed-off-by David Woodhouse <[email protected]>--- arch/x86/pci/xen.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)arch/x86/pci/xen.c +2
-15@@ -152,7 +152,6 @@ static int acpi_register_gsi_xen(struct device *dev, u32 gsi,#if defined(CONFIG_PCI_MSI )#include <linux/msi.h> -#include <asm/msidef.h> struct xen_pci_frontend_ops *xen_pci_frontend;EXPORT_SYMBOL_GPL (xen_pci_frontend);@@ -210,23 +209,20 @@ static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)return ret; } -#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \ - MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0)) -static void xen_msi_compose_msg (struct pci_dev *pdev,unsigned int pirq,struct msi_msg *msg) { -/* We set vector == 0 to tell the hypervisor we don't care about it, -* but we want a pirq setup instead. -* We use the dest_id field to pass the pirq that we want. */ - msg->address_hi =MSI_ADDR_BASE_HI |MSI_ADDR_EXT_DEST_ID (pirq); - msg->address_lo = -MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_MODE_PHYSICAL | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (pirq); - - msg->data =XEN_PIRQ_MSI_DATA ; +/* +* We set vector == 0 to tell the hypervisor we don't care about
arch/x86/pci/xen.c static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type) · line 226
Quoted 10 lines
arch/x86/pci/xen.c +9
-0+* it, but we want a pirq setup instead. We use the dest_id fields +* to pass the pirq that we want. +*/ +memset (msg,0 ,sizeof (*msg)); + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_hi.destid_8_31 = pirq >>8 ; + msg->arch_addr_lo.destid_0_7 = pirq &0xFF ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_data.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; }
Just looking at a random one of these patches...
Does the compiler manage to optimise that reasonably? Or does it generate a lot of shifts and masks as each bitfield is set?
The code generation for bitfields is often a lot worse that that for |= setting bits in a word.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
David Woodhouse
maintainerRe: [PATCH v3 17/35] x86/pci/xen: Use msi_msg shadow structs
Quoted 8 lines · David Laight
Just looking at a random one of these patches...
Does the compiler manage to optimise that reasonably? Or does it generate a lot of shifts and masks as each bitfield is set?
The code generation for bitfields is often a lot worse that that for |= setting bits in a word.
Indeed, it appears to be utterly appalling. That was one of my motivations for doing it with masks and shifts in the first place.
Because in ioapic_setup_msg_from_msi(), for example, these two are consecutive in both source and destination:
entry->vector = msg.arch_data.vector; entry->delivery_mode = msg.arch_data.delivery_mode;
And so are those:
entry->ir_format = msg.arch_addr_lo.dmar_format; entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
But GCC 7.5.0 here is doing them all separately...
00000000000011e0 <ioapic_setup_msg_from_msi>:
{
11e0: 53 push %rbx
11e1: 48 89 f3 mov %rsi,%rbx
11e4: 48 83 ec 18 sub $0x18,%rsp
irq_chip_compose_msi_msg(irq_data, &msg);
11e8: 48 89 e6 mov %rsp,%rsi
{
11eb: 65 48 8b 04 25 28 00 mov %gs:0x28,%rax
11f2: 00 00
11f4: 48 89 44 24 10 mov %rax,0x10(%rsp)
11f9: 31 c0 xor %eax,%eax
irq_chip_compose_msi_msg(irq_data, &msg);
11fb: e8 00 00 00 00 callq 1200 <ioapic_setup_msg_from_msi+0x20>
entry->vector = msg.arch_data.vector;
1200: 0f b6 44 24 08 movzbl 0x8(%rsp),%eax
entry->delivery_mode = msg.arch_data.delivery_mode;
1205: 0f b6 53 01 movzbl 0x1(%rbx),%edx
1209: 0f b6 74 24 09 movzbl 0x9(%rsp),%esi
entry->vector = msg.arch_data.vector;
120e: 88 03 mov %al,(%rbx)
entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
1210: 0f b6 04 24 movzbl (%rsp),%eax
entry->delivery_mode = msg.arch_data.delivery_mode;
1214: 83 e2 f0 and $0xfffffff0,%edx
1217: 83 e6 07 and $0x7,%esi
entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
121a: 09 f2 or %esi,%edx
121c: 8d 0c 00 lea (%rax,%rax,1),%ecx
entry->ir_format = msg.arch_addr_lo.dmar_format;
121f: c0 e8 04 shr $0x4,%al
entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
1222: 83 e1 08 and $0x8,%ecx
1225: 09 ca or %ecx,%edx
1227: 88 53 01 mov %dl,0x1(%rbx)
entry->ir_format = msg.arch_addr_lo.dmar_format;
122a: 89 c2 mov %eax,%edx
entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
122c: 8b 04 24 mov (%rsp),%eax
122f: 83 e2 01 and $0x1,%edx
1232: c1 e8 05 shr $0x5,%eax
1235: 66 25 ff 7f and $0x7fff,%ax
1239: 8d 0c 00 lea (%rax,%rax,1),%ecx
123c: 66 c1 e8 07 shr $0x7,%ax
1240: 88 43 07 mov %al,0x7(%rbx)
1243: 09 ca or %ecx,%edx
}
1245: 48 8b 44 24 10 mov 0x10(%rsp),%rax
124a: 65 48 33 04 25 28 00 xor %gs:0x28,%rax
1251: 00 00
entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
1253: 88 53 06 mov %dl,0x6(%rbx)
}
1256: 75 06 jne 125e <ioapic_setup_msg_from_msi+0x7e>
1258: 48 83 c4 18 add $0x18,%rsp
125c: 5b pop %rbx
125d: c3 retq
125e: e8 00 00 00 00 callq 1263 <ioapic_setup_msg_from_msi+0x83>
1263: 0f 1f 00 nopl (%rax)
1266: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
126d: 00 00 00 Still, this isn't really a fast path so I'm content to file the GCC PR for the really *obvious* misses and let it take its course.
RE: [PATCH v3 17/35] x86/pci/xen: Use msi_msg shadow structs
From: David Woodhouse
Quoted 10 lines
Sent: 25 October 2020 10:26
To: David Laight <[email protected]>; [email protected]Quoted 6 lines · David Laight
Just looking at a random one of these patches...
Does the compiler manage to optimise that reasonably? Or does it generate a lot of shifts and masks as each bitfield is set?
Quoted 5 lines
Quoted 2 lines
The code generation for bitfields is often a lot worse that that for |= setting bits in a word.
Indeed, it appears to be utterly appalling. That was one of my motivations for doing it with masks and shifts in the first place.
I thought it would be.
I'm not even sure using bitfields to map hardware registers makes the code any more readable (or fool proof).
I suspect using #define constants and explicit |= and &= ~ is actually best - provided the names are descriptive enough.
If you set all the fields the compiler will merge the values and do a single write - provided nothing you read might alias the target.
The only way to get strongly typed integers is to cast them to a structure pointer type.
Together with a static inline function to remove the casts and | the values together it might make things fool proof. But I've never tried it.
ISTR once doing something like that with error codes, but it didn't ever make source control.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
[tip: x86/apic] x86/pci/xen: Use msi_msg shadow structs
In reply to David WoodhouseThe following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 41bb2115beec5e318095a89f5ad4a9c343cb21ad Gitweb: https://git.kernel.org/tip/41bb2115beec5e318095a89f5ad4a9c343cb21ad Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:17 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
x86/pci/xen: Use msi_msg shadow structs
Use the msi_msg shadow structs and compose the message with named bitfields instead of the unreadable macro maze.
--- arch/x86/pci/xen.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)
arch/x86/pci/xen.c +11 -15
@@ -152,7 +152,6 @@ static int acpi_register_gsi_xen(struct device *dev, u32 gsi,#if defined(CONFIG_PCI_MSI )#include <linux/msi.h> -#include <asm/msidef.h> struct xen_pci_frontend_ops *xen_pci_frontend;EXPORT_SYMBOL_GPL (xen_pci_frontend);
@@ -210,23 +209,20 @@ free:return ret; } -#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \ - MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0)) -static void xen_msi_compose_msg (struct pci_dev *pdev,unsigned int pirq,struct msi_msg *msg) { -/* We set vector == 0 to tell the hypervisor we don't care about it, -* but we want a pirq setup instead. -* We use the dest_id field to pass the pirq that we want. */ - msg->address_hi =MSI_ADDR_BASE_HI |MSI_ADDR_EXT_DEST_ID (pirq); - msg->address_lo = -MSI_ADDR_BASE_LO | -MSI_ADDR_DEST_MODE_PHYSICAL | -MSI_ADDR_REDIRECTION_CPU | -MSI_ADDR_DEST_ID (pirq); - - msg->data =XEN_PIRQ_MSI_DATA ; +/* +* We set vector == 0 to tell the hypervisor we don't care about +* it, but we want a pirq setup instead. We use the dest_id fields +* to pass the pirq that we want. +*/ +memset (msg,0 ,sizeof (*msg)); + msg->address_hi =X86_MSI_BASE_ADDRESS_HIGH ; + msg->arch_addr_hi.destid_8_31 = pirq >>8 ; + msg->arch_addr_lo.destid_0_7 = pirq &0xFF ; + msg->arch_addr_lo.base_address =X86_MSI_BASE_ADDRESS_LOW ; + msg->arch_data.delivery_mode =APIC_DELIVERY_MODE_EXTINT ; }static int xen_hvm_setup_msi_irqs (struct pci_dev *dev,int nvec,int type)
David Woodhouse
maintainerPATCH v3 18/35 x86/msi: Remove msidef.h
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
Nothing uses the macro maze anymore.
--- arch/x86/include/asm/msidef.h | 57 ----------------------------------- 1 file changed, 57 deletions(-) delete mode 100644 arch/x86/include/asm/msidef.h
arch/x86/include/asm/msidef.h deleted +0 -57
@@ -1,57 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_X86_MSIDEF_H -#define _ASM_X86_MSIDEF_H - -/* -* Constants for Intel APIC based MSI messages. -*/ - -/* -* Shifts for MSI data -*/ - -#define MSI_DATA_VECTOR_SHIFT 0 -#define MSI_DATA_VECTOR_MASK 0x000000ff -#define MSI_DATA_VECTOR (v) (((v) << MSI_DATA_VECTOR_SHIFT) & \ - MSI_DATA_VECTOR_MASK) - -#define MSI_DATA_DELIVERY_MODE_SHIFT 8 -#define MSI_DATA_DELIVERY_FIXED (0 << MSI_DATA_DELIVERY_MODE_SHIFT) -#define MSI_DATA_DELIVERY_LOWPRI (1 << MSI_DATA_DELIVERY_MODE_SHIFT) - -#define MSI_DATA_LEVEL_SHIFT 14 -#define MSI_DATA_LEVEL_DEASSERT (0 << MSI_DATA_LEVEL_SHIFT) -#define MSI_DATA_LEVEL_ASSERT (1 << MSI_DATA_LEVEL_SHIFT) - -#define MSI_DATA_TRIGGER_SHIFT 15 -#define MSI_DATA_TRIGGER_EDGE (0 << MSI_DATA_TRIGGER_SHIFT) -#define MSI_DATA_TRIGGER_LEVEL (1 << MSI_DATA_TRIGGER_SHIFT) - -/* -* Shift/mask fields for msi address -*/ - -#define MSI_ADDR_BASE_HI 0 -#define MSI_ADDR_BASE_LO 0xfee00000 - -#define MSI_ADDR_DEST_MODE_SHIFT 2 -#define MSI_ADDR_DEST_MODE_PHYSICAL (0 << MSI_ADDR_DEST_MODE_SHIFT) -#define MSI_ADDR_DEST_MODE_LOGICAL (1 << MSI_ADDR_DEST_MODE_SHIFT) - -#define MSI_ADDR_REDIRECTION_SHIFT 3 -#define MSI_ADDR_REDIRECTION_CPU (0 << MSI_ADDR_REDIRECTION_SHIFT) -/* dedicated cpu */ -#define MSI_ADDR_REDIRECTION_LOWPRI (1 << MSI_ADDR_REDIRECTION_SHIFT) -/* lowest priority */ - -#define MSI_ADDR_DEST_ID_SHIFT 12 -#define MSI_ADDR_DEST_ID_MASK 0x00ffff0 -#define MSI_ADDR_DEST_ID (dest) (((dest) << MSI_ADDR_DEST_ID_SHIFT) & \ - MSI_ADDR_DEST_ID_MASK) -#define MSI_ADDR_EXT_DEST_ID (dest) ((dest) & 0xffffff00) - -#define MSI_ADDR_IR_EXT_INT (1 << 4) -#define MSI_ADDR_IR_SHV (1 << 3) -#define MSI_ADDR_IR_INDEX1 (index) ((index & 0x8000) >> 13) -#define MSI_ADDR_IR_INDEX2 (index) ((index & 0x7fff) << 5) -#endif /* _ASM_X86_MSIDEF_H */
Signature · 6 lines
2.26.2 _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
[tip: x86/apic] x86/msi: Remove msidef.h
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: 0c1883c1eb9dfa3c72af6e00425eeb1eb171a03e Gitweb: https://git.kernel.org/tip/0c1883c1eb9dfa3c72af6e00425eeb1eb171a03e Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:18 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
x86/msi: Remove msidef.h
Nothing uses the macro maze anymore.
--- arch/x86/include/asm/msidef.h | 57 +---------------------------------- 1 file changed, 57 deletions(-) delete mode 100644 arch/x86/include/asm/msidef.h
arch/x86/include/asm/msidef.h deleted +0 -57
@@ -1,57 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_X86_MSIDEF_H -#define _ASM_X86_MSIDEF_H - -/* -* Constants for Intel APIC based MSI messages. -*/ - -/* -* Shifts for MSI data -*/ - -#define MSI_DATA_VECTOR_SHIFT 0 -#define MSI_DATA_VECTOR_MASK 0x000000ff -#define MSI_DATA_VECTOR (v) (((v) << MSI_DATA_VECTOR_SHIFT) & \ - MSI_DATA_VECTOR_MASK) - -#define MSI_DATA_DELIVERY_MODE_SHIFT 8 -#define MSI_DATA_DELIVERY_FIXED (0 << MSI_DATA_DELIVERY_MODE_SHIFT) -#define MSI_DATA_DELIVERY_LOWPRI (1 << MSI_DATA_DELIVERY_MODE_SHIFT) - -#define MSI_DATA_LEVEL_SHIFT 14 -#define MSI_DATA_LEVEL_DEASSERT (0 << MSI_DATA_LEVEL_SHIFT) -#define MSI_DATA_LEVEL_ASSERT (1 << MSI_DATA_LEVEL_SHIFT) - -#define MSI_DATA_TRIGGER_SHIFT 15 -#define MSI_DATA_TRIGGER_EDGE (0 << MSI_DATA_TRIGGER_SHIFT) -#define MSI_DATA_TRIGGER_LEVEL (1 << MSI_DATA_TRIGGER_SHIFT) - -/* -* Shift/mask fields for msi address -*/ - -#define MSI_ADDR_BASE_HI 0 -#define MSI_ADDR_BASE_LO 0xfee00000 - -#define MSI_ADDR_DEST_MODE_SHIFT 2 -#define MSI_ADDR_DEST_MODE_PHYSICAL (0 << MSI_ADDR_DEST_MODE_SHIFT) -#define MSI_ADDR_DEST_MODE_LOGICAL (1 << MSI_ADDR_DEST_MODE_SHIFT) - -#define MSI_ADDR_REDIRECTION_SHIFT 3 -#define MSI_ADDR_REDIRECTION_CPU (0 << MSI_ADDR_REDIRECTION_SHIFT) -/* dedicated cpu */ -#define MSI_ADDR_REDIRECTION_LOWPRI (1 << MSI_ADDR_REDIRECTION_SHIFT) -/* lowest priority */ - -#define MSI_ADDR_DEST_ID_SHIFT 12 -#define MSI_ADDR_DEST_ID_MASK 0x00ffff0 -#define MSI_ADDR_DEST_ID (dest) (((dest) << MSI_ADDR_DEST_ID_SHIFT) & \ - MSI_ADDR_DEST_ID_MASK) -#define MSI_ADDR_EXT_DEST_ID (dest) ((dest) & 0xffffff00) - -#define MSI_ADDR_IR_EXT_INT (1 << 4) -#define MSI_ADDR_IR_SHV (1 << 3) -#define MSI_ADDR_IR_INDEX1 (index) ((index & 0x8000) >> 13) -#define MSI_ADDR_IR_INDEX2 (index) ((index & 0x7fff) << 5) -#endif /* _ASM_X86_MSIDEF_H */
David Woodhouse
maintainerPATCH v3 19/35 x86/io_apic: Cleanup trigger/polarity helpers
In reply to David WoodhouseFrom: Thomas Gleixner <[email protected]>
'trigger' and 'polarity' are used throughout the I/O-APIC code for handling the trigger type (edge/level) and the active low/high configuration. While there are defines for initializing these variables and struct members, they are not used consequently and the meaning of 'trigger' and 'polarity' is opaque and confusing at best.
Rename them to 'is_level' and 'active_low' and make them boolean in various structs so it's entirely clear what the meaning is.
--- arch/x86/include/asm/hw_irq.h | 6 +- arch/x86/kernel/apic/io_apic.c | 244 +++++++++++++--------------- arch/x86/pci/intel_mid_pci.c | 8 +- drivers/iommu/amd/iommu.c | 10 +- drivers/iommu/intel/irq_remapping.c | 9 +- 5 files changed, 130 insertions(+), 147 deletions(-)
arch/x86/include/asm/hw_irq.h +3 -3
@@ -47,9 +47,9 @@ enum irq_alloc_type {struct ioapic_alloc_info {int pin;int node; -u32 trigger :1 ; -u32 polarity :1 ; -u32 valid :1 ; +u32 is_level :1 ; +u32 active_low :1 ; +u32 valid :1 ;struct IO_APIC_route_entry *entry; };
arch/x86/kernel/apic/io_apic.c +115 -129
389 lines not on this page · raw message
arch/x86/pci/intel_mid_pci.c +4 -4
@@ -214,7 +214,7 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where,static int intel_mid_pci_irq_enable (struct pci_dev *dev) {struct irq_alloc_info info; -int polarity; +bool polarity_low;int ret;u8 gsi;
@@ -229,7 +229,7 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)switch (intel_mid_identify_cpu ()) {case INTEL_MID_CPU_CHIP_TANGIER : - polarity =IOAPIC_POL_HIGH ; + polarity_low = false;/* Special treatment for IRQ0 */ if (gsi ==0 ) {
@@ -251,11 +251,11 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev) }break ;default : - polarity =IOAPIC_POL_LOW ; + polarity_low = true;break ; } -ioapic_set_alloc_attr (&info,dev_to_node (&dev->dev),1 , polarity); +ioapic_set_alloc_attr (&info,dev_to_node (&dev->dev),1 , polarity_low);/* * MRST only have IOAPIC, the PCI irq lines are 1:1 mapped to
drivers/iommu/amd/iommu.c +4 -6
@@ -3687,13 +3687,11 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, entry = info->ioapic.entry; info->ioapic.entry =NULL ;memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; + entry->vector = index; + entry->trigger = info->ioapic.is_level; + entry->polarity = info->ioapic.active_low;/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; + entry->mask = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
drivers/iommu/intel/irq_remapping.c +4 -5
@@ -1306,11 +1306,10 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,* irq handler will do the explicit EOI to the io-apic. */ entry->vector = info->ioapic.pin; - entry->mask =0 ;/* enable IRQ */ - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -if (info->ioapic.trigger) - entry->mask =1 ;/* Mask level triggered irqs. */ + entry->trigger = info->ioapic.is_level; + entry->polarity = info->ioapic.active_low; +/* Mask level triggered irqs. */ + entry->mask = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
Signature · 1 line
2.26.2
[tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
The following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Gitweb: https://git.kernel.org/tip/a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:19 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00
x86/io_apic: Cleanup trigger/polarity helpers
'trigger' and 'polarity' are used throughout the I/O-APIC code for handling the trigger type (edge/level) and the active low/high configuration. While there are defines for initializing these variables and struct members, they are not used consequently and the meaning of 'trigger' and 'polarity' is opaque and confusing at best.
Rename them to 'is_level' and 'active_low' and make them boolean in various structs so it's entirely clear what the meaning is.
--- arch/x86/include/asm/hw_irq.h | 6 +- arch/x86/kernel/apic/io_apic.c | 244 ++++++++++++--------------- arch/x86/pci/intel_mid_pci.c | 8 +- drivers/iommu/amd/iommu.c | 10 +- drivers/iommu/intel/irq_remapping.c | 9 +- 5 files changed, 130 insertions(+), 147 deletions(-)
arch/x86/include/asm/hw_irq.h +3 -3
@@ -47,9 +47,9 @@ enum irq_alloc_type {struct ioapic_alloc_info {int pin;int node; -u32 trigger :1 ; -u32 polarity :1 ; -u32 valid :1 ; +u32 is_level :1 ; +u32 active_low :1 ; +u32 valid :1 ;struct IO_APIC_route_entry *entry; };
arch/x86/kernel/apic/io_apic.c +115 -129
389 lines not on this page · raw message
arch/x86/pci/intel_mid_pci.c +4 -4
@@ -215,7 +215,7 @@ static int pci_write(struct pci_bus *bus, unsigned int devfn, int where,static int intel_mid_pci_irq_enable (struct pci_dev *dev) {struct irq_alloc_info info; -int polarity; +bool polarity_low;int ret;u8 gsi;
@@ -230,7 +230,7 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev)switch (intel_mid_identify_cpu ()) {case INTEL_MID_CPU_CHIP_TANGIER : - polarity =IOAPIC_POL_HIGH ; + polarity_low = false;/* Special treatment for IRQ0 */ if (gsi ==0 ) {
@@ -252,11 +252,11 @@ static int intel_mid_pci_irq_enable(struct pci_dev *dev) }break ;default : - polarity =IOAPIC_POL_LOW ; + polarity_low = true;break ; } -ioapic_set_alloc_attr (&info,dev_to_node (&dev->dev),1 , polarity); +ioapic_set_alloc_attr (&info,dev_to_node (&dev->dev),1 , polarity_low);/* * MRST only have IOAPIC, the PCI irq lines are 1:1 mapped to
drivers/iommu/amd/iommu.c +4 -6
@@ -3687,13 +3687,11 @@ static void irq_remapping_prepare_irte(struct amd_ir_data *data, entry = info->ioapic.entry; info->ioapic.entry =NULL ;memset (entry,0 ,sizeof (*entry)); - entry->vector = index; - entry->mask =0 ; - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; + entry->vector = index; + entry->trigger = info->ioapic.is_level; + entry->polarity = info->ioapic.active_low;/* Mask level triggered irqs. */ -if (info->ioapic.trigger) - entry->mask =1 ; + entry->mask = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
drivers/iommu/intel/irq_remapping.c +4 -5
@@ -1306,11 +1306,10 @@ static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,* irq handler will do the explicit EOI to the io-apic. */ entry->vector = info->ioapic.pin; - entry->mask =0 ;/* enable IRQ */ - entry->trigger = info->ioapic.trigger; - entry->polarity = info->ioapic.polarity; -if (info->ioapic.trigger) - entry->mask =1 ;/* Mask level triggered irqs. */ + entry->trigger = info->ioapic.is_level; + entry->polarity = info->ioapic.active_low; +/* Mask level triggered irqs. */ + entry->mask = info->ioapic.is_level;break ;case X86_IRQ_ALLOC_TYPE_HPET :
Re: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 11 lines · On 10/29/20 7:15 AM, tip-bot2 for Thomas Gleixner wrote:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Gitweb: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/commit/?id=a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:19 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00x86/io_apic: Cleanup trigger/polarity helpers
Quoted 8 lines
'trigger' and 'polarity' are used throughout the I/O-APIC code for handling the trigger type (edge/level) and the active low/high configuration. While there are defines for initializing these variables and struct members, they are not used consequently and the meaning of 'trigger' and 'polarity' is opaque and confusing at best.
Rename them to '
is_level' and 'active_low' and make them boolean in various structs so it's entirely clear what the meaning is.
Running the tip tree on my second generation EPYC system I'm seeing lots of the following:
[ 105.325371] hpet: Lost 9601 RTC interrupts [ 105.485766] hpet: Lost 9600 RTC interrupts [ 105.639182] hpet: Lost 9601 RTC interrupts [ 105.792155] hpet: Lost 9601 RTC interrupts [ 105.947076] hpet: Lost 9601 RTC interrupts [ 106.100876] hpet: Lost 9600 RTC interrupts [ 106.253444] hpet: Lost 9601 RTC interrupts [ 106.406722] hpet: Lost 9601 RTC interrupts
preventing the system from booting. I bisected it to this commit.
Additionally, I'm seeing warnings and error messages (which I haven't bisected, yet) along these lines:
[ 12.790801] WARNING: CPU: 135 PID: 1 at arch/x86/kernel/apic/apic.c:2505 __irq_msi_compose_msg+0x79/0x80 [ 98.121716] irq 3: nobody cared (try booting with the "irqpoll" option) [ 100.692087] irq 15: nobody cared (try booting with the "irqpoll" option) [ 100.800217] irq 11: nobody cared (try booting with the "irqpoll" option) [ 100.800407] irq 10: nobody cared (try booting with the "irqpoll" option)
Thanks,
Tom
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 35 lines · Tom Lendacky
Quoted 11 lines · On 10/29/20 7:15 AM, tip-bot2 for Thomas Gleixner wrote:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Gitweb: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/commit/?id=a27dca645d2c0f31abb7858aa0e10b2fa0f2f659 Author: Thomas Gleixner <[email protected]> AuthorDate: Sat, 24 Oct 2020 22:35:19 +01:00 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 28 Oct 2020 20:26:26 +01:00x86/io_apic: Cleanup trigger/polarity helpers
Quoted 8 lines
'trigger' and 'polarity' are used throughout the I/O-APIC code for handling the trigger type (edge/level) and the active low/high configuration. While there are defines for initializing these variables and struct members, they are not used consequently and the meaning of 'trigger' and 'polarity' is opaque and confusing at best.
Rename them to '
is_level' and 'active_low' and make them boolean in various structs so it's entirely clear what the meaning is.Running the tip tree on my second generation EPYC system I'm seeing lots of the following:
[ 105.325371] hpet: Lost 9601 RTC interrupts [ 105.485766] hpet: Lost 9600 RTC interrupts [ 105.639182] hpet: Lost 9601 RTC interrupts [ 105.792155] hpet: Lost 9601 RTC interrupts [ 105.947076] hpet: Lost 9601 RTC interrupts [ 106.100876] hpet: Lost 9600 RTC interrupts [ 106.253444] hpet: Lost 9601 RTC interrupts [ 106.406722] hpet: Lost 9601 RTC interruptspreventing the system from booting. I bisected it to this commit.
Quoted 8 lines
Additionally, I'm seeing warnings and error messages (which I haven't bisected, yet) along these lines:
[ 12.790801] WARNING: CPU: 135 PID: 1 at arch/x86/kernel/apic/apic.c:2505 __irq_msi_compose_msg+0x79/0x80 [ 98.121716] irq 3: nobody cared (try booting with the "irqpoll" option) [ 100.692087] irq 15: nobody cared (try booting with the "irqpoll" option) [ 100.800217] irq 11: nobody cared (try booting with the "irqpoll" option) [ 100.800407] irq 10: nobody cared (try booting with the "irqpoll" option)
Odd. The warning suggests we're asking __irq_msi_compose_msg() to compose an MSI targeted at an APIC ID above 255, which shouldn't ever happen.
Can we see a full boot log with apic=verbose please?
Borislav Petkov
maintainerRe: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to Tom LendackyQuoted 10 lines · Tom Lendacky
[ 105.325371] hpet: Lost 9601 RTC interrupts [ 105.485766] hpet: Lost 9600 RTC interrupts [ 105.639182] hpet: Lost 9601 RTC interrupts [ 105.792155] hpet: Lost 9601 RTC interrupts [ 105.947076] hpet: Lost 9601 RTC interrupts [ 106.100876] hpet: Lost 9600 RTC interrupts [ 106.253444] hpet: Lost 9601 RTC interrupts [ 106.406722] hpet: Lost 9601 RTC interruptspreventing the system from booting. I bisected it to this commit.
I bisected it to the exact same thing too, on an AMD laptop, after seeing what you're seeing.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 10 lines · Borislav Petkov
Quoted 9 lines · Tom Lendacky
[ 105.325371] hpet: Lost 9601 RTC interrupts [ 105.485766] hpet: Lost 9600 RTC interrupts [ 105.639182] hpet: Lost 9601 RTC interrupts [ 105.792155] hpet: Lost 9601 RTC interrupts [ 105.947076] hpet: Lost 9601 RTC interrupts [ 106.100876] hpet: Lost 9600 RTC interrupts [ 106.253444] hpet: Lost 9601 RTC interrupts [ 106.406722] hpet: Lost 9601 RTC interrupts
Quoted 4 lines
Quoted 1 line
preventing the system from booting. I bisected it to this commit.
I bisected it to the exact same thing too, on an AMD laptop, after seeing what you're seeing.
Bah. I'm a moron.
arch/x86/kernel/apic/io_apic.c +2 -2
@@ -809,9 +809,9 @@ static bool irq_is_level(int idx)case MP_IRQTRIG_DEFAULT :/* * Conforms to spec, ie. bus-type dependent trigger -* mode. PCI defaults to egde, ISA to level. +* mode. PCI defaults to level, ISA to edge. */ - level =test_bit (bus, mp_bus_not_pci); + level = !test_bit (bus, mp_bus_not_pci);/* Take EISA into account */ return eisa_irq_is_level (idx, bus, level);case MP_IRQTRIG_EDGE :
Re: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 24 lines · Thomas Gleixner
Quoted 10 lines · Borislav Petkov
Quoted 9 lines · Tom Lendacky
[ 105.325371] hpet: Lost 9601 RTC interrupts [ 105.485766] hpet: Lost 9600 RTC interrupts [ 105.639182] hpet: Lost 9601 RTC interrupts [ 105.792155] hpet: Lost 9601 RTC interrupts [ 105.947076] hpet: Lost 9601 RTC interrupts [ 106.100876] hpet: Lost 9600 RTC interrupts [ 106.253444] hpet: Lost 9601 RTC interrupts [ 106.406722] hpet: Lost 9601 RTC interruptsQuoted 4 lines
Quoted 1 line
preventing the system from booting. I bisected it to this commit.
I bisected it to the exact same thing too, on an AMD laptop, after seeing what you're seeing.
Bah. I'm a moron.
arch/x86/kernel/apic/io_apic.c +0
-0@@ -809,9 +809,9 @@ static bool irq_is_level(int idx)case MP_IRQTRIG_DEFAULT :
arch/x86/kernel/apic/io_apic.c static bool irq_is_level(int idx) · line 817
Quoted 10 lines
arch/x86/kernel/apic/io_apic.c +2
-2/* * Conforms to spec, ie. bus-type dependent trigger -* mode. PCI defaults to egde, ISA to level. +* mode. PCI defaults to level, ISA to edge. */ - level =test_bit (bus, mp_bus_not_pci); + level = !test_bit (bus, mp_bus_not_pci);/* Take EISA into account */ return eisa_irq_is_level (idx, bus, level);case MP_IRQTRIG_EDGE :
I was about to send the dmesg output when I saw this. A quick test with this change resolves the boot issue, thanks!
I'm still seeing the warning at arch/x86/kernel/apic/apic.c:2527, but I'll start a separate thread on that.
Thanks,
Tom
Re: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 3 lines · Tom Lendacky
On 11/10/20 8:34 AM, Thomas Gleixner wrote:
I was about to send the dmesg output when I saw this. A quick test with this change resolves the boot issue, thanks!
/me feels stupid
Quoted 2 lines
I'm still seeing the warning at
arch/x86/kernel/apic/apic.c:2527, but I'll start a separate thread on that.
Can you please provide the backtrace?
Thanks,
tglx
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 6 lines · Thomas Gleixner
Quoted 4 lines · Tom Lendacky
On 11/10/20 8:34 AM, Thomas Gleixner wrote:
I was about to send the dmesg output when I saw this. A quick test with
this change resolves the boot issue, thanks!
Quoted 7 lines
/me feels stupid
Quoted 3 lines
I'm still seeing the warning at
arch/x86/kernel/apic/apic.c:2527, but I'll
start a separate thread on that.Can you please provide the backtrace?
We'll want something like this, to stop AMD IOMMU from registering its irqdomain even when it shouldn't. But that's the wrong way round; that would give you remapped MSIs when it should be using the x86_vector_domain for them. And you have the converse, it seems.
drivers/iommu/amd/init.c +5 -3
@@ -1601,9 +1601,11 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)if (ret)return ret; - ret =amd_iommu_create_irq_domain (iommu); -if (ret) -return ret; +if (amd_iommu_irq_remap) { + ret =amd_iommu_create_irq_domain (iommu); +if (ret) +return ret; + }/* * Make sure IOMMU is not considered to translate itself. The IVRS
Re: [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to Thomas GleixnerQuoted 11 lines · Thomas Gleixner
Quoted 3 lines · Tom Lendacky
On 11/10/20 8:34 AM, Thomas Gleixner wrote:
I was about to send the dmesg output when I saw this. A quick test with this change resolves the boot issue, thanks!/me feels stupid
Quoted 2 lines
I'm still seeing the warning at
arch/x86/kernel/apic/apic.c:2527, but I'll start a separate thread on that.Can you please provide the backtrace?
Yep. The warning started triggering with:47bea873cf80 ("x86/msi: Only use high bits of MSI address for DMAR unit")
Here's the backtrace:
[ 15.611109] ------------[ cut here ]------------ [ 15.616274] WARNING: CPU: 184 PID: 1 at arch/x86/kernel/apic/apic.c:2505 __irq_msi_compose_msg+0x79/0x80 [ 15.626855] Modules linked in: [ 15.630263] CPU: 184 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc1-sos-custom #1 [ 15.638516] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.647549] RIP: 0010:__irq_msi_compose_msg+0x79/0x80 [ 15.653188] Code: 0f f0 ff 09 d0 89 06 8b 47 04 c7 46 04 00 00 00 00 88 46 08 45 84 c0 74 08 8b 07 30 c0 89 46 04 c3 81 3f ff 00 00 00 77 01 c3 <0f> 0b c3 0f 1f 40 00 0f 1f 44 00 00 8b 05 81 f9 04 02 85 c0 75 16 [ 15.674140] RSP: 0018:ffffc900000c7c30 EFLAGS: 00010212 [ 15.679971] RAX: 0000000000000021 RBX: ffff888143789028 RCX: 0000000000000000 [ 15.687936] RDX: 0000000000000000 RSI: ffffc900000c7c40 RDI: ffff8881447dd1c0 [ 15.695899] RBP: ffff888143789028 R08: 0000000000000000 R09: 0000000000000005 [ 15.703861] R10: 0000000000000002 R11: 0000000000000004 R12: ffff88900e7b80c0 [ 15.711825] R13: 0000000000000000 R14: ffff88907646ba80 R15: 0000000000000000 [ 15.719789] FS: 0000000000000000(0000) GS:ffff88900f000000(0000) knlGS:0000000000000000 [ 15.728821] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 15.735234] CR2: 0000000000000000 CR3: 000000114d40a000 CR4: 0000000000350ee0 [ 15.743197] Call Trace: [ 15.745929] irq_chip_compose_msi_msg+0x2e/0x40 [ 15.750984] msi_domain_activate+0x4b/0x90 [ 15.755556] __irq_domain_activate_irq+0x53/0x80 [ 15.760707] ? irq_set_msi_desc_off+0x5a/0x90 [ 15.765568] irq_domain_activate_irq+0x25/0x40 [ 15.770525] __msi_domain_alloc_irqs+0x16a/0x310 [ 15.775680] __pci_enable_msi_range+0x182/0x2b0 [ 15.780738] ? e820__memblock_setup+0x7d/0x7d [ 15.785597] pci_enable_msi+0x16/0x30 [ 15.789685] iommu_init_msi+0x30/0x190 [ 15.793860] state_next+0x39d/0x665 [ 15.797754] ? e820__memblock_setup+0x7d/0x7d [ 15.802615] iommu_go_to_state+0x24/0x28 [ 15.806993] amd_iommu_init+0x11/0x46 [ 15.811076] pci_iommu_init+0x16/0x3f [ 15.815165] do_one_initcall+0x44/0x1d0 [ 15.819447] kernel_init_freeable+0x1e7/0x249 [ 15.824311] ? rest_init+0xb4/0xb4 [ 15.828097] kernel_init+0xa/0x10c [ 15.831889] ret_from_fork+0x22/0x30 [ 15.835883] CPU: 184 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc1-sos-custom #1 [ 15.836876] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.836876] Call Trace: [ 15.836876] dump_stack+0x6d/0x88 [ 15.836876] __warn.cold+0x24/0x3d [ 15.836876] ? __irq_msi_compose_msg+0x79/0x80 [ 15.836876] report_bug+0xd1/0x100 [ 15.836876] handle_bug+0x35/0x80 [ 15.836876] exc_invalid_op+0x14/0x70 [ 15.836876] asm_exc_invalid_op+0x12/0x20
[ 15.836876] RIP: 0010:__irq_msi_compose_msg+0x79/0x80 [ 15.836876] Code: 0f f0 ff 09 d0 89 06 8b 47 04 c7 46 04 00 00 00 00 88 46 08 45 84 c0 74 08 8b 07 30 c0 89 46 04 c3 81 3f ff 00 00 00 77 01 c3 <0f> 0b c3 0f 1f 40 00 0f 1f 44 00 00 8b 05 81 f9 04 02 85 c0 75 16 [ 15.836876] RSP: 0018:ffffc900000c7c30 EFLAGS: 00010212 [ 15.836876] RAX: 0000000000000021 RBX: ffff888143789028 RCX: 0000000000000000 [ 15.836876] RDX: 0000000000000000 RSI: ffffc900000c7c40 RDI: ffff8881447dd1c0 [ 15.836876] RBP: ffff888143789028 R08: 0000000000000000 R09: 0000000000000005 [ 15.836876] R10: 0000000000000002 R11: 0000000000000004 R12: ffff88900e7b80c0 [ 15.836876] R13: 0000000000000000 R14: ffff88907646ba80 R15: 0000000000000000 [ 15.836876] irq_chip_compose_msi_msg+0x2e/0x40 [ 15.836876] msi_domain_activate+0x4b/0x90 [ 15.836876] __irq_domain_activate_irq+0x53/0x80 [ 15.836876] ? irq_set_msi_desc_off+0x5a/0x90 [ 15.836876] irq_domain_activate_irq+0x25/0x40 [ 15.836876] __msi_domain_alloc_irqs+0x16a/0x310 [ 15.836876] __pci_enable_msi_range+0x182/0x2b0 [ 15.836876] ? e820__memblock_setup+0x7d/0x7d [ 15.836876] pci_enable_msi+0x16/0x30 [ 15.836876] iommu_init_msi+0x30/0x190 [ 15.836876] state_next+0x39d/0x665 [ 15.836876] ? e820__memblock_setup+0x7d/0x7d [ 15.836876] iommu_go_to_state+0x24/0x28 [ 15.836876] amd_iommu_init+0x11/0x46 [ 15.836876] pci_iommu_init+0x16/0x3f [ 15.836876] do_one_initcall+0x44/0x1d0 [ 15.836876] kernel_init_freeable+0x1e7/0x249 [ 15.836876] ? rest_init+0xb4/0xb4 [ 15.836876] kernel_init+0xa/0x10c [ 15.836876] ret_from_fork+0x22/0x30 [ 16.046524] ---[ end trace 2fb272e5ca8df1c8 ]---
Thanks,
Tom
Quoted 5 lines
Thanks,
tglx
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 29 lines · Tom Lendacky
Yep. The warning started triggering with:
47bea873cf80("x86/msi: Only use high bits of MSI address for DMAR unit")Here's the backtrace:
[ 15.611109] ------------[ cut here ]------------ [ 15.616274] WARNING: CPU: 184 PID: 1 at arch/x86/kernel/apic/apic.c:2505 __irq_msi_compose_msg+0x79/0x80 [ 15.626855] Modules linked in: [ 15.630263] CPU: 184 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc1-sos-custom #1 [ 15.638516] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.647549] RIP: 0010:__irq_msi_compose_msg+0x79/0x80 [ 15.653188] Code: 0f f0 ff 09 d0 89 06 8b 47 04 c7 46 04 00 00 00 00 88 46 08 45 84 c0 74 08 8b 07 30 c0 89 46 04 c3 81 3f ff 00 00 00 77 01 c3 <0f> 0b c3 0f 1f 40 00 0f 1f 44 00 00 8b 05 81 f9 04 02 85 c0 75 16 [ 15.674140] RSP: 0018:ffffc900000c7c30 EFLAGS: 00010212 [ 15.679971] RAX: 0000000000000021 RBX: ffff888143789028 RCX: 0000000000000000 [ 15.687936] RDX: 0000000000000000 RSI: ffffc900000c7c40 RDI: ffff8881447dd1c0 [ 15.695899] RBP: ffff888143789028 R08: 0000000000000000 R09: 0000000000000005 [ 15.703861] R10: 0000000000000002 R11: 0000000000000004 R12: ffff88900e7b80c0 [ 15.711825] R13: 0000000000000000 R14: ffff88907646ba80 R15: 0000000000000000 [ 15.719789] FS: 0000000000000000(0000) GS:ffff88900f000000(0000) knlGS:0000000000000000 [ 15.728821] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 15.735234] CR2: 0000000000000000 CR3: 000000114d40a000 CR4: 0000000000350ee0 [ 15.743197] Call Trace: [ 15.745929] irq_chip_compose_msi_msg+0x2e/0x40 [ 15.750984] msi_domain_activate+0x4b/0x90 [ 15.755556] __irq_domain_activate_irq+0x53/0x80 [ 15.760707] ? irq_set_msi_desc_off+0x5a/0x90 [ 15.765568] irq_domain_activate_irq+0x25/0x40 [ 15.770525] __msi_domain_alloc_irqs+0x16a/0x310 [ 15.775680] __pci_enable_msi_range+0x182/0x2b0
Quoted 8 lines
[ 15.780738] ? e820__memblock_setup+0x7d/0x7d [ 15.785597] pci_enable_msi+0x16/0x30 [ 15.789685] iommu_init_msi+0x30/0x190 [ 15.793860] state_next+0x39d/0x665 [ 15.797754] ? e820__memblock_setup+0x7d/0x7d [ 15.802615] iommu_go_to_state+0x24/0x28 [ 15.806993] amd_iommu_init+0x11/0x46 [ 15.811076] pci_iommu_init+0x16/0x3f
Oh joy.
It's asking the core code to generate a PCI MSI message for it and actually program that to the PCI device (since the IOMMU itself is a PCI device).
That isn't actually used for generating MSI, but is instead interpreted to write the intcapxt registers which *do* generate the interrupts.
That wants fixing, preferably not to go via MSI format at all, or maybe just to use the 'dmar' flag to __irq_msi_compose_msg(). Either way by having an irqdomain of its own like the Intel IOMMU does.
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 48 lines · David Woodhouse
Quoted 29 lines · Tom Lendacky
Yep. The warning started triggering with:
47bea873cf80("x86/msi: Only use high bits of MSI address for DMAR unit")Here's the backtrace:
[ 15.611109] ------------[ cut here ]------------ [ 15.616274] WARNING: CPU: 184 PID: 1 at arch/x86/kernel/apic/apic.c:2505 __irq_msi_compose_msg+0x79/0x80 [ 15.626855] Modules linked in: [ 15.630263] CPU: 184 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc1-sos-custom #1 [ 15.638516] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.647549] RIP: 0010:__irq_msi_compose_msg+0x79/0x80 [ 15.653188] Code: 0f f0 ff 09 d0 89 06 8b 47 04 c7 46 04 00 00 00 00 88 46 08 45 84 c0 74 08 8b 07 30 c0 89 46 04 c3 81 3f ff 00 00 00 77 01 c3 <0f> 0b c3 0f 1f 40 00 0f 1f 44 00 00 8b 05 81 f9 04 02 85 c0 75 16 [ 15.674140] RSP: 0018:ffffc900000c7c30 EFLAGS: 00010212 [ 15.679971] RAX: 0000000000000021 RBX: ffff888143789028 RCX: 0000000000000000 [ 15.687936] RDX: 0000000000000000 RSI: ffffc900000c7c40 RDI: ffff8881447dd1c0 [ 15.695899] RBP: ffff888143789028 R08: 0000000000000000 R09: 0000000000000005 [ 15.703861] R10: 0000000000000002 R11: 0000000000000004 R12: ffff88900e7b80c0 [ 15.711825] R13: 0000000000000000 R14: ffff88907646ba80 R15: 0000000000000000 [ 15.719789] FS: 0000000000000000(0000) GS:ffff88900f000000(0000) knlGS:0000000000000000 [ 15.728821] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 15.735234] CR2: 0000000000000000 CR3: 000000114d40a000 CR4: 0000000000350ee0 [ 15.743197] Call Trace: [ 15.745929] irq_chip_compose_msi_msg+0x2e/0x40 [ 15.750984] msi_domain_activate+0x4b/0x90 [ 15.755556] __irq_domain_activate_irq+0x53/0x80 [ 15.760707] ? irq_set_msi_desc_off+0x5a/0x90 [ 15.765568] irq_domain_activate_irq+0x25/0x40 [ 15.770525] __msi_domain_alloc_irqs+0x16a/0x310 [ 15.775680] __pci_enable_msi_range+0x182/0x2b0Quoted 8 lines
[ 15.780738] ? e820__memblock_setup+0x7d/0x7d [ 15.785597] pci_enable_msi+0x16/0x30 [ 15.789685] iommu_init_msi+0x30/0x190 [ 15.793860] state_next+0x39d/0x665 [ 15.797754] ? e820__memblock_setup+0x7d/0x7d [ 15.802615] iommu_go_to_state+0x24/0x28 [ 15.806993] amd_iommu_init+0x11/0x46 [ 15.811076] pci_iommu_init+0x16/0x3fOh joy.
It's asking the core code to generate a PCI MSI message for it and actually program that to the PCI device (since the IOMMU itself is a PCI device).
That isn't actually used for generating MSI, but is instead interpreted to write the intcapxt registers which *do* generate the interrupts.
Quoted 6 lines
That wants fixing, preferably not to go via MSI format at all, or maybe just to use the 'dmar' flag to
__irq_msi_compose_msg(). Either way by having an irqdomain of its own like the Intel IOMMU does.If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
That's odd. I haven't had an issue with or heard of any issues with booting post-5.5 kernels with the AMD IOMMU enabled. What kind of problems are you encountering?
Thanks,
Tom
Quoted 1 line
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to David WoodhouseQuoted 24 lines · David Woodhouse
Quoted 5 lines · Tom Lendacky
Yep. The warning started triggering with:
47bea873cf80("x86/msi: Only use high bits of MSI address for DMAR unit")Here's the backtrace:
Quoted 10 lines
[ 15.745929] irq_chip_compose_msi_msg+0x2e/0x40 [ 15.750984] msi_domain_activate+0x4b/0x90 [ 15.755556] __irq_domain_activate_irq+0x53/0x80 [ 15.760707] ? irq_set_msi_desc_off+0x5a/0x90 [ 15.765568] irq_domain_activate_irq+0x25/0x40 [ 15.770525] __msi_domain_alloc_irqs+0x16a/0x310 [ 15.775680] __pci_enable_msi_range+0x182/0x2b0 [ 15.780738] ? e820__memblock_setup+0x7d/0x7d [ 15.785597] pci_enable_msi+0x16/0x30 [ 15.789685] iommu_init_msi+0x30/0x190It's asking the core code to generate a PCI MSI message for it and actually program that to the PCI device (since the IOMMU itself is a PCI device).
That isn't actually used for generating MSI, but is instead interpreted to write the intcapxt registers which *do* generate the interrupts.
Quoted 6 lines
That wants fixing, preferably not to go via MSI format at all, or maybe just to use the 'dmar' flag to
__irq_msi_compose_msg(). Either way by having an irqdomain of its own like the Intel IOMMU does.If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
It can share the dmar domain code. Let me frob something.
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 5 lines · Thomas Gleixner
Quoted 2 lines · David Woodhouse
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the moment. Something like the untested below should work.
Thanks,
tglx
---
arch/x86/include/asm/apic.h +1 -1
@@ -523,7 +523,7 @@ struct msi_msg;struct irq_cfg ;extern void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, -bool dmar); +bool iommu);extern void ioapic_zap_locks (void );
arch/x86/include/asm/irqdomain.h +2 -0
@@ -63,4 +63,6 @@ static inline void x86_create_pci_msi_do#define x86_pci_msi_default_domain NULL#endif +struct irq_domain *x86_create_iommu_msi_domain (void ); +#endif
arch/x86/kernel/apic/apic.c +2 -2
@@ -2498,7 +2498,7 @@ int hard_smp_processor_id(void) }void __irq_msi_compose_msg (struct irq_cfg *cfg,struct msi_msg *msg, -bool dmar) +bool iommu) {memset (msg,0 ,sizeof (*msg));
@@ -2519,7 +2519,7 @@ void __irq_msi_compose_msg(struct irq_cf* some hypervisors allow the extended destination ID field in bits * 5-11 to be used, giving support for 15 bits of APIC IDs in total. */ -if (dmar) +if (iommu) msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >>8 ;else if (virt_ext_dest_id && cfg->dest_apicid <0x8000 ) msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >>8 ;
arch/x86/kernel/apic/msi.c +48 -11
100 lines not on this page · raw message
drivers/iommu/amd/init.c +8 -0
@@ -1759,6 +1759,7 @@ static const struct attribute_group *amdstatic int __init iommu_init_pci (struct amd_iommu *iommu) { +static struct irq_domain *msidom;int cap_ptr = iommu->cap_ptr;int ret;
@@ -1767,6 +1768,13 @@ static int __init iommu_init_pci(structif (!iommu->dev)return -ENODEV ; +if (!msidom) { + msidom =x86_create_iommu_msi_domain (); +if (!msidom) +return -ENODEV ; + } +dev_set_msi_domain (&iommu->dev->dev, msidom); +/* Prevent binding other PCI device drivers to IOMMU devices */ iommu->dev->match_driver = false;
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 9 lines · Thomas Gleixner
Quoted 5 lines · Thomas Gleixner
Quoted 2 lines · David Woodhouse
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the moment. Something like the untested below should work.
Does it even need its own irqdomain? Can it not just allocate directly from the vector domain then program its own register directly based on the irq_cfg?
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 6 lines · David Woodhouse
Quoted 5 lines · Thomas Gleixner
Quoted 4 lines · Thomas Gleixner
Quoted 2 lines · David Woodhouse
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the moment. Something like the untested below should work.
Does it even need its own irqdomain? Can it not just allocate directly from the vector domain then program its own register directly based on the
irq_cfg?
It uses pci_enable_msi() and I have no clue about that piece of hardware and whether that is actually required or not. If it is, then it needs a domain because that's what pci_enable_msi() uses.
Thanks,
tglx
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 16 lines · Thomas Gleixner
Quoted 1 line · David Woodhouse
On 10 November 2020 18:56:17 GMT, Thomas Gleixner
Quoted 11 lines · <[email protected]> wrote:
Quoted 9 lines · Thomas Gleixner
Quoted 5 lines · David Woodhouse
Quoted 2 lines
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the moment. Something like the untested below should work.
Does it even need its own irqdomain? Can it not just allocate
directly
Quoted 8 lines
Quoted 1 line
from the vector domain then program its own register directly based
on
Quoted 1 line
the
irq_cfg?It uses
pci_enable_msi()and I have no clue about that piece of hardware
and whether that is actually required or not. If it is, then it needs a domain because that's whatpci_enable_msi()uses.
I'd be kind of surprised if it is required, but testing on qemu is obviously not going to cut it. Tom?
Signature · 1 line
Sent from my Android device with K-9 Mail. Please excuse my brevity.
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 21 lines · David Woodhouse
Quoted 18 lines · Thomas Gleixner
Quoted 1 line · David Woodhouse
On 10 November 2020 18:56:17 GMT, Thomas Gleixner
Quoted 11 lines · <[email protected]> wrote:
Quoted 9 lines · Thomas Gleixner
Quoted 5 lines · David Woodhouse
Quoted 2 lines
If I could get post-5.5 kernels to boot at all with the AMD IOMMU enabled, I'd have a go at throwing that together now...
It can share the dmar domain code. Let me frob something.
Not much to share there and I can't access my AMD machine at the moment. Something like the untested below should work.
Does it even need its own irqdomain? Can it not just allocate
directly
Quoted 1 line
from the vector domain then program its own register directly based
on
Quoted 8 lines
Quoted 6 lines
Quoted 1 line
the
irq_cfg?It uses
pci_enable_msi()and I have no clue about that piece of hardware
and whether that is actually required or not. If it is, then it needs a domain because that's whatpci_enable_msi()uses.I'd be kind of surprised if it is required, but testing on qemu is obviously not going to cut it. Tom?
Was just in the process of testing it... I still get a warning. Here's the new backtrace:
[ 15.581115] WARNING: CPU: 6 PID: 1 at arch/x86/kernel/apic/apic.c:2527 __irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] Modules linked in: [ 15.581115] CPU: 6 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc3-sos-custom #1 [ 15.581115] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.581115] RIP: 0010:__irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] Code: 01 00 74 1e 3d ff 7f 00 00 77 1f 0f b7 16 c1 e8 08 83 e0 7f c1 e0 05 66 81 e2 1f f0 09 d0 66 89 06 c3 3d ff 00 00 00 77 01 c3 <0f> 0b c3 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 00 0f 1f 44 00 00 [ 15.581115] RSP: 0018:ffffc900000c7c18 EFLAGS: 00010012 [ 15.581115] RAX: 0000000000000100 RBX: 0000000000000000 RCX: 0000000000000000 [ 15.581115] RDX: 0000000000000000 RSI: ffffc900000c7c20 RDI: ffff8881088341c0 [ 15.581115] RBP: ffff8881599e0428 R08: 0000000000000000 R09: ffffffffffffffff [ 15.581115] R10: 0000000000000003 R11: 0000000000000004 R12: ffff8881088341c0 [ 15.581115] R13: 0000000000000000 R14: 0000000000000004 R15: ffff888108834180 [ 15.581115] FS: 0000000000000000(0000) GS:ffff88900d380000(0000) knlGS:0000000000000000 [ 15.581115] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 15.581115] CR2: 0000000000000000 CR3: 000000015240a000 CR4: 0000000000350ee0 [ 15.581115] Call Trace: [ 15.581115] irq_msi_update_msg+0x4d/0x80 [ 15.581115] msi_set_affinity+0x160/0x190 [ 15.581115] irq_do_set_affinity+0x52/0x190 [ 15.581115] irq_setup_affinity+0xd7/0x170 [ 15.581115] irq_startup+0x5d/0xf0 [ 15.581115] __setup_irq+0x6b9/0x700 [ 15.581115] request_threaded_irq+0xf8/0x160 [ 15.581115] ? irq_remapping_alloc+0x4d0/0x4d0 [ 15.581115] ? e820__memblock_setup+0x7d/0x7d [ 15.581115] iommu_init_msi+0x60/0x190 [ 15.581115] state_next+0x39d/0x665 [ 15.581115] ? e820__memblock_setup+0x7d/0x7d [ 15.581115] iommu_go_to_state+0x24/0x28 [ 15.581115] amd_iommu_init+0x11/0x46 [ 15.581115] pci_iommu_init+0x16/0x3f [ 15.581115] do_one_initcall+0x44/0x1d0 [ 15.581115] kernel_init_freeable+0x1e7/0x249 [ 15.581115] ? rest_init+0xb4/0xb4 [ 15.581115] kernel_init+0xa/0x10c [ 15.581115] ret_from_fork+0x22/0x30 [ 15.581115] CPU: 6 PID: 1 Comm: swapper/0 Not tainted 5.10.0-rc3-sos-custom #1 [ 15.581115] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS REX1006G 01/25/2020 [ 15.581115] Call Trace: [ 15.581115] dump_stack+0x6d/0x88 [ 15.581115] __warn.cold+0x24/0x3d [ 15.581115] ? __irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] report_bug+0xd1/0x100 [ 15.581115] handle_bug+0x35/0x80 [ 15.581115] exc_invalid_op+0x14/0x70 [ 15.581115] asm_exc_invalid_op+0x12/0x20 [ 15.581115] RIP: 0010:__irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] Code: 01 00 74 1e 3d ff 7f 00 00 77 1f 0f b7 16 c1 e8 08 83 e0 7f c1 e0 05 66 81 e2 1f f0 09 d0 66 89 06 c3 3d ff 00 00 00 77 01 c3 <0f> 0b c3 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 00 0f 1f 44 00 00 [ 15.581115] RSP: 0018:ffffc900000c7c18 EFLAGS: 00010012 [ 15.581115] RAX: 0000000000000100 RBX: 0000000000000000 RCX: 0000000000000000 [ 15.581115] RDX: 0000000000000000 RSI: ffffc900000c7c20 RDI: ffff8881088341c0 [ 15.581115] RBP: ffff8881599e0428 R08: 0000000000000000 R09: ffffffffffffffff [ 15.581115] R10: 0000000000000003 R11: 0000000000000004 R12: ffff8881088341c0 [ 15.581115] R13: 0000000000000000 R14: 0000000000000004 R15: ffff888108834180 [ 15.581115] irq_msi_update_msg+0x4d/0x80 [ 15.581115] msi_set_affinity+0x160/0x190 [ 15.581115] irq_do_set_affinity+0x52/0x190 [ 15.581115] irq_setup_affinity+0xd7/0x170 [ 15.581115] irq_startup+0x5d/0xf0 [ 15.581115] __setup_irq+0x6b9/0x700 [ 15.581115] request_threaded_irq+0xf8/0x160 [ 15.581115] ? irq_remapping_alloc+0x4d0/0x4d0 [ 15.581115] ? e820__memblock_setup+0x7d/0x7d [ 15.581115] iommu_init_msi+0x60/0x190 [ 15.581115] state_next+0x39d/0x665 [ 15.581115] ? e820__memblock_setup+0x7d/0x7d [ 15.581115] iommu_go_to_state+0x24/0x28 [ 15.581115] amd_iommu_init+0x11/0x46 [ 15.581115] pci_iommu_init+0x16/0x3f [ 15.581115] do_one_initcall+0x44/0x1d0 [ 15.581115] kernel_init_freeable+0x1e7/0x249 [ 15.581115] ? rest_init+0xb4/0xb4 [ 15.581115] kernel_init+0xa/0x10c [ 15.581115] ret_from_fork+0x22/0x30 [ 15.581115] ---[ end trace 05c9465e30ba6a20 ]---
Thanks,
Tom
Quoted 1 line
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 5 lines · Tom Lendacky
On 11/10/20 3:30 PM, David Woodhouse wrote:
[ 15.581115] WARNING: CPU: 6 PID: 1 at arch/x86/kernel/apic/apic.c:2527 __irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] Call Trace: [ 15.581115] irq_msi_update_msg+0x4d/0x80 [ 15.581115] msi_set_affinity+0x160/0x190
Duh. Yes, that want's some love as well. Delta patch below.
Thanks,
tglx
---
arch/x86/kernel/apic/msi.c +5 -4
@@ -24,10 +24,11 @@ struct irq_domain *x86_pci_msi_default_dstatic void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) { +struct irq_chip *chip =irq_data_get_irq_chip (irqd);struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg, false); -irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); +__irq_msi_compose_msg (cfg, msg, chip->flags &IRQCHIP_MSI_EXTID ); + chip->irq_write_msi_msg (irqd, msg); } static int
@@ -271,7 +272,7 @@ static struct irq_chip iommu_msi_control .irq_retrigger = irq_chip_retrigger_hierarchy, .irq_set_affinity = msi_set_affinity, .irq_compose_msi_msg = iommu_msi_compose_msg, - .flags = IRQCHIP_SKIP_SET_WAKE, + .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MSI_EXTID, };static struct msi_domain_info iommu_msi_domain_info = {
@@ -310,7 +311,7 @@ static struct irq_chip dmar_msi_controll .irq_retrigger = irq_chip_retrigger_hierarchy, .irq_compose_msi_msg = iommu_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, - .flags = IRQCHIP_SKIP_SET_WAKE, + .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MSI_EXTID, }; static int dmar_msi_init(struct irq_domain *domain,
include/linux/irq.h +3 -0
@@ -567,6 +567,8 @@ struct irq_chip {* IRQCHIP_SUPPORTS_NMI: Chip can deliver NMIs, only for root irqchips * IRQCHIP_ENABLE_WAKEUP_ON_SUSPEND: Invokes __enable_irq()/__disable_irq() for wake irqs * in the suspend path if they are in disabled state +* IRQCHIP_MSI_EXTID The MSI message created for this chip can +* have an otherwise forbidden extended ID */ enum {IRQCHIP_SET_TYPE_MASKED = (1 <<0 ),
@@ -579,6 +581,7 @@ enum {IRQCHIP_SUPPORTS_LEVEL_MSI = (1 <<7 ),IRQCHIP_SUPPORTS_NMI = (1 <<8 ),IRQCHIP_ENABLE_WAKEUP_ON_SUSPEND = (1 <<9 ), + IRQCHIP_MSI_EXTID = (1 << 10), };#include <linux/irqdesc.h>
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 7 lines · Thomas Gleixner
Quoted 5 lines · Tom Lendacky
On 11/10/20 3:30 PM, David Woodhouse wrote:
[ 15.581115] WARNING: CPU: 6 PID: 1 at arch/x86/kernel/apic/apic.c:2527 __irq_msi_compose_msg+0x9f/0xb0 [ 15.581115] Call Trace: [ 15.581115] irq_msi_update_msg+0x4d/0x80 [ 15.581115] msi_set_affinity+0x160/0x190
Quoted 5 lines
Duh. Yes, that want's some love as well. Delta patch below.
Thanks,
tglx
That fixed it.
Thanks,
Tom
Quoted 55 lines
---
arch/x86/kernel/apic/msi.c +5
-4@@ -24,10 +24,11 @@ struct irq_domain *x86_pci_msi_default_dstatic void irq_msi_update_msg (struct irq_data *irqd,struct irq_cfg *cfg) { +struct irq_chip *chip =irq_data_get_irq_chip (irqd);struct msi_msg msg[2 ] = { [1 ] = { }, }; -__irq_msi_compose_msg (cfg, msg, false); -irq_data_get_irq_chip (irqd)->irq_write_msi_msg (irqd, msg); +__irq_msi_compose_msg (cfg, msg, chip->flags &IRQCHIP_MSI_EXTID ); + chip->irq_write_msi_msg (irqd, msg); } static int@@ -271,7 +272,7 @@ static struct irq_chip iommu_msi_control .irq_retrigger = irq_chip_retrigger_hierarchy, .irq_set_affinity = msi_set_affinity, .irq_compose_msi_msg = iommu_msi_compose_msg, - .flags = IRQCHIP_SKIP_SET_WAKE, + .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MSI_EXTID, };static struct msi_domain_info iommu_msi_domain_info = {@@ -310,7 +311,7 @@ static struct irq_chip dmar_msi_controll .irq_retrigger = irq_chip_retrigger_hierarchy, .irq_compose_msi_msg = iommu_msi_compose_msg, .irq_write_msi_msg = dmar_msi_write_msg, - .flags = IRQCHIP_SKIP_SET_WAKE, + .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MSI_EXTID, }; static int dmar_msi_init(struct irq_domain *domain,include/linux/irq.h +3
-0@@ -567,6 +567,8 @@ struct irq_chip {* IRQCHIP_SUPPORTS_NMI: Chip can deliver NMIs, only for root irqchips * IRQCHIP_ENABLE_WAKEUP_ON_SUSPEND: Invokes __enable_irq()/__disable_irq() for wake irqs * in the suspend path if they are in disabled state +* IRQCHIP_MSI_EXTID The MSI message created for this chip can +* have an otherwise forbidden extended ID */ enum {IRQCHIP_SET_TYPE_MASKED = (1 <<0 ),@@ -579,6 +581,7 @@ enum {IRQCHIP_SUPPORTS_LEVEL_MSI = (1 <<7 ),IRQCHIP_SUPPORTS_NMI = (1 <<8 ),IRQCHIP_ENABLE_WAKEUP_ON_SUSPEND = (1 <<9 ), + IRQCHIP_MSI_EXTID = (1 << 10), };#include <linux/irqdesc.h>
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to Thomas GleixnerQuoted 2 lines · Thomas Gleixner
+ *
IRQCHIP_MSI_EXTIDThe MSI message created for this chip can
+ * have an otherwise forbidden extended ID
If we're going to do that then we could ditch the separate iommu_compose_msi_msg() function too, right?
But actually I'd be more inclined to fix this differently, in a way that doesn't still leave AMD's iommu_init_intcapxt() having to set use irq_set_affinity_notifier() to update its own registers. That's icky.
Given that this is *its* irqdomain in the first place, it should just sit at ->set_affinity() for itself, and call its parent as usual without having to use a notifier.
We should also leave it using the basic PCI MSI support in the case where the IOMMU doesn't have XTSUP support. It doesn't need its own irqdomain for that.
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 11 lines · David Woodhouse
Quoted 2 lines · Thomas Gleixner
+ *
IRQCHIP_MSI_EXTIDThe MSI message created for this chip can
+ * have an otherwise forbidden extended IDIf we're going to do that then we could ditch the separate
iommu_compose_msi_msg()function too, right?But actually I'd be more inclined to fix this differently, in a way that doesn't still leave AMD's
iommu_init_intcapxt()having to set useirq_set_affinity_notifier()to update its own registers. That's icky.
Quoted 7 lines
Given that this is *its* irqdomain in the first place, it should just sit at ->set_affinity() for itself, and call its parent as usual without having to use a notifier.
We should also leave it using the basic PCI MSI support in the case where the IOMMU doesn't have XTSUP support. It doesn't need its own irqdomain for that.
Looking at it now with brain awake, the XTSUP stuff is pretty much the same as DMAR, which I didn't realize yesterday. The affinity notifier muck is not needed when we have a write_msg() function which twiddles the bits into those other locations.
Thanks,
tglx
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 4 lines · Thomas Gleixner
Looking at it now with brain awake, the XTSUP stuff is pretty much the same as DMAR, which I didn't realize yesterday. The affinity notifier muck is not needed when we have a
write_msg()function which twiddles the bits into those other locations.
I kind of hate the fact that it's swizzling those bits through invalid MSI messages, so I did it as its own irqdomain using irq_domain_create_hierarchy() directly instead of msi_create_irq_domain().
Looks something like this, although I believe I'm missing a call to irq_domain_set_info() somewhere which means that the irq_chip 'intcapxt_controller' isn't being used at all...
drivers/iommu/amd/init.c +104 -56
@@ -16,6 +16,7 @@#include <linux/syscore_ops.h> #include <linux/interrupt.h> #include <linux/msi.h> +#include <linux/irq.h> #include <linux/amd-iommu.h> #include <linux/export.h> #include <linux/kmemleak.h>
@@ -1563,8 +1564,7 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)* for MSI address lo/hi/data, we need to check both * EFR[XtSup] and EFR[MsiCapMmioSup] for x2APIC support. */ -if ((h->efr_reg &BIT (IOMMU_EFR_XTSUP_SHIFT )) && - (h->efr_reg &BIT (IOMMU_EFR_MSICAPMMIOSUP_SHIFT ))) +if (h->efr_reg &BIT (IOMMU_EFR_XTSUP_SHIFT )) amd_iommu_xt_mode =IRQ_REMAP_X2APIC_MODE ;break ;default :
@@ -1978,28 +1978,27 @@ union intcapxt { destid_24_31 :8 ; } __attribute__ ((packed)); -/* -* Setup the IntCapXT registers with interrupt routing information -* based on the PCI MSI capability block registers, accessed via -* MMIO MSI address low/hi and MSI data registers. -*/ -static void iommu_update_intcapxt (struct amd_iommu *iommu) +/* We weren't actually doing anything before. Should we? */ +static void intcapxt_unmask_irq (struct irq_data *data) { -struct msi_msg msg; -union intcapxt xt; -u32 destid; +} +static void intcapxt_mask_irq (struct irq_data *data) +{ +} - msg.address_lo =readl (iommu->mmio_base +MMIO_MSI_ADDR_LO_OFFSET ); - msg.address_hi =readl (iommu->mmio_base +MMIO_MSI_ADDR_HI_OFFSET ); - msg.data =readl (iommu->mmio_base +MMIO_MSI_DATA_OFFSET ); - destid =x86_msi_msg_get_destid (&msg,x2apic_enabled ()); +static int intcapxt_irqdomain_activate (struct irq_domain *domain, +struct irq_data *irqd,bool reserve) +{ +struct amd_iommu *iommu = domain->host_data; +struct irq_cfg *cfg =irqd_cfg (irqd); +union intcapxt xt; xt.capxt =0ULL ; - xt.dest_mode_logical = msg.arch_data.dest_mode_logical; - xt.vector = msg.arch_data.vector; - xt.destid_0_23 = destid &GENMASK (23 ,0 ); - xt.destid_24_31 = destid >>24 ; + xt.dest_mode_logical = apic->dest_mode_logical; + xt.vector = cfg->vector; + xt.destid_0_23 = cfg->dest_apicid &GENMASK (23 ,0 ); + xt.destid_24_31 = cfg->dest_apicid >>24 ;/** * Current IOMMU implemtation uses the same IRQ for all
@@ -2008,64 +2007,117 @@ static void iommu_update_intcapxt(struct amd_iommu *iommu)writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_EVT_OFFSET );writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_PPR_OFFSET );writeq (xt.capxt, iommu->mmio_base +MMIO_INTCAPXT_GALOG_OFFSET ); + +return 0 ; } -static void _irq_notifier_notify (struct irq_affinity_notify *notify, -const cpumask_t *mask) +static void intcapxt_irqdomain_deactivate (struct irq_domain *domain, +struct irq_data *irqd) { -struct amd_iommu *iommu; +intcapxt_mask_irq (irqd); +} -for_each_iommu (iommu) { -if (iommu->dev->irq == notify->irq) { -iommu_update_intcapxt (iommu); -break ; - } - } + +static int intcapxt_irqdomain_alloc (struct irq_domain *domain,unsigned int virq, +unsigned int nr_irqs,void *arg) +{ +return irq_domain_alloc_irqs_parent (domain, virq, nr_irqs, arg); } -static void _irq_notifier_release (struct kref *ref) +static void intcapxt_irqdomain_free (struct irq_domain *domain,unsigned int virq, +unsigned int nr_irqs) { +irq_domain_free_irqs_top (domain, virq, nr_irqs); } -static int iommu_init_intcapxt (struct amd_iommu *iommu) +static int intcapxt_set_affinity (struct irq_data *irqd, +const struct cpumask *mask,bool force) { +struct irq_data *parent = irqd->parent_data;int ret; -struct irq_affinity_notify *notify = &iommu->intcapxt_notify; -/** -* IntCapXT requires XTSup=1 and MsiCapMmioSup=1, -* which can be inferred from amd_iommu_xt_mode. -*/ -if (amd_iommu_xt_mode !=IRQ_REMAP_X2APIC_MODE ) -return 0 ; + ret = parent->chip->irq_set_affinity (parent, mask, force); +if (ret <0 || ret ==IRQ_SET_MASK_OK_DONE ) +return ret; -/** -* Also, we need to setup notifier to update the IntCapXT registers -* whenever the irq affinity is changed from user-space. -*/ - notify->irq = iommu->dev->irq; - notify->notify = _irq_notifier_notify, - notify->release = _irq_notifier_release, - ret =irq_set_affinity_notifier (iommu->dev->irq, notify); +return intcapxt_irqdomain_activate (irqd->domain, irqd, false); +} + +static struct irq_chip intcapxt_controller = { + .name ="IOMMU-MSI" , + .irq_unmask = intcapxt_unmask_irq, + .irq_mask = intcapxt_mask_irq, + .irq_ack = irq_chip_ack_parent, + .irq_retrigger = irq_chip_retrigger_hierarchy, + .irq_set_affinity = intcapxt_set_affinity, + .flags =IRQCHIP_SKIP_SET_WAKE , +}; + +static const struct irq_domain_ops intcapxt_domain_ops = { + .alloc = intcapxt_irqdomain_alloc, + .free = intcapxt_irqdomain_free, + .activate = intcapxt_irqdomain_activate, + .deactivate = intcapxt_irqdomain_deactivate, +}; + +static struct irq_domain *iommu_get_irqdomain (struct amd_iommu *iommu) +{ +struct irq_domain *domain =NULL ; +struct fwnode_handle *fn; + + fn =irq_domain_alloc_named_id_fwnode ("AMD-Vi-MSI" , iommu->index); +if (!fn) +return NULL ; + + domain =irq_domain_create_hierarchy (x86_vector_domain,0 ,0 , + fn, &intcapxt_domain_ops, + iommu); +if (!domain) +irq_domain_free_fwnode (fn); + +return domain; +} + +static int iommu_setup_intcapxt (struct amd_iommu *iommu) +{ +struct irq_domain *domain; +int irq, ret; + + domain =iommu_get_irqdomain (iommu); +if (!domain) +return -ENXIO ; + + irq =irq_domain_alloc_irqs (domain,1 ,NUMA_NO_NODE ,NULL ); +if (irq <0 ) { +irq_domain_remove (domain); +return irq; + } + + ret =request_threaded_irq (irq, + amd_iommu_int_handler, + amd_iommu_int_thread, +0 ,"AMD-Vi" , + iommu);if (ret) { -pr_err ("Failed to register irq affinity notifier (devid=%#x, irq %d)\n" , - iommu->devid, iommu->dev->irq); +irq_domain_free_irqs (irq,1 ); +irq_domain_remove (domain);return ret; } -iommu_update_intcapxt (iommu);iommu_feature_enable (iommu,CONTROL_INTCAPXT_EN ); -return ret; +return 0 ; } -static int iommu_init_msi (struct amd_iommu *iommu) +static int iommu_init_irq (struct amd_iommu *iommu) {int ret;if (iommu->int_enabled) goto enable_faults; -if (iommu->dev->msi_cap) +if (amd_iommu_xt_mode ==IRQ_REMAP_X2APIC_MODE ) + ret =iommu_setup_intcapxt (iommu); +else if (iommu->dev->msi_cap) ret =iommu_setup_msi (iommu);else ret = -ENODEV ;
@@ -2074,10 +2126,6 @@ static int iommu_init_msi(struct amd_iommu *iommu)return ret; enable_faults: - ret =iommu_init_intcapxt (iommu); -if (ret) -return ret; -iommu_feature_enable (iommu,CONTROL_EVT_INT_EN );if (iommu->ppr_log !=NULL )
@@ -2700,7 +2748,7 @@ static int amd_iommu_enable_interrupts(void)int ret =0 ;for_each_iommu (iommu) { - ret =iommu_init_msi (iommu); + ret =iommu_init_irq (iommu);if (ret) goto out; }
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 10 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
Looking at it now with brain awake, the XTSUP stuff is pretty much the same as DMAR, which I didn't realize yesterday. The affinity notifier muck is not needed when we have a
write_msg()function which twiddles the bits into those other locations.I kind of hate the fact that it's swizzling those bits through invalid MSI messages, so I did it as its own irqdomain using
irq_domain_create_hierarchy()directly instead ofmsi_create_irq_domain().
Please give this a spin:
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/amdvi
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 7 lines · David Woodhouse
Quoted 6 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
Looking at it now with brain awake, the XTSUP stuff is pretty much the same as DMAR, which I didn't realize yesterday. The affinity notifier muck is not needed when we have a
write_msg()function which twiddles the bits into those other locations.
Quoted 8 lines
Quoted 4 lines
I kind of hate the fact that it's swizzling those bits through invalid MSI messages, so I did it as its own irqdomain using
irq_domain_create_hierarchy()directly instead ofmsi_create_irq_domain().Please give this a spin:
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/amdvi
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Thanks,
Tom
Quoted 1 line
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 13 lines · Tom Lendacky
Quoted 12 lines · David Woodhouse
Quoted 10 lines · David Woodhouse
Quoted 4 lines · Thomas Gleixner
Looking at it now with brain awake, the XTSUP stuff is pretty much the same as DMAR, which I didn't realize yesterday. The affinity notifier muck is not needed when we have a
write_msg()function which twiddles the bits into those other locations.I kind of hate the fact that it's swizzling those bits through invalid MSI messages, so I did it as its own irqdomain using
irq_domain_create_hierarchy()directly instead ofmsi_create_irq_domain().
Quoted 7 lines
Quoted 3 lines
Please give this a spin:
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/amdvi
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Thanks. Exercising it by actually triggering faults and observing the IOMMU interrupt really working would be quite useful — on hardware with both INTCAPXT and the older MSI delivery.
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to Tom LendackyQuoted 3 lines · Tom Lendacky
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Thanks. I see Thomas has taken the first two into the tip.git x86/apic branch already, so we're just looking for an ack on the third. Which is this one...
From 49ee4fa51b8c06d14b7c4c74d15a7d76f865a8ea Mon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Wed, 11 Nov 2020 12:09:01 +0000
Subject: [PATCH] iommu/amd: Fix IOMMU interrupt generation in X2APIC mode
The AMD IOMMU has two modes for generating its own interrupts.
The first is very much based on PCI MSI, and can be configured by Linux precisely that way. But like legacy unmapped PCI MSI it's limited to 8 bits of APIC ID.
The second method does not use PCI MSI at all in hardawre, and instead configures the INTCAPXT registers in the IOMMU directly with the APIC ID and vector.
In the latter case, the IOMMU driver would still use pci_enable_msi(), read back (through MMIO) the MSI message that Linux wrote to the PCI MSI table, then swizzle those bits into the appropriate register.
Historically, this worked because__irq_compose_msi_msg() would silently generate an invalid MSI message with the high bits of the APIC ID in the high bits of the MSI address. That hack was intended only for the Intel IOMMU, and I recently enforced that, introducing a warning in __irq_msi_compose_msg() if it was invoked with an APIC ID above 255.
Fix the AMD IOMMU not to depend on that hack any more, by having its own irqdomain and directly putting the bits from the irq_cfg into the right place in its ->activate() method.
--- arch/x86/include/asm/hw_irq.h | 1 + drivers/iommu/amd/init.c | 190 +++++++++++++++++++++++----------- 2 files changed, 132 insertions(+), 59 deletions(-)
arch/x86/include/asm/hw_irq.h +1 -0
@@ -39,6 +39,7 @@ enum irq_alloc_type {X86_IRQ_ALLOC_TYPE_PCI_MSI ,X86_IRQ_ALLOC_TYPE_PCI_MSIX ,X86_IRQ_ALLOC_TYPE_DMAR , +X86_IRQ_ALLOC_TYPE_AMDVI , X86_IRQ_ALLOC_TYPE_UV, };
drivers/iommu/amd/init.c +131 -59
259 lines not on this page · raw message
Signature · 1 line
2.17.1
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 5 lines · David Woodhouse
Quoted 3 lines · Tom Lendacky
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Quoted 8 lines
Thanks. I see Thomas has taken the first two into the tip.git x86/apic branch already, so we're just looking for an ack on the third. Which is this one...
From
49ee4fa51b8c06d14b7c4c74d15a7d76f865a8eaMon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Wed, 11 Nov 2020 12:09:01 +0000
Subject: [PATCH] iommu/amd: Fix IOMMU interrupt generation in X2APIC mode
Ping?
Suravee Suthikulpanit
reviewerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to David WoodhouseDavid,
Quoted 34 lines · David Woodhouse
Quoted 3 lines · Tom Lendacky
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Thanks. I see Thomas has taken the first two into the tip.git x86/apic branch already, so we're just looking for an ack on the third. Which is this one...
From
49ee4fa51b8c06d14b7c4c74d15a7d76f865a8eaMon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Wed, 11 Nov 2020 12:09:01 +0000
Subject: [PATCH] iommu/amd: Fix IOMMU interrupt generation in X2APIC modeThe AMD IOMMU has two modes for generating its own interrupts.
The first is very much based on PCI MSI, and can be configured by Linux precisely that way. But like legacy unmapped PCI MSI it's limited to 8 bits of APIC ID.
The second method does not use PCI MSI at all in hardawre, and instead configures the INTCAPXT registers in the IOMMU directly with the APIC ID and vector.
In the latter case, the IOMMU driver would still use
pci_enable_msi(), read back (through MMIO) the MSI message that Linux wrote to the PCI MSI table, then swizzle those bits into the appropriate register.Historically, this worked
because__irq_compose_msi_msg()would silently generate an invalid MSI message with the high bits of the APIC ID in the high bits of the MSI address. That hack was intended only for the Intel IOMMU, and I recently enforced that, introducing a warning in__irq_msi_compose_msg()if it was invoked with an APIC ID above 255.
Quoted 6 lines
Fix the AMD IOMMU not to depend on that hack any more, by having its own irqdomain and directly putting the bits from the
irq_cfginto the right place in its ->activate() method.Fixes 47bea873cf80 "x86/msi: Only use high bits of MSI address for DMAR unit")Signed-off-by David Woodhouse <[email protected]>
I'm still working on testing this series using IO_PAGE_FAULT injection to trigger the IOMMU interrupts. I am still debugging some issues, and I'll keep you updated on the findings.
Thanks,
Suravee
Suravee Suthikulpanit
reviewerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
David
Quoted 41 lines · Suravee Suthikulpanit
David,
Quoted 38 lines · David Woodhouse
Quoted 3 lines · Tom Lendacky
I had trouble cloning your tree for some reason, so just took the top three patches and applied them to the tip tree. This all appears to be working. I'll let the IOMMU experts take a closer look (adding Suravee).
Thanks. I see Thomas has taken the first two into the tip.git x86/apic branch already, so we're just looking for an ack on the third. Which is this one...
From
49ee4fa51b8c06d14b7c4c74d15a7d76f865a8eaMon Sep 17 00:00:00 2001
From: David Woodhouse <[email protected]>
Date: Wed, 11 Nov 2020 12:09:01 +0000
Subject: [PATCH] iommu/amd: Fix IOMMU interrupt generation in X2APIC modeThe AMD IOMMU has two modes for generating its own interrupts.
The first is very much based on PCI MSI, and can be configured by Linux precisely that way. But like legacy unmapped PCI MSI it's limited to 8 bits of APIC ID.
The second method does not use PCI MSI at all in hardawre, and instead configures the INTCAPXT registers in the IOMMU directly with the APIC ID and vector.
In the latter case, the IOMMU driver would still use
pci_enable_msi(), read back (through MMIO) the MSI message that Linux wrote to the PCI MSI table, then swizzle those bits into the appropriate register.Historically, this worked
because__irq_compose_msi_msg()would silently generate an invalid MSI message with the high bits of the APIC ID in the high bits of the MSI address. That hack was intended only for the Intel IOMMU, and I recently enforced that, introducing a warning in__irq_msi_compose_msg()if it was invoked with an APIC ID above 255.Fix the AMD IOMMU not to depend on that hack any more, by having its own irqdomain and directly putting the bits from the
irq_cfginto the right place in its ->activate() method.
Quoted 7 lines
Quoted 2 lines
Fixes 47bea873cf80 "x86/msi: Only use high bits of MSI address for DMAR unit")Signed-off-by David Woodhouse <[email protected]>I'm still working on testing this series using
IO_PAGE_FAULTinjection to trigger the IOMMU interrupts. I am still debugging some issues, and I'll keep you updated on the findings.Thanks,
Suravee
I might need your help debugging this issue. I'm seeing the following error:
[ 14.005937] irq 29, desc: 00000000d200500b, depth: 0, count: 0, unhandled: 0 [ 14.006234] ->handle_irq(): 00000000eab4b6eb, handle_bad_irq+0x0/0x230 [ 14.006234] ->irq_data.chip(): 000000001cce6d6b, intcapxt_controller+0x0/0x120 [ 14.006234] ->action(): 0000000083bfd734 [ 14.006234] ->action->handler(): 0000000094806345, amd_iommu_int_handler+0x0/0x10 [ 14.006234] unexpected IRQ trap at vector 1d
Do you have any idea what might have gone wrong here?
Thanks,
Suravee
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 4 lines · Suravee Suthikulpanit
I might need your help debugging this issue. I'm seeing the following error:
[ 14.005937] irq 29, desc: 00000000d200500b, depth: 0, count: 0, unhandled: 0 [ 14.006234] ->handle_irq(): 00000000eab4b6eb, handle_bad_irq+0x0/0x230
Where's that line coming from?
Quoted 6 lines
[ 14.006234] ->irq_data.chip(): 000000001cce6d6b, intcapxt_controller+0x0/0x120 [ 14.006234] ->action(): 0000000083bfd734 [ 14.006234] ->action->handler(): 0000000094806345, amd_iommu_int_handler+0x0/0x10 [ 14.006234] unexpected IRQ trap at vector 1dDo you have any idea what might have gone wrong here?
Hm, vector 0x1d is vector 29. It's also IRQ 29. I wonder if that's a coincidence.
Can you make intcapxt_irqdomain_activate() print the 64-bit value that it's writing to the intcapxt registers?
Re: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
In reply to Suravee SuthikulpanitSuravee,
Quoted 4 lines · Suravee Suthikulpanit
On 11/17/20 9:00 AM, Suravee Suthikulpanit wrote:
I might need your help debugging this issue. I'm seeing the following error:
Quoted 8 lines
[ 14.005937] irq 29, desc: 00000000d200500b, depth: 0, count: 0, unhandled: 0 [ 14.006234] ->handle_irq(): 00000000eab4b6eb, handle_bad_irq+0x0/0x230 [ 14.006234] ->irq_data.chip(): 000000001cce6d6b, intcapxt_controller+0x0/0x120 [ 14.006234] ->action(): 0000000083bfd734 [ 14.006234] ->action->handler(): 0000000094806345, amd_iommu_int_handler+0x0/0x10 [ 14.006234] unexpected IRQ trap at vector 1dDo you have any idea what might have gone wrong here?
Yes. This lacks setting up the low level flow handler. Delta patch below.
Thanks,
tglx
---
drivers/iommu/amd/init.c +1 -0
@@ -2033,6 +2033,7 @@ static int intcapxt_irqdomain_alloc(stru irqd->chip = &intcapxt_controller; irqd->chip_data = info->data; +__irq_set_handler (i, handle_edge_irq,0 ,"edge" ); }return ret;
Suravee Suthikulpanit
reviewerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Tglx,
Quoted 27 lines · Thomas Gleixner
Suravee,
Quoted 4 lines · Suravee Suthikulpanit
On 11/17/20 9:00 AM, Suravee Suthikulpanit wrote:
I might need your help debugging this issue. I'm seeing the following error:
Quoted 8 lines
[ 14.005937] irq 29, desc: 00000000d200500b, depth: 0, count: 0, unhandled: 0 [ 14.006234] ->handle_irq(): 00000000eab4b6eb, handle_bad_irq+0x0/0x230 [ 14.006234] ->irq_data.chip(): 000000001cce6d6b, intcapxt_controller+0x0/0x120 [ 14.006234] ->action(): 0000000083bfd734 [ 14.006234] ->action->handler(): 0000000094806345, amd_iommu_int_handler+0x0/0x10 [ 14.006234] unexpected IRQ trap at vector 1dDo you have any idea what might have gone wrong here?
Yes. This lacks setting up the low level flow handler. Delta patch below.
Thanks,
tglx
---
drivers/iommu/amd/init.c static int intcapxt_irqdomain_alloc(stru · line 2039
Quoted 7 lines
drivers/iommu/amd/init.c +1
-0@@ -2033,6 +2033,7 @@ static int intcapxt_irqdomain_alloc(stru irqd->chip = &intcapxt_controller; irqd->chip_data = info->data; +__irq_set_handler (i, handle_edge_irq,0 ,"edge" ); }return ret;
Yes, this fixes the issue. Now I can receive the IOMMU event log interrupts for IO_PAGE_FAULT event, which is triggered using the injection interface via debugfs.
Thanks,
Suravee
David Woodhouse
maintainerRe: [EXTERNAL] [tip: x86/apic] x86/io_apic: Cleanup trigger/polarity helpers
Quoted 3 lines · Suravee Suthikulpanit
Yes, this fixes the issue. Now I can receive the IOMMU event log interrupts for
IO_PAGE_FAULTevent, which is triggered using the injection interface via debugfs.
Thanks, Suravee.
[tip: x86/apic] iommu/amd: Fix IOMMU interrupt generation in X2APIC mode
In reply to David WoodhouseThe following commit has been merged into the x86/apic branch of tip:
The following commit has been merged into the x86/apic branch of tip:
Commit-ID: d1adcfbb520c43c10fc22fcdccdd4204e014fb53 Gitweb: https://git.kernel.org/tip/d1adcfbb520c43c10fc22fcdccdd4204e014fb53 Author: David Woodhouse <[email protected]> AuthorDate: Wed, 11 Nov 2020 12:09:01 Committer: Thomas Gleixner <[email protected]> CommitterDate: Wed, 18 Nov 2020 20:55:59 +01:00
iommu/amd: Fix IOMMU interrupt generation in X2APIC mode
The AMD IOMMU has two modes for generating its own interrupts.
The first is very much based on PCI MSI, and can be configured by Linux precisely that way. But like legacy unmapped PCI MSI it's limited to 8 bits of APIC ID.
The second method does not use PCI MSI at all in hardawre, and instead configures the INTCAPXT registers in the IOMMU directly with the APIC ID and vector.
In the latter case, the IOMMU driver would still use pci_enable_msi(), read back (through MMIO) the MSI message that Linux wrote to the PCI MSI table, then swizzle those bits into the appropriate register.
Historically, this worked because__irq_compose_msi_msg() would silently generate an invalid MSI message with the high bits of the APIC ID in the high bits of the MSI address. That hack was intended only for the Intel IOMMU, and I recently enforced that, introducing a warning in __irq_msi_compose_msg() if it was invoked with an APIC ID above 255.
Fix the AMD IOMMU not to depend on that hack any more, by having its own irqdomain and directly putting the bits from the irq_cfg into the right place in its ->activate() method.
--- arch/x86/include/asm/hw_irq.h | 1 +- drivers/iommu/amd/init.c | 191 ++++++++++++++++++++++----------- 2 files changed, 133 insertions(+), 59 deletions(-)
arch/x86/include/asm/hw_irq.h +1 -0
@@ -39,6 +39,7 @@ enum irq_alloc_type {X86_IRQ_ALLOC_TYPE_PCI_MSI ,X86_IRQ_ALLOC_TYPE_PCI_MSIX ,X86_IRQ_ALLOC_TYPE_DMAR , +X86_IRQ_ALLOC_TYPE_AMDVI , X86_IRQ_ALLOC_TYPE_UV, };
drivers/iommu/amd/init.c +132 -59
260 lines not on this page · raw message
David Woodhouse
maintainerPATCH 1/3 iommu/amd: Don't register interrupt remapping irqdomain when IR is disabled
In reply to Thomas GleixnerFrom: David Woodhouse <[email protected]>
This was potentially allowing I/OAPIC and MSI interrupts to be parented in the IOMMU IR domain even when IR was disabled. Don't do that.
--- drivers/iommu/amd/init.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/iommu/amd/init.c +5 -3
@@ -1601,9 +1601,11 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)if (ret)return ret; - ret =amd_iommu_create_irq_domain (iommu); -if (ret) -return ret; +if (amd_iommu_irq_remap) { + ret =amd_iommu_create_irq_domain (iommu); +if (ret) +return ret; + }/* * Make sure IOMMU is not considered to translate itself. The IVRS
Signature · 1 line
2.26.2