irqchip/gic-v3-its: Don't confuse get_vlpi_map() by writing DB config
Subsystems (3)
Zenghui Yu
reviewerPATCH 1/1 irqchip/gic-v3-its: Don't confuse get_vlpi_map() by writing DB config
When we're writing config for the doorbell interrupt, get_vlpi_map() will get confused by doorbell's d->parent_data hack and find the wrong its_dev as chip data and the wrong event.
Fix this issue by making sure no doorbells will be involved before invoking get_vlpi_map(), which restore some of the logic in lpi_write_config().
---
This is based on mainline and can't be directly applied to the current irqchip-next.
drivers/irqchip/irq-gic-v3-its.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/irqchip/irq-gic-v3-its.c +3 -2
@@ -1181,12 +1181,13 @@ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d)static void lpi_write_config (struct irq_data *d,u8 clr,u8 set) { -struct its_vlpi_map *map =get_vlpi_map (d);irq_hw_number_t hwirq;void *va;u8 *cfg; -if (map) { +if (irqd_is_forwarded_to_vcpu (d)) { +struct its_vlpi_map *map =get_vlpi_map (d); + va =page_address (map->vm->vprop_page); hwirq = map->vintid;
Signature · 1 line
2.19.1
Marc Zyngier
maintainerHi Zenghui,
Thanks for this.
Quoted 27 lines · Zenghui Yu
When we're writing config for the doorbell interrupt,
get_vlpi_map()will get confused by doorbell'sd->parent_datahack and find the wrongits_devas chip data and the wrong event.Fix this issue by making sure no doorbells will be involved before invoking
get_vlpi_map(), which restore some of the logic inlpi_write_config().Fixes c1d4d5cd203c ("irqchip/gic-v3-its: Add its_vlpi_map helpers")Signed-off-by Zenghui Yu <[email protected]>---
This is based on mainline and can't be directly applied to the current irqchip-next.
drivers/irqchip/irq-gic-v3-its.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/irqchip/irq-gic-v3-its.c +0
-1@@ -1181,12 +1181,13 @@ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d)static void lpi_write_config (struct irq_data *d,u8 clr,u8 set) { -struct its_vlpi_map *map =get_vlpi_map (d);
drivers/irqchip/irq-gic-v3-its.c static struct its_vlpi_map · line 1193
Quoted 10 lines
drivers/irqchip/irq-gic-v3-its.c +3
-1irq_hw_number_t hwirq;void *va;u8 *cfg; -if (map) { +if (irqd_is_forwarded_to_vcpu (d)) { +struct its_vlpi_map *map =get_vlpi_map (d); + va =page_address (map->vm->vprop_page); hwirq = map->vintid;
Shouldn't we fix get_vlpi_map() instead? Something like (untested):
drivers/irqchip/irq-gic-v3-its.c +6 -5
@@ -1170,13 +1170,14 @@ static void its_send_vclear(struct its_device *dev, u32 event_id)*/ static struct its_vlpi_map *get_vlpi_map (struct irq_data *d) { -struct its_device *its_dev =irq_data_get_irq_chip_data (d); -u32 event =its_get_event_id (d); +if (irqd_is_forwarded_to_vcpu (d)) { +struct its_device *its_dev =irq_data_get_irq_chip_data (d); +u32 event =its_get_event_id (d); -if (!irqd_is_forwarded_to_vcpu (d)) -return NULL ; +return dev_event_to_vlpi_map (its_dev, event); + } -return dev_event_to_vlpi_map (its_dev, event); +return NULL ; }static void lpi_write_config (struct irq_data *d,u8 clr,u8 set)
Or am I missing the actual problem?
Overall, I'm starting to hate that ->parent hack as it's been the source of a number of bugs.
The main issue is that the VPE hierarchy is missing one level (it has no ITS domain, and goes directly from the VPE domain to the low-level GIC domain). It means we end-up special-casing things, and that's never good...
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
Zenghui Yu
reviewerHi Marc,
Quoted 72 lines · Marc Zyngier
Hi Zenghui,
Thanks for this.
Quoted 27 lines · Zenghui Yu
When we're writing config for the doorbell interrupt,
get_vlpi_map()will get confused by doorbell'sd->parent_datahack and find the wrongits_devas chip data and the wrong event.Fix this issue by making sure no doorbells will be involved before invoking
get_vlpi_map(), which restore some of the logic inlpi_write_config().Fixes c1d4d5cd203c ("irqchip/gic-v3-its: Add its_vlpi_map helpers")Signed-off-by Zenghui Yu <[email protected]>---
This is based on mainline and can't be directly applied to the current irqchip-next.
drivers/irqchip/irq-gic-v3-its.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)drivers/irqchip/irq-gic-v3-its.c +0
-1@@ -1181,12 +1181,13 @@ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d) static void lpi_write_config(struct irq_data *d, u8 clr, u8 set) { - struct its_vlpi_map *map = get_vlpi_map(d);drivers/irqchip/irq-gic-v3-its.c static struct its_vlpi_map · line 1193
Quoted 10 lines
drivers/irqchip/irq-gic-v3-its.c +3
-1irq_hw_number_t hwirq; void *va; u8 *cfg; - if (map) { + if (irqd_is_forwarded_to_vcpu(d)) { + struct its_vlpi_map *map = get_vlpi_map(d); + va = page_address(map->vm->vprop_page); hwirq = map->vintid;Shouldn't we fix
get_vlpi_map()instead? Something like (untested):drivers/irqchip/irq-gic-v3-its.c +6
-5@@ -1170,13 +1170,14 @@ static void its_send_vclear(struct its_device*dev, u32 event_id) */ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d) { - struct its_device *its_dev = irq_data_get_irq_chip_data(d); - u32 event = its_get_event_id(d); + if (irqd_is_forwarded_to_vcpu(d)) { + struct its_device *its_dev = irq_data_get_irq_chip_data(d); + u32 event = its_get_event_id(d); - if (!irqd_is_forwarded_to_vcpu(d)) - return NULL; + return dev_event_to_vlpi_map(its_dev, event); + } - return dev_event_to_vlpi_map(its_dev, event); + return NULL; }static void
lpi_write_config(struct irq_data *d, u8 clr, u8 set)
Quoted 1 line
Or am I missing the actual problem?
No. I also thought about fixing the issue by this way and I'm OK with both approaches.
Quoted 3 lines
Overall, I'm starting to hate that ->parent hack as it's been the source of a number of bugs.
(thankfully it's rarely used and we've so far handled them carefully, except the lpi_write_config issue in this patch...)
Quoted 5 lines
The main issue is that the VPE hierarchy is missing one level (it has no ITS domain, and goes directly from the VPE domain to the low-level GIC domain). It means we end-up special-casing things, and that's never good...
Yeah, this may comes from the fact that the per-vPE doorbell is not served by ITS and can be asserted by redistributor directly. And the special doorbell is programmed together with those normal LPI (translated from MSI by ITS).
Thanks,
Zenghui
Marc Zyngier
maintainerQuoted 72 lines · Zenghui Yu
Hi Marc,
Quoted 69 lines · Marc Zyngier
Hi Zenghui,
Thanks for this.
Quoted 27 lines · Zenghui Yu
When we're writing config for the doorbell interrupt,
get_vlpi_map()will get confused by doorbell'sd->parent_datahack and find the wrongits_devas chip data and the wrong event.Fix this issue by making sure no doorbells will be involved before invoking
get_vlpi_map(), which restore some of the logic inlpi_write_config().Fixes c1d4d5cd203c ("irqchip/gic-v3-its: Add its_vlpi_map helpers")Signed-off-by Zenghui Yu <[email protected]>---
This is based on mainline and can't be directly applied to the current irqchip-next.
drivers/irqchip/irq-gic-v3-its.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)drivers/irqchip/irq-gic-v3-its.c +0
-1@@ -1181,12 +1181,13 @@ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d) static void lpi_write_config(struct irq_data *d, u8 clr, u8 set) { - struct its_vlpi_map *map = get_vlpi_map(d);drivers/irqchip/irq-gic-v3-its.c static struct its_vlpi_map · line 1193
Quoted 10 lines
drivers/irqchip/irq-gic-v3-its.c +3
-1irq_hw_number_t hwirq; void *va; u8 *cfg; - if (map) { + if (irqd_is_forwarded_to_vcpu(d)) { + struct its_vlpi_map *map = get_vlpi_map(d); + va = page_address(map->vm->vprop_page); hwirq = map->vintid;Shouldn't we fix
get_vlpi_map()instead? Something like (untested):drivers/irqchip/irq-gic-v3-its.c +6
-5@@ -1170,13 +1170,14 @@ static void its_send_vclear(struct its_device*dev, u32 event_id) */ static struct its_vlpi_map *get_vlpi_map(struct irq_data *d) { - struct its_device *its_dev = irq_data_get_irq_chip_data(d); - u32 event = its_get_event_id(d); + if (irqd_is_forwarded_to_vcpu(d)) { + struct its_device *its_dev = irq_data_get_irq_chip_data(d); + u32 event = its_get_event_id(d); - if (!irqd_is_forwarded_to_vcpu(d)) - return NULL; + return dev_event_to_vlpi_map(its_dev, event); + } - return dev_event_to_vlpi_map(its_dev, event); + return NULL; }
Quoted 7 lines
Quoted 4 lines
static void
lpi_write_config(struct irq_data *d, u8 clr, u8 set)Or am I missing the actual problem?
No. I also thought about fixing the issue by this way and I'm OK with both approaches.
OK, thanks. I've added this to irqchip-next[1], and rebased the v4.1 series on top of it. That way, the fix will trickle down to stable without conflicts.
I've also given it a go on D05 with GICv4 enabled, and nothing caught fire.
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
[tip: irq/core] irqchip/gic-v3-its: Fix get_vlpi_map() breakage with doorbells
In reply to Zenghui YuThe following commit has been merged into the irq/core branch of tip:
The following commit has been merged into the irq/core branch of tip:
Commit-ID: 093bf439fee0d40ade7e309c1288b409cdc3b38f Gitweb: https://git.kernel.org/tip/093bf439fee0d40ade7e309c1288b409cdc3b38f Author: Marc Zyngier <[email protected]> AuthorDate: Wed, 22 Jan 2020 13:53:44 Committer: Marc Zyngier <[email protected]> CommitterDate: Wed, 22 Jan 2020 14:21:07
irqchip/gic-v3-its: Fix get_vlpi_map() breakage with doorbells
When updating an LPI configuration, get_vlpi_map() may be passed a irq_data structure relative to an ITS domain (the normal case) or one that is relative to the core GICv3 domain in the case of a GICv4 doorbell.
In the latter case, special care must be take not to dereference the irq_chip data as an its_dev structure, as that isn't what is stored there. Instead, check *first* whether the IRQ is forwarded to a vcpu, and only then try to obtain the vlpi mapping.
--- drivers/irqchip/irq-gic-v3-its.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/irqchip/irq-gic-v3-its.c +6 -5
@@ -1170,13 +1170,14 @@ static void its_send_vclear(struct its_device *dev, u32 event_id)*/ static struct its_vlpi_map *get_vlpi_map (struct irq_data *d) { -struct its_device *its_dev =irq_data_get_irq_chip_data (d); -u32 event =its_get_event_id (d); +if (irqd_is_forwarded_to_vcpu (d)) { +struct its_device *its_dev =irq_data_get_irq_chip_data (d); +u32 event =its_get_event_id (d); -if (!irqd_is_forwarded_to_vcpu (d)) -return NULL ; +return dev_event_to_vlpi_map (its_dev, event); + } -return dev_event_to_vlpi_map (its_dev, event); +return NULL ; }static void lpi_write_config (struct irq_data *d,u8 clr,u8 set)