Updated: GICv3 emulation by kent-mcleod · Pull Request #58 · seL4/seL4_projects_libs

kent-mcleod · 2022-06-16T13:26:07Z

This PR adds an emulated GIC500 (GICv3) to libsel4vmm as a virtual device.

This branch is based off of an earlier PR, #8, by @pingerino which already included commits pulled together from a couple authors.
This has been tested with the imx8mm-evk platform to run a Linux guest VM with several pass-through peripherals.

Marking the PR as a draft for now, as some minor cleanup is still needed.

Some imx8mm changes: kent-mcleod@bd4b06c

axel-h · 2022-07-04T22:28:22Z

+    uint32_t res4;                /* 0x0054 */
+    uint32_t clrspi_sr;           /* 0x0058 */
+    uint32_t res5[9];             /* [0x005C, 0x0080) */
+    uint32_t irq_group0[CONFIG_MAX_NUM_NODES];          /* [0x080, 0x84) */


This copies the same sloppiness than in the GICv2. We are not adding 1 filed here, but multiple in case there are multiple codes. So the offsets mentioned in the comments are not really right. It does not matter practically, but it might be a bit misleading of one really does offset calculation here - they wont always work. But that's a separate thing to clean up one day

This code was originally written several years ago. I've tried to fold through the updates to the original files as I've rebased it, but some things may have been missed. Thanks for pointing out any changes that were missed.

axel-h · 2022-07-04T22:31:35Z

In some commit message you say

Co-authored-by: Kent McLeod <kent@kry10.com>
Signed-off-by: Kent McLeod <kent@kry10.com>

I wonder, is this a toolin quirk or is Co-Author missing?

kent-mcleod · 2022-07-04T22:37:18Z

In some commit message you say
Co-authored-by: Kent McLeod <kent@kry10.com>
Signed-off-by: Kent McLeod <kent@kry10.com>
I wonder, is this a toolin quirk or is Co-Author missing?

I'm not sure what you mean. Co-authored-by: is how some Git applications can track multiple authors of a commit (https://stackoverflow.com/a/7442255)

axel-h · 2022-07-04T23:31:48Z

Could you split this up and make the fixes a PR to merge first (remove unused function, vgic cleanups - calloc, use correct index in VGIC fault) before the refactoring and adding GICv3?

axel-h · 2022-07-04T23:33:58Z

In some commit message you say
Co-authored-by: Kent McLeod <kent@kry10.com>
Signed-off-by: Kent McLeod <kent@kry10.com>
I wonder, is this a toolin quirk or is Co-Author missing?
I'm not sure what you mean. Co-authored-by: is how some Git applications can track multiple authors of a commit (https://stackoverflow.com/a/7442255)

Ah ok, I did not see that you kept the original commits.

axel-h · 2022-07-04T23:43:40Z

        assert(!"Unable to calloc memory for VGIC");
        return -1;
    }
-    vgic_virq_init(vgic);


Agree, I found the same thing in one of my dev branches.

axel-h · 2022-07-04T23:46:13Z

+#define IRQ_IDX(irq) ((irq) / 32)
+#define IRQ_BIT(irq) (1U << ((irq) % 32))
+
+extern const struct vgic_dist_device dev_vgic_dist;


if we put this here, for consistency we should also include a header that has the type definition.

This doesn't need to be here anyway.

axel-h · 2022-07-04T23:53:41Z

    return is_spi_enabled(gic_dist, irq);
 }

-static inline void set_sgi_ppi_active(struct gic_dist_map *gic_dist, int irq, bool set_active, int vcpu_id)


Yes, this seems unused, we are only accessing the field directly. I was a assuming this was kept in case a VMM needs to be able to manage something here. But activating interrupts unter the hood of a VM seems an odd thing to do.

axel-h · 2022-07-05T12:00:32Z

Could you split this up and make the fixes a PR to merge first (remove unused function, vgic cleanups - calloc, use correct index in VGIC fault) before the refactoring and adding GICv3?

I have created #65 for this.

axel-h · 2022-07-10T21:12:59Z

+        return -1;
+    }
+
+    registers->dist = calloc(1, sizeof(struct gic_dist_map));


I wonder, why do we allocate dist and redist separately and not make this a fixed part of vgic_reg_t, so we only have to allocate things once? That saves us 2 calls to calloc() and there would not be much difference because neither of these structs gets a special 4K alignment anyway.

There are a few subtle differences in the gic distributor between gicv2 and v3. Refactor these into the gicv2 header to prepare for a gicv3 implementation. Co-authored-by: Kent McLeod <kent@kry10.com> Signed-off-by: Kent McLeod <kent@kry10.com>

This commit adds support for emulating the gicv3. Work for this commit was pulled from prototype works by the following authors: Co-authored-by: Kent McLeod <kent@kry10.com> Co-authored-by: Yanyan Shen <yanyan.shen@data61.csiro.au> Co-authored-by: Chris Guikema <chris.guikema@dornerworks.com> Signed-off-by: Kent McLeod <kent@kry10.com>

wom-bat · 2022-08-16T23:31:47Z

How is this going? Are you @kent-mcleod still working on it?

kent-mcleod · 2022-08-18T11:36:40Z

How is this going? Are you @kent-mcleod still working on it?

Most of the refactoring was merged in #68.
The remaining commits requires a little bit more cleanup, but I've not had time to finish it recently unfortunately.

kent-mcleod force-pushed the kent/gicv3 branch from de54c04 to 16fcfb0 Compare June 16, 2022 13:26

kent-mcleod requested a review from axel-h June 28, 2022 00:20

axel-h reviewed Jul 4, 2022

View reviewed changes

axel-h mentioned this pull request Jul 5, 2022

Code cleanup (cherry-picked from GICv3 support branch) #65

Merged

axel-h mentioned this pull request Jul 6, 2022

Reduce dependency on vgic_dist_device #66

Merged

kent-mcleod force-pushed the kent/gicv3 branch 6 times, most recently from 3e88b18 to 94d29d5 Compare July 10, 2022 00:24

kent-mcleod mentioned this pull request Jul 10, 2022

GICv3 emulation preparation PR #68

Merged

kent-mcleod force-pushed the kent/gicv3 branch 2 times, most recently from f785b5f to 615fdab Compare July 10, 2022 02:58

axel-h mentioned this pull request Jul 10, 2022

GICv3 emulation #8

Closed

axel-h reviewed Jul 10, 2022

View reviewed changes

kent-mcleod force-pushed the kent/gicv3 branch 2 times, most recently from f01cdbc to 051079c Compare July 13, 2022 02:45

kent-mcleod and others added 2 commits August 13, 2022 08:01

libsel4vm: prep vdist for gicv3

f9e97e5

There are a few subtle differences in the gic distributor between gicv2 and v3. Refactor these into the gicv2 header to prepare for a gicv3 implementation. Co-authored-by: Kent McLeod <kent@kry10.com> Signed-off-by: Kent McLeod <kent@kry10.com>

kent-mcleod force-pushed the kent/gicv3 branch from 051079c to aad2643 Compare August 18, 2022 11:34

Conversation

kent-mcleod commented Jun 16, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

axel-h Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

kent-mcleod Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

axel-h commented Jul 4, 2022

Uh oh!

kent-mcleod commented Jul 4, 2022

Uh oh!

axel-h commented Jul 4, 2022

Uh oh!

axel-h commented Jul 4, 2022

Uh oh!

axel-h Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

axel-h Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

kent-mcleod Jul 9, 2022

Choose a reason for hiding this comment

Uh oh!

axel-h Jul 4, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

axel-h commented Jul 5, 2022

Uh oh!

axel-h Jul 10, 2022

Choose a reason for hiding this comment

Uh oh!

wom-bat commented Aug 16, 2022

Uh oh!

kent-mcleod commented Aug 18, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

kent-mcleod commented Jun 16, 2022 •

edited

Loading

axel-h Jul 4, 2022 •

edited

Loading