Page MenuHomeFreeBSD
Differential D42187

arm64: Teach bus_dma on arm64 about NUMA
ClosedPublic
Actions

Authored by andrew on Oct 13 2023, 10:38 AM.
Tags
None
Referenced Files
Unknown Object (File)
Sun, Apr 6, 11:57 PM
Unknown Object (File)
Sat, Apr 5, 1:15 AM
Unknown Object (File)
Mon, Mar 31, 10:48 PM
Unknown Object (File)
Mon, Mar 31, 12:55 PM
Unknown Object (File)
Feb 21 2025, 1:24 PM
Unknown Object (File)
Feb 21 2025, 1:18 PM
Unknown Object (File)
Feb 21 2025, 1:06 PM
Unknown Object (File)
Feb 21 2025, 1:03 PM
View All 47 Files
Subscribers
emaste

Details

Reviewers
jhb
imp
gallatin
markj
manu
Group Reviewers
arm64
Commits
rG63b2b7cca22d: arm64: Teach bus_dma on arm64 about NUMA
rG271e669ed5e5: arm64: Teach bus_dma on arm64 about NUMA
Summary

When allocating memory we should try to allocate from the NUMA node
closest to the device to reduce cross domain memory traffic. Teach the
arm64 bus_dma code to do this.

While here use mallocarray to guard against an unlikely integer
overflow.

Sponsored by: Arm Ltd

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Skipped
Unit
Tests Skipped
Build Status
Buildable 53966
Build 50856: arc lint + arc unit

Event Timeline

andrew created this revision.Oct 13 2023, 10:38 AM
Herald added a reviewer: manu. · View Herald TranscriptOct 13 2023, 10:38 AM
Herald added a subscriber: emaste. · View Herald Transcript
andrew requested review of this revision.Oct 13 2023, 10:38 AM
Harbormaster completed remote builds in B53958: Diff 128710.Oct 13 2023, 10:38 AM
markj added a comment.Oct 13 2023, 2:04 PM

Nothing is setting the domain for tags belonging to the root bus devices. On amd64 this happens in acpi_pcib_acpi_attach() but I believe that implementation isn't used on arm64.

sys/arm64/arm64/busdma_bounce.c
411–413
522–524
589

Why did this change?

andrew added a comment.Oct 13 2023, 3:01 PM

See D42186 for the pci change

sys/arm64/arm64/busdma_bounce.c
589

Because I switched to malloc_domainset_aligned so we could explicitly set the alignment.

I can sync the above comment with x86 as needed.

andrew updated this revision to Diff 128723.Oct 13 2023, 3:12 PM

Remove unneeded casts and update the comment in bounce_bus_dmamem_alloc

Harbormaster completed remote builds in B53966: Diff 128723.Oct 13 2023, 3:13 PM
markj accepted this revision.Oct 13 2023, 3:36 PM
This revision is now accepted and ready to land.Oct 13 2023, 3:36 PM
Closed by commit rG271e669ed5e5: arm64: Teach bus_dma on arm64 about NUMA (authored by andrew). · Explain WhyOct 23 2023, 10:54 AM
This revision was automatically updated to reflect the committed changes.
andrew added a commit: rG271e669ed5e5: arm64: Teach bus_dma on arm64 about NUMA.
andrew added a commit: rG63b2b7cca22d: arm64: Teach bus_dma on arm64 about NUMA.Feb 19 2024, 4:47 PM

Revision Contents
Changeset List

PathSize
sys/
arm64/
arm64/
126 lines
13 lines

Diff 128723

View Options

sys/arm64/arm64/busdma_bounce.c

Loading...
View Options

sys/arm64/arm64/busdma_machdep.c

Show All 38 Lines
#include <sys/kernel.h> #include <sys/kernel.h>
#include <sys/ktr.h> #include <sys/ktr.h>
#include <sys/lock.h> #include <sys/lock.h>
#include <sys/memdesc.h> #include <sys/memdesc.h>
#include <sys/mutex.h> #include <sys/mutex.h>
#include <sys/uio.h> #include <sys/uio.h>
#include <vm/vm.h> #include <vm/vm.h>
#include <vm/vm_extern.h> #include <vm/vm_extern.h>
#include <vm/vm_phys.h>
#include <vm/pmap.h> #include <vm/pmap.h>
#include <machine/bus.h> #include <machine/bus.h>
#include <arm64/include/bus_dma_impl.h> #include <arm64/include/bus_dma_impl.h>
/* /*
* Return true if a match is made. * Return true if a match is made.
* *
▲ Show 20 Lines • Show All 82 Lines • ▼ Show 20 Lines if (common->filter == NULL) {
/* /*
* Short circuit looking at our parent directly * Short circuit looking at our parent directly
* since we have encapsulated all of its information * since we have encapsulated all of its information
*/ */
common->filter = parent->filter; common->filter = parent->filter;
common->filterarg = parent->filterarg; common->filterarg = parent->filterarg;
common->parent = parent->parent; common->parent = parent->parent;
} }
common->domain = parent->domain;
atomic_add_int(&parent->ref_count, 1); atomic_add_int(&parent->ref_count, 1);
} }
common->domain = vm_phys_domain_match(common->domain, 0ul,
common->lowaddr);
*dmat = common; *dmat = common;
return (0); return (0);
} }
/* /*
* Allocate a device specific dma_tag. * Allocate a device specific dma_tag.
*/ */
int int
▲ Show 20 Lines • Show All 49 Lines • ▼ Show 20 Linesbus_dma_tag_destroy(bus_dma_tag_t dmat)
tc = (struct bus_dma_tag_common *)dmat; tc = (struct bus_dma_tag_common *)dmat;
return (tc->impl->tag_destroy(dmat)); return (tc->impl->tag_destroy(dmat));
} }
int int
bus_dma_tag_set_domain(bus_dma_tag_t dmat, int domain) bus_dma_tag_set_domain(bus_dma_tag_t dmat, int domain)
{ {
struct bus_dma_tag_common *tc;
tc = (struct bus_dma_tag_common *)dmat;
domain = vm_phys_domain_match(domain, 0ul, tc->lowaddr);
/* Only call the callback if it changes. */
if (domain == tc->domain)
return (0); return (0);
tc->domain = domain;
return (tc->impl->tag_set_domain(dmat));
} }