| Message ID | 20250320173931.1583800-4-surenb@google.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Guaranteed CMA | expand |
On Thu, Mar 20, 2025 at 10:39:31AM -0700, Suren Baghdasaryan wrote: > This patch introduces a new "guarantee" property for shared-dma-pool. > With this property, admin can create specific memory pool as > GCMA-based CMA if they care about allocation success rate and latency. > The downside of GCMA is that it can host only clean file-backed pages > since it's using cleancache as its secondary user. > > Signed-off-by: Minchan Kim <minchan@google.com> > Signed-off-by: Suren Baghdasaryan <surenb@google.com> > --- > arch/powerpc/kernel/fadump.c | 2 +- > include/linux/cma.h | 2 +- > kernel/dma/contiguous.c | 11 ++++++++++- > mm/cma.c | 33 ++++++++++++++++++++++++++------- > mm/cma.h | 1 + > mm/cma_sysfs.c | 10 ++++++++++ > 6 files changed, 49 insertions(+), 10 deletions(-) > > diff --git a/arch/powerpc/kernel/fadump.c b/arch/powerpc/kernel/fadump.c > index 4b371c738213..4eb7be0cdcdb 100644 > --- a/arch/powerpc/kernel/fadump.c > +++ b/arch/powerpc/kernel/fadump.c > @@ -111,7 +111,7 @@ void __init fadump_cma_init(void) > return; > } > > - rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma); > + rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma, false); > if (rc) { > pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc); > /* > diff --git a/include/linux/cma.h b/include/linux/cma.h > index 62d9c1cf6326..3207db979e94 100644 > --- a/include/linux/cma.h > +++ b/include/linux/cma.h > @@ -46,7 +46,7 @@ extern int __init cma_declare_contiguous_multi(phys_addr_t size, > extern int cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, > unsigned int order_per_bit, > const char *name, > - struct cma **res_cma); > + struct cma **res_cma, bool gcma); > extern struct page *cma_alloc(struct cma *cma, unsigned long count, unsigned int align, > bool no_warn); > extern bool cma_pages_valid(struct cma *cma, const struct page *pages, unsigned long count); > diff --git a/kernel/dma/contiguous.c b/kernel/dma/contiguous.c > index 055da410ac71..a68b3123438c 100644 > --- a/kernel/dma/contiguous.c > +++ b/kernel/dma/contiguous.c > @@ -459,6 +459,7 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) > unsigned long node = rmem->fdt_node; > bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL); > struct cma *cma; > + bool gcma; > int err; > > if (size_cmdline != -1 && default_cma) { > @@ -476,7 +477,15 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) > return -EINVAL; > } > > - err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma); > + gcma = !!of_get_flat_dt_prop(node, "guarantee", NULL); When this (or if I guess) this goes !RFC, you will need to document this new property that you're adding.
On Fri, Mar 21, 2025 at 7:06 AM Conor Dooley <conor@kernel.org> wrote: > > On Thu, Mar 20, 2025 at 10:39:31AM -0700, Suren Baghdasaryan wrote: > > This patch introduces a new "guarantee" property for shared-dma-pool. > > With this property, admin can create specific memory pool as > > GCMA-based CMA if they care about allocation success rate and latency. > > The downside of GCMA is that it can host only clean file-backed pages > > since it's using cleancache as its secondary user. > > > > Signed-off-by: Minchan Kim <minchan@google.com> > > Signed-off-by: Suren Baghdasaryan <surenb@google.com> > > --- > > arch/powerpc/kernel/fadump.c | 2 +- > > include/linux/cma.h | 2 +- > > kernel/dma/contiguous.c | 11 ++++++++++- > > mm/cma.c | 33 ++++++++++++++++++++++++++------- > > mm/cma.h | 1 + > > mm/cma_sysfs.c | 10 ++++++++++ > > 6 files changed, 49 insertions(+), 10 deletions(-) > > > > diff --git a/arch/powerpc/kernel/fadump.c b/arch/powerpc/kernel/fadump.c > > index 4b371c738213..4eb7be0cdcdb 100644 > > --- a/arch/powerpc/kernel/fadump.c > > +++ b/arch/powerpc/kernel/fadump.c > > @@ -111,7 +111,7 @@ void __init fadump_cma_init(void) > > return; > > } > > > > - rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma); > > + rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma, false); > > if (rc) { > > pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc); > > /* > > diff --git a/include/linux/cma.h b/include/linux/cma.h > > index 62d9c1cf6326..3207db979e94 100644 > > --- a/include/linux/cma.h > > +++ b/include/linux/cma.h > > @@ -46,7 +46,7 @@ extern int __init cma_declare_contiguous_multi(phys_addr_t size, > > extern int cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, > > unsigned int order_per_bit, > > const char *name, > > - struct cma **res_cma); > > + struct cma **res_cma, bool gcma); > > extern struct page *cma_alloc(struct cma *cma, unsigned long count, unsigned int align, > > bool no_warn); > > extern bool cma_pages_valid(struct cma *cma, const struct page *pages, unsigned long count); > > diff --git a/kernel/dma/contiguous.c b/kernel/dma/contiguous.c > > index 055da410ac71..a68b3123438c 100644 > > --- a/kernel/dma/contiguous.c > > +++ b/kernel/dma/contiguous.c > > @@ -459,6 +459,7 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) > > unsigned long node = rmem->fdt_node; > > bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL); > > struct cma *cma; > > + bool gcma; > > int err; > > > > if (size_cmdline != -1 && default_cma) { > > @@ -476,7 +477,15 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) > > return -EINVAL; > > } > > > > - err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma); > > + gcma = !!of_get_flat_dt_prop(node, "guarantee", NULL); > > When this (or if I guess) this goes !RFC, you will need to document this > new property that you're adding. Definitely. I'll document the cleancache and GCMA as well. Thanks!
diff --git a/arch/powerpc/kernel/fadump.c b/arch/powerpc/kernel/fadump.c index 4b371c738213..4eb7be0cdcdb 100644 --- a/arch/powerpc/kernel/fadump.c +++ b/arch/powerpc/kernel/fadump.c @@ -111,7 +111,7 @@ void __init fadump_cma_init(void) return; } - rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma); + rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma, false); if (rc) { pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc); /* diff --git a/include/linux/cma.h b/include/linux/cma.h index 62d9c1cf6326..3207db979e94 100644 --- a/include/linux/cma.h +++ b/include/linux/cma.h @@ -46,7 +46,7 @@ extern int __init cma_declare_contiguous_multi(phys_addr_t size, extern int cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, unsigned int order_per_bit, const char *name, - struct cma **res_cma); + struct cma **res_cma, bool gcma); extern struct page *cma_alloc(struct cma *cma, unsigned long count, unsigned int align, bool no_warn); extern bool cma_pages_valid(struct cma *cma, const struct page *pages, unsigned long count); diff --git a/kernel/dma/contiguous.c b/kernel/dma/contiguous.c index 055da410ac71..a68b3123438c 100644 --- a/kernel/dma/contiguous.c +++ b/kernel/dma/contiguous.c @@ -459,6 +459,7 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) unsigned long node = rmem->fdt_node; bool default_cma = of_get_flat_dt_prop(node, "linux,cma-default", NULL); struct cma *cma; + bool gcma; int err; if (size_cmdline != -1 && default_cma) { @@ -476,7 +477,15 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem) return -EINVAL; } - err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, &cma); + gcma = !!of_get_flat_dt_prop(node, "guarantee", NULL); +#ifndef CONFIG_GCMA + if (gcma) { + pr_err("Reserved memory: unable to setup GCMA region, GCMA is not enabled\n"); + return -EINVAL; + } +#endif + err = cma_init_reserved_mem(rmem->base, rmem->size, 0, rmem->name, + &cma, gcma); if (err) { pr_err("Reserved memory: unable to setup CMA region\n"); return err; diff --git a/mm/cma.c b/mm/cma.c index b06d5fe73399..f12cef849e58 100644 --- a/mm/cma.c +++ b/mm/cma.c @@ -26,6 +26,7 @@ #include <linux/cma.h> #include <linux/highmem.h> #include <linux/io.h> +#include <linux/gcma.h> #include <linux/kmemleak.h> #include <trace/events/cma.h> @@ -165,11 +166,18 @@ static void __init cma_activate_area(struct cma *cma) count = cmr->early_pfn - cmr->base_pfn; bitmap_count = cma_bitmap_pages_to_bits(cma, count); bitmap_set(cmr->bitmap, 0, bitmap_count); + } else { + count = 0; } - for (pfn = cmr->early_pfn; pfn < cmr->base_pfn + cmr->count; - pfn += pageblock_nr_pages) - init_cma_reserved_pageblock(pfn_to_page(pfn)); + if (cma->gcma) { + gcma_register_area(cma->name, cmr->early_pfn, + cma->count - count); + } else { + for (pfn = cmr->early_pfn; pfn < cmr->base_pfn + cmr->count; + pfn += pageblock_nr_pages) + init_cma_reserved_pageblock(pfn_to_page(pfn)); + } } spin_lock_init(&cma->lock); @@ -270,7 +278,7 @@ static void __init cma_drop_area(struct cma *cma) int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, unsigned int order_per_bit, const char *name, - struct cma **res_cma) + struct cma **res_cma, bool gcma) { struct cma *cma; int ret; @@ -301,6 +309,7 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, cma->ranges[0].count = cma->count; cma->nranges = 1; cma->nid = NUMA_NO_NODE; + cma->gcma = gcma; *res_cma = cma; @@ -721,7 +730,8 @@ static int __init __cma_declare_contiguous_nid(phys_addr_t base, base = addr; } - ret = cma_init_reserved_mem(base, size, order_per_bit, name, res_cma); + ret = cma_init_reserved_mem(base, size, order_per_bit, name, res_cma, + false); if (ret) memblock_phys_free(base, size); @@ -815,7 +825,13 @@ static int cma_range_alloc(struct cma *cma, struct cma_memrange *cmr, pfn = cmr->base_pfn + (bitmap_no << cma->order_per_bit); mutex_lock(&cma->alloc_mutex); - ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA, gfp); + if (cma->gcma) { + gcma_alloc_range(pfn, count); + ret = 0; + } else { + ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA, + gfp); + } mutex_unlock(&cma->alloc_mutex); if (ret == 0) { page = pfn_to_page(pfn); @@ -992,7 +1008,10 @@ bool cma_release(struct cma *cma, const struct page *pages, if (r == cma->nranges) return false; - free_contig_range(pfn, count); + if (cma->gcma) + gcma_free_range(pfn, count); + else + free_contig_range(pfn, count); cma_clear_bitmap(cma, cmr, pfn, count); cma_sysfs_account_release_pages(cma, count); trace_cma_release(cma->name, pfn, pages, count); diff --git a/mm/cma.h b/mm/cma.h index 41a3ab0ec3de..c2a5576d7987 100644 --- a/mm/cma.h +++ b/mm/cma.h @@ -47,6 +47,7 @@ struct cma { char name[CMA_MAX_NAME]; int nranges; struct cma_memrange ranges[CMA_MAX_RANGES]; + bool gcma; #ifdef CONFIG_CMA_SYSFS /* the number of CMA page successful allocations */ atomic64_t nr_pages_succeeded; diff --git a/mm/cma_sysfs.c b/mm/cma_sysfs.c index 97acd3e5a6a5..4ecc36270a4d 100644 --- a/mm/cma_sysfs.c +++ b/mm/cma_sysfs.c @@ -80,6 +80,15 @@ static ssize_t available_pages_show(struct kobject *kobj, } CMA_ATTR_RO(available_pages); +static ssize_t gcma_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + struct cma *cma = cma_from_kobj(kobj); + + return sysfs_emit(buf, "%d\n", cma->gcma); +} +CMA_ATTR_RO(gcma); + static void cma_kobj_release(struct kobject *kobj) { struct cma *cma = cma_from_kobj(kobj); @@ -95,6 +104,7 @@ static struct attribute *cma_attrs[] = { &release_pages_success_attr.attr, &total_pages_attr.attr, &available_pages_attr.attr, + &gcma_attr.attr, NULL, }; ATTRIBUTE_GROUPS(cma);