Message ID | 20240102095138.17933-10-carlo.nonato@minervasys.tech (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | Arm cache coloring | expand |
On 02.01.2024 10:51, Carlo Nonato wrote: > This commit adds a new memory page allocator that implements the cache > coloring mechanism. The allocation algorithm enforces equal frequency > distribution of cache partitions, following the coloring configuration of a > domain. This allows an even utilization of cache sets for every domain. > > Pages are stored in a color-indexed array of lists. Those lists are filled > by a simple init function which computes the color of each page. > When a domain requests a page, the allocator extract the page from the list > with the maximum number of free pages between those that the domain can > access, given its coloring configuration. > > The allocator can only handle requests of order-0 pages. This allows for > easier implementation and since cache coloring targets only embedded systems, > it's assumed not to be a major problem. I'm curious about the specific properties of embedded systems that makes the performance implications of deeper page walks less of an issue for them. Nothing is said about address-constrained allocations. Are such entirely of no interest to domains on Arm, not even to Dom0 (e.g. for filling Linux'es swiotlb)? Certainly alloc_color_heap_page() should at least fail when it can't satisfy the passed in memflags. > --- > v5: > - Carlo Nonato as the new author > - the colored allocator balances color usage for each domain and it searches > linearly only in the number of colors (FIXME removed) While this addresses earlier concerns, meanwhile NUMA work has also been progressing. What's the plan of interaction of coloring with it? > --- a/xen/arch/Kconfig > +++ b/xen/arch/Kconfig > @@ -47,3 +47,15 @@ config NR_LLC_COLORS > bound. The runtime value is autocomputed or manually set via cmdline. > The default value corresponds to an 8 MiB 16-ways LLC, which should be > more than what needed in the general case. > + > +config BUDDY_ALLOCATOR_SIZE > + int "Buddy allocator reserved memory size (MiB)" > + default "64" > + depends on LLC_COLORING > + help > + Amount of memory reserved for the buddy allocator to work alongside > + the colored one. The colored allocator is meant as an alternative to > + the buddy allocator because its allocation policy is by definition > + incompatible with the generic one. Since the Xen heap is not colored > + yet, we need to support the coexistence of the two allocators and some > + memory must be left for the buddy one. Imo help text should be about the specific option, not general documentation. How about Amount of memory reserved for the buddy allocator, to serve Xen's heap, to work alongside the colored one. or some such? > --- a/xen/arch/arm/llc-coloring.c > +++ b/xen/arch/arm/llc-coloring.c > @@ -30,6 +30,9 @@ static unsigned int __ro_after_init nr_colors = CONFIG_NR_LLC_COLORS; > static unsigned int __ro_after_init dom0_colors[CONFIG_NR_LLC_COLORS]; > static unsigned int __ro_after_init dom0_num_colors; > > +#define mfn_color_mask (nr_colors - 1) This is used solely ... > +#define mfn_to_color(mfn) (mfn_x(mfn) & mfn_color_mask) ... here, and this one in turn is used solely ... > @@ -312,6 +315,16 @@ int domain_set_llc_colors_from_str(struct domain *d, const char *str) > return domain_check_colors(d); > } > > +unsigned int page_to_llc_color(const struct page_info *pg) > +{ > + return mfn_to_color(page_to_mfn(pg)); > +} ... here. What's the point in having those (private) macros? > @@ -1946,6 +1951,162 @@ static unsigned long avail_heap_pages( > return free_pages; > } > > +/************************* > + * COLORED SIDE-ALLOCATOR > + * > + * Pages are grouped by LLC color in lists which are globally referred to as the > + * color heap. Lists are populated in end_boot_allocator(). > + * After initialization there will be N lists where N is the number of > + * available colors on the platform. > + */ > +static struct page_list_head *__ro_after_init _color_heap; > +static unsigned long *__ro_after_init free_colored_pages; It's "just" two pointers, but still - what use are they when ... > +/* Memory required for buddy allocator to work with colored one */ > +#ifdef CONFIG_LLC_COLORING ... this isn't defined? > +static unsigned long __initdata buddy_alloc_size = > + MB(CONFIG_BUDDY_ALLOCATOR_SIZE); > +size_param("buddy-alloc-size", buddy_alloc_size); > + > +#define domain_num_llc_colors(d) ((d)->num_llc_colors) > +#define domain_llc_color(d, i) ((d)->llc_colors[(i)]) Nit: No need to parenthesize i when used like this. > +#else > +static unsigned long __initdata buddy_alloc_size; > + > +#define domain_num_llc_colors(d) 0 > +#define domain_llc_color(d, i) 0 > +#define page_to_llc_color(p) 0 > +#define get_nr_llc_colors() 0 > +#endif > + > +#define color_heap(color) (&_color_heap[color]) > + > +void free_color_heap_page(struct page_info *pg, bool need_scrub) Likely applicable further down as well - this is dead code when !CONFIG_LLC_COLORING. Besides me, Misra also won't like this. The function also looks to want to be static, at which point DCE would apparently take care of removing it (and others, and then hopefully also the two static variables commented on above). > +struct page_info *alloc_color_heap_page(unsigned int memflags, struct domain *d) I don't think d is written through in the function, so it wants to be pointer-to-const. > +void __init init_color_heap_pages(struct page_info *pg, unsigned long nr_pages) > +{ > + unsigned int i; > + bool need_scrub = (system_state < SYS_STATE_active && Can this part of the condition be false, seeing we're in an __init function? > + opt_bootscrub == BOOTSCRUB_IDLE); > + > + if ( buddy_alloc_size ) > + { > + unsigned long buddy_pages = min(PFN_DOWN(buddy_alloc_size), nr_pages); > + > + init_heap_pages(pg, buddy_pages); > + nr_pages -= buddy_pages; > + buddy_alloc_size -= buddy_pages << PAGE_SHIFT; > + pg += buddy_pages; > + } So whatever is passed into this function first is going to fill the Xen heap, without regard to address. I expect you're sure this won't cause issues on Arm. On x86 certain constraints exist which would require lower address ranges to be preferred. > +void dump_color_heap(void) > +{ > + unsigned int color; > + > + printk("Dumping color heap info\n"); > + for ( color = 0; color < get_nr_llc_colors(); color++ ) > + if ( free_colored_pages[color] > 0 ) > + printk("Color heap[%u]: %lu pages\n", > + color, free_colored_pages[color]); > +} What's a typical range of number of colors on a system? I expect more than 9, but I'm not sure about a reasonable upper bound. For the output to be easy to consume, [%u] may want to become at least [%2u]. Jan
On 09.01.2024 11:28, Jan Beulich wrote: > On 02.01.2024 10:51, Carlo Nonato wrote: >> v5: >> - Carlo Nonato as the new author >> - the colored allocator balances color usage for each domain and it searches >> linearly only in the number of colors (FIXME removed) > > While this addresses earlier concerns, meanwhile NUMA work has also > been progressing. What's the plan of interaction of coloring with it? Thinking of interactions - what about static memory? Is coloring incompatible with that? If so, should the two features be excluded to both be used at the same time? Jan
On Tue, 9 Jan 2024, Jan Beulich wrote: > On 09.01.2024 11:28, Jan Beulich wrote: > > On 02.01.2024 10:51, Carlo Nonato wrote: > >> v5: > >> - Carlo Nonato as the new author > >> - the colored allocator balances color usage for each domain and it searches > >> linearly only in the number of colors (FIXME removed) > > > > While this addresses earlier concerns, meanwhile NUMA work has also > > been progressing. What's the plan of interaction of coloring with it? > > Thinking of interactions - what about static memory? Is coloring incompatible > with that? If so, should the two features be excluded to both be used at the > same time? Yes the two features are incompatible. It makes sense to reject attempts to enable both at the same time.
On Tue, 9 Jan 2024, Jan Beulich wrote: > On 02.01.2024 10:51, Carlo Nonato wrote: > > This commit adds a new memory page allocator that implements the cache > > coloring mechanism. The allocation algorithm enforces equal frequency > > distribution of cache partitions, following the coloring configuration of a > > domain. This allows an even utilization of cache sets for every domain. > > > > Pages are stored in a color-indexed array of lists. Those lists are filled > > by a simple init function which computes the color of each page. > > When a domain requests a page, the allocator extract the page from the list > > with the maximum number of free pages between those that the domain can > > access, given its coloring configuration. > > > > The allocator can only handle requests of order-0 pages. This allows for > > easier implementation and since cache coloring targets only embedded systems, > > it's assumed not to be a major problem. > > I'm curious about the specific properties of embedded systems that makes > the performance implications of deeper page walks less of an issue for > them. I think Carlo meant to say that embedded systems tend to have a smaller amount of RAM (our boards today have 4-8GB of total memory). So higher level allocations (2MB/1GB) might not be possible. Also, domains that care about interrupt latency tend to be RTOSes (e.g. Zephyr, FreeRTOS) and RTOSes are happy to run with less than 1MB of total memory available. This is so true that I vaguely remember hitting a bug in xl/libxl when I tried to create a domain with 128KB of memory. > Nothing is said about address-constrained allocations. Are such entirely > of no interest to domains on Arm, not even to Dom0 (e.g. for filling > Linux'es swiotlb)? Cache coloring is useful if you can use an IOMMU with all the dma-capable devices. If that is not the case, then not even Dom0 would be able to boot with cache coloring enabled (because it wouldn't be 1:1 mapped). On ARM we only support booting Dom0 1:1 mapped, or not-1:1-mapped but relying on the IOMMU. > Certainly alloc_color_heap_page() should at least fail when it can't > satisfy the passed in memflags. [...] > > +void dump_color_heap(void) > > +{ > > + unsigned int color; > > + > > + printk("Dumping color heap info\n"); > > + for ( color = 0; color < get_nr_llc_colors(); color++ ) > > + if ( free_colored_pages[color] > 0 ) > > + printk("Color heap[%u]: %lu pages\n", > > + color, free_colored_pages[color]); > > +} > > What's a typical range of number of colors on a system? I expect more > than 9, but I'm not sure about a reasonable upper bound. For the > output to be easy to consume, [%u] may want to become at least [%2u]. 16 colors on our boards
On 10.01.2024 01:46, Stefano Stabellini wrote: > On Tue, 9 Jan 2024, Jan Beulich wrote: >> On 02.01.2024 10:51, Carlo Nonato wrote: >>> This commit adds a new memory page allocator that implements the cache >>> coloring mechanism. The allocation algorithm enforces equal frequency >>> distribution of cache partitions, following the coloring configuration of a >>> domain. This allows an even utilization of cache sets for every domain. >>> >>> Pages are stored in a color-indexed array of lists. Those lists are filled >>> by a simple init function which computes the color of each page. >>> When a domain requests a page, the allocator extract the page from the list >>> with the maximum number of free pages between those that the domain can >>> access, given its coloring configuration. >>> >>> The allocator can only handle requests of order-0 pages. This allows for >>> easier implementation and since cache coloring targets only embedded systems, >>> it's assumed not to be a major problem. >> >> I'm curious about the specific properties of embedded systems that makes >> the performance implications of deeper page walks less of an issue for >> them. > > I think Carlo meant to say that embedded systems tend to have a smaller > amount of RAM (our boards today have 4-8GB of total memory). So higher > level allocations (2MB/1GB) might not be possible. > > Also, domains that care about interrupt latency tend to be RTOSes (e.g. > Zephyr, FreeRTOS) and RTOSes are happy to run with less than 1MB of > total memory available. This is so true that I vaguely remember hitting > a bug in xl/libxl when I tried to create a domain with 128KB of memory. > > >> Nothing is said about address-constrained allocations. Are such entirely >> of no interest to domains on Arm, not even to Dom0 (e.g. for filling >> Linux'es swiotlb)? > > Cache coloring is useful if you can use an IOMMU with all the > dma-capable devices. If that is not the case, then not even Dom0 would > be able to boot with cache coloring enabled (because it wouldn't be 1:1 > mapped). > > On ARM we only support booting Dom0 1:1 mapped, or not-1:1-mapped but > relying on the IOMMU. So another constraint to be enforced both at the Kconfig level and at runtime? That said, Linux'es swiotlb allocation can't know whether an IOMMU is in use by Xen. If something like that was done in a Dom0, the respective allocations still wouldn't really work correctly (and the kernel may or may not choke on this). Jan
On Wed, 10 Jan 2024, Jan Beulich wrote: > On 10.01.2024 01:46, Stefano Stabellini wrote: > > On Tue, 9 Jan 2024, Jan Beulich wrote: > >> On 02.01.2024 10:51, Carlo Nonato wrote: > >>> This commit adds a new memory page allocator that implements the cache > >>> coloring mechanism. The allocation algorithm enforces equal frequency > >>> distribution of cache partitions, following the coloring configuration of a > >>> domain. This allows an even utilization of cache sets for every domain. > >>> > >>> Pages are stored in a color-indexed array of lists. Those lists are filled > >>> by a simple init function which computes the color of each page. > >>> When a domain requests a page, the allocator extract the page from the list > >>> with the maximum number of free pages between those that the domain can > >>> access, given its coloring configuration. > >>> > >>> The allocator can only handle requests of order-0 pages. This allows for > >>> easier implementation and since cache coloring targets only embedded systems, > >>> it's assumed not to be a major problem. > >> > >> I'm curious about the specific properties of embedded systems that makes > >> the performance implications of deeper page walks less of an issue for > >> them. > > > > I think Carlo meant to say that embedded systems tend to have a smaller > > amount of RAM (our boards today have 4-8GB of total memory). So higher > > level allocations (2MB/1GB) might not be possible. > > > > Also, domains that care about interrupt latency tend to be RTOSes (e.g. > > Zephyr, FreeRTOS) and RTOSes are happy to run with less than 1MB of > > total memory available. This is so true that I vaguely remember hitting > > a bug in xl/libxl when I tried to create a domain with 128KB of memory. > > > > > >> Nothing is said about address-constrained allocations. Are such entirely > >> of no interest to domains on Arm, not even to Dom0 (e.g. for filling > >> Linux'es swiotlb)? > > > > Cache coloring is useful if you can use an IOMMU with all the > > dma-capable devices. If that is not the case, then not even Dom0 would > > be able to boot with cache coloring enabled (because it wouldn't be 1:1 > > mapped). > > > > On ARM we only support booting Dom0 1:1 mapped, or not-1:1-mapped but > > relying on the IOMMU. > > So another constraint to be enforced both at the Kconfig level and at > runtime? Yeah potentially > That said, Linux'es swiotlb allocation can't know whether an > IOMMU is in use by Xen. Well, not exactly but we have XENFEAT_direct_mapped and XENFEAT_not_direct_mapped, that is how normally the kernel knows how to behave > If something like that was done in a Dom0, the > respective allocations still wouldn't really work correctly (and the > kernel may or may not choke on this).
Hi Jan, On Tue, Jan 9, 2024 at 11:28 AM Jan Beulich <jbeulich@suse.com> wrote: > > On 02.01.2024 10:51, Carlo Nonato wrote: > > This commit adds a new memory page allocator that implements the cache > > coloring mechanism. The allocation algorithm enforces equal frequency > > distribution of cache partitions, following the coloring configuration of a > > domain. This allows an even utilization of cache sets for every domain. > > > > Pages are stored in a color-indexed array of lists. Those lists are filled > > by a simple init function which computes the color of each page. > > When a domain requests a page, the allocator extract the page from the list > > with the maximum number of free pages between those that the domain can > > access, given its coloring configuration. > > > > The allocator can only handle requests of order-0 pages. This allows for > > easier implementation and since cache coloring targets only embedded systems, > > it's assumed not to be a major problem. > > I'm curious about the specific properties of embedded systems that makes > the performance implications of deeper page walks less of an issue for > them. > > Nothing is said about address-constrained allocations. Are such entirely > of no interest to domains on Arm, not even to Dom0 (e.g. for filling > Linux'es swiotlb)? Certainly alloc_color_heap_page() should at least > fail when it can't satisfy the passed in memflags. > > > --- > > v5: > > - Carlo Nonato as the new author > > - the colored allocator balances color usage for each domain and it searches > > linearly only in the number of colors (FIXME removed) > > While this addresses earlier concerns, meanwhile NUMA work has also > been progressing. What's the plan of interaction of coloring with it? > > > --- a/xen/arch/Kconfig > > +++ b/xen/arch/Kconfig > > @@ -47,3 +47,15 @@ config NR_LLC_COLORS > > bound. The runtime value is autocomputed or manually set via cmdline. > > The default value corresponds to an 8 MiB 16-ways LLC, which should be > > more than what needed in the general case. > > + > > +config BUDDY_ALLOCATOR_SIZE > > + int "Buddy allocator reserved memory size (MiB)" > > + default "64" > > + depends on LLC_COLORING > > + help > > + Amount of memory reserved for the buddy allocator to work alongside > > + the colored one. The colored allocator is meant as an alternative to > > + the buddy allocator because its allocation policy is by definition > > + incompatible with the generic one. Since the Xen heap is not colored > > + yet, we need to support the coexistence of the two allocators and some > > + memory must be left for the buddy one. > > Imo help text should be about the specific option, not general > documentation. How about > > Amount of memory reserved for the buddy allocator, to serve Xen's > heap, to work alongside the colored one. > > or some such? > > > --- a/xen/arch/arm/llc-coloring.c > > +++ b/xen/arch/arm/llc-coloring.c > > @@ -30,6 +30,9 @@ static unsigned int __ro_after_init nr_colors = CONFIG_NR_LLC_COLORS; > > static unsigned int __ro_after_init dom0_colors[CONFIG_NR_LLC_COLORS]; > > static unsigned int __ro_after_init dom0_num_colors; > > > > +#define mfn_color_mask (nr_colors - 1) > > This is used solely ... > > > +#define mfn_to_color(mfn) (mfn_x(mfn) & mfn_color_mask) > > ... here, and this one in turn is used solely ... > > > @@ -312,6 +315,16 @@ int domain_set_llc_colors_from_str(struct domain *d, const char *str) > > return domain_check_colors(d); > > } > > > > +unsigned int page_to_llc_color(const struct page_info *pg) > > +{ > > + return mfn_to_color(page_to_mfn(pg)); > > +} > > ... here. What's the point in having those (private) macros? They will be used in later patches (#13). > > @@ -1946,6 +1951,162 @@ static unsigned long avail_heap_pages( > > return free_pages; > > } > > > > +/************************* > > + * COLORED SIDE-ALLOCATOR > > + * > > + * Pages are grouped by LLC color in lists which are globally referred to as the > > + * color heap. Lists are populated in end_boot_allocator(). > > + * After initialization there will be N lists where N is the number of > > + * available colors on the platform. > > + */ > > +static struct page_list_head *__ro_after_init _color_heap; > > +static unsigned long *__ro_after_init free_colored_pages; > > It's "just" two pointers, but still - what use are they when ... > > > +/* Memory required for buddy allocator to work with colored one */ > > +#ifdef CONFIG_LLC_COLORING > > ... this isn't defined? > > > +static unsigned long __initdata buddy_alloc_size = > > + MB(CONFIG_BUDDY_ALLOCATOR_SIZE); > > +size_param("buddy-alloc-size", buddy_alloc_size); > > + > > +#define domain_num_llc_colors(d) ((d)->num_llc_colors) > > +#define domain_llc_color(d, i) ((d)->llc_colors[(i)]) > > Nit: No need to parenthesize i when used like this. > > > +#else > > +static unsigned long __initdata buddy_alloc_size; > > + > > +#define domain_num_llc_colors(d) 0 > > +#define domain_llc_color(d, i) 0 > > +#define page_to_llc_color(p) 0 > > +#define get_nr_llc_colors() 0 > > +#endif > > + > > +#define color_heap(color) (&_color_heap[color]) > > + > > +void free_color_heap_page(struct page_info *pg, bool need_scrub) > > Likely applicable further down as well - this is dead code when > !CONFIG_LLC_COLORING. Besides me, Misra also won't like this. The > function also looks to want to be static, at which point DCE would > apparently take care of removing it (and others, and then hopefully > also the two static variables commented on above). > > > +struct page_info *alloc_color_heap_page(unsigned int memflags, struct domain *d) > > I don't think d is written through in the function, so it wants to > be pointer-to-const. > > > +void __init init_color_heap_pages(struct page_info *pg, unsigned long nr_pages) > > +{ > > + unsigned int i; > > + bool need_scrub = (system_state < SYS_STATE_active && > > Can this part of the condition be false, seeing we're in an __init > function? Nope. I'll drop it. > > + opt_bootscrub == BOOTSCRUB_IDLE); > > + > > + if ( buddy_alloc_size ) > > + { > > + unsigned long buddy_pages = min(PFN_DOWN(buddy_alloc_size), nr_pages); > > + > > + init_heap_pages(pg, buddy_pages); > > + nr_pages -= buddy_pages; > > + buddy_alloc_size -= buddy_pages << PAGE_SHIFT; > > + pg += buddy_pages; > > + } > > So whatever is passed into this function first is going to fill the > Xen heap, without regard to address. I expect you're sure this won't > cause issues on Arm. On x86 certain constraints exist which would > require lower address ranges to be preferred. > > > +void dump_color_heap(void) > > +{ > > + unsigned int color; > > + > > + printk("Dumping color heap info\n"); > > + for ( color = 0; color < get_nr_llc_colors(); color++ ) > > + if ( free_colored_pages[color] > 0 ) > > + printk("Color heap[%u]: %lu pages\n", > > + color, free_colored_pages[color]); > > +} > > What's a typical range of number of colors on a system? I expect more > than 9, but I'm not sure about a reasonable upper bound. For the > output to be easy to consume, [%u] may want to become at least [%2u]. 16 or 32 colors are pretty typical. In the past we set an upper bound at 128 colors. Thanks. > Jan
Hi Jan, On Tue, Jan 9, 2024 at 11:33 AM Jan Beulich <jbeulich@suse.com> wrote: > > On 09.01.2024 11:28, Jan Beulich wrote: > > On 02.01.2024 10:51, Carlo Nonato wrote: > >> v5: > >> - Carlo Nonato as the new author > >> - the colored allocator balances color usage for each domain and it searches > >> linearly only in the number of colors (FIXME removed) > > > > While this addresses earlier concerns, meanwhile NUMA work has also > > been progressing. What's the plan of interaction of coloring with it? > > Thinking of interactions - what about static memory? Is coloring incompatible > with that? If so, should the two features be excluded to both be used at the > same time? This was done in one of the earlier revisions. Then it was dropped because we thought that it would have been better to have a binary that supported both static memory and coloring. The only checks are done at runtime. > Jan Thanks.
Hi Carlo, On 18/01/2024 14:27, Carlo Nonato wrote: > On Tue, Jan 9, 2024 at 11:33 AM Jan Beulich <jbeulich@suse.com> wrote: >> >> On 09.01.2024 11:28, Jan Beulich wrote: >>> On 02.01.2024 10:51, Carlo Nonato wrote: >>>> v5: >>>> - Carlo Nonato as the new author >>>> - the colored allocator balances color usage for each domain and it searches >>>> linearly only in the number of colors (FIXME removed) >>> >>> While this addresses earlier concerns, meanwhile NUMA work has also >>> been progressing. What's the plan of interaction of coloring with it? >> >> Thinking of interactions - what about static memory? Is coloring incompatible >> with that? If so, should the two features be excluded to both be used at the >> same time? > > This was done in one of the earlier revisions. Then it was dropped because > we thought that it would have been better to have a binary that supported both > static memory and coloring. The only checks are done at runtime. I actually prefer the runtime checks if they are not too complicated. This would allow distro to ship a generic Xen where user can play with various (incompatible) configuration without any rebuild. Cheers,
diff --git a/docs/misc/arm/cache-coloring.rst b/docs/misc/arm/cache-coloring.rst index ae1dd8f4af..c5582340da 100644 --- a/docs/misc/arm/cache-coloring.rst +++ b/docs/misc/arm/cache-coloring.rst @@ -9,6 +9,9 @@ To compile LLC coloring support set ``CONFIG_LLC_COLORING=y``. If needed, change the maximum number of colors with ``CONFIG_NR_LLC_COLORS=<n>``. +If needed, change the buddy allocator reserved size with +``CONFIG_BUDDY_ALLOCATOR_SIZE=<n>``. + Compile Xen and the toolstack and then configure it via `Command line parameters`_. For DomUs follow `DomUs configuration`_. @@ -86,6 +89,8 @@ More specific documentation is available at `docs/misc/xen-command-line.pandoc`. +----------------------+-------------------------------+ | ``dom0-llc-colors`` | Dom0 color configuration | +----------------------+-------------------------------+ +| ``buddy-alloc-size`` | Buddy allocator reserved size | ++----------------------+-------------------------------+ Colors selection format *********************** @@ -160,6 +165,17 @@ DomUs colors can be set via Device Tree, also for Dom0less configurations **Note:** If no color configuration is provided for a domain, the default one, which corresponds to all available colors, is used instead. +Colored allocator and buddy allocator +************************************* + +The colored allocator distributes pages based on color configurations of +domains so that each domains only gets pages of its own colors. +The colored allocator is meant as an alternative to the buddy allocator because +its allocation policy is by definition incompatible with the generic one. Since +the Xen heap is not colored yet, we need to support the coexistence of the two +allocators and some memory must be left for the buddy one. Buddy memory +reservation is configured via Kconfig or via command-line. + Known issues and limitations **************************** @@ -170,3 +186,24 @@ In the domain configuration, "xen,static-mem" allows memory to be statically allocated to the domain. This isn't possibile when LLC coloring is enabled, because that memory can't be guaranteed to use only colors assigned to the domain. + +Cache coloring is intended only for embedded systems +#################################################### + +The current implementation aims to satisfy the need of predictability in +embedded systems with small amount of memory to be managed in a colored way. +Given that, some shortcuts are taken in the development. Expect worse +performances on larger systems. + +Colored allocator can only make use of order-0 pages +#################################################### + +The cache coloring technique relies on memory mappings and on the smallest +amount of memory that can be mapped to achieve the maximum number of colors +(cache partitions) possible. This amount is what is normally called a page and, +in Xen terminology, the order-0 page is the smallest one. The fairly simple +colored allocator currently implemented, makes use only of such pages. +It must be said that a more complex one could, in theory, adopt higher order +pages if the colors selection contained adjacent colors. Two subsequent colors, +for example, can be represented by an order-1 page, four colors correspond to +an order-2 page, etc. diff --git a/docs/misc/xen-command-line.pandoc b/docs/misc/xen-command-line.pandoc index 51f6adf035..163fe7bcb1 100644 --- a/docs/misc/xen-command-line.pandoc +++ b/docs/misc/xen-command-line.pandoc @@ -270,6 +270,20 @@ and not running softirqs. Reduce this if softirqs are not being run frequently enough. Setting this to a high value may cause boot failure, particularly if the NMI watchdog is also enabled. +### buddy-alloc-size (arm64) +> `= <size>` + +> Default: `64M` + +Amount of memory reserved for the buddy allocator when colored allocator is +active. This options is parsed only when LLC coloring support is enabled. +The colored allocator is meant as an alternative to the buddy allocator, +because its allocation policy is by definition incompatible with the generic +one. Since the Xen heap systems is not colored yet, we need to support the +coexistence of the two allocators for now. This parameter, which is optional +and for expert only, it's used to set the amount of memory reserved to the +buddy allocator. + ### cet = List of [ shstk=<bool>, ibt=<bool> ] diff --git a/xen/arch/Kconfig b/xen/arch/Kconfig index aad7e9da38..f4c1406c02 100644 --- a/xen/arch/Kconfig +++ b/xen/arch/Kconfig @@ -47,3 +47,15 @@ config NR_LLC_COLORS bound. The runtime value is autocomputed or manually set via cmdline. The default value corresponds to an 8 MiB 16-ways LLC, which should be more than what needed in the general case. + +config BUDDY_ALLOCATOR_SIZE + int "Buddy allocator reserved memory size (MiB)" + default "64" + depends on LLC_COLORING + help + Amount of memory reserved for the buddy allocator to work alongside + the colored one. The colored allocator is meant as an alternative to + the buddy allocator because its allocation policy is by definition + incompatible with the generic one. Since the Xen heap is not colored + yet, we need to support the coexistence of the two allocators and some + memory must be left for the buddy one. diff --git a/xen/arch/arm/include/asm/mm.h b/xen/arch/arm/include/asm/mm.h index cbcf3bf147..1829c559d6 100644 --- a/xen/arch/arm/include/asm/mm.h +++ b/xen/arch/arm/include/asm/mm.h @@ -145,6 +145,11 @@ struct page_info #else #define PGC_static 0 #endif +#ifdef CONFIG_LLC_COLORING +/* Page is cache colored */ +#define _PGC_colored PG_shift(4) +#define PGC_colored PG_mask(1, 4) +#endif /* ... */ /* Page is broken? */ #define _PGC_broken PG_shift(7) diff --git a/xen/arch/arm/llc-coloring.c b/xen/arch/arm/llc-coloring.c index d3de5f14cb..526129cc43 100644 --- a/xen/arch/arm/llc-coloring.c +++ b/xen/arch/arm/llc-coloring.c @@ -30,6 +30,9 @@ static unsigned int __ro_after_init nr_colors = CONFIG_NR_LLC_COLORS; static unsigned int __ro_after_init dom0_colors[CONFIG_NR_LLC_COLORS]; static unsigned int __ro_after_init dom0_num_colors; +#define mfn_color_mask (nr_colors - 1) +#define mfn_to_color(mfn) (mfn_x(mfn) & mfn_color_mask) + /* * Parse the coloring configuration given in the buf string, following the * syntax below. @@ -312,6 +315,16 @@ int domain_set_llc_colors_from_str(struct domain *d, const char *str) return domain_check_colors(d); } +unsigned int page_to_llc_color(const struct page_info *pg) +{ + return mfn_to_color(page_to_mfn(pg)); +} + +unsigned int get_nr_llc_colors(void) +{ + return nr_colors; +} + /* * Local variables: * mode: C diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c index 3bf3120287..54f52e8677 100644 --- a/xen/common/page_alloc.c +++ b/xen/common/page_alloc.c @@ -126,6 +126,7 @@ #include <xen/irq.h> #include <xen/keyhandler.h> #include <xen/lib.h> +#include <xen/llc-coloring.h> #include <xen/mm.h> #include <xen/nodemask.h> #include <xen/numa.h> @@ -158,7 +159,11 @@ #define PGC_static 0 #endif -#define preserved_flags (PGC_extra | PGC_static) +#ifndef PGC_colored +#define PGC_colored 0 +#endif + +#define preserved_flags (PGC_extra | PGC_static | PGC_colored) #ifndef PGT_TYPE_INFO_INITIALIZER #define PGT_TYPE_INFO_INITIALIZER 0 @@ -1946,6 +1951,162 @@ static unsigned long avail_heap_pages( return free_pages; } +/************************* + * COLORED SIDE-ALLOCATOR + * + * Pages are grouped by LLC color in lists which are globally referred to as the + * color heap. Lists are populated in end_boot_allocator(). + * After initialization there will be N lists where N is the number of + * available colors on the platform. + */ +static struct page_list_head *__ro_after_init _color_heap; +static unsigned long *__ro_after_init free_colored_pages; + +/* Memory required for buddy allocator to work with colored one */ +#ifdef CONFIG_LLC_COLORING +static unsigned long __initdata buddy_alloc_size = + MB(CONFIG_BUDDY_ALLOCATOR_SIZE); +size_param("buddy-alloc-size", buddy_alloc_size); + +#define domain_num_llc_colors(d) ((d)->num_llc_colors) +#define domain_llc_color(d, i) ((d)->llc_colors[(i)]) +#else +static unsigned long __initdata buddy_alloc_size; + +#define domain_num_llc_colors(d) 0 +#define domain_llc_color(d, i) 0 +#define page_to_llc_color(p) 0 +#define get_nr_llc_colors() 0 +#endif + +#define color_heap(color) (&_color_heap[color]) + +void free_color_heap_page(struct page_info *pg, bool need_scrub) +{ + unsigned int color = page_to_llc_color(pg); + struct page_list_head *head = color_heap(color); + + spin_lock(&heap_lock); + + mark_page_free(pg, page_to_mfn(pg)); + + if ( need_scrub ) + { + pg->count_info |= PGC_need_scrub; + poison_one_page(pg); + } + + pg->count_info |= PGC_colored; + free_colored_pages[color]++; + page_list_add(pg, head); + + spin_unlock(&heap_lock); +} + +struct page_info *alloc_color_heap_page(unsigned int memflags, struct domain *d) +{ + struct page_info *pg = NULL; + unsigned int i, color; + unsigned long max = 0; + bool need_tlbflush = false; + uint32_t tlbflush_timestamp = 0; + bool scrub = !(memflags & MEMF_no_scrub); + + spin_lock(&heap_lock); + + for ( i = 0; i < domain_num_llc_colors(d); i++ ) + { + unsigned long free = free_colored_pages[domain_llc_color(d, i)]; + + if ( free > max ) + { + color = domain_llc_color(d, i); + pg = page_list_first(color_heap(color)); + max = free; + } + } + + if ( !pg ) + { + spin_unlock(&heap_lock); + return NULL; + } + + pg->count_info = PGC_state_inuse | PGC_colored | + (pg->count_info & PGC_need_scrub); + free_colored_pages[color]--; + page_list_del(pg, color_heap(color)); + + if ( !(memflags & MEMF_no_tlbflush) ) + accumulate_tlbflush(&need_tlbflush, pg, &tlbflush_timestamp); + + init_free_page_fields(pg); + + spin_unlock(&heap_lock); + + if ( test_and_clear_bit(_PGC_need_scrub, &pg->count_info) && scrub ) + scrub_one_page(pg); + else if ( scrub ) + check_one_page(pg); + + if ( need_tlbflush ) + filtered_flush_tlb_mask(tlbflush_timestamp); + + flush_page_to_ram(mfn_x(page_to_mfn(pg)), + !(memflags & MEMF_no_icache_flush)); + + return pg; +} + +void __init init_color_heap_pages(struct page_info *pg, unsigned long nr_pages) +{ + unsigned int i; + bool need_scrub = (system_state < SYS_STATE_active && + opt_bootscrub == BOOTSCRUB_IDLE); + + if ( buddy_alloc_size ) + { + unsigned long buddy_pages = min(PFN_DOWN(buddy_alloc_size), nr_pages); + + init_heap_pages(pg, buddy_pages); + nr_pages -= buddy_pages; + buddy_alloc_size -= buddy_pages << PAGE_SHIFT; + pg += buddy_pages; + } + + if ( !_color_heap ) + { + unsigned int nr_colors = get_nr_llc_colors(); + + _color_heap = xmalloc_array(struct page_list_head, nr_colors); + free_colored_pages = xzalloc_array(unsigned long, nr_colors); + if ( !_color_heap || !free_colored_pages ) + panic("Can't allocate colored heap. Buddy reserved size is too low"); + + for ( i = 0; i < nr_colors; i++ ) + INIT_PAGE_LIST_HEAD(color_heap(i)); + } + + if ( nr_pages ) + printk(XENLOG_DEBUG + "Init color heap with %lu pages starting from: %#"PRIx64"\n", + nr_pages, page_to_maddr(pg)); + + for ( i = 0; i < nr_pages; i++ ) + free_color_heap_page(&pg[i], need_scrub); +} + +void dump_color_heap(void) +{ + unsigned int color; + + printk("Dumping color heap info\n"); + for ( color = 0; color < get_nr_llc_colors(); color++ ) + if ( free_colored_pages[color] > 0 ) + printk("Color heap[%u]: %lu pages\n", + color, free_colored_pages[color]); +} + void __init end_boot_allocator(void) { unsigned int i; @@ -1965,7 +2126,13 @@ void __init end_boot_allocator(void) for ( i = nr_bootmem_regions; i-- > 0; ) { struct bootmem_region *r = &bootmem_region_list[i]; - if ( r->s < r->e ) + + if ( r->s >= r->e ) + continue; + + if ( llc_coloring_enabled ) + init_color_heap_pages(mfn_to_page(_mfn(r->s)), r->e - r->s); + else init_heap_pages(mfn_to_page(_mfn(r->s)), r->e - r->s); } nr_bootmem_regions = 0; @@ -2461,7 +2628,14 @@ struct page_info *alloc_domheap_pages( if ( memflags & MEMF_no_owner ) memflags |= MEMF_no_refcount; - if ( !dma_bitsize ) + /* Only domains are supported for coloring */ + if ( d && is_domain_llc_colored(d) ) + { + /* Colored allocation must be done on 0 order */ + if ( order || (pg = alloc_color_heap_page(memflags, d)) == NULL ) + return NULL; + } + else if ( !dma_bitsize ) memflags &= ~MEMF_no_dma; else if ( (dma_zone = bits_to_zone(dma_bitsize)) < zone_hi ) pg = alloc_heap_pages(dma_zone + 1, zone_hi, order, memflags, d); @@ -2486,7 +2660,10 @@ struct page_info *alloc_domheap_pages( } if ( assign_page(pg, order, d, memflags) ) { - free_heap_pages(pg, order, memflags & MEMF_no_scrub); + if ( pg->count_info & PGC_colored ) + free_color_heap_page(pg, memflags & MEMF_no_scrub); + else + free_heap_pages(pg, order, memflags & MEMF_no_scrub); return NULL; } } @@ -2569,7 +2746,10 @@ void free_domheap_pages(struct page_info *pg, unsigned int order) scrub = 1; } - free_heap_pages(pg, order, scrub); + if ( pg->count_info & PGC_colored ) + free_color_heap_page(pg, scrub); + else + free_heap_pages(pg, order, scrub); } if ( drop_dom_ref ) @@ -2676,6 +2856,9 @@ static void cf_check dump_heap(unsigned char key) continue; printk("Node %d has %lu unscrubbed pages\n", i, node_need_scrub[i]); } + + if ( llc_coloring_enabled ) + dump_color_heap(); } static __init int cf_check register_heap_trigger(void) diff --git a/xen/include/xen/llc-coloring.h b/xen/include/xen/llc-coloring.h index fa2edc8ad8..7cd481e955 100644 --- a/xen/include/xen/llc-coloring.h +++ b/xen/include/xen/llc-coloring.h @@ -36,6 +36,10 @@ void domain_dump_llc_colors(struct domain *d); int domain_set_llc_colors_domctl(struct domain *d, const struct xen_domctl_set_llc_colors *config); +struct page_info; +unsigned int page_to_llc_color(const struct page_info *pg); +unsigned int get_nr_llc_colors(void); + #endif /* __COLORING_H__ */ /*