Date: Sun, 16 Aug 2015 11:57:17 -0500 From: Alan Cox <alc@rice.edu> To: Konstantin Belousov <kostikbel@gmail.com>, Roger Pau Monn?? <royger@FreeBSD.org> Cc: Jason Harmening <jason.harmening@gmail.com>, "Jason A. Harmening" <jah@FreeBSD.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r286787 - head/sys/x86/x86 Message-ID: <55D0C0ED.807@rice.edu> In-Reply-To: <20150816095004.GX2072@kib.kiev.ua> References: <201508142008.t7EK8Hkt037329@repo.freebsd.org> <55CF390F.5010407@FreeBSD.org> <55CF5B13.1040501@gmail.com> <55D046F5.60601@FreeBSD.org> <20150816090358.GW2072@kib.kiev.ua> <20150816095004.GX2072@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On 08/16/2015 04:50, Konstantin Belousov wrote:
> On Sun, Aug 16, 2015 at 12:03:58PM +0300, Konstantin Belousov wrote:
>> On Sun, Aug 16, 2015 at 10:16:53AM +0200, Roger Pau Monn?? wrote:
>>> pmap_map_io_transient contains some of this logic, but it uses
>>> vmem_alloc (with M_WAITOK) instead of a pcpu pageframe, which defeats
>>> part of the purpose of this change and cannot be used as-is.
>> This logic can be repeated, but it is probably too much for the purpose.
>> It would be enough to have single frame (we cannot reuse CMAP1),
>> protected by a spin mutex. I do not see much sense in providing
>> optimized per-cpu frames for this case.
> Like this. I only compiled the patch.
>
> diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
> index 1e64fc8..9d2b2d9 100644
> --- a/sys/amd64/amd64/pmap.c
> +++ b/sys/amd64/amd64/pmap.c
> @@ -390,6 +390,8 @@ static struct md_page *pv_table;
> */
> pt_entry_t *CMAP1 = 0;
> caddr_t CADDR1 = 0;
> +static vm_offset_t qframe = 0;
> +static struct mtx qframe_mtx;
>
> static int pmap_flags = PMAP_PDE_SUPERPAGE; /* flags for x86 pmaps */
>
> @@ -1031,7 +1033,7 @@ pmap_init(void)
> struct pmap_preinit_mapping *ppim;
> vm_page_t mpte;
> vm_size_t s;
> - int i, pv_npg;
> + int error, i, pv_npg;
>
> /*
> * Initialize the vm page array entries for the kernel pmap's
> @@ -1112,6 +1114,12 @@ pmap_init(void)
> printf("PPIM %u: PA=%#lx, VA=%#lx, size=%#lx, mode=%#x\n", i,
> ppim->pa, ppim->va, ppim->sz, ppim->mode);
> }
> +
> + mtx_init(&qframe_mtx, "qfrmlk", NULL, MTX_SPIN);
> + error = vmem_alloc(kernel_arena, PAGE_SIZE, M_BESTFIT | M_WAITOK,
> + (vmem_addr_t *)&qframe);
> + if (error != 0)
> + panic("qframe allocation failed");
> }
>
> static SYSCTL_NODE(_vm_pmap, OID_AUTO, pde, CTLFLAG_RD, 0,
> @@ -7019,13 +7027,28 @@ pmap_unmap_io_transient(vm_page_t page[], vm_offset_t vaddr[], int count,
> vm_offset_t
> pmap_quick_enter_page(vm_page_t m)
> {
> + vm_paddr_t paddr;
>
> - return (PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)));
> + paddr = VM_PAGE_TO_PHYS(m);
> + if (paddr < dmaplimit)
> + return (PHYS_TO_DMAP(paddr));
> + mtx_lock_spin(&qframe_mtx);
> + KASSERT(*vtopte(qframe) == 0, ("qframe busy"));
> + pte_store(vtopte(qframe), paddr | X86_PG_RW | X86_PG_V | X86_PG_A |
> + X86_PG_M | pmap_cache_bits(kernel_pmap, m->md.pat_mode, 0));
> + invlpg(qframe);
This invlpg is unnecessary.
> + return (qframe);
> }
>
> void
> pmap_quick_remove_page(vm_offset_t addr)
> {
> +
> + if (addr != qframe)
> + return;
> + pte_store(vtopte(qframe), 0);
> + invlpg(qframe);
> + mtx_unlock_spin(&qframe_mtx);
> }
>
> #include "opt_ddb.h"
>
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?55D0C0ED.807>