From owner-svn-src-all@freebsd.org Wed Oct 21 04:53:36 2015 Return-Path: Delivered-To: svn-src-all@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 2C67AA1AEDB; Wed, 21 Oct 2015 04:53:36 +0000 (UTC) (envelope-from jah@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 01F3C8CE; Wed, 21 Oct 2015 04:53:35 +0000 (UTC) (envelope-from jah@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id t9L4rZTO025020; Wed, 21 Oct 2015 04:53:35 GMT (envelope-from jah@FreeBSD.org) Received: (from jah@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id t9L4rYek025019; Wed, 21 Oct 2015 04:53:34 GMT (envelope-from jah@FreeBSD.org) Message-Id: <201510210453.t9L4rYek025019@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: jah set sender to jah@FreeBSD.org using -f From: "Jason A. Harmening" Date: Wed, 21 Oct 2015 04:53:34 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r289675 - head/sys/arm/arm X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 21 Oct 2015 04:53:36 -0000 Author: jah Date: Wed Oct 21 04:53:34 2015 New Revision: 289675 URL: https://svnweb.freebsd.org/changeset/base/289675 Log: Use pmap_quick* for out-of-context bounce buffers and (limited) cache maintenance of unmapped buffers in armv5 busdma. Tested by: Mattia Rossi Differential Revision: https://reviews.freebsd.org/D3522 Modified: head/sys/arm/arm/busdma_machdep.c Modified: head/sys/arm/arm/busdma_machdep.c ============================================================================== --- head/sys/arm/arm/busdma_machdep.c Wed Oct 21 02:50:22 2015 (r289674) +++ head/sys/arm/arm/busdma_machdep.c Wed Oct 21 04:53:34 2015 (r289675) @@ -124,14 +124,16 @@ struct bounce_page { vm_offset_t vaddr; /* kva of bounce buffer */ bus_addr_t busaddr; /* Physical address */ vm_offset_t datavaddr; /* kva of client data */ - bus_addr_t dataaddr; /* client physical address */ + vm_page_t datapage; /* physical page of client data */ + vm_offset_t dataoffs; /* page offset of client data */ bus_size_t datacount; /* client data count */ STAILQ_ENTRY(bounce_page) links; }; struct sync_list { - vm_offset_t vaddr; /* kva of bounce buffer */ - bus_addr_t busaddr; /* Physical address */ + vm_offset_t vaddr; /* kva of client data */ + vm_page_t pages; /* starting page of client data */ + vm_offset_t dataoffs; /* page offset of client data */ bus_size_t datacount; /* client data count */ }; @@ -197,6 +199,8 @@ static bus_addr_t add_bounce_page(bus_dm vm_offset_t vaddr, bus_addr_t addr, bus_size_t size); static void free_bounce_page(bus_dma_tag_t dmat, struct bounce_page *bpage); +static void bus_dmamap_sync_sl(struct sync_list *sl, bus_dmasync_op_t op, + int bufaligned); /* Default tag, as most drivers provide no parent tag. */ bus_dma_tag_t arm_root_dma_tag; @@ -819,7 +823,8 @@ _bus_dmamap_count_phys(bus_dma_tag_t dma while (buflen != 0) { sgsize = MIN(buflen, dmat->maxsegsz); if (run_filter(dmat, curaddr) != 0) { - sgsize = MIN(sgsize, PAGE_SIZE); + sgsize = MIN(sgsize, + PAGE_SIZE - (curaddr & PAGE_MASK)); map->pagesneeded++; } curaddr += sgsize; @@ -949,8 +954,10 @@ int _bus_dmamap_load_phys(bus_dma_tag_t dmat, bus_dmamap_t map, vm_paddr_t buf, bus_size_t buflen, int flags, bus_dma_segment_t *segs, int *segp) { + struct sync_list *sl; bus_size_t sgsize; bus_addr_t curaddr; + bus_addr_t sl_end = 0; int error; if (segs == NULL) @@ -965,14 +972,35 @@ _bus_dmamap_load_phys(bus_dma_tag_t dmat } } + sl = map->slist + map->sync_count - 1; + while (buflen > 0) { curaddr = buf; sgsize = MIN(buflen, dmat->maxsegsz); if (((dmat->flags & BUS_DMA_COULD_BOUNCE) != 0) && map->pagesneeded != 0 && run_filter(dmat, curaddr)) { - sgsize = MIN(sgsize, PAGE_SIZE); + sgsize = MIN(sgsize, PAGE_SIZE - (curaddr & PAGE_MASK)); curaddr = add_bounce_page(dmat, map, 0, curaddr, sgsize); + } else { + if (map->sync_count > 0) + sl_end = VM_PAGE_TO_PHYS(sl->pages) + + sl->dataoffs + sl->datacount; + + if (map->sync_count == 0 || curaddr != sl_end) { + if (++map->sync_count > dmat->nsegments) + break; + sl++; + sl->vaddr = 0; + sl->datacount = sgsize; + /* + * PHYS_TO_VM_PAGE() will truncate + * unaligned addresses. + */ + sl->pages = PHYS_TO_VM_PAGE(curaddr); + sl->dataoffs = curaddr & PAGE_MASK; + } else + sl->datacount += sgsize; } sgsize = _bus_dmamap_addseg(dmat, map, curaddr, sgsize, segs, segp); @@ -1013,8 +1041,11 @@ _bus_dmamap_load_buffer(bus_dma_tag_t dm { bus_size_t sgsize; bus_addr_t curaddr; + bus_addr_t sl_pend = 0; struct sync_list *sl; + vm_offset_t kvaddr; vm_offset_t vaddr = (vm_offset_t)buf; + vm_offset_t sl_vend = 0; int error = 0; if (segs == NULL) @@ -1033,21 +1064,25 @@ _bus_dmamap_load_buffer(bus_dma_tag_t dm CTR3(KTR_BUSDMA, "lowaddr= %d boundary= %d, " "alignment= %d", dmat->lowaddr, dmat->boundary, dmat->alignment); + sl = map->slist + map->sync_count - 1; + while (buflen > 0) { /* * Get the physical address for this segment. */ if (__predict_true(pmap == kernel_pmap)) { curaddr = pmap_kextract(vaddr); + kvaddr = vaddr; } else { curaddr = pmap_extract(pmap, vaddr); map->flags &= ~DMAMAP_COHERENT; + kvaddr = 0; } /* * Compute the segment size, and adjust counts. */ - sgsize = PAGE_SIZE - ((u_long)curaddr & PAGE_MASK); + sgsize = PAGE_SIZE - (curaddr & PAGE_MASK); if (sgsize > dmat->maxsegsz) sgsize = dmat->maxsegsz; if (buflen < sgsize) @@ -1055,18 +1090,30 @@ _bus_dmamap_load_buffer(bus_dma_tag_t dm if (((dmat->flags & BUS_DMA_COULD_BOUNCE) != 0) && map->pagesneeded != 0 && run_filter(dmat, curaddr)) { - curaddr = add_bounce_page(dmat, map, vaddr, curaddr, + curaddr = add_bounce_page(dmat, map, kvaddr, curaddr, sgsize); } else { - sl = &map->slist[map->sync_count - 1]; + if (map->sync_count > 0) { + sl_pend = VM_PAGE_TO_PHYS(sl->pages) + + sl->dataoffs + sl->datacount; + sl_vend = sl->vaddr + sl->datacount; + } + if (map->sync_count == 0 || - vaddr != sl->vaddr + sl->datacount) { + (kvaddr != 0 && kvaddr != sl_vend) || + (kvaddr == 0 && curaddr != sl_pend)) { + if (++map->sync_count > dmat->nsegments) goto cleanup; sl++; - sl->vaddr = vaddr; + sl->vaddr = kvaddr; sl->datacount = sgsize; - sl->busaddr = curaddr; + /* + * PHYS_TO_VM_PAGE() will truncate + * unaligned addresses. + */ + sl->pages = PHYS_TO_VM_PAGE(curaddr); + sl->dataoffs = curaddr & PAGE_MASK; } else sl->datacount += sgsize; } @@ -1198,18 +1245,66 @@ bus_dmamap_sync_buf(vm_offset_t buf, int } static void +bus_dmamap_sync_sl(struct sync_list *sl, bus_dmasync_op_t op, + int bufaligned) +{ + vm_offset_t tempvaddr; + vm_page_t curpage; + size_t npages; + + if (sl->vaddr != 0) { + bus_dmamap_sync_buf(sl->vaddr, sl->datacount, op, bufaligned); + return; + } + + tempvaddr = 0; + npages = atop(round_page(sl->dataoffs + sl->datacount)); + + for (curpage = sl->pages; curpage != sl->pages + npages; ++curpage) { + /* + * If the page is mapped to some other VA that hasn't + * been supplied to busdma, then pmap_quick_enter_page() + * will find all duplicate mappings and mark them + * uncacheable. + * That will also do any necessary wb/inv. Otherwise, + * if the page is truly unmapped, then we don't actually + * need to do cache maintenance. + * XXX: May overwrite DMA'ed data in the POSTREAD + * case where the CPU has written to a cacheline not + * completely covered by the DMA region. + */ + KASSERT(VM_PAGE_TO_PHYS(curpage) == VM_PAGE_TO_PHYS(sl->pages) + + ptoa(curpage - sl->pages), + ("unexpected vm_page_t phys: 0x%08x != 0x%08x", + VM_PAGE_TO_PHYS(curpage), VM_PAGE_TO_PHYS(sl->pages) + + ptoa(curpage - sl->pages))); + tempvaddr = pmap_quick_enter_page(curpage); + pmap_quick_remove_page(tempvaddr); + } +} + +static void _bus_dmamap_sync_bp(bus_dma_tag_t dmat, bus_dmamap_t map, bus_dmasync_op_t op) { struct bounce_page *bpage; + vm_offset_t datavaddr, tempvaddr; + + if ((op & (BUS_DMASYNC_PREWRITE | BUS_DMASYNC_POSTREAD)) == 0) + return; STAILQ_FOREACH(bpage, &map->bpages, links) { + tempvaddr = 0; + datavaddr = bpage->datavaddr; if (op & BUS_DMASYNC_PREWRITE) { - if (bpage->datavaddr != 0) - bcopy((void *)bpage->datavaddr, - (void *)bpage->vaddr, bpage->datacount); - else - physcopyout(bpage->dataaddr, - (void *)bpage->vaddr,bpage->datacount); + if (datavaddr == 0) { + tempvaddr = + pmap_quick_enter_page(bpage->datapage); + datavaddr = tempvaddr | bpage->dataoffs; + } + bcopy((void *)datavaddr, + (void *)bpage->vaddr, bpage->datacount); + if (tempvaddr != 0) + pmap_quick_remove_page(tempvaddr); cpu_dcache_wb_range(bpage->vaddr, bpage->datacount); cpu_l2cache_wb_range(bpage->vaddr, bpage->datacount); dmat->bounce_zone->total_bounced++; @@ -1217,12 +1312,15 @@ _bus_dmamap_sync_bp(bus_dma_tag_t dmat, if (op & BUS_DMASYNC_POSTREAD) { cpu_dcache_inv_range(bpage->vaddr, bpage->datacount); cpu_l2cache_inv_range(bpage->vaddr, bpage->datacount); - if (bpage->datavaddr != 0) - bcopy((void *)bpage->vaddr, - (void *)bpage->datavaddr, bpage->datacount); - else - physcopyin((void *)bpage->vaddr, - bpage->dataaddr, bpage->datacount); + if (datavaddr == 0) { + tempvaddr = + pmap_quick_enter_page(bpage->datapage); + datavaddr = tempvaddr | bpage->dataoffs; + } + bcopy((void *)bpage->vaddr, + (void *)datavaddr, bpage->datacount); + if (tempvaddr != 0) + pmap_quick_remove_page(tempvaddr); dmat->bounce_zone->total_bounced++; } } @@ -1245,8 +1343,7 @@ _bus_dmamap_sync(bus_dma_tag_t dmat, bus if (map->sync_count) { end = &map->slist[map->sync_count]; for (sl = &map->slist[0]; sl != end; sl++) - bus_dmamap_sync_buf(sl->vaddr, sl->datacount, op, - bufaligned); + bus_dmamap_sync_sl(sl, op, bufaligned); } drain: @@ -1444,7 +1541,9 @@ add_bounce_page(bus_dma_tag_t dmat, bus_ bpage->busaddr |= addr & PAGE_MASK; } bpage->datavaddr = vaddr; - bpage->dataaddr = addr; + /* PHYS_TO_VM_PAGE() will truncate unaligned addresses. */ + bpage->datapage = PHYS_TO_VM_PAGE(addr); + bpage->dataoffs = addr & PAGE_MASK; bpage->datacount = size; STAILQ_INSERT_TAIL(&(map->bpages), bpage, links); return (bpage->busaddr);