From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pf0-f193.google.com (mail-pf0-f193.google.com [209.85.192.193]) by dpdk.org (Postfix) with ESMTP id 6726429D9 for ; Fri, 14 Jul 2017 12:36:08 +0200 (CEST) Received: by mail-pf0-f193.google.com with SMTP id e199so10570390pfh.0 for ; Fri, 14 Jul 2017 03:36:08 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=fridaylinux-org.20150623.gappssmtp.com; s=20150623; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=p6FINxzXBOfLf7KdeavetezGdBbTDK4tVlicQqJXWnk=; b=0CjbeBHz4qKcIpZxrUyGjyeBHAL4fjPU1G6u2J8SLbgRdckQ3+Ue39Ulgkcw6MTNbd kPX2lAw49MvCrR87eihQtMtby1i6AedRvxV04beqTFBHkMtxo9k4NkuaYXRX6ouuF8gy 2r8SKrByaS5yV4wcJrxlZxjvsgt25Wp+VDKrNBh5BxMTa9wAV68XhGRAuToOHz9qDLge K4pbQbptjUsH1VhxUmcJ0NAr36iD8oPLZAWy0fxJ23qLCljVTYDqNQ+6FEPySiJF1Qnq dqJqdIRU3OGkG9vqTlgsLEHUT7j304VerPYl3+hJVY7kBoGL0vl469Wc5Io9CdcU/JKK Fw/g== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=p6FINxzXBOfLf7KdeavetezGdBbTDK4tVlicQqJXWnk=; b=fuhAbgBFo2G+2oClZfXOnEclYr0Ymn8+8ibe6bpwcxYutudBMgc3Zu50Wt6RCHuKX0 TNCu5D3twFs6y+4joWGeshfeh6jLBrqGGZmqMBL6u3ZeOIDN6osuz+vwm1gjmA5DnIdw 7j0YHuxCaCIg0NJmBoI3XSrD2YsjCOnctpt9DTxPvEJp1eA5cgt0dlGyMtE7oXUOCmsD dz95cBNUzTZvFtOaUsRG8lETZnVc6PaSV0WZclcVQz7dVGpfQkpRwDnw6lOxAgdgz1MU wqip9wZyMqgG8StVWTWQl2fayadWaiicLNX+4y67yslN7ZXCFI+c+J4SvxhYB6RiXH7i B4fQ== X-Gm-Message-State: AIVw113lYwir/MIc0GBGx+/vZTlSZTazZLSwX3xkFRPeHP6vOBNSQ/3H TmyNuCT5kOmS04jx X-Received: by 10.98.83.130 with SMTP id h124mr4432671pfb.13.1500028567681; Fri, 14 Jul 2017 03:36:07 -0700 (PDT) Received: from localhost.localdomain ([180.158.48.171]) by smtp.gmail.com with ESMTPSA id y205sm15273553pfb.124.2017.07.14.03.36.04 (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Fri, 14 Jul 2017 03:36:06 -0700 (PDT) From: Yuanhan Liu To: Tiwei Bie Cc: Bruce Richardson , dpdk stable Date: Fri, 14 Jul 2017 18:33:50 +0800 Message-Id: <1500028450-25989-26-git-send-email-yliu@fridaylinux.org> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1500028450-25989-1-git-send-email-yliu@fridaylinux.org> References: <1500028450-25989-1-git-send-email-yliu@fridaylinux.org> Subject: [dpdk-stable] patch 'contigmem: do not zero pages during each mmap' has been queued to LTS release 16.11.3 X-BeenThere: stable@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches for DPDK stable branches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 14 Jul 2017 10:36:08 -0000 Hi, FYI, your patch has been queued to LTS release 16.11.3 Note it hasn't been pushed to http://dpdk.org/browse/dpdk-stable yet. It will be pushed if I get no objections before 07/19/17. So please shout if anyone has objections. Thanks. --yliu --- >>From 156fe3d7888a680342a6c3345284fa3a4a10b82b Mon Sep 17 00:00:00 2001 From: Tiwei Bie Date: Sun, 4 Jun 2017 13:53:24 +0800 Subject: [PATCH] contigmem: do not zero pages during each mmap [ upstream commit 190ce8645e1ebeb0733f9dbc7fa854b50d1dac2e ] Don't zero the pages during each mmap. Instead, only zero the pages when they are not already mmapped. Otherwise, the multi-process support will be broken, as the pages will be zeroed when secondary processes map the memory. Besides, track the open and mmap operations on the cdev, and prevent the module from being unloaded when it is still in use. Fixes: 82f931805506 ("contigmem: zero all pages during mmap") Signed-off-by: Tiwei Bie Acked-by: Bruce Richardson --- lib/librte_eal/bsdapp/contigmem/contigmem.c | 186 ++++++++++++++++++++++++---- 1 file changed, 160 insertions(+), 26 deletions(-) diff --git a/lib/librte_eal/bsdapp/contigmem/contigmem.c b/lib/librte_eal/bsdapp/contigmem/contigmem.c index 03e3e8d..e8fb908 100644 --- a/lib/librte_eal/bsdapp/contigmem/contigmem.c +++ b/lib/librte_eal/bsdapp/contigmem/contigmem.c @@ -50,24 +50,37 @@ __FBSDID("$FreeBSD$"); #include #include +#include #include #include #include +#include + +struct contigmem_buffer { + void *addr; + int refcnt; + struct mtx mtx; +}; + +struct contigmem_vm_handle { + int buffer_index; +}; static int contigmem_load(void); static int contigmem_unload(void); static int contigmem_physaddr(SYSCTL_HANDLER_ARGS); -static d_mmap_t contigmem_mmap; static d_mmap_single_t contigmem_mmap_single; static d_open_t contigmem_open; +static d_close_t contigmem_close; static int contigmem_num_buffers = RTE_CONTIGMEM_DEFAULT_NUM_BUFS; static int64_t contigmem_buffer_size = RTE_CONTIGMEM_DEFAULT_BUF_SIZE; static eventhandler_tag contigmem_eh_tag; -static void *contigmem_buffers[RTE_CONTIGMEM_MAX_NUM_BUFS]; +static struct contigmem_buffer contigmem_buffers[RTE_CONTIGMEM_MAX_NUM_BUFS]; static struct cdev *contigmem_cdev = NULL; +static int contigmem_refcnt; TUNABLE_INT("hw.contigmem.num_buffers", &contigmem_num_buffers); TUNABLE_QUAD("hw.contigmem.buffer_size", &contigmem_buffer_size); @@ -78,6 +91,8 @@ SYSCTL_INT(_hw_contigmem, OID_AUTO, num_buffers, CTLFLAG_RD, &contigmem_num_buffers, 0, "Number of contigmem buffers allocated"); SYSCTL_QUAD(_hw_contigmem, OID_AUTO, buffer_size, CTLFLAG_RD, &contigmem_buffer_size, 0, "Size of each contiguous buffer"); +SYSCTL_INT(_hw_contigmem, OID_AUTO, num_references, CTLFLAG_RD, + &contigmem_refcnt, 0, "Number of references to contigmem"); static SYSCTL_NODE(_hw_contigmem, OID_AUTO, physaddr, CTLFLAG_RD, 0, "physaddr"); @@ -114,9 +129,10 @@ MODULE_VERSION(contigmem, 1); static struct cdevsw contigmem_ops = { .d_name = "contigmem", .d_version = D_VERSION, - .d_mmap = contigmem_mmap, + .d_flags = D_TRACKCLOSE, .d_mmap_single = contigmem_mmap_single, .d_open = contigmem_open, + .d_close = contigmem_close, }; static int @@ -124,6 +140,7 @@ contigmem_load() { char index_string[8], description[32]; int i, error = 0; + void *addr; if (contigmem_num_buffers > RTE_CONTIGMEM_MAX_NUM_BUFS) { printf("%d buffers requested is greater than %d allowed\n", @@ -141,18 +158,20 @@ contigmem_load() } for (i = 0; i < contigmem_num_buffers; i++) { - contigmem_buffers[i] = - contigmalloc(contigmem_buffer_size, M_CONTIGMEM, M_ZERO, 0, - BUS_SPACE_MAXADDR, contigmem_buffer_size, 0); - - if (contigmem_buffers[i] == NULL) { + addr = contigmalloc(contigmem_buffer_size, M_CONTIGMEM, M_ZERO, + 0, BUS_SPACE_MAXADDR, contigmem_buffer_size, 0); + if (addr == NULL) { printf("contigmalloc failed for buffer %d\n", i); error = ENOMEM; goto error; } - printf("%2u: virt=%p phys=%p\n", i, contigmem_buffers[i], - (void *)pmap_kextract((vm_offset_t)contigmem_buffers[i])); + printf("%2u: virt=%p phys=%p\n", i, addr, + (void *)pmap_kextract((vm_offset_t)addr)); + + mtx_init(&contigmem_buffers[i].mtx, "contigmem", NULL, MTX_DEF); + contigmem_buffers[i].addr = addr; + contigmem_buffers[i].refcnt = 0; snprintf(index_string, sizeof(index_string), "%d", i); snprintf(description, sizeof(description), @@ -170,10 +189,13 @@ contigmem_load() return 0; error: - for (i = 0; i < contigmem_num_buffers; i++) - if (contigmem_buffers[i] != NULL) - contigfree(contigmem_buffers[i], contigmem_buffer_size, - M_CONTIGMEM); + for (i = 0; i < contigmem_num_buffers; i++) { + if (contigmem_buffers[i].addr != NULL) + contigfree(contigmem_buffers[i].addr, + contigmem_buffer_size, M_CONTIGMEM); + if (mtx_initialized(&contigmem_buffers[i].mtx)) + mtx_destroy(&contigmem_buffers[i].mtx); + } return error; } @@ -183,16 +205,22 @@ contigmem_unload() { int i; + if (contigmem_refcnt > 0) + return EBUSY; + if (contigmem_cdev != NULL) destroy_dev(contigmem_cdev); if (contigmem_eh_tag != NULL) EVENTHANDLER_DEREGISTER(process_exit, contigmem_eh_tag); - for (i = 0; i < RTE_CONTIGMEM_MAX_NUM_BUFS; i++) - if (contigmem_buffers[i] != NULL) - contigfree(contigmem_buffers[i], contigmem_buffer_size, - M_CONTIGMEM); + for (i = 0; i < RTE_CONTIGMEM_MAX_NUM_BUFS; i++) { + if (contigmem_buffers[i].addr != NULL) + contigfree(contigmem_buffers[i].addr, + contigmem_buffer_size, M_CONTIGMEM); + if (mtx_initialized(&contigmem_buffers[i].mtx)) + mtx_destroy(&contigmem_buffers[i].mtx); + } return 0; } @@ -203,7 +231,7 @@ contigmem_physaddr(SYSCTL_HANDLER_ARGS) uint64_t physaddr; int index = (int)(uintptr_t)arg1; - physaddr = (uint64_t)vtophys(contigmem_buffers[index]); + physaddr = (uint64_t)vtophys(contigmem_buffers[index].addr); return sysctl_handle_64(oidp, &physaddr, 0, req); } @@ -211,22 +239,121 @@ static int contigmem_open(struct cdev *cdev, int fflags, int devtype, struct thread *td) { + + atomic_add_int(&contigmem_refcnt, 1); + + return 0; +} + +static int +contigmem_close(struct cdev *cdev, int fflags, int devtype, + struct thread *td) +{ + + atomic_subtract_int(&contigmem_refcnt, 1); + return 0; } static int -contigmem_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr, - int prot, vm_memattr_t *memattr) +contigmem_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, + vm_ooffset_t foff, struct ucred *cred, u_short *color) { + struct contigmem_vm_handle *vmh = handle; + struct contigmem_buffer *buf; + + buf = &contigmem_buffers[vmh->buffer_index]; + + atomic_add_int(&contigmem_refcnt, 1); + + mtx_lock(&buf->mtx); + if (buf->refcnt == 0) + memset(buf->addr, 0, contigmem_buffer_size); + buf->refcnt++; + mtx_unlock(&buf->mtx); - *paddr = offset; return 0; } +static void +contigmem_cdev_pager_dtor(void *handle) +{ + struct contigmem_vm_handle *vmh = handle; + struct contigmem_buffer *buf; + + buf = &contigmem_buffers[vmh->buffer_index]; + + mtx_lock(&buf->mtx); + buf->refcnt--; + mtx_unlock(&buf->mtx); + + free(vmh, M_CONTIGMEM); + + atomic_subtract_int(&contigmem_refcnt, 1); +} + +static int +contigmem_cdev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot, + vm_page_t *mres) +{ + vm_paddr_t paddr; + vm_page_t m_paddr, page; + vm_memattr_t memattr, memattr1; + + memattr = object->memattr; + + VM_OBJECT_WUNLOCK(object); + + paddr = offset; + + m_paddr = vm_phys_paddr_to_vm_page(paddr); + if (m_paddr != NULL) { + memattr1 = pmap_page_get_memattr(m_paddr); + if (memattr1 != memattr) + memattr = memattr1; + } + + if (((*mres)->flags & PG_FICTITIOUS) != 0) { + /* + * If the passed in result page is a fake page, update it with + * the new physical address. + */ + page = *mres; + VM_OBJECT_WLOCK(object); + vm_page_updatefake(page, paddr, memattr); + } else { + vm_page_t mret; + /* + * Replace the passed in reqpage page with our own fake page and + * free up the original page. + */ + page = vm_page_getfake(paddr, memattr); + VM_OBJECT_WLOCK(object); + mret = vm_page_replace(page, object, (*mres)->pindex); + KASSERT(mret == *mres, + ("invalid page replacement, old=%p, ret=%p", *mres, mret)); + vm_page_lock(mret); + vm_page_free(mret); + vm_page_unlock(mret); + *mres = page; + } + + page->valid = VM_PAGE_BITS_ALL; + + return VM_PAGER_OK; +} + +static struct cdev_pager_ops contigmem_cdev_pager_ops = { + .cdev_pg_ctor = contigmem_cdev_pager_ctor, + .cdev_pg_dtor = contigmem_cdev_pager_dtor, + .cdev_pg_fault = contigmem_cdev_pager_fault, +}; + static int contigmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t size, struct vm_object **obj, int nprot) { + struct contigmem_vm_handle *vmh; uint64_t buffer_index; /* @@ -238,10 +365,17 @@ contigmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t size, if (buffer_index >= contigmem_num_buffers) return EINVAL; - memset(contigmem_buffers[buffer_index], 0, contigmem_buffer_size); - *offset = (vm_ooffset_t)vtophys(contigmem_buffers[buffer_index]); - *obj = vm_pager_allocate(OBJT_DEVICE, cdev, size, nprot, *offset, - curthread->td_ucred); + if (size > contigmem_buffer_size) + return EINVAL; + + vmh = malloc(sizeof(*vmh), M_CONTIGMEM, M_NOWAIT | M_ZERO); + if (vmh == NULL) + return ENOMEM; + vmh->buffer_index = buffer_index; + + *offset = (vm_ooffset_t)vtophys(contigmem_buffers[buffer_index].addr); + *obj = cdev_pager_allocate(vmh, OBJT_DEVICE, &contigmem_cdev_pager_ops, + size, nprot, *offset, curthread->td_ucred); return 0; } -- 2.7.4