[ofa-general] Re: [PATCH ofed-1.2-rc3 2/4] ehca: backport for rhel-4.5 - mmap functonality

Stefan Roscher ossrosch at linux.vnet.ibm.com
Fri May 11 08:38:12 PDT 2007


Hi Michael,

thanks for reviewing. I make the changes and send the new patch below.

On Friday 11 May 2007 16:35, Michael S. Tsirkin wrote:
> Some questions:

> + MODULE_DESCRIPTION("IBM eServer HCA InfiniBand Device Driver");
> +-MODULE_VERSION("SVNEHCA_0022");
> ++MODULE_VERSION("SVNEHCA_0019");
> + 
> + int ehca_open_aqp1     = 0;
> + int ehca_debug_level   = 0;
> 
> Is this intentional?

no, i just delete this hunk now
> 
> + 	shca->ib_device.detach_mcast	    = ehca_detach_mcast;
> +-	/* shca->ib_device.process_mad	    = ehca_process_mad;     */
> ++	/* shca->ib_device.process_mad	    = ehca_process_mad;	    */
> + 	shca->ib_device.mmap		    = ehca_mmap;
> + 
> + 	return ret;
> 
> Is this really necessary?

no I think there was an unantentional change in spaces, i delete this hunk
> 
> +@@ -811,7 +811,7 @@ int __init ehca_module_init(void)
> + 	int ret;
> + 
> + 	printk(KERN_INFO "eHCA Infiniband Device Driver "
> +-	                 "(Rel.: SVNEHCA_0022)\n");
> ++	                 "(Rel.: SVNEHCA_0019)\n");
> + 	idr_init(&ehca_qp_idr);
> + 	idr_init(&ehca_cq_idr);
> + 	spin_lock_init(&ehca_qp_idr_lock);
> 
> Is this intentional?

no, i delete this hunk too

below is the new patch.

Signed-off-by: Stefan Roscher <stefan.roscher at de.ibm.com>
---
backport_ehca_2_rhel45_umap.patch |  823 ++++++++++++++++++++++++++++++++++++++
1 files changed, 823 insertions(+)


diff -Nurp ofa_kernel-1.2_old/kernel_patches/backport/2.6.9_U5/backport_ehca_2_rhel45_umap.patch ofa_kernel-1.2_new/kernel_patches/backport/2.6.9_U5/backport_ehca_2_rhel45_umap.patch
--- ofa_kernel-1.2_old/kernel_patches/backport/2.6.9_U5/backport_ehca_2_rhel45_umap.patch	1970-01-01 01:00:00.000000000 +0100
+++ ofa_kernel-1.2_new/kernel_patches/backport/2.6.9_U5/backport_ehca_2_rhel45_umap.patch	2007-05-11 19:13:51.000000000 +0200
@@ -0,0 +1,823 @@
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_classes.h ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_classes.h
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_classes.h	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_classes.h	2007-05-04 10:40:06.000000000 +0200
+@@ -126,14 +126,13 @@ struct ehca_qp {
+ 	struct ipz_qp_handle ipz_qp_handle;
+ 	struct ehca_pfqp pf;
+ 	struct ib_qp_init_attr init_attr;
++	u64 uspace_squeue;
++	u64 uspace_rqueue;
++	u64 uspace_fwh;
+ 	struct ehca_cq *send_cq;
+ 	struct ehca_cq *recv_cq;
+ 	unsigned int sqerr_purgeflag;
+ 	struct hlist_node list_entries;
+-	/* mmap counter for resources mapped into user space */
+-	u32 mm_count_squeue;
+-	u32 mm_count_rqueue;
+-	u32 mm_count_galpa;
+ };
+ 
+ /* must be power of 2 */
+@@ -150,6 +149,8 @@ struct ehca_cq {
+ 	struct ipz_cq_handle ipz_cq_handle;
+ 	struct ehca_pfcq pf;
+ 	spinlock_t cb_lock;
++	u64 uspace_queue;
++	u64 uspace_fwh;
+ 	struct hlist_head qp_hashtab[QP_HASHTAB_LEN];
+ 	struct list_head entry;
+ 	u32 nr_callbacks; /* #events assigned to cpu by scaling code */
+@@ -157,9 +158,6 @@ struct ehca_cq {
+ 	wait_queue_head_t wait_completion;
+ 	spinlock_t task_lock;
+ 	u32 ownpid;
+-	/* mmap counter for resources mapped into user space */
+-	u32 mm_count_queue;
+-	u32 mm_count_galpa;
+ };
+ 
+ enum ehca_mr_flag {
+@@ -259,6 +257,20 @@ struct ehca_ucontext {
+ 	struct ib_ucontext ib_ucontext;
+ };
+ 
++struct ehca_module *ehca_module_new(void);
++
++int ehca_module_delete(struct ehca_module *me);
++
++int ehca_eq_ctor(struct ehca_eq *eq);
++
++int ehca_eq_dtor(struct ehca_eq *eq);
++
++struct ehca_shca *ehca_shca_new(void);
++
++int ehca_shca_delete(struct ehca_shca *me);
++
++struct ehca_sport *ehca_sport_new(struct ehca_shca *anchor);
++
+ int ehca_init_pd_cache(void);
+ void ehca_cleanup_pd_cache(void);
+ int ehca_init_cq_cache(void);
+@@ -282,6 +294,7 @@ extern int ehca_use_hp_mr;
+ extern int ehca_scaling_code;
+ 
+ struct ipzu_queue_resp {
++	u64 queue;        /* points to first queue entry */
+ 	u32 qe_size;      /* queue entry size */
+ 	u32 act_nr_of_sg;
+ 	u32 queue_length; /* queue length allocated in bytes */
+@@ -294,6 +307,7 @@ struct ehca_create_cq_resp {
+ 	u32 cq_number;
+ 	u32 token;
+ 	struct ipzu_queue_resp ipz_queue;
++	struct h_galpas galpas;
+ };
+ 
+ struct ehca_create_qp_resp {
+@@ -306,6 +320,7 @@ struct ehca_create_qp_resp {
+ 	u32 dummy; /* padding for 8 byte alignment */
+ 	struct ipzu_queue_resp ipz_squeue;
+ 	struct ipzu_queue_resp ipz_rqueue;
++	struct h_galpas galpas;
+ };
+ 
+ struct ehca_alloc_cq_parms {
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_cq.c ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_cq.c
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_cq.c	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_cq.c	2007-05-04 10:40:06.000000000 +0200
+@@ -268,6 +268,7 @@ struct ib_cq *ehca_create_cq(struct ib_d
+ 	if (context) {
+ 		struct ipz_queue *ipz_queue = &my_cq->ipz_queue;
+ 		struct ehca_create_cq_resp resp;
++		struct vm_area_struct *vma;
+ 		memset(&resp, 0, sizeof(resp));
+ 		resp.cq_number = my_cq->cq_number;
+ 		resp.token = my_cq->token;
+@@ -276,14 +277,40 @@ struct ib_cq *ehca_create_cq(struct ib_d
+ 		resp.ipz_queue.queue_length = ipz_queue->queue_length;
+ 		resp.ipz_queue.pagesize = ipz_queue->pagesize;
+ 		resp.ipz_queue.toggle_state = ipz_queue->toggle_state;
++		ret = ehca_mmap_nopage(((u64)(my_cq->token) << 32) | 0x12000000,
++				       ipz_queue->queue_length,
++				       (void**)&resp.ipz_queue.queue,
++				       &vma);
++		if (ret) {
++			ehca_err(device, "Could not mmap queue pages");
++			cq = ERR_PTR(ret);
++			goto create_cq_exit4;
++		}
++		my_cq->uspace_queue = resp.ipz_queue.queue;
++		resp.galpas = my_cq->galpas;
++		ret = ehca_mmap_register(my_cq->galpas.user.fw_handle,
++					 (void**)&resp.galpas.kernel.fw_handle,
++					 &vma);
++		if (ret) {
++			ehca_err(device, "Could not mmap fw_handle");
++			cq = ERR_PTR(ret);
++			goto create_cq_exit5;
++		}
++		my_cq->uspace_fwh = (u64)resp.galpas.kernel.fw_handle;
+ 		if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
+ 			ehca_err(device, "Copy to udata failed.");
+-			goto create_cq_exit4;
++			goto create_cq_exit6;
+ 		}
+ 	}
+ 
+ 	return cq;
+ 
++create_cq_exit6:
++	ehca_munmap(my_cq->uspace_fwh, EHCA_PAGESIZE);
++
++create_cq_exit5:
++	ehca_munmap(my_cq->uspace_queue, my_cq->ipz_queue.queue_length);
++
+ create_cq_exit4:
+ 	ipz_queue_dtor(&my_cq->ipz_queue);
+ 
+@@ -317,6 +344,7 @@ static int get_cq_nr_events(struct ehca_
+ int ehca_destroy_cq(struct ib_cq *cq)
+ {
+ 	u64 h_ret;
++	int ret;
+ 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
+ 	int cq_num = my_cq->cq_number;
+ 	struct ib_device *device = cq->device;
+@@ -326,20 +354,6 @@ int ehca_destroy_cq(struct ib_cq *cq)
+ 	u32 cur_pid = current->tgid;
+ 	unsigned long flags;
+ 
+-	if (cq->uobject) {
+-		if (my_cq->mm_count_galpa || my_cq->mm_count_queue) {
+-			ehca_err(device, "Resources still referenced in "
+-				 "user space cq_num=%x", my_cq->cq_number);
+-			return -EINVAL;
+-		}
+-		if (my_cq->ownpid != cur_pid) {
+-			ehca_err(device, "Invalid caller pid=%x ownpid=%x "
+-				 "cq_num=%x",
+-				 cur_pid, my_cq->ownpid, my_cq->cq_number);
+-			return -EINVAL;
+-		}
+-	}
+-
+ 	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+ 	while (my_cq->nr_events) {
+ 		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+@@ -351,6 +365,25 @@ int ehca_destroy_cq(struct ib_cq *cq)
+ 	idr_remove(&ehca_cq_idr, my_cq->token);
+ 	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+ 
++	if (my_cq->uspace_queue && my_cq->ownpid != cur_pid) {
++		ehca_err(device, "Invalid caller pid=%x ownpid=%x",
++			 cur_pid, my_cq->ownpid);
++		return -EINVAL;
++	}
++
++	/* un-mmap if vma alloc */
++	if (my_cq->uspace_queue ) {
++		ret = ehca_munmap(my_cq->uspace_queue,
++				  my_cq->ipz_queue.queue_length);
++		if (ret)
++			ehca_err(device, "Could not munmap queue ehca_cq=%p "
++				 "cq_num=%x", my_cq, cq_num);
++		ret = ehca_munmap(my_cq->uspace_fwh, EHCA_PAGESIZE);
++		if (ret)
++			ehca_err(device, "Could not munmap fwh ehca_cq=%p "
++				 "cq_num=%x", my_cq, cq_num);
++	}
++
+ 	h_ret = hipz_h_destroy_cq(adapter_handle, my_cq, 0);
+ 	if (h_ret == H_R_STATE) {
+ 		/* cq in err: read err data and destroy it forcibly */
+@@ -379,7 +412,7 @@ int ehca_resize_cq(struct ib_cq *cq, int
+ 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
+ 	u32 cur_pid = current->tgid;
+ 
+-	if (cq->uobject && my_cq->ownpid != cur_pid) {
++	if (my_cq->uspace_queue && my_cq->ownpid != cur_pid) {
+ 		ehca_err(cq->device, "Invalid caller pid=%x ownpid=%x",
+ 			 cur_pid, my_cq->ownpid);
+ 		return -EINVAL;
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_iverbs.h ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_iverbs.h
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_iverbs.h	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_iverbs.h	2007-04-29 15:10:56.000000000 +0200
+@@ -171,11 +171,19 @@ int ehca_mmap(struct ib_ucontext *contex
+ 
+ void ehca_poll_eqs(unsigned long data);
+ 
++int ehca_mmap_nopage(u64 foffset,u64 length,void **mapped,
++		     struct vm_area_struct **vma);
++
++int ehca_mmap_register(u64 physical,void **mapped,
++		       struct vm_area_struct **vma);
++
++int ehca_munmap(unsigned long addr, size_t len);
++
+ #ifdef CONFIG_PPC_64K_PAGES
+ void *ehca_alloc_fw_ctrlblock(gfp_t flags);
+ void ehca_free_fw_ctrlblock(void *ptr);
+ #else
+-#define ehca_alloc_fw_ctrlblock(flags) ((void*) get_zeroed_page(flags))
++#define ehca_alloc_fw_ctrlblock(flags) ((void *) get_zeroed_page(flags))
+ #define ehca_free_fw_ctrlblock(ptr) free_page((unsigned long)(ptr))
+ #endif
+ 
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_main.c ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_main.c
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_main.c	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_main.c	2007-05-04 10:40:06.000000000 +0200
+@@ -293,7 +293,7 @@ int ehca_init_device(struct ehca_shca *s
+ 	strlcpy(shca->ib_device.name, "ehca%d", IB_DEVICE_NAME_MAX);
+ 	shca->ib_device.owner               = THIS_MODULE;
+ 
+-	shca->ib_device.uverbs_abi_ver	    = 6;
++	shca->ib_device.uverbs_abi_ver	    = 5;
+ 	shca->ib_device.uverbs_cmd_mask	    =
+ 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
+ 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_qp.c ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_qp.c
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_qp.c	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_qp.c	2007-04-29 15:10:56.000000000 +0200
+@@ -637,6 +637,7 @@ struct ib_qp *ehca_create_qp(struct ib_p
+ 		struct ipz_queue *ipz_rqueue = &my_qp->ipz_rqueue;
+ 		struct ipz_queue *ipz_squeue = &my_qp->ipz_squeue;
+ 		struct ehca_create_qp_resp resp;
++		struct vm_area_struct * vma;
+ 		memset(&resp, 0, sizeof(resp));
+ 
+ 		resp.qp_num = my_qp->real_qp_num;
+@@ -650,21 +651,59 @@ struct ib_qp *ehca_create_qp(struct ib_p
+ 		resp.ipz_rqueue.queue_length = ipz_rqueue->queue_length;
+ 		resp.ipz_rqueue.pagesize = ipz_rqueue->pagesize;
+ 		resp.ipz_rqueue.toggle_state = ipz_rqueue->toggle_state;
++		ret = ehca_mmap_nopage(((u64)(my_qp->token) << 32) | 0x22000000,
++				       ipz_rqueue->queue_length,
++				       (void**)&resp.ipz_rqueue.queue,
++				       &vma);
++		if (ret) {
++			ehca_err(pd->device, "Could not mmap rqueue pages");
++			goto create_qp_exit3;
++		}
++		my_qp->uspace_rqueue = resp.ipz_rqueue.queue;
+ 		/* squeue properties */
+ 		resp.ipz_squeue.qe_size = ipz_squeue->qe_size;
+ 		resp.ipz_squeue.act_nr_of_sg = ipz_squeue->act_nr_of_sg;
+ 		resp.ipz_squeue.queue_length = ipz_squeue->queue_length;
+ 		resp.ipz_squeue.pagesize = ipz_squeue->pagesize;
+ 		resp.ipz_squeue.toggle_state = ipz_squeue->toggle_state;
++		ret = ehca_mmap_nopage(((u64)(my_qp->token) << 32) | 0x23000000,
++				       ipz_squeue->queue_length,
++				       (void**)&resp.ipz_squeue.queue,
++				       &vma);
++		if (ret) {
++			ehca_err(pd->device, "Could not mmap squeue pages");
++			goto create_qp_exit4;
++		}
++		my_qp->uspace_squeue = resp.ipz_squeue.queue;
++		/* fw_handle */
++		resp.galpas = my_qp->galpas;
++		ret = ehca_mmap_register(my_qp->galpas.user.fw_handle,
++					 (void**)&resp.galpas.kernel.fw_handle,
++					 &vma);
++		if (ret) {
++			ehca_err(pd->device, "Could not mmap fw_handle");
++			goto create_qp_exit5;
++		}
++		my_qp->uspace_fwh = (u64)resp.galpas.kernel.fw_handle;
++
+ 		if (ib_copy_to_udata(udata, &resp, sizeof resp)) {
+ 			ehca_err(pd->device, "Copy to udata failed");
+ 			ret = -EINVAL;
+-			goto create_qp_exit3;
++			goto create_qp_exit6;
+ 		}
+ 	}
+ 
+ 	return &my_qp->ib_qp;
+ 
++create_qp_exit6:
++	ehca_munmap(my_qp->uspace_fwh, EHCA_PAGESIZE);
++
++create_qp_exit5:
++	ehca_munmap(my_qp->uspace_squeue, my_qp->ipz_squeue.queue_length);
++
++create_qp_exit4:
++	ehca_munmap(my_qp->uspace_rqueue, my_qp->ipz_rqueue.queue_length);
++
+ create_qp_exit3:
+ 	ipz_queue_dtor(&my_qp->ipz_rqueue);
+ 	ipz_queue_dtor(&my_qp->ipz_squeue);
+@@ -892,7 +931,7 @@ static int internal_modify_qp(struct ib_
+ 	     my_qp->qp_type == IB_QPT_SMI) &&
+ 	    statetrans == IB_QPST_SQE2RTS) {
+ 		/* mark next free wqe if kernel */
+-		if (!ibqp->uobject) {
++		if (my_qp->uspace_squeue == 0) {
+ 			struct ehca_wqe *wqe;
+ 			/* lock send queue */
+ 			spin_lock_irqsave(&my_qp->spinlock_s, spl_flags);
+@@ -1378,18 +1417,11 @@ int ehca_destroy_qp(struct ib_qp *ibqp)
+ 	enum ib_qp_type	qp_type;
+ 	unsigned long flags;
+ 
+-	if (ibqp->uobject) {
+-		if (my_qp->mm_count_galpa ||
+-		    my_qp->mm_count_rqueue || my_qp->mm_count_squeue) {
+-			ehca_err(ibqp->device, "Resources still referenced in "
+-				 "user space qp_num=%x", ibqp->qp_num);
+-			return -EINVAL;
+-		}
+-		if (my_pd->ownpid != cur_pid) {
+-			ehca_err(ibqp->device, "Invalid caller pid=%x ownpid=%x",
+-				 cur_pid, my_pd->ownpid);
+-			return -EINVAL;
+-		}
++	if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context &&
++	    my_pd->ownpid != cur_pid) {
++		ehca_err(ibqp->device, "Invalid caller pid=%x ownpid=%x",
++			 cur_pid, my_pd->ownpid);
++		return -EINVAL;
+ 	}
+ 
+ 	if (my_qp->send_cq) {
+@@ -1407,6 +1439,24 @@ int ehca_destroy_qp(struct ib_qp *ibqp)
+ 	idr_remove(&ehca_qp_idr, my_qp->token);
+ 	spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
+ 
++	/* un-mmap if vma alloc */
++	if (my_qp->uspace_rqueue) {
++		ret = ehca_munmap(my_qp->uspace_rqueue,
++				  my_qp->ipz_rqueue.queue_length);
++		if (ret)
++			ehca_err(ibqp->device, "Could not munmap rqueue "
++				 "qp_num=%x", qp_num);
++		ret = ehca_munmap(my_qp->uspace_squeue,
++				  my_qp->ipz_squeue.queue_length);
++		if (ret)
++			ehca_err(ibqp->device, "Could not munmap squeue "
++				 "qp_num=%x", qp_num);
++		ret = ehca_munmap(my_qp->uspace_fwh, EHCA_PAGESIZE);
++		if (ret)
++			ehca_err(ibqp->device, "Could not munmap fwh qp_num=%x",
++				 qp_num);
++	}
++
+ 	h_ret = hipz_h_destroy_qp(shca->ipz_hca_handle, my_qp);
+ 	if (h_ret != H_SUCCESS) {
+ 		ehca_err(ibqp->device, "hipz_h_destroy_qp() failed rc=%lx "
+diff -Nurp ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_uverbs.c ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_uverbs.c
+--- ofa_kernel-1.2_old/drivers/infiniband/hw/ehca/ehca_uverbs.c	2007-05-04 10:38:23.000000000 +0200
++++ ofa_kernel-1.2_new/drivers/infiniband/hw/ehca/ehca_uverbs.c	2007-04-29 15:10:56.000000000 +0200
+@@ -68,183 +68,105 @@ int ehca_dealloc_ucontext(struct ib_ucon
+ 	return 0;
+ }
+ 
+-static void ehca_mm_open(struct vm_area_struct *vma)
++struct page *ehca_nopage(struct vm_area_struct *vma,
++			 unsigned long address, int *type)
+ {
+-	u32 *count = (u32*)vma->vm_private_data;
+-	if (!count) {
+-		ehca_gen_err("Invalid vma struct vm_start=%lx vm_end=%lx",
+-			     vma->vm_start, vma->vm_end);
+-		return;
+-	}
+-	(*count)++;
+-	if (!(*count))
+-		ehca_gen_err("Use count overflow vm_start=%lx vm_end=%lx",
+-			     vma->vm_start, vma->vm_end);
+-	ehca_gen_dbg("vm_start=%lx vm_end=%lx count=%x",
+-		     vma->vm_start, vma->vm_end, *count);
+-}
+-
+-static void ehca_mm_close(struct vm_area_struct *vma)
+-{
+-	u32 *count = (u32*)vma->vm_private_data;
+-	if (!count) {
+-		ehca_gen_err("Invalid vma struct vm_start=%lx vm_end=%lx",
+-			     vma->vm_start, vma->vm_end);
+-		return;
+-	}
+-	(*count)--;
+-	ehca_gen_dbg("vm_start=%lx vm_end=%lx count=%x",
+-		     vma->vm_start, vma->vm_end, *count);
+-}
+-
+-static struct vm_operations_struct vm_ops = {
+-	.open =	ehca_mm_open,
+-	.close = ehca_mm_close,
+-};
+-
+-static int ehca_mmap_fw(struct vm_area_struct *vma, struct h_galpas *galpas,
+-			u32 *mm_count)
+-{
+-	int ret;
+-	u64 vsize, physical;
+-
+-	vsize = vma->vm_end - vma->vm_start;
+-	if (vsize != EHCA_PAGESIZE) {
+-		ehca_gen_err("invalid vsize=%lx", vma->vm_end - vma->vm_start);
+-		return -EINVAL;
+-	}
+-
+-	physical = galpas->user.fw_handle;
+-	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+-	ehca_gen_dbg("vsize=%lx physical=%lx", vsize, physical);
+-	/* VM_IO | VM_RESERVED are set by remap_pfn_range() */
+-	ret = remap_pfn_range(vma, vma->vm_start, physical >> PAGE_SHIFT,
+-			      vsize, vma->vm_page_prot);
+-	if (unlikely(ret)) {
+-		ehca_gen_err("remap_pfn_range() failed ret=%x", ret);
+-		return -ENOMEM;
+-	}
+-
+-	vma->vm_private_data = mm_count;
+-	(*mm_count)++;
+-	vma->vm_ops = &vm_ops;
+-
+-	return 0;
+-}
++	struct page *mypage = NULL;
++	u64 fileoffset = vma->vm_pgoff << PAGE_SHIFT;
++	u32 idr_handle = fileoffset >> 32;
++	u32 q_type = (fileoffset >> 28) & 0xF;	  /* CQ, QP,...        */
++	u32 rsrc_type = (fileoffset >> 24) & 0xF; /* sq,rq,cmnd_window */
++	u32 cur_pid = current->tgid;
++	unsigned long flags;
++	struct ehca_cq *cq;
++	struct ehca_qp *qp;
++	struct ehca_pd *pd;
++	u64 offset;
++	void *vaddr;
+ 
+-static int ehca_mmap_queue(struct vm_area_struct *vma, struct ipz_queue *queue,
+-			   u32 *mm_count)
+-{
+-	int ret;
+-	u64 start, ofs;
+-	struct page *page;
++	switch (q_type) {
++	case 1: /* CQ */
++		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
++		cq = idr_find(&ehca_cq_idr, idr_handle);
++		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+ 
+-	vma->vm_flags |= VM_RESERVED;
+-	start = vma->vm_start;
+-	for (ofs = 0; ofs < queue->queue_length; ofs += PAGE_SIZE) {
+-		u64 virt_addr = (u64)ipz_qeit_calc(queue, ofs);
+-		page = virt_to_page(virt_addr);
+-		ret = vm_insert_page(vma, start, page);
+-		if (unlikely(ret)) {
+-			ehca_gen_err("vm_insert_page() failed rc=%x", ret);
+-			return ret;
++		/* make sure this mmap really belongs to the authorized user */
++		if (!cq) {
++			ehca_gen_err("cq is NULL ret=NOPAGE_SIGBUS");
++			return NOPAGE_SIGBUS;
+ 		}
+-		start +=  PAGE_SIZE;
+-	}
+-	vma->vm_private_data = mm_count;
+-	(*mm_count)++;
+-	vma->vm_ops = &vm_ops;
+ 
+-	return 0;
+-}
+-
+-static int ehca_mmap_cq(struct vm_area_struct *vma, struct ehca_cq *cq,
+-			u32 rsrc_type)
+-{
+-	int ret;
+-
+-	switch (rsrc_type) {
+-	case 1: /* galpa fw handle */
+-		ehca_dbg(cq->ib_cq.device, "cq_num=%x fw", cq->cq_number);
+-		ret = ehca_mmap_fw(vma, &cq->galpas, &cq->mm_count_galpa);
+-		if (unlikely(ret)) {
++		if (cq->ownpid != cur_pid) {
+ 			ehca_err(cq->ib_cq.device,
+-				 "ehca_mmap_fw() failed rc=%x cq_num=%x",
+-				 ret, cq->cq_number);
+-			return ret;
++				 "Invalid caller pid=%x ownpid=%x",
++				 cur_pid, cq->ownpid);
++			return NOPAGE_SIGBUS;
+ 		}
+-		break;
+ 
+-	case 2: /* cq queue_addr */
+-		ehca_dbg(cq->ib_cq.device, "cq_num=%x queue", cq->cq_number);
+-		ret = ehca_mmap_queue(vma, &cq->ipz_queue, &cq->mm_count_queue);
+-		if (unlikely(ret)) {
+-			ehca_err(cq->ib_cq.device,
+-				 "ehca_mmap_queue() failed rc=%x cq_num=%x",
+-				 ret, cq->cq_number);
+-			return ret;
++		if (rsrc_type == 2) {
++			ehca_dbg(cq->ib_cq.device, "cq=%p cq queuearea", cq);
++			offset = address - vma->vm_start;
++			vaddr = ipz_qeit_calc(&cq->ipz_queue, offset);
++			ehca_dbg(cq->ib_cq.device, "offset=%lx vaddr=%p",
++				 offset, vaddr);
++			mypage = virt_to_page(vaddr);
+ 		}
+ 		break;
+ 
+-	default:
+-		ehca_err(cq->ib_cq.device, "bad resource type=%x cq_num=%x",
+-			 rsrc_type, cq->cq_number);
+-		return -EINVAL;
+-	}
+-
+-	return 0;
+-}
+-
+-static int ehca_mmap_qp(struct vm_area_struct *vma, struct ehca_qp *qp,
+-			u32 rsrc_type)
+-{
+-	int ret;
++	case 2: /* QP */
++		spin_lock_irqsave(&ehca_qp_idr_lock, flags);
++		qp = idr_find(&ehca_qp_idr, idr_handle);
++		spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
+ 
+-	switch (rsrc_type) {
+-	case 1: /* galpa fw handle */
+-		ehca_dbg(qp->ib_qp.device, "qp_num=%x fw", qp->ib_qp.qp_num);
+-		ret = ehca_mmap_fw(vma, &qp->galpas, &qp->mm_count_galpa);
+-		if (unlikely(ret)) {
+-			ehca_err(qp->ib_qp.device,
+-				 "remap_pfn_range() failed ret=%x qp_num=%x",
+-				 ret, qp->ib_qp.qp_num);
+-			return -ENOMEM;
++		/* make sure this mmap really belongs to the authorized user */
++		if (!qp) {
++			ehca_gen_err("qp is NULL ret=NOPAGE_SIGBUS");
++			return NOPAGE_SIGBUS;
+ 		}
+-		break;
+ 
+-	case 2: /* qp rqueue_addr */
+-		ehca_dbg(qp->ib_qp.device, "qp_num=%x rqueue",
+-			 qp->ib_qp.qp_num);
+-		ret = ehca_mmap_queue(vma, &qp->ipz_rqueue, &qp->mm_count_rqueue);
+-		if (unlikely(ret)) {
++		pd = container_of(qp->ib_qp.pd, struct ehca_pd, ib_pd);
++		if (pd->ownpid != cur_pid) {
+ 			ehca_err(qp->ib_qp.device,
+-				 "ehca_mmap_queue(rq) failed rc=%x qp_num=%x",
+-				 ret, qp->ib_qp.qp_num);
+-			return ret;
++				 "Invalid caller pid=%x ownpid=%x",
++				 cur_pid, pd->ownpid);
++			return NOPAGE_SIGBUS;
+ 		}
+-		break;
+ 
+-	case 3: /* qp squeue_addr */
+-		ehca_dbg(qp->ib_qp.device, "qp_num=%x squeue",
+-			 qp->ib_qp.qp_num);
+-		ret = ehca_mmap_queue(vma, &qp->ipz_squeue, &qp->mm_count_squeue);
+-		if (unlikely(ret)) {
+-			ehca_err(qp->ib_qp.device,
+-				 "ehca_mmap_queue(sq) failed rc=%x qp_num=%x",
+-				 ret, qp->ib_qp.qp_num);
+-			return ret;
++		if (rsrc_type == 2) {	/* rqueue */
++			ehca_dbg(qp->ib_qp.device, "qp=%p qp rqueuearea", qp);
++			offset = address - vma->vm_start;
++			vaddr = ipz_qeit_calc(&qp->ipz_rqueue, offset);
++			ehca_dbg(qp->ib_qp.device, "offset=%lx vaddr=%p",
++				 offset, vaddr);
++			mypage = virt_to_page(vaddr);
++		} else if (rsrc_type == 3) {	/* squeue */
++			ehca_dbg(qp->ib_qp.device, "qp=%p qp squeuearea", qp);
++			offset = address - vma->vm_start;
++			vaddr = ipz_qeit_calc(&qp->ipz_squeue, offset);
++			ehca_dbg(qp->ib_qp.device, "offset=%lx vaddr=%p",
++				 offset, vaddr);
++			mypage = virt_to_page(vaddr);
+ 		}
+ 		break;
+ 
+ 	default:
+-		ehca_err(qp->ib_qp.device, "bad resource type=%x qp=num=%x",
+-			 rsrc_type, qp->ib_qp.qp_num);
+-		return -EINVAL;
++		ehca_gen_err("bad queue type %x", q_type);
++		return NOPAGE_SIGBUS;
+ 	}
+ 
+-	return 0;
++	if (!mypage) {
++		ehca_gen_err("Invalid page adr==NULL ret=NOPAGE_SIGBUS");
++		return NOPAGE_SIGBUS;
++	}
++	get_page(mypage);
++
++	return mypage;
+ }
+ 
++static struct vm_operations_struct ehcau_vm_ops = {
++	.nopage = ehca_nopage,
++};
++
+ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+ {
+ 	u64 fileoffset = vma->vm_pgoff << PAGE_SHIFT;
+@@ -253,6 +175,7 @@ int ehca_mmap(struct ib_ucontext *contex
+ 	u32 rsrc_type = (fileoffset >> 24) & 0xF; /* sq,rq,cmnd_window */
+ 	u32 cur_pid = current->tgid;
+ 	u32 ret;
++	u64 vsize, physical;
+ 	unsigned long flags;
+ 	struct ehca_cq *cq;
+ 	struct ehca_qp *qp;
+@@ -278,12 +201,44 @@ int ehca_mmap(struct ib_ucontext *contex
+ 		if (!cq->ib_cq.uobject || cq->ib_cq.uobject->context != context)
+ 			return -EINVAL;
+ 
+-		ret = ehca_mmap_cq(vma, cq, rsrc_type);
+-		if (unlikely(ret)) {
+-			ehca_err(cq->ib_cq.device,
+-				 "ehca_mmap_cq() failed rc=%x cq_num=%x",
+-				 ret, cq->cq_number);
+-			return ret;
++		switch (rsrc_type) {
++		case 1: /* galpa fw handle */
++			ehca_dbg(cq->ib_cq.device, "cq=%p cq triggerarea", cq);
++			vma->vm_flags |= VM_RESERVED;
++			vsize = vma->vm_end - vma->vm_start;
++			if (vsize != EHCA_PAGESIZE) {
++				ehca_err(cq->ib_cq.device, "invalid vsize=%lx",
++					 vma->vm_end - vma->vm_start);
++				return -EINVAL;
++			}
++
++			physical = cq->galpas.user.fw_handle;
++			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
++			vma->vm_flags |= VM_IO | VM_RESERVED;
++
++			ehca_dbg(cq->ib_cq.device,
++				 "vsize=%lx physical=%lx", vsize, physical);
++			ret = remap_pfn_range(vma, vma->vm_start,
++					      physical >> PAGE_SHIFT, vsize,
++					      vma->vm_page_prot);
++			if (ret) {
++				ehca_err(cq->ib_cq.device,
++					 "remap_pfn_range() failed ret=%x",
++					 ret);
++				return -ENOMEM;
++			}
++			break;
++
++		case 2: /* cq queue_addr */
++			ehca_dbg(cq->ib_cq.device, "cq=%p cq q_addr", cq);
++			vma->vm_flags |= VM_RESERVED;
++			vma->vm_ops = &ehcau_vm_ops;
++			break;
++
++		default:
++			ehca_err(cq->ib_cq.device, "bad resource type %x",
++				 rsrc_type);
++			return -EINVAL;
+ 		}
+ 		break;
+ 
+@@ -307,12 +262,50 @@ int ehca_mmap(struct ib_ucontext *contex
+ 		if (!qp->ib_qp.uobject || qp->ib_qp.uobject->context != context)
+ 			return -EINVAL;
+ 
+-		ret = ehca_mmap_qp(vma, qp, rsrc_type);
+-		if (unlikely(ret)) {
+-			ehca_err(qp->ib_qp.device,
+-				 "ehca_mmap_qp() failed rc=%x qp_num=%x",
+-				 ret, qp->ib_qp.qp_num);
+-			return ret;
++		switch (rsrc_type) {
++		case 1: /* galpa fw handle */
++			ehca_dbg(qp->ib_qp.device, "qp=%p qp triggerarea", qp);
++			vma->vm_flags |= VM_RESERVED;
++			vsize = vma->vm_end - vma->vm_start;
++			if (vsize != EHCA_PAGESIZE) {
++				ehca_err(qp->ib_qp.device, "invalid vsize=%lx",
++					 vma->vm_end - vma->vm_start);
++				return -EINVAL;
++			}
++
++			physical = qp->galpas.user.fw_handle;
++			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
++			vma->vm_flags |= VM_IO | VM_RESERVED;
++
++			ehca_dbg(qp->ib_qp.device, "vsize=%lx physical=%lx",
++				 vsize, physical);
++			ret = remap_pfn_range(vma, vma->vm_start,
++					      physical >> PAGE_SHIFT, vsize,
++					      vma->vm_page_prot);
++			if (ret) {
++				ehca_err(qp->ib_qp.device,
++					 "remap_pfn_range() failed ret=%x",
++					 ret);
++				return -ENOMEM;
++			}
++			break;
++
++		case 2: /* qp rqueue_addr */
++			ehca_dbg(qp->ib_qp.device, "qp=%p qp rqueue_addr", qp);
++			vma->vm_flags |= VM_RESERVED;
++			vma->vm_ops = &ehcau_vm_ops;
++			break;
++
++		case 3: /* qp squeue_addr */
++			ehca_dbg(qp->ib_qp.device, "qp=%p qp squeue_addr", qp);
++			vma->vm_flags |= VM_RESERVED;
++			vma->vm_ops = &ehcau_vm_ops;
++			break;
++
++		default:
++			ehca_err(qp->ib_qp.device, "bad resource type %x",
++				 rsrc_type);
++			return -EINVAL;
+ 		}
+ 		break;
+ 
+@@ -323,3 +316,77 @@ int ehca_mmap(struct ib_ucontext *contex
+ 
+ 	return 0;
+ }
++
++int ehca_mmap_nopage(u64 foffset, u64 length, void **mapped,
++		     struct vm_area_struct **vma)
++{
++	down_write(&current->mm->mmap_sem);
++	*mapped = (void*)do_mmap(NULL,0, length, PROT_WRITE,
++				 MAP_SHARED | MAP_ANONYMOUS,
++				 foffset);
++	up_write(&current->mm->mmap_sem);
++	if (!(*mapped)) {
++		ehca_gen_err("couldn't mmap foffset=%lx length=%lx",
++			     foffset, length);
++		return -EINVAL;
++	}
++
++	*vma = find_vma(current->mm, (u64)*mapped);
++	if (!(*vma)) {
++		down_write(&current->mm->mmap_sem);
++		do_munmap(current->mm, 0, length);
++		up_write(&current->mm->mmap_sem);
++		ehca_gen_err("couldn't find vma queue=%p", *mapped);
++		return -EINVAL;
++	}
++	(*vma)->vm_flags |= VM_RESERVED;
++	(*vma)->vm_ops = &ehcau_vm_ops;
++
++	return 0;
++}
++
++int ehca_mmap_register(u64 physical, void **mapped,
++		       struct vm_area_struct **vma)
++{
++	int ret;
++	unsigned long vsize;
++	/* ehca hw supports only 4k page */
++	ret = ehca_mmap_nopage(0, EHCA_PAGESIZE, mapped, vma);
++	if (ret) {
++		ehca_gen_err("could'nt mmap physical=%lx", physical);
++		return ret;
++	}
++
++	(*vma)->vm_flags |= VM_RESERVED;
++	vsize = (*vma)->vm_end - (*vma)->vm_start;
++	if (vsize != EHCA_PAGESIZE) {
++		ehca_gen_err("invalid vsize=%lx",
++			     (*vma)->vm_end - (*vma)->vm_start);
++		return -EINVAL;
++	}
++
++	(*vma)->vm_page_prot = pgprot_noncached((*vma)->vm_page_prot);
++	(*vma)->vm_flags |= VM_IO | VM_RESERVED;
++
++	ret = remap_pfn_range((*vma), (*vma)->vm_start,
++			      physical >> PAGE_SHIFT, vsize,
++			      (*vma)->vm_page_prot);
++	if (ret) {
++		ehca_gen_err("remap_pfn_range() failed ret=%x", ret);
++		return -ENOMEM;
++	}
++
++	return 0;
++
++}
++
++int ehca_munmap(unsigned long addr, size_t len) {
++	int ret = 0;
++	struct mm_struct *mm = current->mm;
++	if (mm) {
++		down_write(&mm->mmap_sem);
++		ret = do_munmap(mm, addr, len);
++		up_write(&mm->mmap_sem);
++	}
++	return ret;
++}






More information about the general mailing list