From cd1334f03f7b799bc6893b511daf2080e8f73863 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:19 -0700
Subject: gru: bug fixes for GRU exception handling

Bug fixes for GRU exception handling.  Additional fields from the CBR must
be returned to the user to allow the user to correctly diagnose GRU
exceptions.

Handle endcase in TFH TLB miss handling.  Verify that TFH actually
indicates a pending exception.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c | 25 +++++++++++++++++++++----
 1 file changed, 21 insertions(+), 4 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index ab118558552..4089f862aa2 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -334,6 +334,8 @@ static int gru_try_dropin(struct gru_thread_state *gts,
 	 * Might be a hardware race OR a stupid user. Ignore FMM because FMM
 	 * is a transient state.
 	 */
+	if (tfh->status != TFHSTATUS_EXCEPTION)
+		goto failnoexception;
 	if (tfh->state == TFHSTATE_IDLE)
 		goto failidle;
 	if (tfh->state == TFHSTATE_MISS_FMM && cb)
@@ -401,8 +403,17 @@ failfmm:
 	gru_dbg(grudev, "FAILED fmm tfh: 0x%p, state %d\n", tfh, tfh->state);
 	return 0;
 
+failnoexception:
+	/* TFH status did not show exception pending */
+	gru_flush_cache(tfh);
+	if (cb)
+		gru_flush_cache(cb);
+	STAT(tlb_dropin_fail_no_exception);
+	gru_dbg(grudev, "FAILED non-exception tfh: 0x%p, status %d, state %d\n", tfh, tfh->status, tfh->state);
+	return 0;
+
 failidle:
-	/* TFH was idle  - no miss pending */
+	/* TFH state was idle  - no miss pending */
 	gru_flush_cache(tfh);
 	if (cb)
 		gru_flush_cache(cb);
@@ -472,7 +483,8 @@ irqreturn_t gru_intr(int irq, void *dev_id)
 		 * This is running in interrupt context. Trylock the mmap_sem.
 		 * If it fails, retry the fault in user context.
 		 */
-		if (down_read_trylock(&gts->ts_mm->mmap_sem)) {
+		if (!gts->ts_force_cch_reload &&
+					down_read_trylock(&gts->ts_mm->mmap_sem)) {
 			gru_try_dropin(gts, tfh, NULL);
 			up_read(&gts->ts_mm->mmap_sem);
 		} else {
@@ -595,14 +607,19 @@ int gru_get_exception_detail(unsigned long arg)
 		excdet.ecause = cbe->ecause;
 		excdet.exceptdet0 = cbe->idef1upd;
 		excdet.exceptdet1 = cbe->idef3upd;
+		excdet.cbrstate = cbe->cbrstate;
+		excdet.cbrexecstatus = cbe->cbrexecstatus;
 		ret = 0;
 	} else {
 		ret = -EAGAIN;
 	}
 	gru_unlock_gts(gts);
 
-	gru_dbg(grudev, "address 0x%lx, ecause 0x%x\n", excdet.cb,
-		excdet.ecause);
+	gru_dbg(grudev,
+		"cb 0x%lx, op %d, exopc %d, cbrstate %d, cbrexecstatus 0x%x, ecause 0x%x, "
+		"exdet0 0x%lx, exdet1 0x%x\n",
+		excdet.cb, excdet.opc, excdet.exopc, excdet.cbrstate, excdet.cbrexecstatus,
+		excdet.ecause, excdet.exceptdet0, excdet.exceptdet1);
 	if (!ret && copy_to_user((void __user *)arg, &excdet, sizeof(excdet)))
 		ret = -EFAULT;
 	return ret;
-- 
cgit v1.2.3-70-g09d2


From d57c82b10709bbb1deb7eb26cf42abcde8851e4d Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:20 -0700
Subject: gru: change context load and unload

Remove "static" from the functions for loading/unloading GRU contexts.
These functions will be called from other GRU files.  Fix bug in unlocking
gru context.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c  |  4 ++--
 drivers/misc/sgi-gru/grumain.c   | 10 +++++-----
 drivers/misc/sgi-gru/grutables.h |  4 ++++
 3 files changed, 11 insertions(+), 7 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 4089f862aa2..f15152165a9 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -558,8 +558,8 @@ int gru_handle_user_call_os(unsigned long cb)
 	 * CCH may contain stale data if ts_force_cch_reload is set.
 	 */
 	if (gts->ts_gru && gts->ts_force_cch_reload) {
-		gru_update_cch(gts, 0);
 		gts->ts_force_cch_reload = 0;
+		gru_update_cch(gts, 0);
 	}
 
 	ret = -EAGAIN;
@@ -644,7 +644,7 @@ static int gru_unload_all_contexts(void)
 			if (gts && mutex_trylock(&gts->ts_ctxlock)) {
 				spin_unlock(&gru->gs_lock);
 				gru_unload_context(gts, 1);
-				gru_unlock_gts(gts);
+				mutex_unlock(&gts->ts_ctxlock);
 				spin_lock(&gru->gs_lock);
 			}
 		}
diff --git a/drivers/misc/sgi-gru/grumain.c b/drivers/misc/sgi-gru/grumain.c
index 14baabc79da..6358244f392 100644
--- a/drivers/misc/sgi-gru/grumain.c
+++ b/drivers/misc/sgi-gru/grumain.c
@@ -533,7 +533,7 @@ void gru_unload_context(struct gru_thread_state *gts, int savestate)
  * Load a GRU context by copying it from the thread data structure in memory
  * to the GRU.
  */
-static void gru_load_context(struct gru_thread_state *gts)
+void gru_load_context(struct gru_thread_state *gts)
 {
 	struct gru_state *gru = gts->ts_gru;
 	struct gru_context_configuration_handle *cch;
@@ -600,8 +600,8 @@ int gru_update_cch(struct gru_thread_state *gts, int force_unload)
 			gts->ts_tlb_int_select = gru_cpu_fault_map_id();
 			cch->tlb_int_select = gru_cpu_fault_map_id();
 			cch->tfm_fault_bit_enable =
-	    		    (gts->ts_user_options == GRU_OPT_MISS_FMM_POLL
-	     		    || gts->ts_user_options == GRU_OPT_MISS_FMM_INTR);
+				(gts->ts_user_options == GRU_OPT_MISS_FMM_POLL
+				|| gts->ts_user_options == GRU_OPT_MISS_FMM_INTR);
 		} else {
 			for (i = 0; i < 8; i++)
 				cch->asid[i] = 0;
@@ -645,7 +645,7 @@ static int gru_retarget_intr(struct gru_thread_state *gts)
 #define next_gru(b, g)	(((g) < &(b)->bs_grus[GRU_CHIPLETS_PER_BLADE - 1]) ?  \
 				 ((g)+1) : &(b)->bs_grus[0])
 
-static void gru_steal_context(struct gru_thread_state *gts, int blade_id)
+void gru_steal_context(struct gru_thread_state *gts, int blade_id)
 {
 	struct gru_blade_state *blade;
 	struct gru_state *gru, *gru0;
@@ -711,7 +711,7 @@ static void gru_steal_context(struct gru_thread_state *gts, int blade_id)
 /*
  * Scan the GRUs on the local blade & assign a GRU context.
  */
-static struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts,
+struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts,
 						int blade)
 {
 	struct gru_state *gru, *grux;
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index c69086c9b98..4ddb5b92acb 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -611,6 +611,10 @@ extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct
 				*vma, int tsid);
 extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct
 				*vma, int tsid);
+extern struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts,
+		int blade);
+extern void gru_load_context(struct gru_thread_state *gts);
+extern void gru_steal_context(struct gru_thread_state *gts, int blade_id);
 extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
 extern int gru_update_cch(struct gru_thread_state *gts, int force_unload);
 extern void gts_drop(struct gru_thread_state *gts);
-- 
cgit v1.2.3-70-g09d2


From 4a7a17c1188a878e9f00e4ca8dc724c7cff17606 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:25 -0700
Subject: gru: support instruction completion interrupts

Add support for interrupts generated by GRU instruction completion.
Previously, the only interrupts were for TLB misses.  The hardware also
supports interrupts on instruction completion.  This will be supported for
instructions issued by the kernel.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c  | 27 ++++++++++++++++++++-------
 drivers/misc/sgi-gru/grumain.c   |  4 ++++
 drivers/misc/sgi-gru/grutables.h |  5 +++++
 3 files changed, 29 insertions(+), 7 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index f15152165a9..3220e95be6b 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -166,7 +166,8 @@ static inline struct gru_state *irq_to_gru(int irq)
  * the GRU, atomic operations must be used to clear bits.
  */
 static void get_clear_fault_map(struct gru_state *gru,
-				struct gru_tlb_fault_map *map)
+				struct gru_tlb_fault_map *imap,
+				struct gru_tlb_fault_map *dmap)
 {
 	unsigned long i, k;
 	struct gru_tlb_fault_map *tfm;
@@ -177,7 +178,11 @@ static void get_clear_fault_map(struct gru_state *gru,
 		k = tfm->fault_bits[i];
 		if (k)
 			k = xchg(&tfm->fault_bits[i], 0UL);
-		map->fault_bits[i] = k;
+		imap->fault_bits[i] = k;
+		k = tfm->done_bits[i];
+		if (k)
+			k = xchg(&tfm->done_bits[i], 0UL);
+		dmap->fault_bits[i] = k;
 	}
 
 	/*
@@ -449,7 +454,7 @@ failactive:
 irqreturn_t gru_intr(int irq, void *dev_id)
 {
 	struct gru_state *gru;
-	struct gru_tlb_fault_map map;
+	struct gru_tlb_fault_map imap, dmap;
 	struct gru_thread_state *gts;
 	struct gru_tlb_fault_handle *tfh = NULL;
 	int cbrnum, ctxnum;
@@ -462,11 +467,19 @@ irqreturn_t gru_intr(int irq, void *dev_id)
 			raw_smp_processor_id(), irq);
 		return IRQ_NONE;
 	}
-	get_clear_fault_map(gru, &map);
-	gru_dbg(grudev, "irq %d, gru %x, map 0x%lx\n", irq, gru->gs_gid,
-		map.fault_bits[0]);
+	get_clear_fault_map(gru, &imap, &dmap);
+	gru_dbg(grudev,
+		"irq %d, gid %d, imap %016lx %016lx, dmap %016lx %016lx\n",
+		irq, gru->gs_gid, dmap.fault_bits[0], dmap.fault_bits[1],
+		dmap.fault_bits[0], dmap.fault_bits[1]);
+
+	for_each_cbr_in_tfm(cbrnum, dmap.fault_bits) {
+		complete(gru->gs_blade->bs_async_wq);
+		gru_dbg(grudev, "gid %d, cbr_done %d, done %d\n",
+			gru->gs_gid, cbrnum, gru->gs_blade->bs_async_wq->done);
+	}
 
-	for_each_cbr_in_tfm(cbrnum, map.fault_bits) {
+	for_each_cbr_in_tfm(cbrnum, imap.fault_bits) {
 		tfh = get_tfh_by_index(gru, cbrnum);
 		prefetchw(tfh);	/* Helps on hdw, required for emulator */
 
diff --git a/drivers/misc/sgi-gru/grumain.c b/drivers/misc/sgi-gru/grumain.c
index afc4c473c79..e38a0f1775f 100644
--- a/drivers/misc/sgi-gru/grumain.c
+++ b/drivers/misc/sgi-gru/grumain.c
@@ -572,8 +572,12 @@ void gru_load_context(struct gru_thread_state *gts)
 
 	if (is_kernel_context(gts)) {
 		cch->unmap_enable = 1;
+		cch->tfm_done_bit_enable = 1;
+		cch->cb_int_enable = 1;
 	} else {
 		cch->unmap_enable = 0;
+		cch->tfm_done_bit_enable = 0;
+		cch->cb_int_enable = 0;
 		asid = gru_load_mm_tracker(gru, gts);
 		for (i = 0; i < 8; i++) {
 			cch->asid[i] = asid + i;
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index 5f8f3bda2fa..ca81800146f 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -462,6 +462,11 @@ struct gru_blade_state {
 	struct rw_semaphore	bs_kgts_sema;		/* lock for kgts */
 	struct gru_thread_state *bs_kgts;		/* GTS for kernel use */
 
+	/* ---- the following are used for managing kernel async GRU CBRs --- */
+	int			bs_async_dsr_bytes;	/* DSRs for async */
+	int			bs_async_cbrs;		/* CBRs AU for async */
+	struct completion	*bs_async_wq;
+
 	/* ---- the following are protected by the bs_lock spinlock ---- */
 	spinlock_t		bs_lock;		/* lock used for
 							   stealing contexts */
-- 
cgit v1.2.3-70-g09d2


From 9120dec47f150636d85b3dba03318ccecd181c79 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:25 -0700
Subject: gru: support for asynchronous gru instructions

Add support for asynchronous GRU instructions.  Currently, asynchronous
instructions are supported only for GRU instructions issued by the kernel.

[akpm@linux-foundation.org: build fix]
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c     |   4 -
 drivers/misc/sgi-gru/grukservices.c | 178 ++++++++++++++++++++++++++++++++----
 drivers/misc/sgi-gru/grukservices.h |  51 +++++++++++
 3 files changed, 213 insertions(+), 20 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 3220e95be6b..8443e90f9f6 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -468,10 +468,6 @@ irqreturn_t gru_intr(int irq, void *dev_id)
 		return IRQ_NONE;
 	}
 	get_clear_fault_map(gru, &imap, &dmap);
-	gru_dbg(grudev,
-		"irq %d, gid %d, imap %016lx %016lx, dmap %016lx %016lx\n",
-		irq, gru->gs_gid, dmap.fault_bits[0], dmap.fault_bits[1],
-		dmap.fault_bits[0], dmap.fault_bits[1]);
 
 	for_each_cbr_in_tfm(cbrnum, dmap.fault_bits) {
 		complete(gru->gs_blade->bs_async_wq);
diff --git a/drivers/misc/sgi-gru/grukservices.c b/drivers/misc/sgi-gru/grukservices.c
index a0f981022a6..9dff33cb72e 100644
--- a/drivers/misc/sgi-gru/grukservices.c
+++ b/drivers/misc/sgi-gru/grukservices.c
@@ -52,7 +52,53 @@
  * loaded on demand & can be stolen by a user if the user demand exceeds the
  * kernel demand. The kernel can always reload the kernel context but
  * a SLEEP may be required!!!.
+ *
+ * Async Overview:
+ *
+ * 	Each blade has one "kernel context" that owns GRU kernel resources
+ * 	located on the blade. Kernel drivers use GRU resources in this context
+ * 	for sending messages, zeroing memory, etc.
+ *
+ * 	The kernel context is dynamically loaded on demand. If it is not in
+ * 	use by the kernel, the kernel context can be unloaded & given to a user.
+ * 	The kernel context will be reloaded when needed. This may require that
+ * 	a context be stolen from a user.
+ * 		NOTE: frequent unloading/reloading of the kernel context is
+ * 		expensive. We are depending on batch schedulers, cpusets, sane
+ * 		drivers or some other mechanism to prevent the need for frequent
+ *	 	stealing/reloading.
+ *
+ * 	The kernel context consists of two parts:
+ * 		- 1 CB & a few DSRs that are reserved for each cpu on the blade.
+ * 		  Each cpu has it's own private resources & does not share them
+ * 		  with other cpus. These resources are used serially, ie,
+ * 		  locked, used & unlocked  on each call to a function in
+ * 		  grukservices.
+ * 		  	(Now that we have dynamic loading of kernel contexts, I
+ * 		  	 may rethink this & allow sharing between cpus....)
+ *
+ *		- Additional resources can be reserved long term & used directly
+ *		  by UV drivers located in the kernel. Drivers using these GRU
+ *		  resources can use asynchronous GRU instructions that send
+ *		  interrupts on completion.
+ *		  	- these resources must be explicitly locked/unlocked
+ *		  	- locked resources prevent (obviously) the kernel
+ *		  	  context from being unloaded.
+ *			- drivers using these resource directly issue their own
+ *			  GRU instruction and must wait/check completion.
+ *
+ * 		  When these resources are reserved, the caller can optionally
+ * 		  associate a wait_queue with the resources and use asynchronous
+ * 		  GRU instructions. When an async GRU instruction completes, the
+ * 		  driver will do a wakeup on the event.
+ *
  */
+
+
+#define ASYNC_HAN_TO_BID(h)	((h) - 1)
+#define ASYNC_BID_TO_HAN(b)	((b) + 1)
+#define ASYNC_HAN_TO_BS(h)	gru_base[ASYNC_HAN_TO_BID(h)]
+
 #define GRU_NUM_KERNEL_CBR	1
 #define GRU_NUM_KERNEL_DSR_BYTES 256
 #define GRU_NUM_KERNEL_DSR_CL	(GRU_NUM_KERNEL_DSR_BYTES /		\
@@ -98,20 +144,6 @@ struct message_header {
 
 #define HSTATUS(mq, h)	((mq) + offsetof(struct message_queue, hstatus[h]))
 
-/*
- * Allocate a kernel context (GTS) for the specified blade.
- * 	- protected by writelock on bs_kgts_sema.
- */
-static void gru_alloc_kernel_context(struct gru_blade_state *bs, int blade_id)
-{
-	int cbr_au_count, dsr_au_count, ncpus;
-
-	ncpus = uv_blade_nr_possible_cpus(blade_id);
-	cbr_au_count = GRU_CB_COUNT_TO_AU(GRU_NUM_KERNEL_CBR * ncpus);
-	dsr_au_count = GRU_DS_BYTES_TO_AU(GRU_NUM_KERNEL_DSR_BYTES * ncpus);
-	bs->bs_kgts = gru_alloc_gts(NULL, cbr_au_count, dsr_au_count, 0, 0);
-}
-
 /*
  * Reload the blade's kernel context into a GRU chiplet. Called holding
  * the bs_kgts_sema for READ. Will steal user contexts if necessary.
@@ -121,17 +153,23 @@ static void gru_load_kernel_context(struct gru_blade_state *bs, int blade_id)
 	struct gru_state *gru;
 	struct gru_thread_state *kgts;
 	void *vaddr;
-	int ctxnum;
+	int ctxnum, ncpus;
 
 	up_read(&bs->bs_kgts_sema);
 	down_write(&bs->bs_kgts_sema);
 
 	if (!bs->bs_kgts)
-		gru_alloc_kernel_context(bs, blade_id);
+		bs->bs_kgts = gru_alloc_gts(NULL, 0, 0, 0, 0);
 	kgts = bs->bs_kgts;
 
 	if (!kgts->ts_gru) {
 		STAT(load_kernel_context);
+		ncpus = uv_blade_nr_possible_cpus(blade_id);
+		kgts->ts_cbr_au_count = GRU_CB_COUNT_TO_AU(
+			GRU_NUM_KERNEL_CBR * ncpus + bs->bs_async_cbrs);
+		kgts->ts_dsr_au_count = GRU_DS_BYTES_TO_AU(
+			GRU_NUM_KERNEL_DSR_BYTES * ncpus +
+				bs->bs_async_dsr_bytes);
 		while (!gru_assign_gru_context(kgts, blade_id)) {
 			msleep(1);
 			gru_steal_context(kgts, blade_id);
@@ -203,6 +241,114 @@ static void gru_free_cpu_resources(void *cb, void *dsr)
 	preempt_enable();
 }
 
+/*
+ * Reserve GRU resources to be used asynchronously.
+ *   Note: currently supports only 1 reservation per blade.
+ *
+ * 	input:
+ * 		blade_id  - blade on which resources should be reserved
+ * 		cbrs	  - number of CBRs
+ * 		dsr_bytes - number of DSR bytes needed
+ *	output:
+ *		handle to identify resource
+ *		(0 = async resources already reserved)
+ */
+unsigned long gru_reserve_async_resources(int blade_id, int cbrs, int dsr_bytes,
+			struct completion *cmp)
+{
+	struct gru_blade_state *bs;
+	struct gru_thread_state *kgts;
+	int ret = 0;
+
+	bs = gru_base[blade_id];
+
+	down_write(&bs->bs_kgts_sema);
+
+	/* Verify no resources already reserved */
+	if (bs->bs_async_dsr_bytes + bs->bs_async_cbrs)
+		goto done;
+	bs->bs_async_dsr_bytes = dsr_bytes;
+	bs->bs_async_cbrs = cbrs;
+	bs->bs_async_wq = cmp;
+	kgts = bs->bs_kgts;
+
+	/* Resources changed. Unload context if already loaded */
+	if (kgts && kgts->ts_gru)
+		gru_unload_context(kgts, 0);
+	ret = ASYNC_BID_TO_HAN(blade_id);
+
+done:
+	up_write(&bs->bs_kgts_sema);
+	return ret;
+}
+
+/*
+ * Release async resources previously reserved.
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+void gru_release_async_resources(unsigned long han)
+{
+	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);
+
+	down_write(&bs->bs_kgts_sema);
+	bs->bs_async_dsr_bytes = 0;
+	bs->bs_async_cbrs = 0;
+	bs->bs_async_wq = NULL;
+	up_write(&bs->bs_kgts_sema);
+}
+
+/*
+ * Wait for async GRU instructions to complete.
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+void gru_wait_async_cbr(unsigned long han)
+{
+	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);
+
+	wait_for_completion(bs->bs_async_wq);
+	mb();
+}
+
+/*
+ * Lock previous reserved async GRU resources
+ *
+ *	input:
+ *		han - handle to identify resources
+ *	output:
+ *		cb  - pointer to first CBR
+ *		dsr - pointer to first DSR
+ */
+void gru_lock_async_resource(unsigned long han,  void **cb, void **dsr)
+{
+	struct gru_blade_state *bs = ASYNC_HAN_TO_BS(han);
+	int blade_id = ASYNC_HAN_TO_BID(han);
+	int ncpus;
+
+	gru_lock_kernel_context(blade_id);
+	ncpus = uv_blade_nr_possible_cpus(blade_id);
+	if (cb)
+		*cb = bs->kernel_cb + ncpus * GRU_HANDLE_STRIDE;
+	if (dsr)
+		*dsr = bs->kernel_dsr + ncpus * GRU_NUM_KERNEL_DSR_BYTES;
+}
+
+/*
+ * Unlock previous reserved async GRU resources
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+void gru_unlock_async_resource(unsigned long han)
+{
+	int blade_id = ASYNC_HAN_TO_BID(han);
+
+	gru_unlock_kernel_context(blade_id);
+}
+
 /*----------------------------------------------------------------------*/
 int gru_get_cb_exception_detail(void *cb,
 		struct control_block_extended_exc_detail *excdet)
diff --git a/drivers/misc/sgi-gru/grukservices.h b/drivers/misc/sgi-gru/grukservices.h
index 747ed315d56..d60d34bca44 100644
--- a/drivers/misc/sgi-gru/grukservices.h
+++ b/drivers/misc/sgi-gru/grukservices.h
@@ -146,4 +146,55 @@ extern void *gru_get_next_message(struct gru_message_queue_desc *mqd);
 extern int gru_copy_gpa(unsigned long dest_gpa, unsigned long src_gpa,
 							unsigned int bytes);
 
+/*
+ * Reserve GRU resources to be used asynchronously.
+ *
+ * 	input:
+ * 		blade_id  - blade on which resources should be reserved
+ * 		cbrs	  - number of CBRs
+ * 		dsr_bytes - number of DSR bytes needed
+ * 		cmp	  - completion structure for waiting for
+ * 			    async completions
+ *	output:
+ *		handle to identify resource
+ *		(0 = no resources)
+ */
+extern unsigned long gru_reserve_async_resources(int blade_id, int cbrs, int dsr_bytes,
+				struct completion *cmp);
+
+/*
+ * Release async resources previously reserved.
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+extern void gru_release_async_resources(unsigned long han);
+
+/*
+ * Wait for async GRU instructions to complete.
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+extern void gru_wait_async_cbr(unsigned long han);
+
+/*
+ * Lock previous reserved async GRU resources
+ *
+ *	input:
+ *		han - handle to identify resources
+ *	output:
+ *		cb  - pointer to first CBR
+ *		dsr - pointer to first DSR
+ */
+extern void gru_lock_async_resource(unsigned long han,  void **cb, void **dsr);
+
+/*
+ * Unlock previous reserved async GRU resources
+ *
+ *	input:
+ *		han - handle to identify resources
+ */
+extern void gru_unlock_async_resource(unsigned long han);
+
 #endif 		/* __GRU_KSERVICES_H_ */
-- 
cgit v1.2.3-70-g09d2


From 270952a907220c0331fdaecbb55df892921c5e2d Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:27 -0700
Subject: gru: update to rev 0.9 of gru spec

Update GRU driver to the latest version of the GRU spec. This consists
of minor updates:
	- changes & additions to error status bits
	- new restriction on handling of TLB misses while in FMM mode
	- new field (not used by software) in TFH

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/gru_instructions.h | 19 ++++++++++++-------
 drivers/misc/sgi-gru/grufault.c         |  8 ++++++--
 drivers/misc/sgi-gru/gruhandles.h       |  3 ++-
 drivers/misc/sgi-gru/grukservices.c     |  3 ++-
 drivers/misc/sgi-gru/gruprocfs.c        |  1 +
 drivers/misc/sgi-gru/grutables.h        |  1 +
 6 files changed, 24 insertions(+), 11 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/gru_instructions.h b/drivers/misc/sgi-gru/gru_instructions.h
index 2feb885ca96..eb9140d32e6 100644
--- a/drivers/misc/sgi-gru/gru_instructions.h
+++ b/drivers/misc/sgi-gru/gru_instructions.h
@@ -253,32 +253,37 @@ struct gru_instruction {
 #define CBE_CAUSE_HA_RESPONSE_FATAL		(1 << 13)
 #define CBE_CAUSE_HA_RESPONSE_NON_FATAL		(1 << 14)
 #define CBE_CAUSE_ADDRESS_SPACE_DECODE_ERROR	(1 << 15)
-#define CBE_CAUSE_RESPONSE_DATA_ERROR		(1 << 16)
-#define CBE_CAUSE_PROTOCOL_STATE_DATA_ERROR	(1 << 17)
+#define CBE_CAUSE_PROTOCOL_STATE_DATA_ERROR	(1 << 16)
+#define CBE_CAUSE_RA_RESPONSE_DATA_ERROR	(1 << 17)
+#define CBE_CAUSE_HA_RESPONSE_DATA_ERROR	(1 << 18)
 
 /* CBE cbrexecstatus bits */
 #define CBR_EXS_ABORT_OCC_BIT			0
 #define CBR_EXS_INT_OCC_BIT			1
 #define CBR_EXS_PENDING_BIT			2
 #define CBR_EXS_QUEUED_BIT			3
-#define CBR_EXS_TLBHW_BIT			4
+#define CBR_EXS_TLB_INVAL_BIT			4
 #define CBR_EXS_EXCEPTION_BIT			5
 
 #define CBR_EXS_ABORT_OCC			(1 << CBR_EXS_ABORT_OCC_BIT)
 #define CBR_EXS_INT_OCC				(1 << CBR_EXS_INT_OCC_BIT)
 #define CBR_EXS_PENDING				(1 << CBR_EXS_PENDING_BIT)
 #define CBR_EXS_QUEUED				(1 << CBR_EXS_QUEUED_BIT)
-#define CBR_EXS_TLBHW				(1 << CBR_EXS_TLBHW_BIT)
+#define CBR_TLB_INVAL				(1 << CBR_EXS_TLB_INVAL_BIT)
 #define CBR_EXS_EXCEPTION			(1 << CBR_EXS_EXCEPTION_BIT)
 
 /*
  * Exceptions are retried for the following cases. If any OTHER bits are set
  * in ecause, the exception is not retryable.
  */
-#define EXCEPTION_RETRY_BITS (CBE_CAUSE_RESPONSE_DATA_ERROR |		\
-			      CBE_CAUSE_RA_REQUEST_TIMEOUT |		\
+#define EXCEPTION_RETRY_BITS (CBE_CAUSE_EXECUTION_HW_ERROR |		\
 			      CBE_CAUSE_TLBHW_ERROR |			\
-			      CBE_CAUSE_HA_REQUEST_TIMEOUT)
+			      CBE_CAUSE_RA_REQUEST_TIMEOUT |		\
+			      CBE_CAUSE_RA_RESPONSE_NON_FATAL |		\
+			      CBE_CAUSE_HA_RESPONSE_NON_FATAL |		\
+			      CBE_CAUSE_RA_RESPONSE_DATA_ERROR |	\
+			      CBE_CAUSE_HA_RESPONSE_DATA_ERROR		\
+			      )
 
 /* Message queue head structure */
 union gru_mesqhead {
diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 8443e90f9f6..a489807613f 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -339,8 +339,12 @@ static int gru_try_dropin(struct gru_thread_state *gts,
 	 * Might be a hardware race OR a stupid user. Ignore FMM because FMM
 	 * is a transient state.
 	 */
-	if (tfh->status != TFHSTATUS_EXCEPTION)
-		goto failnoexception;
+	if (tfh->status != TFHSTATUS_EXCEPTION) {
+		gru_flush_cache(tfh);
+		if (tfh->status != TFHSTATUS_EXCEPTION)
+			goto failnoexception;
+		STAT(tfh_stale_on_fault);
+	}
 	if (tfh->state == TFHSTATE_IDLE)
 		goto failidle;
 	if (tfh->state == TFHSTATE_MISS_FMM && cb)
diff --git a/drivers/misc/sgi-gru/gruhandles.h b/drivers/misc/sgi-gru/gruhandles.h
index 9f41e2cc09d..99ec82678f5 100644
--- a/drivers/misc/sgi-gru/gruhandles.h
+++ b/drivers/misc/sgi-gru/gruhandles.h
@@ -255,7 +255,8 @@ struct gru_tlb_fault_handle {
 	unsigned int state:3;
 	unsigned int fill3:1;
 
-	unsigned int cause:7;
+	unsigned int cause:6;
+	unsigned int cb_int:1;
 	unsigned int fill4:1;
 
 	unsigned int indexway:12;	/* DW 0 - high 32 */
diff --git a/drivers/misc/sgi-gru/grukservices.c b/drivers/misc/sgi-gru/grukservices.c
index 7d7952b27e0..ba6fcd963f3 100644
--- a/drivers/misc/sgi-gru/grukservices.c
+++ b/drivers/misc/sgi-gru/grukservices.c
@@ -406,7 +406,8 @@ static int gru_retry_exception(void *cb)
 			return CBS_IDLE;
 
 		gru_get_cb_exception_detail(cb, &excdet);
-		if (excdet.ecause & ~EXCEPTION_RETRY_BITS)
+		if ((excdet.ecause & ~EXCEPTION_RETRY_BITS) ||
+				(excdet.cbrexecstatus & CBR_EXS_ABORT_OCC))
 			break;
 		if (retry-- == 0)
 			break;
diff --git a/drivers/misc/sgi-gru/gruprocfs.c b/drivers/misc/sgi-gru/gruprocfs.c
index 6ef4cb4b84c..b5b9cf5c182 100644
--- a/drivers/misc/sgi-gru/gruprocfs.c
+++ b/drivers/misc/sgi-gru/gruprocfs.c
@@ -89,6 +89,7 @@ static int statistics_show(struct seq_file *s, void *p)
 	printstat(s, tlb_dropin_fail_fmm);
 	printstat(s, tlb_dropin_fail_no_exception);
 	printstat(s, tlb_dropin_fail_no_exception_war);
+	printstat(s, tfh_stale_on_fault);
 	printstat(s, mmu_invalidate_range);
 	printstat(s, mmu_invalidate_page);
 	printstat(s, mmu_clear_flush_young);
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index 6dfb3e69411..246c63883eb 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -212,6 +212,7 @@ struct gru_stats_s {
 	atomic_long_t tlb_dropin_fail_fmm;
 	atomic_long_t tlb_dropin_fail_no_exception;
 	atomic_long_t tlb_dropin_fail_no_exception_war;
+	atomic_long_t tfh_stale_on_fault;
 	atomic_long_t mmu_invalidate_range;
 	atomic_long_t mmu_invalidate_page;
 	atomic_long_t mmu_clear_flush_young;
-- 
cgit v1.2.3-70-g09d2


From 1a2c09e3b41e334b6651d53b39cfe8ceefbc45f8 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:28 -0700
Subject: gru: fix cache coherency issues with instruction retry

Fix two problems related to GRU instruction failures.  Cache coherency is
not maintained for CBEs except when loading or unloading contexts.  When
reading a CBE to extract error information, the CBE must first be flushed
from the cache.

The function that reads kerrnel CBEs was reading the wrong CBE.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c     |  3 ++-
 drivers/misc/sgi-gru/grufile.c      |  2 +-
 drivers/misc/sgi-gru/grukservices.c | 12 ++++++++++--
 drivers/misc/sgi-gru/grutables.h    |  1 +
 4 files changed, 14 insertions(+), 4 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index a489807613f..6d0681236db 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -614,7 +614,7 @@ int gru_get_exception_detail(unsigned long arg)
 	} else if (gts->ts_gru) {
 		cbrnum = thread_cbr_number(gts, ucbnum);
 		cbe = get_cbe_by_index(gts->ts_gru, cbrnum);
-		prefetchw(cbe);/* Harmless on hardware, required for emulator */
+		gru_flush_cache(cbe);	/* CBE not coherent */
 		excdet.opc = cbe->opccpy;
 		excdet.exopc = cbe->exopccpy;
 		excdet.ecause = cbe->ecause;
@@ -622,6 +622,7 @@ int gru_get_exception_detail(unsigned long arg)
 		excdet.exceptdet1 = cbe->idef3upd;
 		excdet.cbrstate = cbe->cbrstate;
 		excdet.cbrexecstatus = cbe->cbrexecstatus;
+		gru_flush_cache(cbe);
 		ret = 0;
 	} else {
 		ret = -EAGAIN;
diff --git a/drivers/misc/sgi-gru/grufile.c b/drivers/misc/sgi-gru/grufile.c
index 796ac704795..bfc88d1b2a5 100644
--- a/drivers/misc/sgi-gru/grufile.c
+++ b/drivers/misc/sgi-gru/grufile.c
@@ -46,6 +46,7 @@
 
 struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
 unsigned long gru_start_paddr __read_mostly;
+void *gru_start_vaddr __read_mostly;
 unsigned long gru_end_paddr __read_mostly;
 unsigned int gru_max_gids __read_mostly;
 struct gru_stats_s gru_stats;
@@ -376,7 +377,6 @@ static int __init gru_init(void)
 {
 	int ret, irq, chip;
 	char id[10];
-	void *gru_start_vaddr;
 
 	if (!is_uv_system())
 		return 0;
diff --git a/drivers/misc/sgi-gru/grukservices.c b/drivers/misc/sgi-gru/grukservices.c
index ba6fcd963f3..7586b89fd0d 100644
--- a/drivers/misc/sgi-gru/grukservices.c
+++ b/drivers/misc/sgi-gru/grukservices.c
@@ -98,6 +98,9 @@
 #define ASYNC_HAN_TO_BID(h)	((h) - 1)
 #define ASYNC_BID_TO_HAN(b)	((b) + 1)
 #define ASYNC_HAN_TO_BS(h)	gru_base[ASYNC_HAN_TO_BID(h)]
+#define KCB_TO_GID(cb)		((cb - gru_start_vaddr) /		\
+					(GRU_SIZE * GRU_CHIPLETS_PER_BLADE))
+#define KCB_TO_BS(cb)		gru_base[KCB_TO_GID(cb)]
 
 #define GRU_NUM_KERNEL_CBR	1
 #define GRU_NUM_KERNEL_DSR_BYTES 256
@@ -354,14 +357,19 @@ int gru_get_cb_exception_detail(void *cb,
 		struct control_block_extended_exc_detail *excdet)
 {
 	struct gru_control_block_extended *cbe;
+	struct gru_blade_state *bs;
+	int cbrnum;
 
-	cbe = get_cbe(GRUBASE(cb), get_cb_number(cb));
-	prefetchw(cbe);	/* Harmless on hardware, required for emulator */
+	bs = KCB_TO_BS(cb);
+	cbrnum = thread_cbr_number(bs->bs_kgts, get_cb_number(cb));
+	cbe = get_cbe(GRUBASE(cb), cbrnum);
+	gru_flush_cache(cbe);	/* CBE not coherent */
 	excdet->opc = cbe->opccpy;
 	excdet->exopc = cbe->exopccpy;
 	excdet->ecause = cbe->ecause;
 	excdet->exceptdet0 = cbe->idef1upd;
 	excdet->exceptdet1 = cbe->idef3upd;
+	gru_flush_cache(cbe);
 	return 0;
 }
 
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index 246c63883eb..665704683ab 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -153,6 +153,7 @@
 extern struct gru_stats_s gru_stats;
 extern struct gru_blade_state *gru_base[];
 extern unsigned long gru_start_paddr, gru_end_paddr;
+extern void *gru_start_vaddr;
 extern unsigned int gru_max_gids;
 
 #define GRU_MAX_BLADES		MAX_NUMNODES
-- 
cgit v1.2.3-70-g09d2


From 7e796a72a2691d7094fd62da61097294d0d59ce4 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:30 -0700
Subject: gru: collect per-context user statistics

Collect GRU statistics for each user GRU context.  Statistics are kept for
TLB misses & content resource contention.  Add user request for retrieving
the statistics.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c  | 27 +++++++++++++++++++++++++++
 drivers/misc/sgi-gru/grufile.c   |  3 +++
 drivers/misc/sgi-gru/grulib.h    | 18 ++++++++++++++++++
 drivers/misc/sgi-gru/grumain.c   |  1 +
 drivers/misc/sgi-gru/grutables.h |  3 +++
 5 files changed, 52 insertions(+)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 6d0681236db..cdd151b30dc 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -498,6 +498,7 @@ irqreturn_t gru_intr(int irq, void *dev_id)
 		 */
 		if (!gts->ts_force_cch_reload &&
 					down_read_trylock(&gts->ts_mm->mmap_sem)) {
+			gts->ustats.fmm_tlbdropin++;
 			gru_try_dropin(gts, tfh, NULL);
 			up_read(&gts->ts_mm->mmap_sem);
 		} else {
@@ -516,6 +517,7 @@ static int gru_user_dropin(struct gru_thread_state *gts,
 	struct gru_mm_struct *gms = gts->ts_gms;
 	int ret;
 
+	gts->ustats.upm_tlbdropin++;
 	while (1) {
 		wait_event(gms->ms_wait_queue,
 			   atomic_read(&gms->ms_range_active) == 0);
@@ -718,6 +720,31 @@ int gru_user_flush_tlb(unsigned long arg)
 	return 0;
 }
 
+/*
+ * Fetch GSEG statisticss
+ */
+long gru_get_gseg_statistics(unsigned long arg)
+{
+	struct gru_thread_state *gts;
+	struct gru_get_gseg_statistics_req req;
+
+	if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
+		return -EFAULT;
+
+	gts = gru_find_lock_gts(req.gseg);
+	if (gts) {
+		memcpy(&req.stats, &gts->ustats, sizeof(gts->ustats));
+		gru_unlock_gts(gts);
+	} else {
+		memset(&req.stats, 0, sizeof(gts->ustats));
+	}
+
+	if (copy_to_user((void __user *)arg, &req, sizeof(req)))
+		return -EFAULT;
+
+	return 0;
+}
+
 /*
  * Register the current task as the user of the GSEG slice.
  * Needed for TLB fault interrupt targeting.
diff --git a/drivers/misc/sgi-gru/grufile.c b/drivers/misc/sgi-gru/grufile.c
index e22012db239..0d1c8b8c1c1 100644
--- a/drivers/misc/sgi-gru/grufile.c
+++ b/drivers/misc/sgi-gru/grufile.c
@@ -251,6 +251,9 @@ static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
 	case GRU_USER_CALL_OS:
 		err = gru_handle_user_call_os(arg);
 		break;
+	case GRU_GET_GSEG_STATISTICS:
+		err = gru_get_gseg_statistics(arg);
+		break;
 	case GRU_KTEST:
 		err = gru_ktest(arg);
 		break;
diff --git a/drivers/misc/sgi-gru/grulib.h b/drivers/misc/sgi-gru/grulib.h
index 87586551d16..c5865dd19ee 100644
--- a/drivers/misc/sgi-gru/grulib.h
+++ b/drivers/misc/sgi-gru/grulib.h
@@ -50,6 +50,9 @@
 /* For dumpping GRU chiplet state */
 #define GRU_DUMP_CHIPLET_STATE		_IOWR(GRU_IOCTL_NUM, 11, void *)
 
+/* For getting gseg statistics */
+#define GRU_GET_GSEG_STATISTICS		_IOWR(GRU_IOCTL_NUM, 12, void *)
+
 /* For user TLB flushing (primarily for tests) */
 #define GRU_USER_FLUSH_TLB		_IOWR(GRU_IOCTL_NUM, 50, void *)
 
@@ -61,6 +64,21 @@
 
 #define CONTEXT_WINDOW_BYTES(th)        (GRU_GSEG_PAGESIZE * (th))
 #define THREAD_POINTER(p, th)		(p + GRU_GSEG_PAGESIZE * (th))
+#define GSEG_START(cb)			((void *)((unsigned long)(cb) & ~(GRU_GSEG_PAGESIZE - 1)))
+
+/*
+ * Statictics kept on a per-GTS basis.
+ */
+struct gts_statistics {
+	unsigned long	fmm_tlbdropin;
+	unsigned long	upm_tlbdropin;
+	unsigned long	context_stolen;
+};
+
+struct gru_get_gseg_statistics_req {
+	unsigned long		gseg;
+	struct gts_statistics	stats;
+};
 
 /*
  * Structure used to pass TLB flush parameters to the driver
diff --git a/drivers/misc/sgi-gru/grumain.c b/drivers/misc/sgi-gru/grumain.c
index e38a0f1775f..347004e4f59 100644
--- a/drivers/misc/sgi-gru/grumain.c
+++ b/drivers/misc/sgi-gru/grumain.c
@@ -744,6 +744,7 @@ void gru_steal_context(struct gru_thread_state *gts, int blade_id)
 	spin_unlock(&blade->bs_lock);
 
 	if (ngts) {
+		gts->ustats.context_stolen++;
 		ngts->ts_steal_jiffies = jiffies;
 		gru_unload_context(ngts, is_kernel_context(ngts) ? 0 : 1);
 		gts_stolen(ngts, blade);
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index 9761bfee866..63b76e2732f 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -148,6 +148,7 @@
 #include <linux/wait.h>
 #include <linux/mmu_notifier.h>
 #include "gru.h"
+#include "grulib.h"
 #include "gruhandles.h"
 
 extern struct gru_stats_s gru_stats;
@@ -388,6 +389,7 @@ struct gru_thread_state {
 							  allocated CB */
 	int			ts_data_valid;	/* Indicates if ts_gdata has
 						   valid data */
+	struct gts_statistics	ustats;		/* User statistics */
 	unsigned long		ts_gdata[0];	/* save area for GRU data (CB,
 						   DS, CBE) */
 };
@@ -641,6 +643,7 @@ extern void gru_tgh_flush_init(struct gru_state *gru);
 extern int gru_kservices_init(void);
 extern void gru_kservices_exit(void);
 extern int gru_dump_chiplet_request(unsigned long arg);
+extern long gru_get_gseg_statistics(unsigned long arg);
 extern irqreturn_t gru_intr(int irq, void *dev_id);
 extern int gru_handle_user_call_os(unsigned long address);
 extern int gru_user_flush_tlb(unsigned long arg);
-- 
cgit v1.2.3-70-g09d2


From 92b39388eeb45326feb0fa8bd69dbbce66c9efbf Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:32 -0700
Subject: gru: generic infrastructure for context options

Change the user GRU request for specifying the "task_slice" option to use
a generic infrastructure that can be expanded in the future to include
additional context options.  No new capabilities are added with this
patch.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c  | 24 ++++++++++++++++++------
 drivers/misc/sgi-gru/grufile.c   |  4 ++--
 drivers/misc/sgi-gru/grulib.h    | 14 ++++++++++++--
 drivers/misc/sgi-gru/gruprocfs.c |  2 +-
 drivers/misc/sgi-gru/grutables.h |  4 ++--
 5 files changed, 35 insertions(+), 13 deletions(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index cdd151b30dc..b894b7ed9c3 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -749,18 +749,30 @@ long gru_get_gseg_statistics(unsigned long arg)
  * Register the current task as the user of the GSEG slice.
  * Needed for TLB fault interrupt targeting.
  */
-int gru_set_task_slice(long address)
+int gru_set_context_option(unsigned long arg)
 {
 	struct gru_thread_state *gts;
+	struct gru_set_context_option_req req;
+	int ret = 0;
 
-	STAT(set_task_slice);
-	gru_dbg(grudev, "address 0x%lx\n", address);
-	gts = gru_alloc_locked_gts(address);
+	STAT(set_context_option);
+	if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
+		return -EFAULT;
+	gru_dbg(grudev, "op %d, gseg 0x%lx, value1 0x%lx\n", req.op, req.gseg, req.val1);
+
+	gts = gru_alloc_locked_gts(req.gseg);
 	if (!gts)
 		return -EINVAL;
 
-	gts->ts_tgid_owner = current->tgid;
+	switch (req.op) {
+	case sco_gseg_owner:
+ 		/* Register the current task as the GSEG owner */
+		gts->ts_tgid_owner = current->tgid;
+		break;
+	default:
+		ret = -EINVAL;
+	}
 	gru_unlock_gts(gts);
 
-	return 0;
+	return ret;
 }
diff --git a/drivers/misc/sgi-gru/grufile.c b/drivers/misc/sgi-gru/grufile.c
index 2b72629db91..fa2d93a9fb8 100644
--- a/drivers/misc/sgi-gru/grufile.c
+++ b/drivers/misc/sgi-gru/grufile.c
@@ -198,8 +198,8 @@ static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
 	case GRU_CREATE_CONTEXT:
 		err = gru_create_new_context(arg);
 		break;
-	case GRU_SET_TASK_SLICE:
-		err = gru_set_task_slice(arg);
+	case GRU_SET_CONTEXT_OPTION:
+		err = gru_set_context_option(arg);
 		break;
 	case GRU_USER_GET_EXCEPTION_DETAIL:
 		err = gru_get_exception_detail(arg);
diff --git a/drivers/misc/sgi-gru/grulib.h b/drivers/misc/sgi-gru/grulib.h
index 8ed6acbc47c..8615b904a7c 100644
--- a/drivers/misc/sgi-gru/grulib.h
+++ b/drivers/misc/sgi-gru/grulib.h
@@ -32,8 +32,8 @@
 /* Set Number of Request Blocks */
 #define GRU_CREATE_CONTEXT		_IOWR(GRU_IOCTL_NUM, 1, void *)
 
-/* Register task as using the slice */
-#define GRU_SET_TASK_SLICE		_IOWR(GRU_IOCTL_NUM, 5, void *)
+/*  Set Context Options */
+#define GRU_SET_CONTEXT_OPTION		_IOWR(GRU_IOCTL_NUM, 4, void *)
 
 /* Fetch exception detail */
 #define GRU_USER_GET_EXCEPTION_DETAIL	_IOWR(GRU_IOCTL_NUM, 6, void *)
@@ -95,6 +95,16 @@ struct gru_unload_context_req {
 	unsigned long	gseg;
 };
 
+/*
+ * Structure used to set context options
+ */
+enum {sco_gseg_owner};
+struct gru_set_context_option_req {
+	unsigned long	gseg;
+	int		op;
+	unsigned long	val1;
+};
+
 /*
  * Structure used to pass TLB flush parameters to the driver
  */
diff --git a/drivers/misc/sgi-gru/gruprocfs.c b/drivers/misc/sgi-gru/gruprocfs.c
index b5b9cf5c182..9cbf95bedce 100644
--- a/drivers/misc/sgi-gru/gruprocfs.c
+++ b/drivers/misc/sgi-gru/gruprocfs.c
@@ -73,7 +73,7 @@ static int statistics_show(struct seq_file *s, void *p)
 	printstat(s, user_flush_tlb);
 	printstat(s, user_unload_context);
 	printstat(s, user_exception);
-	printstat(s, set_task_slice);
+	printstat(s, set_context_option);
 	printstat(s, migrate_check);
 	printstat(s, migrated_retarget);
 	printstat(s, migrated_unload);
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index 63b76e2732f..ee2f4121db2 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -198,7 +198,7 @@ struct gru_stats_s {
 	atomic_long_t user_flush_tlb;
 	atomic_long_t user_unload_context;
 	atomic_long_t user_exception;
-	atomic_long_t set_task_slice;
+	atomic_long_t set_context_option;
 	atomic_long_t migrate_check;
 	atomic_long_t migrated_retarget;
 	atomic_long_t migrated_unload;
@@ -649,7 +649,7 @@ extern int gru_handle_user_call_os(unsigned long address);
 extern int gru_user_flush_tlb(unsigned long arg);
 extern int gru_user_unload_context(unsigned long arg);
 extern int gru_get_exception_detail(unsigned long arg);
-extern int gru_set_task_slice(long address);
+extern int gru_set_context_option(unsigned long address);
 extern int gru_cpu_fault_map_id(void);
 extern struct vm_area_struct *gru_find_vma(unsigned long vaddr);
 extern void gru_flush_all_tlb(struct gru_state *gru);
-- 
cgit v1.2.3-70-g09d2


From b1b19fcfa417cf62447413d6e8b9b6598adf00b9 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:33 -0700
Subject: gru: add user request to specify gru slice

Add a user request to specify the gru instruction slice parameter for user
contexts.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c  | 4 ++++
 drivers/misc/sgi-gru/grulib.h    | 2 +-
 drivers/misc/sgi-gru/grumain.c   | 7 +++++++
 drivers/misc/sgi-gru/grutables.h | 1 +
 4 files changed, 13 insertions(+), 1 deletion(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index b894b7ed9c3..1ad360cd318 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -769,6 +769,10 @@ int gru_set_context_option(unsigned long arg)
  		/* Register the current task as the GSEG owner */
 		gts->ts_tgid_owner = current->tgid;
 		break;
+	case sco_cch_req_slice:
+ 		/* Set the CCH slice option */
+		gts->ts_cch_req_slice = req.val1 & 3;
+		break;
 	default:
 		ret = -EINVAL;
 	}
diff --git a/drivers/misc/sgi-gru/grulib.h b/drivers/misc/sgi-gru/grulib.h
index 8615b904a7c..a484a9fee3c 100644
--- a/drivers/misc/sgi-gru/grulib.h
+++ b/drivers/misc/sgi-gru/grulib.h
@@ -98,7 +98,7 @@ struct gru_unload_context_req {
 /*
  * Structure used to set context options
  */
-enum {sco_gseg_owner};
+enum {sco_gseg_owner, sco_cch_req_slice};
 struct gru_set_context_option_req {
 	unsigned long	gseg;
 	int		op;
diff --git a/drivers/misc/sgi-gru/grumain.c b/drivers/misc/sgi-gru/grumain.c
index 347004e4f59..0c20be007b2 100644
--- a/drivers/misc/sgi-gru/grumain.c
+++ b/drivers/misc/sgi-gru/grumain.c
@@ -321,6 +321,7 @@ struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
 	gts->ts_tsid = tsid;
 	gts->ts_ctxnum = NULLCTX;
 	gts->ts_tlb_int_select = -1;
+	gts->ts_cch_req_slice = -1;
 	gts->ts_sizeavail = GRU_SIZEAVAIL(PAGE_SHIFT);
 	if (vma) {
 		gts->ts_mm = current->mm;
@@ -566,6 +567,12 @@ void gru_load_context(struct gru_thread_state *gts)
 		gts->ts_tlb_int_select = gru_cpu_fault_map_id();
 		cch->tlb_int_select = gts->ts_tlb_int_select;
 	}
+	if (gts->ts_cch_req_slice >= 0) {
+		cch->req_slice_set_enable = 1;
+		cch->req_slice = gts->ts_cch_req_slice;
+	} else {
+		cch->req_slice_set_enable =0;
+	}
 	cch->tfm_done_bit_enable = 0;
 	cch->dsr_allocation_map = gts->ts_dsr_map;
 	cch->cbr_allocation_map = gts->ts_cbr_map;
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index ee2f4121db2..34ab3d45391 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -380,6 +380,7 @@ struct gru_thread_state {
 						   required for contest */
 	unsigned char		ts_cbr_au_count;/* Number of CBR resources
 						   required for contest */
+	char			ts_cch_req_slice;/* CCH packet slice */
 	char			ts_blade;	/* If >= 0, migrate context if
 						   ref from diferent blade */
 	char			ts_force_cch_reload;
-- 
cgit v1.2.3-70-g09d2


From 1926ee85a903d189c5702eed6531be321e33eb47 Mon Sep 17 00:00:00 2001
From: Jack Steiner <steiner@sgi.com>
Date: Wed, 17 Jun 2009 16:28:33 -0700
Subject: gru: fix potential use-after-free when purging GRU tlbs

Fix potential SGI GRU bug that could cause a use-after-free.  If one
thread in a task is flushing the GRU and another thread destroys the GRU
context, there is the potential to access a table after it has been freed.

Copy the gms pointer to a local variable before unlocking the gts table.
Note that no refcnt is needed for the gms - the reference is held
indirectly by the task's mm_struct.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---
 drivers/misc/sgi-gru/grufault.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

(limited to 'drivers/misc/sgi-gru/grufault.c')

diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 1ad360cd318..679e0177828 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -702,6 +702,7 @@ int gru_user_flush_tlb(unsigned long arg)
 {
 	struct gru_thread_state *gts;
 	struct gru_flush_tlb_req req;
+	struct gru_mm_struct *gms;
 
 	STAT(user_flush_tlb);
 	if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
@@ -714,8 +715,9 @@ int gru_user_flush_tlb(unsigned long arg)
 	if (!gts)
 		return -EINVAL;
 
-	gru_flush_tlb_range(gts->ts_gms, req.vaddr, req.len);
+	gms = gts->ts_gms;
 	gru_unlock_gts(gts);
+	gru_flush_tlb_range(gms, req.vaddr, req.len);
 
 	return 0;
 }
-- 
cgit v1.2.3-70-g09d2