#include <linux/mm.h>
#include <linux/vmalloc.h>
#include <linux/smp.h>
-#include <linux/smp_lock.h>
#include <linux/stddef.h>
#include <linux/unistd.h>
}
#endif /* debug */
-#define POLL_WHILE_FALSE(_c) POLL_WHILE_TRUE(!(_c))
+#define POLL_WHILE_FALSE(_c) POLL_WHILE_TRUE(!(_c))
static inline void acquire_spu_lock(struct spu *spu)
{
case MFC_CNTL_SUSPEND_COMPLETE:
if (csa) {
csa->priv2.mfc_control_RW =
- in_be64(&priv2->mfc_control_RW) |
+ MFC_CNTL_SUSPEND_MASK |
MFC_CNTL_SUSPEND_DMA_QUEUE;
}
break;
MFC_CNTL_SUSPEND_DMA_STATUS_MASK) ==
MFC_CNTL_SUSPEND_COMPLETE);
if (csa) {
- csa->priv2.mfc_control_RW =
- in_be64(&priv2->mfc_control_RW) &
- ~MFC_CNTL_SUSPEND_DMA_QUEUE;
+ csa->priv2.mfc_control_RW = 0;
}
break;
}
* Read MFC_CNTL[Ds]. Update saved copy of
* CSA.MFC_CNTL[Ds].
*/
- if (in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING) {
- csa->priv2.mfc_control_RW |= MFC_CNTL_DECREMENTER_RUNNING;
- csa->suspend_time = get_cycles();
- out_be64(&priv2->spu_chnlcntptr_RW, 7ULL);
- eieio();
- csa->spu_chnldata_RW[7] = in_be64(&priv2->spu_chnldata_RW);
- eieio();
- } else {
- csa->priv2.mfc_control_RW &= ~MFC_CNTL_DECREMENTER_RUNNING;
- }
+ csa->priv2.mfc_control_RW |=
+ in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING;
}
static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
* Write MFC_CNTL[Dh] set to a '1' to halt
* the decrementer.
*/
- out_be64(&priv2->mfc_control_RW, MFC_CNTL_DECREMENTER_HALTED);
+ out_be64(&priv2->mfc_control_RW,
+ MFC_CNTL_DECREMENTER_HALTED | MFC_CNTL_SUSPEND_MASK);
eieio();
}
csa->prob.dma_querytype_RW = in_be32(&prob->dma_querytype_RW);
}
+static inline void save_ppu_tagstatus(struct spu_state *csa, struct spu *spu)
+{
+ struct spu_problem __iomem *prob = spu->problem;
+
+ /* Save the Prxy_TagStatus register in the CSA.
+ *
+ * It is unnecessary to restore dma_tagstatus_R, however,
+ * dma_tagstatus_R in the CSA is accessed via backing_ops, so
+ * we must save it.
+ */
+ csa->prob.dma_tagstatus_R = in_be32(&prob->dma_tagstatus_R);
+}
+
static inline void save_mfc_csr_tsq(struct spu_state *csa, struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
static inline void save_ch_part1(struct spu_state *csa, struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
- u64 idx, ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+ u64 idx, ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
int i;
/* Save, Step 42:
csa->spu_chnldata_RW[1] = in_be64(&priv2->spu_chnldata_RW);
/* Save the following CH: [0,3,4,24,25,27] */
- for (i = 0; i < 7; i++) {
+ for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
idx = ch_indices[i];
out_be64(&priv2->spu_chnlcntptr_RW, idx);
eieio();
*/
}
-static inline void suspend_mfc(struct spu_state *csa, struct spu *spu)
+static inline void suspend_mfc_and_halt_decr(struct spu_state *csa,
+ struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
/* Restore, Step 7:
- * Restore, Step 47.
- * Write MFC_Cntl[Dh,Sc]='1','1' to suspend
+ * Write MFC_Cntl[Dh,Sc,Sm]='1','1','0' to suspend
* the queue and halt the decrementer.
*/
out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE |
static inline void reset_ch_part1(struct spu_state *csa, struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
- u64 ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+ u64 ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
u64 idx;
int i;
out_be64(&priv2->spu_chnldata_RW, 0UL);
/* Reset the following CH: [0,3,4,24,25,27] */
- for (i = 0; i < 7; i++) {
+ for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
idx = ch_indices[i];
out_be64(&priv2->spu_chnlcntptr_RW, idx);
eieio();
cycles_t resume_time = get_cycles();
cycles_t delta_time = resume_time - csa->suspend_time;
+ csa->lscsa->decr_status.slot[0] = SPU_DECR_STATUS_RUNNING;
+ if (csa->lscsa->decr.slot[0] < delta_time) {
+ csa->lscsa->decr_status.slot[0] |=
+ SPU_DECR_STATUS_WRAPPED;
+ }
+
csa->lscsa->decr.slot[0] -= delta_time;
+ } else {
+ csa->lscsa->decr_status.slot[0] = 0;
}
}
send_mfc_dma(spu, addr, ls_offset, size, tag, rclass, cmd);
}
+static inline void suspend_mfc(struct spu_state *csa, struct spu *spu)
+{
+ struct spu_priv2 __iomem *priv2 = spu->priv2;
+
+ /* Restore, Step 47.
+ * Write MFC_Cntl[Sc,Sm]='1','0' to suspend
+ * the queue.
+ */
+ out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE);
+ eieio();
+}
+
static inline void clear_interrupts(struct spu_state *csa, struct spu *spu)
{
/* Restore, Step 49:
* "wrapped" flag is set, OR in a '1' to
* CSA.SPU_Event_Status[Tm].
*/
- if (csa->lscsa->decr_status.slot[0] == 1) {
+ if (csa->lscsa->decr_status.slot[0] & SPU_DECR_STATUS_WRAPPED) {
csa->spu_chnldata_RW[0] |= 0x20;
}
- if ((csa->lscsa->decr_status.slot[0] == 1) &&
+ if ((csa->lscsa->decr_status.slot[0] & SPU_DECR_STATUS_WRAPPED) &&
(csa->spu_chnlcnt_RW[0] == 0 &&
((csa->spu_chnldata_RW[2] & 0x20) == 0x0) &&
((csa->spu_chnldata_RW[0] & 0x20) != 0x1))) {
static inline void restore_ch_part1(struct spu_state *csa, struct spu *spu)
{
struct spu_priv2 __iomem *priv2 = spu->priv2;
- u64 idx, ch_indices[7] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
+ u64 idx, ch_indices[] = { 0UL, 3UL, 4UL, 24UL, 25UL, 27UL };
int i;
/* Restore, Step 59:
+ * Restore the following CH: [0,3,4,24,25,27]
*/
-
- /* Restore CH 1 without count */
- out_be64(&priv2->spu_chnlcntptr_RW, 1);
- out_be64(&priv2->spu_chnldata_RW, csa->spu_chnldata_RW[1]);
-
- /* Restore the following CH: [0,3,4,24,25,27] */
- for (i = 0; i < 7; i++) {
+ for (i = 0; i < ARRAY_SIZE(ch_indices); i++) {
idx = ch_indices[i];
out_be64(&priv2->spu_chnlcntptr_RW, idx);
eieio();
save_mfc_queues(prev, spu); /* Step 19. */
save_ppu_querymask(prev, spu); /* Step 20. */
save_ppu_querytype(prev, spu); /* Step 21. */
+ save_ppu_tagstatus(prev, spu); /* NEW. */
save_mfc_csr_tsq(prev, spu); /* Step 22. */
save_mfc_csr_cmd(prev, spu); /* Step 23. */
save_mfc_csr_ato(prev, spu); /* Step 24. */
set_switch_pending(prev, spu); /* Step 5. */
stop_spu_isolate(spu); /* NEW. */
remove_other_spu_access(prev, spu); /* Step 6. */
- suspend_mfc(prev, spu); /* Step 7. */
+ suspend_mfc_and_halt_decr(prev, spu); /* Step 7. */
wait_suspend_mfc_complete(prev, spu); /* Step 8. */
if (!suspend_spe(prev, spu)) /* Step 9. */
clear_spu_status(prev, spu); /* Step 10. */
reset_spu_privcntl(prev, spu); /* Step 16. */
reset_spu_lslr(prev, spu); /* Step 17. */
setup_mfc_sr1(prev, spu); /* Step 18. */
- spu_invalidate_slbs(spu); /* Step 19. */
+ spu_invalidate_slbs(spu); /* Step 19. */
reset_ch_part1(prev, spu); /* Step 20. */
reset_ch_part2(prev, spu); /* Step 21. */
enable_interrupts(prev, spu); /* Step 22. */
}
EXPORT_SYMBOL_GPL(spu_restore);
-/**
- * spu_harvest - SPU harvest (reset) operation
- * @spu: pointer to SPU iomem structure.
- *
- * Perform SPU harvest (reset) operation.
- */
-void spu_harvest(struct spu *spu)
-{
- acquire_spu_lock(spu);
- harvest(NULL, spu);
- release_spu_lock(spu);
-}
-
static void init_prob(struct spu_state *csa)
{
csa->spu_chnlcnt_RW[9] = 1;
* as it is by far the largest of the context save regions,
* and may need to be pinned or otherwise specially aligned.
*/
-void spu_init_csa(struct spu_state *csa)
+int spu_init_csa(struct spu_state *csa)
{
- struct spu_lscsa *lscsa;
- unsigned char *p;
+ int rc;
if (!csa)
- return;
+ return -EINVAL;
memset(csa, 0, sizeof(struct spu_state));
- lscsa = vmalloc(sizeof(struct spu_lscsa));
- if (!lscsa)
- return;
+ rc = spu_alloc_lscsa(csa);
+ if (rc)
+ return rc;
- memset(lscsa, 0, sizeof(struct spu_lscsa));
- csa->lscsa = lscsa;
spin_lock_init(&csa->register_lock);
- /* Set LS pages reserved to allow for user-space mapping. */
- for (p = lscsa->ls; p < lscsa->ls + LS_SIZE; p += PAGE_SIZE)
- SetPageReserved(vmalloc_to_page(p));
-
init_prob(csa);
init_priv1(csa);
init_priv2(csa);
+
+ return 0;
}
EXPORT_SYMBOL_GPL(spu_init_csa);
void spu_fini_csa(struct spu_state *csa)
{
- /* Clear reserved bit before vfree. */
- unsigned char *p;
- for (p = csa->lscsa->ls; p < csa->lscsa->ls + LS_SIZE; p += PAGE_SIZE)
- ClearPageReserved(vmalloc_to_page(p));
-
- vfree(csa->lscsa);
+ spu_free_lscsa(csa);
}
EXPORT_SYMBOL_GPL(spu_fini_csa);