All STM32 host drivers. In IN endpoint retry, delay for a clock tick to give some breathing space for the CPU. EXPERIMENTAL change.

arch/arm/src/stm32/stm32_otgfshost.c

@@ -1906,6 +1906,14 @@ static ssize_t stm32_in_transfer(FAR struct stm32_usbhost_s *priv, int chidx,
 
                       return (ssize_t)ret;
                     }
+
+                  /* Wait a bit before retrying after a NAK.
+                   *
+                   * REVISIT:  This is intended to give the CPU a break from
+                   * the tight polling loop.  But are there performance issues?
+                   */
+
+                  usleep(1000);
                 }
             }
           else

arch/arm/src/stm32/stm32_otghshost.c

@@ -1911,6 +1911,14 @@ static ssize_t stm32_in_transfer(FAR struct stm32_usbhost_s *priv, int chidx,
 
                       return (ssize_t)ret;
                     }
+
+                  /* Wait a bit before retrying after a NAK.
+                   *
+                   * REVISIT:  This is intended to give the CPU a break from
+                   * the tight polling loop.  But are there performance issues?
+                   */
+
+                  usleep(1000);
                 }
             }
           else

arch/arm/src/stm32f7/stm32_otghost.c

@@ -1905,6 +1905,14 @@ static ssize_t stm32_in_transfer(FAR struct stm32_usbhost_s *priv, int chidx,
 
                       return (ssize_t)ret;
                     }
+
+                  /* Wait a bit before retrying after a NAK.
+                   *
+                   * REVISIT:  This is intended to give the CPU a break from
+                   * the tight polling loop.  But are there performance issues?
+                   */
+
+                  usleep(1000);
                 }
             }
           else

arch/arm/src/stm32l4/stm32l4_otgfshost.c

@@ -1910,6 +1910,14 @@ static ssize_t stm32l4_in_transfer(FAR struct stm32l4_usbhost_s *priv,
 
                       return (ssize_t)ret;
                     }
+
+                  /* Wait a bit before retrying after a NAK.
+                   *
+                   * REVISIT:  This is intended to give the CPU a break from
+                   * the tight polling loop.  But are there performance issues?
+                   */
+
+                  usleep(1000);
                 }
             }
           else